Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The impact of epilepsy surgery on the structural connectome and its relation to outcome

The impact of epilepsy surgery on the structural connectome and its relation to outcome NeuroImage: Clinical 18 (2018) 202–214 Contents lists available at ScienceDirect NeuroImage: Clinical journal homepage: www.elsevier.com/locate/ynicl The impact of epilepsy surgery on the structural connectome and its relation to outcome a,b,c, a,b a,b,c d,e c Peter N. Taylor , Nishant Sinha , Yujiang Wang , Sjoerd B. Vos , Jane de Tisi , c c c,e,1 c,e,1 Anna Miserocchi , Andrew W. McEvoy , Gavin P. Winston , John S. Duncan Interdisciplinary Computing and Complex BioSystems Group, School of Computing Science, Newcastle University, UK Institute of Neuroscience, Faculty of Medical Science, Newcastle University, UK NIHR University College London Hospitals Biomedical Research Centre, UCL Institute of Neurology, Queen Square, London WC1N 3BG, UK Translational Imaging Group, Centre for Medical Image Computing, University College London, UK Chalfont Centre for Epilepsy, Chalfont St Peter SL9 0LR, UK ARTICLE I NFO ABSTRACT Keywords: Background: Temporal lobe surgical resection brings seizure remission in up to 80% of patients, with long-term Connectome complete seizure freedom in 41%. However, it is unclear how surgery impacts on the structural white matter Network network, and how the network changes relate to seizure outcome. Temporal lobe epilepsy Methods: We used white matter fibre tractography on preoperative diffusion MRI to generate a structural white Surgery matter network, and postoperative T1-weighted MRI to retrospectively infer the impact of surgical resection on Machine learning this network. We then applied graph theory and machine learning to investigate the properties of change be- Support vector machine (SVM) tween the preoperative and predicted postoperative networks. Results: Temporal lobe surgery had a modest impact on global network efficiency, despite the disruption caused. This was due to alternative shortest paths in the network leading to widespread increases in betweenness centrality post-surgery. Measurements of network change could retrospectively predict seizure outcomes with 79% accuracy and 65% specificity, which is twice as high as the empirical distribution. Fifteen connections which changed due to surgery were identified as useful for prediction of outcome, eight of which connected to the ipsilateral temporal pole. Conclusion: Our results suggest that the use of network change metrics may have clinical value for predicting seizure outcome. This approach could be used to prospectively predict outcomes given a suggested resection mask using preoperative data only. 1. Introduction subjected to quantitative analysis techniques, which measure local and global properties in networks (see Bernhardt et al. (2015) for review). Epilepsy is a serious neurological disorder characterised by re- Network measures that have been found to be altered in temporal lobe current unprovoked seizures affecting 1% of the population. epilepsy (TLE) include the clustering coefficient of a region, which Neurosurgical resection can bring remission in up to 80% of those with captures the connectedness of neighbours of a region (Bernhardt et al., refractory focal epilepsy, with 41% remaining entirely seizure free for 2011). Furthermore, the strength of a connection (e.g. the number of years (De Tisi et al., 2011). The most common type of epilepsy surgery streamlines connecting two areas), or the strength of a region's con- is anterior temporal lobe resection, in which the amygdala, anterior nectivity (e.g. the number of streamlines connecting a region to all hippocampus, and anterior temporal neocortex are removed. The other regions) may also be altered in TLE (Besson et al., 2014a, 2014b; commonest neurological sequelae of temporal lobe surgery are memory Taylor et al., 2015). Another measure of a network is its efficiency, impairment, visual field deficits and word-finding difficulties (Jutila which is a measure of network integration - i.e. how easy it is to travel et al., 2002; Gooneratne et al., 2017). between one region to another via direct and indirect paths, and has Recent studies have investigated surgical outcome by considering been shown to be altered in patients with TLE (Liu et al., 2014). Finally, the brain as a network of connected regions. Such networks can then be regression analysis and machine learning approaches have also been Corresponding author at: Interdisciplinary Computing and Complex BioSystems Group, School of Computing Science, Newcastle University, UK. E-mail address: peter.taylor@newcastle.ac.uk (P.N. Taylor). Equal contribution. https://doi.org/10.1016/j.nicl.2018.01.028 Received 26 July 2017; Received in revised form 5 December 2017; Accepted 21 January 2018 2213-1582/ © 2018 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/). P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 applied to brain networks of TLE to relate them to surgical outcome Table 1 Patient demographics and relation to outcome group. (Bonilha et al., 2013; Munsell et al., 2015; Bonilha et al., 2015; Ji et al., 2015). ILAE 1 ILAE 2–6 Significance A challenge in comparing networks across subjects is the choice of an appropriate baseline or benchmark. There are two common ap- N 36 (68%) 17 (32%) proaches to this. One is to threshold the connectivity so that all subjects Male/female 16/20 4/13 p = 0.3597, have the same number of connections. A range of thresholds are then χ =0.839 checked, and the most significant results reported across thresholds Left/right TLE 22/14 8/9 p = 0.3353, (Zhang et al., 2011). This has the drawback of removing ‘weak’ but χ =0.923 Age (mean, S.D./median, 37, 11.6/ 41.5, 10.6/ p = 0.2374 potentially important connections. A second approach is to compare the I.Q.R.) 39.6, 19.25 42.3, 10.8 network to a random network with the same number of regions and Hippocampal sclerosis 25 (69%) 10 (59%) p = 0.4460, connections. Typically this is done by either rewiring the existing net- χ =0.5808 work (Maslov and Sneppen, 2002), or by generating a new network according to predefined rules (Betzel et al., 2016; Bauer and Kaiser, 2017). Many different types of baseline networks can be used and this Electric, Waukesha, Milwaukee, WI). Standard imaging gradients with a −1 −1 −1 will therefore influence results. maximum strength of 40mT m and slew rate 150 T m s were Recently Kuceyeski et al. (2013) introduced the network modifica- used. All data were acquired using a body coil for transmission, and 8- tion (NeMo) tool in the context of stroke (Kuceyeski et al., 2015a, channel phased array coil for reception. Standard clinical sequences 2015b) and multiple sclerosis (Kuceyeski et al., 2015a, 2015b). The were performed including a coronal T1-weighted volumetric acquisi- NeMo tool is a method to enable a direct comparison between networks tion with 170 contiguous 1.1 mm-thick slices (matrix, 256 × 256; in- that undergo change. For example, in their study of stroke, the authors plane resolution, 0.9375 × 0.9375 mm). drew masks over stroke affected areas and overlaid this mask with data Diffusion MRI data were acquired using a cardiac-triggered single- from healthy subjects. Normal connectivity from the healthy subjects, shot spin-echo planar imaging sequence (Wheeler-Kingshott et al., and altered connectivity (i.e. tracts which pass through the stroke 2002) with echo time = 73 ms. Sets of 60 contiguous 2.4 mm-thick mask) were calculated. This approach allowed the authors to calculate a axial slices were obtained covering the whole brain, with diffusion change in connectivity metric (ChaCo), which was shown to correlate sensitizing gradients applied in each of 52 noncollinear directions (b 2 −1 with outcomes. Since the authors use the pre-stroke network as a value of 1,200mm s [δ = 21 ms, Δ = 29 ms, using full gradient −1 baseline to investigate the implied post-stroke differences, the analysis strength of 40 mT m ]) along with 6 non-diffusion weighted scans. is possible without the need to generate random networks or threshold The gradient directions were calculated and ordered as described the connectivities. This obviates the need for arbitrarily chosen surro- elsewhere (Cook et al., 2007). The field of view was 24 cm, and the gate networks by effectively using the patient's own network as the acquisition matrix size was 96 × 96, zero filled to 128 × 128 during surrogate instead – a distinct advantage of the technique. A drawback of reconstruction, giving a reconstructed voxel size of that study is that the tractography was derived from a cohort of healthy 1.875 × 1.875 × 2.4 mm. The DTI acquisition time for a total of 3480 controls, rather than the stroke patients. Nonetheless, this framework is image slices was approximately 25 min (depending on subject heart ideally suited to investigate changes in networks, given a well-defined rate). alteration such as a stroke or surgery. In this study we used a ChaCo-like approach in the context of epi- 2.2. Image processing lepsy surgery and addressed the following questions: What is the impact of surgery on the patient's network? How does this impact graph the- 2.2.1. T1 processing oretic properties such as region strength, network efficiency? Do these Preoperative anatomical MRI was used to generate parcellated re- changes to patient networks correlate with surgical outcome? gions of interest (network nodes: ROIs). We used two different ap- Although the resection masks we use in this study are derived ret- proaches to do this, generating two different parcellation schemes. rospectively from postoperative data, our methods could in future be First, we used the FreeSurfer recon-all pipeline (https://surfer.nmr. applied preoperatively using a mask of the intended resection. mgh.harvard.edu/), which performs intensity normalization, skull stripping, subcortical volume generation, gray/white segmentation, 2. Materials and methods and parcellation (Fischl, 2012). The default parcellation scheme from FreeSurfer (the Desikan-Killiany atlas (Fischl et al., 2002; Desikan et al., 2.1. Patients & MRI acquisition 2006)) contains 82 cortical ROIs and subcortical ROIs and is widely used in the literature (e.g. Munsell et al., 2015; Taylor et al., 2015). The We retrospectively studied 53 patients who underwent temporal method FreeSurfer uses to generate its ROIs uses anatomical priors lobe epilepsy surgery at the National Hospital for Neurology and based on a manually annotated dataset from healthy controls. However, Neurosurgery, London, United Kingdom. Full patient details can be this may be suboptimal in the case of disease and therefore, we use a found in Table S11, a summary is given in Table 1. Patient outcomes second approach based on geodesic information flow (GIF) to generate were defined at 12 months postoperatively, according to the ILAE ROIs which has the advantage of performing well even in the presence classification of surgical outcomes (Wieser et al., 2001) and separated of neuropathology (Cardoso et al., 2015). Using the GIF approach, we into two groups. Group 1 includes patients who were completely sei- generate 114 cortical and subcortical ROIs (Table 2). A drawback of zure free (ILAE 1), and group 2 incorporates all other possibilities (ILAE using the GIF approach is comparison to previous studies is less 2–6). No patient had any prior history of neurosurgery. We used a χ straightforward since most previous work use alternative atlases. The test to check for differences between outcome groups in gender, side of results presented in the main manuscript use the GIF derived ROIs, surgery, and evidence of hippocampal sclerosis. We applied Kruskal- while we include results using FreeSurfer derived ROIs in supplemen- Wallis test to check for differences in age between outcome groups. tary materials to aid comparison to previous studies. All patients underwent preoperative anatomical T1-weighted MRI and preoperative diffusion MRI. Postoperative T1-weighted MRI was obtained within 12 months of surgery with the exception of one patient, 2.2.2. DWI processing who was rescanned later. Preoperative diffusion MRI data were first corrected for signal drift MRI studies were performed on a 3T GE Signa HDx scanner (General (Vos et al., 2016), then eddy current and movement artefacts were 203 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Table 2 MRI to the preoperative MRI using the FSL FLIRT tool (Jenkinson and Full names of the abbreviated regions of interest (ROIs). Smith, 2001; Jenkinson et al., 2002). Resection masks were then manually drawn using the FslView software by overlaying the post- Full name Abbreviation Full name Abbreviation operative MRI with the preoperative MRI starting at the most anterior Anterior cingulate ACgG Occipital pole OCP coronal slice, then proceeding posteriorly every three slices. Attention gyrus was given to ensure masks did not extend beyond the Sylvian fissure Anterior insula AIns Occipital fusiform gyrus OFuG into inferior frontal lobe since this is unrealistic for anterior temporal Anterior orbital AOrG Opercular part of the OpIFG resection and there was no evidence for this on post-operative MRI. gyrus inferior frontal gyrus Once complete, coronal slices were then joined by drawing in every Angular gyrus AnG Orbital part of the inferior OrIFG frontal gyrus sagittal slice. Masks were saved in the preoperative T1 space. Linear Calcarine cortex Calc Posterior cingulate gyrus PCgG registration is justified in this instance because nonlinear deformation Central operculum CO Precuneus PCu of the tissue around the resection site may lead to misleading bound- Cuneus Cun Parahippocampal gyrus PHG aries due to local deformations induced as part of the processing. All Entorhinal area Ent Posterior insula PIns Frontal operculum FO Parietal operculum PO registrations were visually inspected in detail during the drawing of the Frontal pole FRP Postcentral gyrus PoG resection mask. Fusiform gyrus FuG Posterior orbital gyrus POrG Gyrus rectus GRe Planum polare PP 2.3. Network generation Inferior occipital IOG Precentral gyrus PrG gyrus Inferior temporal ITG Planum temporale PT To align the tracts with the ROIs we linearly registered the pre- gyrus operative T1 image to the first non-diffusion-weighted image and saved Lingual gyrus LiG Subcallosal area SCA the transformation matrix using FSL FLIRT. We then multiplied every Lateral orbital gyrus LOrG Superior frontal gyrus SFG coordinate in every tract by the inverse of this transformation matrix to Middle cingulate MCgG Supplementary motor SMC gyrus cortex get the tracts in T1 space. The quality of the registration between tracts, Medial frontal cortex MFC Supramarginal gyrus SMG ROIs, and the resection mask was confirmed through visual inspection Middle frontal gyrus MFG Superior occipital gyrus SOG for all subjects. Since networks are constructed in native space, this Middle occipital MOG Superior parietal lobule SPL removes any mismatching of track types due to potential nonlinear gyrus registration issues which is advantageous compared to previous studies Medial orbital gyrus MOrG Superior temporal gyrus STG Postcentral gyrus MPoG Temporal pole TMP of network change (Kuceyeski et al., 2016). To generate preoperative medial segment connectivity matrices, we looped through all tracts and deemed two Precentral gyrus MPrG Triangular part of the TrIFG regions as connected if the two endpoints of the tract terminate in those medial segment inferior frontal gyrus regions. This generated a weighted connectivity (adjacency) matrix in Superior frontal MSFG Transverse temporal gyrus TTG which each entry in the matrix represents the number of streamlines gyrus medial segment connecting two regions. To generate predicted postoperative con- Middle temporal MTG nectivity matrices we performed the same process as above with one gyrus exception. Any tract that had any point within the resection mask was excluded from building the matrix. The inferred postoperative network therefore always had fewer streamlines than the preoperative network corrected using the FSL eddy_correct tool (Andersson and Sotiropoulos, and makes the assumption that remaining portions of removed tracts 2016). The b vectors were then rotated appropriately using the ‘fdt- subserve no functionality. This reduction in streamlines alone did not rotate-bvecs’ tool as part of FSL (Jenkinson et al., 2012; Leemans and explain outcome (Fig. S1). For analysis we applied a log transfor- Jones, 2009). The diffusion data were reconstructed using generalized mation to connection weights. Following this, right hemisphere regions q-sampling imaging (Yeh et al., 2010) with a diffusion sampling length (rows and columns in the matrix) were flipped in patients with a right ratio of 1.2. A deterministic fibre tracking algorithm (Yeh et al., 2013) sided resection as we investigated ipsi- and contralateral differences. All was then used, allowing for crossing fibres within voxels, with seeds subjects had 72 regions per hemisphere. placed at the whole brain. Probabilistic approaches to fibre tracking A summary of the processing pipeline to generate the GIF con- (e.g. Tournier et al., 2012) have been shown to perform less well with nectivity matrices is shown in Fig. 1. respect to accuracy and number of false positive connections inferred. The choice of tractography algorithm is therefore related to whether 2.4. Visualisation sensitivity or specificity is more important. It was recently shown (Zalesky et al., 2016) that the introduction of false positive connections To visualise the resection mask we first linearly registered the pre- is substantially more detrimental to the calculation of graph theoretic operative T1 to the MNI brain template using FSL FLIRT to generate a measures such as efficiency and clustering than the introduction of false transformation matrix. Next, we nonlinearly registered the preoperative negatives. We therefore use deterministic tractography since this has T1-weighted image to the MNI template using FSL FNIRT initialised been shown to have fewer false positive connections . Default tracto- with the aforementioned transformation to generate a nonlinear warp. graphy parameters from the 14 February 2017 build of DSI studio Finally, we applied this warp to both the preoperative T1-weighted software were used as follows. The angular threshold used was 60 de- image, and mask. All images were visually inspected to check for re- grees and the step size was set to 0.9375 mm. The anisotropy threshold gistration quality. We repeated this for all patients to generate a mask in was determined automatically by DSI Studio. Tracks with length < 10 MNI space. These masks were then loaded into MatLab, binarised mm and > 300 mm were discarded. A total of 1,000,000 tracts were (thresholded at 0.5), and summed across all subjects. This was then calculated per subject and saved in diffusion space. saved, and overlaid with the MNI brain using FSLView. Masks in MNI space were generated for visualisation purposes only and not used in 2.2.3. Resection mask the analysis. To draw resection masks we linearly registered the postoperative T1 Three-dimensional projections of brain regions were produced using the centre of mass for each region and visualised using BrainNet Viewer (Xia et al., 2013). We created the scatter plots using the UniVarScatter function http://www.tractometer.org/downloads/downloads/ismrm_presentation/Challenge_ ResultTractometer_updated_for_pdf_generation.pdf. in Matlab (https://github.com/GRousselet/matlab_visualisation). 204 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 1. Processing pipeline summary for GIF matrix generation. The pipeline is applied to each subject. 2.5. Network analysis ((−+yc xa )) i i min ∑we log(1++ ) xx+ λ i=1 1 (1) We investigated properties of the networks using well established graph theory measures that assessed the properties of a region, the where, y =(y ,y ,⋯,y ) is the n-dimensional vector representing the 1 2 n properties of a connection, and the properties of the network as a whole surgical outcomes (+1 for seizure-free outcome and −1 for nonseizure- (i.e. global). We excluded self-connections from all measures while n×m free outcome); a denotes the i− th row of feature vector A∈ ℝ ; w i i applying the weighted and undirected versions of the functions in the is the weight for the i− th sample (all samples were equally weighted Brain Connectivity Toolbox (Rubinov and Sporns, 2010). We in- in our implementation); c is the scalar intercept; λ∥ x∥ is the l reg- 1 1 vestigated measures of strength, regional communicability (sum of ularisation (sparsity) term; and x is the l regularisation (smooth- communicability matrix Estrada and Hatano, 2008), betweenness cen- ness) term. We optimised the cost function in Eq. 1 by applying the trality, clustering, and global efficiency, since several of these have LogisticR method in the sparse learning with efficient projections been suggested to be altered in previous studies (Bernhardt et al., 2011; (SLEP) software package (Liu et al., 2009a, 2009b; Liu and Ye, 2009; Besson et al., 2014a, 2014b; Taylor et al., 2015; Liu et al., 2014). Liu and Ye, 2009; Duchi et al., 2008). Region volume, region strength, connection strength changes were The weight vector x computed from Eq. 1 is a sparse representation computed by dividing the post-operative network property by the pre- of the training data set, i.e., the zero weights indicate the features operative network property. For example, if node strength reduces from which are not associated with surgical outcome, whereas the non-zero 10 pre-operatively to 2 in the inferred post-operative network the weights indicate the features associated with the surgical outcome. We change measurement will be 0.2. Other measures (betweenness, clus- derived a binary mask by setting all the non-zero weights in x to 1. tering) can increase or decrease after surgery and so the difference was Finally, we applied this mask to the high-dimensional feature space computed by subtracting the pre-operative value from the post-opera- while selecting only the features associated with surgical outcome, re- tive value. sulting in a low-dimensional feature space. In the second step of our classification framework, we incorporated the aforementioned reduced feature representation in a support vector 2.6. Elastic net regularisation and support vector machine classification machine (SVM) classifier with a linear kernel. We designed the SVM framework classifier in MATLAB using the ‘fitcsvm’ class. Specifically, we divided our data samples into test and training sets, incorporated a leave-one- We implemented a machine learning framework to investigate the out cross-validation scheme, and computed the average accuracy, sen- association of metrics indicating changes in the region strength, region sitivity, and specificity. To prevent the classifier from evaluating volumes, and connection strength with seizure free (ILAE 1) and non- skewed classification performance due to the between-class im- seizure free (ILAE 2–6) outcomes. We computed the aforementioned balance—36 in the seizure-free class as opposed to 17 in the non-sei- metrics and treated them as feature vectors. Specifically, the feature zure-free class—we implemented a class-weighted SVM by setting all space included 380, 85, and 14 features computed from proportional class prior probabilities as uniform similar to previous studies (e.g. reduction in the connection strength, region strength, and region vo- Wang et al., 2012). This sets a higher misclassification penalty for the lumes respectively for each subject. This resulted in a high-dimensional n×m minority class as compared to the majority class, thereby preventing the feature matrix, A ∈ ℝ , where m = 479 indicates the total number of classifier from learning simply from the underlying empirical data features and n = 53 denotes the total number of subjects in our study. distribution (Osuna et al., 1997; Veropoulos et al., 1999; Wu and Since the number of features, m, is much larger than the number of Rohini, 2004). subjects, n, we implemented a two-step classification paradigm Since our objective was to determine a minimum set of features that (Munsell et al., 2015; Casanova et al., 2011). In the first step, we per- would most accurately separate the seizure-free outcome and the formed feature selection by implementing logistic regression with nonseizure-free outcome classes, we incorporated a leave-one-out cross- elastic net regularisation (Zou and Hastie, 2005). This gives a sparse m- validated grid search on the l and l regularisation parameters (λ,ρ) dimensional weight vector, x ∈ ℝ , which minimises the following 1 2 (Munsell et al., 2015; Casanova et al., 2011). We varied the λ parameter regularised logistic regression problem: 205 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 2. Resection mask and region volume changes after surgery. (a) Region mask in MNI space for visualisation only (all other analyses were performed in preoperative T1 space) shows similar spatial profile between outcome groups. (b) Absolute amount of tissue resected does not relate with outcome. (c) Percentage of whole brain removed, does not relate to outcome. (d) Proportion of tissue remaining after surgery varies between subjects and regions. Here, 1 represents that the region remained intact, 0 represents that region was removed entirely. Full region names can be found in Table 2 Each dot per region is a single subject (53 per region in total). sequentially from 0.01 to 0.96 in steps of 0.05 and the ρ parameter from The sum of all streamlines directly connected to a region is termed the 0.01 to 1.96 in steps of 0.05. At each grid point, we incorporated the region strength - i.e. how strongly a region is directly connected to all machine learning framework detailed above, obtained a reduced fea- other regions. Fig. 3 shows the impact of removing all streamlines that ture space, and computed the performance metrics indicating the ac- pass into the resected tissue by plotting the resultant change in strength. curacy, sensitivity, and specificity. As expected, those regions that are partially removed (i.e. those showing a reduction in Fig. 2d) also have reduced strengths. The effect of surgery on other brain regions can be demonstrated with this ap- 3. Results proach. For example, the ipsilateral basal forebrain, although it is ex- tratemporal and not resected, has a substantial reduction in strength for Firstly, we investigate the inferred impact of surgery on network many patients. This suggests that preoperatively the basal forebrain has regions and secondly, on network connections. Finally, using machine many of its connections with the resected tissue, rather than other learning we investigate how network change metrics relate to outcomes areas. Several other areas beyond the resected regions also undergo from surgery. alterations in strength in many subjects including the ipsilateral lateral, medial and posterior, orbital gyri among others. These results are 3.1. Impact of surgery on network regions shown in detail in Table S12(a, b), and repeated for the FreeSurfer regions in Fig. S4. It is worth noting that region strength changes shown 3.1.1. Region volumes here may not be limited to only regions partly, or wholly resected, but All patients had similar surgical resections with the most variation may also change if a portion of a tract travels through the resection site, between subjects on the boundary of the resection site (Fig. 2a). The effectively disconnecting two non-resected areas. absolute (Fig. 2b), and relative (Fig. 2c) volume of the resected tissue Broadly speaking, the strengths of contralateral regions were not as did vary between subjects, but did not explain outcome (p = 0.56 and substantially affected by the surgery which suggests most connections p = 0.62 respectively – permutation test for difference between mean, are intrahemispheric. None of the changes in node strength met sig- 10,000 permutations used). Fig. 2d shows the proportion of tissue re- nificance for association with seizure-free outcome or not. maining after resection in different brain regions. Unsurprisingly the regions which were most reduced in volume were amygdala, hippo- campus, entorhinal cortex, parahippocampal gyrus, and the temporal 3.1.3. Region network measures pole. This suggests partial disruption to multiple regions, rather than The region volume change (Fig. 2) captures information solely complete disruption to specific regions. Although there is variation about individual regions, while the region strength change (Fig. 3) between subjects in the proportion of each region removed, this did not captures information about a region and its directly connected neigh- predict outcome (Fig. S2). These results are reproduced for the Free- bours. Other measures of network regions capture information re- Surfer atlas in Fig. S3. garding aspects of their role in the wider network. The betweenness centrality of a region represents how many of the 3.1.2. Region strength shortest paths between other regions traverse the region. Hub regions Each brain region is connected to other regions with streamlines. tend to occur in many of the shortest paths and have been implicated in 206 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 3. Changes in region strength are predominantly ipsilateral. (a) Proportion of strength remaining after surgery. A value of 1 (0) indicates all streamlines connecting to that region are kept (removed). (b) An example region shows ipsilateral hippocampal connection change in the ILAE > 1 group, and the ILAE 1 group. (c) Substantially affected regions visualised in 3D MNI space. Size and colour correspond to median reduction across all patients. Colour bars in panel (c) and (a) are consistent and correspond to the proportion of strength remaining between 0 and 1. a wide range of neurological disorders (Crossley et al., 2014). If a region measure beyond the resection site following surgery (Fig. S6), which is surgically removed, the shortest path between regions may be via did not relate to outcome. Reductions of regional network communic- other areas. The betweenness centrality of a region can therefore either ability were most pronounced in ipsilateral amygdala, entorhinal, and increase or decrease following surgery. temporal pole regions but also did not relate to outcome (p > 0.05, Fig. 4 shows the change in betweenness centrality after surgery. Kruskal-Wallis test, supplementary results S7). Regions involved in the resection have reductions in betweenness centrality suggesting reduced integration of those areas with the wider 3.2. Impact of surgery on network connections network. However, if those regions are no longer occurring in shortest paths, the following question remains: which ones are? Fig. 4a shows 3.2.1. Connection strength that many regions have substantial increase in betweenness centrality. Fig. 5a shows the connections which are affected by surgery in all The redirection of shortest paths in the network following surgery ap- patients. These connections typically involve those areas which are pears to be widely distributed among many other regions, rather than resected. Panel (b) shows the median reduction in connectivity fol- just the hubs - suggesting there are many alternative pathways (i.e. lowing surgery. Connections in panel (a) are a subset of those in panel redundancy). Panel (b) in Fig. 4 shows the spatial distribution of the (b). There are many connections which are only partly (as opposed to median change across all patients. This confirms that the increases entirely) reduced in strength. Partial reduction in strength can be (shown in blue) are widespread, even in contralateral regions. considered as removing some, but not all, of the connectivity between Because of the widespread redistribution of the shortest paths, we two areas. The greatest impact was seen in connections into the ipsi- also evaluated global efficiency, which is the inverse of the average lateral posterior temporal lobe and the inferior frontal lobe, with in shortest path length of the network. Fig. 4c demonstrates the reduction some patients, connections to the ipsilateral parietal lobe and con- in global efficiency following surgery. Given the redirection of shortest tralateral temporal lobe and medial frontal and parietal lobes as would paths shown in panel (a) of the same figure, it is now easy to see why be expected given Fig. 3. the reduction on global efficiency is only < 10% in most cases. This change does not relate to seizure outcome (p = 0.49 – permutation test 3.2.2. Connection network measures for difference between mean, 10,000 permutations used). We observed Similar to the regional betweenness centrality, connection (edge) similar results when using the FreeSurfer atlas (Fig. S5). betweenness centrality captures information regarding how often a We additionally investigated the change in regional clustering connection occurs in the shortest path between other regions – i.e. how coefficient, a measure of the interconnectedness of neighbours of a frequently that connection is traversed in the shortest paths. This value region, following surgery. We found very subtle differences in this can increase or decrease following surgery, as alternative paths is taken. 207 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 4. Widespread changes in region betweenness centrality following surgery. (a) Enhancements (positive values) and reductions (negative values) in betweenness centrality following surgery. A value of 0 indicates no change. (b) Median change of betweenness centrality shows that decreases are generally constrained to the resected regions (red) whereas increases are widespread across many other regions (blue). Colour bars in panel (b) and (a) are consistent and correspond to the change in betweenness centrality (y axis in panel a). Red (blue) indicates a decrease (increase) in betweenness centrality after surgery. (c) Global network efficiency is typically reduced by < 10% following surgery, and does not predict outcome. Fig. 6 shows the median change in betweenness centrality across all 3.3. Network change association with seizure outcome subjects. Decreases in connection betweenness centrality were broadly limited to ipsilateral temporal areas whereas the increases were more As an outlook to prospective applications, we used a machine widely distributed. learning strategy to retrospectively investigate the association between Fig. 5. Connectivity is disrupted by surgery. (a) Connections which are reduced in strength following surgery in all patients. (b) Median reduction in connectivity strength across all patients. A value of 1 (0) indicates that the median change to a connection following surgery is a 0% (100%) reduction in the number of streamlines. Only values < 1 are shown and therefore shows only connections which are reduced in strength in the majority of patients (as opposed to connections which are reduced in all patients - shown in panel a). Connections in panel (b) are a superset of those in panel (a). 208 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 6. Widespread changes in connection betweenness centrality following surgery. Median decreases (increases) in connection betweenness centrality across all subjects shown in red (blue). network changes – region volume, region strength, and connection incorporated all changes in volume, region strength, and connection strength – with seizure outcome. strength and applied an elastic net algorithm to select those which were Our machine learning approach minimises the number of features most useful. Fig. 7 shows those features selected by the algorithm. that give a high accuracy of outcome classification. As input features we Fifteen features were selected, all of which were connections. This Fig. 7. Features derived with the elastic net algorithm. (a) Fifteen features which were found to be most informative using our machine learning pipeline. All are connections which are reduced in strength following surgery. Colour coding shows the difference in mean reduction between groups. Connections in red are reduced by surgery more in seizure-free patients than in not-seizure-free patients. (b) Correspondence with resected tissue. 209 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 suggests that incorporating knowledge of regional properties (strength networks inferred from diffusion MRI to investigate network properties and volume change) does not improve the prediction substantially. All with respect to surgical outcomes (Bonilha et al., 2013). This includes a fifteen connections in Fig. 7 are coloured according to the difference in multi-centre study using machine learning which is perhaps most si- mean value between groups. In other words, a negative value shows milar to our work (Munsell et al., 2015). That study analysed pre- that the mean ILAE > 1 patient has a greater reduction in connectivity operative structural networks and compared several machine learning than the mean ILAE 1 patient (and vice-versa). The greatest reduction algorithms for feature selection and classification for prediction of any group can have is 1 (i.e. a 100% reduction). Values are therefore surgical outcomes. They found that the best performing algorithm uti- bound between ± 1, and consequently not normally distributed. We lised an elastic net/support vector machine combination, and achieved therefore show the difference in mean, rather than the effect size. Eight a cross-site accuracy of 66%, with a same-site accuracy of 70%. We of the features are connections involving the temporal pole. The list of therefore chose to use broadly the same technique for classification selected features is given in Table S13 and are visualised in movie S14. here. Instead of investigating preoperative networks, however, we in- Replication of these results using normalised data, preoperative data, or vestigate the difference networks - i.e. the inferred changes brought the alternative FreeSurfer parcellation did not lead to substantial im- about by patient-specific surgery. These difference networks have led us provements in accuracy (Supplementary results S10). Those who be- to achieve a same-site accuracy of 79% in our data. came seizure free had greater reductions of connectivity into the ipsi- We have performed a retrospective analysis of connectivity change lateral superior temporal lobe and frontal lobe. using a mask drawn from postoperative data – i.e. the tissue that was Table 3 shows our findings including accuracy, sensitivity and resected. However, we envisage that in future a prospective prediction specificity when using these features. Our specificity of 64.7% reflects could be performed. This should be possible using the preoperative data the finding that of all patients that the model predicted would have a and a mask of what the surgeon intends to resect (e.g. through the not-seizure-free outcome, 64.7% actually had a not-seizure-free out- manual adaptation of an ‘average mask’ in the case of a standardised come. This is substantially higher than the empirical distribution surgery such as anterior temporal lobe resection). Indeed, in theory a (32%). prediction is possible for any given presurgical resection plan. Fig. 8 suggests how this could be incorporated into such an evaluation, with the aim to address the question of, ‘will this proposed resection render 4. Discussion the patient seizure free?’ We do not address the separate question of, ‘what is the optimal resection for this patient?’, which has to integrate In this study we have used detailed neuroanatomical information the need to avoid damage to critical structures (e.g. Winston et al., from both pre- and postoperative MRI to infer the impact of surgery on 2012). This study has developed a pipeline for answering the first of patient-specific brain networks. Our three main contributions are as those two questions, and retrospectively validated it with 79% accu- follows. racy. Approaches to investigate the second of those questions have been First, we find that the impact of surgery leads to a reduction proposed recently (Sinha et al., 2017; Goodfellow et al., 2016). of < 10% in efficiency in the majority of patients, due to a redirection of shortest paths. Second, there is no single feature which fully accounts for outcome with 100% accuracy. This means that there is likely not a 4.1. Regional change following surgery single ‘target’ to aim for with surgery in this patient group that is captured by our data. Third, we have demonstrated that machine A potentially surprising result (Fig. 2), is that we found that the learning, in conjunction with connectivity change metrics, can produce amount of resected tissue did not appear to predict outcome. This is in predictions with high accuracy, and specificity which is twice as high as contrast to the work of (Wyler et al., 1995) and (Shamim et al., 2009) the empirical distribution. This means that around two thirds of the who have shown that larger resections do correlate with seizure out- patients we predict are not seizure free, are actually not seizure free. comes. However, other studies have found conflicting results which This could help preoperatively in identifying which patients might have accord with our own findings (Hardy et al., 2003; Keller et al., 2015; lower chances of becoming seizure free and put extra consideration in Liao et al., 2016). The variations in the literature may be due to dif- offering surgery or when consenting those patients. ferent populations of left/right TLE, evidence of hippocampal sclerosis, Most previous network-based studies in epilepsy typically perform and follow-up duration. Here we have controlled for these aspects group comparisons of patients relative to controls using either struc- where possible with no significant differences between outcome groups tural networks (Kamiya et al., 2016; Taylor et al., 2015; Lemkaddem (Table 1). et al., 2014; de Salvo et al., 2014; Bonilha et al., 2012), functional Our finding that there is variation between subjects, in terms of networks (Chavez et al., 2010; Liao et al., 2010), or both (Zhang et al., proportion of regions removed, can be explained by two potential fac- 2011; Douw et al., 2015; Wirsich et al., 2016). Few have investigated tors. First, there are differences in tissue removed between subjects, and alterations in global network properties with respect to surgical out- second, there are differences in regions between subjects. The latter was come (although see He et al., 2017 and Morgan et al., 2017 for recent one of our motivations for using two different parcellations to aid examples). Bonilha and colleagues utilised preoperative structural comparison (the second of which is shown in supplementary materials). Table 3 Confusion matrix indicating performance of machine learning framework for predicting surgical outcomes. Actual surgical outcome Seizure free = 36 Not seizure free = 17 Predicted Seizure free = 37 True positive = 31 False positive = 6 outcome (Type I Error) Not seizure free = 16 False negative = 5 True negative = 11 (Type II Error) True Positive Rate, False Positive Rate or Sensitivity = 0.861 or Fall out = 0.353 Accuracy = 0.792 False Negative Rate, True Negative Rate, or Miss rate = 0.139 or Specificity = 0.647 210 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 8. Potential incorporation of pipeline for prospective evaluation. Suggestion of how our approach could be used as part of the pre-surgical evaluation. From left to right: Multimodal data is acquired and evaluated at the multi-disciplinary team meeting. If surgery is recommended a proposed surgical resection mask is drawn, then fed into the connection change pipeline. The pipeline then gives a predicted post-operative network given the patient-specific mask and patient-specific presurgical DTI & T1w MRI. The pipeline uses these predicted connectivity changes, along with the pre-existing feature set to predict seizure outcome. In this example the predicted outcome would be seizure free. Furthermore, the resolution of the parcellation (i.e. the number of re- paths, leading to increased FA and thus avoiding major cognitive def- gions that the brain network is divided into) will also affect results. icits following surgery. Here we use fairly low resolution parcellations of ≈ 100 regions in The relationship between network efficiency, shortest path lengths, contrast to recently proposed higher resolutions (Besson et al., 2014a, and function is not entirely clear. However, path transitivity, which is 2014b; Taylor et al., 2017; Besson et al., 2017). Our reasoning for this is strongly related to path length has been shown to strongly correlate to that in this study we compare regional properties between subjects; functional connectivity (Goñi et al., 2014). Network efficiency itself has thus it is important that we are confident that one region in one subject also been shown to relate to intelligence in functional networks in is the same region in another, and can be reliably reproduced. One of healthy controls (van den Heuvel et al., 2009). It remains to be seen the reasons that the FreeSurfer software is so popular for this is that the how changes to network efficiency metrics following surgery relate to authors have demonstrated confidence maps at region boundaries patient changes in brain function beyond seizure outcome. (Fischl et al., 2004). One of the drawbacks of using such low resolution In addition to widespread network changes such as those described connectivity matrices however is the inability to investigate detailed above, the network change pipeline also allows the inference of more within-area architecture as in (Taylor et al., 2017). Indeed, this may be localised differences. As shown in Fig. 3, we found decreases in node an important and productive area of future research. strength in extratemporal areas, particularly the basal forebrain in the frontal lobe. This leads us to suggest that frontal lobe function may be altered following surgery. Several studies are in agreement with this. 4.2. Network change following surgery For example, Stretton et al. (2014) found improvements in frontal lobe working memory function after surgery, while Martin et al. (2000) We suggest that our finding of only a < 10% reduction in network showed significant increases in verbal fluency following surgery. efficiency following surgery (Fig. 4c) may explain why surgery usually However, other studies have also found conflicting results (see Stretton leads to relatively few severe side effects. The reason for only a subtle and Thompson, 2012 for review). We hypothesise these improvements change in efficiency is the redirection of shortest paths via other con- are related to post-surgical alterations in connectivity, and the varia- nections as shown in Fig. 6. We speculate that the decreases in edge bility in the results may be down to patient variation (i.e. the large betweenness correspond to those connections which undergo Wallerian spread in Fig. 3). This effect is typically referred to as a process of degeneration. Furthermore, we speculate that connections with in- functional improvement once nociferous cortex has been removed. creased betweenness centrality will undergo ‘strengthening’, potentially Future studies correlating these changes in connectivity with changes in reflected by increases in FA on structural imaging postoperatively. frontal lobe function will help to elucidate these relationships, poten- Qualitative evidence to support this has been shown by (Winston et al., tially predicting not only seizure outcome, but also functional out- 2014) who found decreases in FA in tracts involving the resected tissue comes. Indeed, this has already been demonstrated in stroke where and increased FA in more distant ipsilateral superior tracts within fairly connectivity change metrics have been found to relate to functional short timescales of three months. Further support for this hypothesis outcomes (Kuceyeski et al., 2015a, 2015b). comes from a study by (Jeong et al., 2016) who report increases in the FA of connections following frontal lobe epilepsy surgery far from the surgical site. Other studies have also made similar findings (Yogarajah 4.3. Machine learning for outcome prediction et al., 2010; Pustina et al., 2014; Faber et al., 2013). Taken together, this suggests a potential rerouting of connectivity, via new shortest Although very few have used brain network data, several studies in 211 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 recent years have begun to apply machine learning strategies for sur- giving confidence in the reproducibility of the masks drawn here. Our gical outcome prediction. The benefits of the approach appear pro- choice of tractography algorithm was influenced by the intention to mising. For example, (Bernhardt et al., 2015), using hippocampal limit false positive connections as suggested by Zalesky et al. (2016). measurements as features, made outcome predictions with 81% accu- Despite this, there are known limitations with all approaches and this racy. In another study of surgical outcomes in temporal lobe epilepsy study should also be considered with the known limitations of the (Memarian et al., 2015) demonstrated accuracy of 95% using various imaging technologies in mind. For example, the inability to resolve features derived from both imaging and demographic data. In addition, connection direction and potential bias for tractography to favour (Feis et al., 2013) achieved accuracy of outcome prediction of up to shorter, straighter connections (Jones, 2010). Our use of number of 96% using preoperative T1 measurements of white matter volumes. streamlines as a connectivity metric is based on the extensive prior Other studies have also used machine learning algorithms, along with literature of ChaCo analysis by Kuceyeski et al. (2013, 2015a, 2015b). imaging data, to classify lateralisation of seizure onset (Focke et al., However, other connection strength metrics, such as density (Taylor 2012; Bennett et al., 2017). Our prediction accuracy is in a broadly et al., 2015), or volume (Irimia and Van Horn, 2016) may also yield similar range to many of these and may be improved by including other informative results. features such as those used for generating predictive nomograms (Jehi Our network change pipeline allows to infer the immediate network et al., 2015). change brought about by surgery. It is worth highlighting that this does Multiple approaches exist in machine learning to attempt to im- not necessarily represent the long-term brain network changes in the prove the accuracy of the outcome prediction by selecting the most months/years that occur as a result of e.g. plasticity or degeneration. predictive features from a high-dimensional feature space (Tibshirani, This is both a strength and a weakness of our approach and is one 1996, Hoerl and Kennard, 2004, Breiman et al., 1984, Breiman et al., reason why we assess outcomes at only 12 months. The mechanisms 1984). Here we have adopted a fairly conservative approach in which underlying why some patients relapse to seizure recurrence only at we implement elastic net regularisation (Zou and Hastie, 2005) while several years after surgery (De Tisi et al., 2011) are unclear and would incorporating regularisation parameters which identify a minimum require longitudinal diffusion MRI data gathered over several years to number of features important for the prediction. Even using this con- investigate in this framework. The potential to use our approach pro- servative approach we obtain a prediction accuracy of 79.2%. It is spectively (Fig. 8) is a distinct advantage however. worth noting that we impose a prior probability expectation of 50%. If we did not impose this, and instead used the empirical distribution of 4.5. Outlook and wider implications outcomes (68%: 32% - row 1 in Table 1), it would be easy to get ac- curacies of at least 68% by simply suggesting all patients will be seizure- Although we have focussed exclusively on temporal lobe epilepsy free. The prediction accuracy of 79.2%, even using our highly con- surgery here, this method could be generalized to other types of neu- servative approach, is therefore encouraging. rosurgery such as those for other epilepsies such as neocortical epilepsy, tumours or movement disorders. Additionally, other outcome criteria 4.4. Strengths & weaknesses could also be predicted. For example, instead of predicting seizure outcome, functional deficits arising due to damage to known functional There are several strengths to this work. For example, here we have subnetworks (e.g. somatomotor, attention, visual) could be predicted a fairly large cohort of 53 patients which were all scanned on the same using existing atlases (Yeo et al., 2011). Our approach would be pos- scanner using the same scanning protocol. We also have good con- sible if preoperative imaging data is available and a proposed resection sistency in terms of operation type (all anterior temporal lobe resection mask drawn before the surgery. Deficits for the given mask (surgery) performed by the same 2 surgeons using the same technique), and could then be predicted for individual subjects. Additionally, multiple consistency in follow-up duration for outcome - 12 months. The con- proposed masks (surgeries) could be tested against each other in-silico, sistency in follow-up duration is particularly important because out- the results for each compared, and the optimal surgery predicted. Al- comes can change over time (De Tisi et al., 2011). A further strength of gorithms for a similar approach are already being developed for in- our study, in comparison to some other network connectivity change vasive electrode recordings (Rodionov et al., 2013; Sinha et al., 2017). studies, is that our analysis is performed in subject (native) space. This means nonlinear deformations of the scans to a common space, which 4.6. Conclusion may decrease the accuracy of the region mapping, are not necessary. Our investigation of changes in structural connectivity following surgery It is to be expected that outcome from surgery will depend not only using a ChaCo-like approach is, to our knowledge, novel in epilepsy. on the network connectivity of the brain before surgery, but the dis- Moreover, the ChaCo approach has the advantage of not requiring ar- ruption to this network caused by the surgery itself. This can be mea- bitrary thresholding of the network, or comparison to arbitrarily chosen sured as the change in connectivity, which can be predicted for a given random networks as a baseline – instead the patient's own pre-operative resection preoperatively. Our analysis shows that such connectivity network is used. change metrics can be used to retrospectively predict seizure outcome A possible weakness of this study is that we have flipped the con- with high accuracy, and may relate to functional outcomes. nectivity matrices in right TLE patients, simplifying our analysis into Supplementary data to this article can be found online at https:// ipsi- and contralateral hemispheres. Although there is precedence for doi.org/10.1016/j.nicl.2018.01.028. this approach (Bonilha et al., 2013; Munsell et al., 2015; Ji et al., 2015) it has been suggested that left and right TLE are not simply the mirror Acknowledgments image of each other (Besson et al., 2014a, 2014b), especially with re- spect to post-surgical cognitive deficits and resection size which is ty- We are grateful to the Wolfson Foundation and the Epilepsy Society pically larger in the non-dominant hemisphere. We did try to mitigate for supporting the Epilepsy Society MRI scanner. this however by ensuring the proportion of left:right patients in each PNT was funded by Wellcome Trust (105617/Z/14/Z). PNT thanks outcome group are similar (p = 0.3353; Table 1). Rob Forsyth and Helen Marley for discussion. GPW was supported by Another potential bias is the resection mask, which was manually the MRC (G0802012, MR/M00841X/1). SBV is funded by the National drawn for all subjects by the same rater, with a majority subset drawn Institute for Health Research University College London Hospitals by a second rater (results S9). Variations between raters exist, but are Biomedical Research Centre (NIHR BRC UCLH/UCL High Impact small. The correlation of network features between raters is > 0.98, Initiative). 212 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 This study was carried out in part at UCLH and the support of the Estrada, E., Hatano, N., 2008. Communicability in complex networks. Phys. Rev. E 77 (3), NIHR-funded UCLH/UCL BRC is gratefully acknowledged. JD is sup- Faber, Jennifer, Schoene-Bake, Jan-Christoph, Trautner, Peter, Lehe, Marec, Elger, ported by the NIHR Senior investigator scheme. Christian E., Weber, Bernd, 2013. Progressive fiber tract affections after temporal lobe surgery. Epilepsia 54 (4), e53-e57. Feis, Delia-Lisa, Schoene-Bake, Jan-Christoph, Elger, Christian, Wagner, Jan, References Tittgemeyer, Marc, Weber, Bernd, 2013. Prediction of post-surgical seizure outcome in left mesial temporal lobe epilepsy. Neurol. Clin. 2, 903–911 (Elsevier). Andersson, Jesper L.R., Sotiropoulos, Stamatios N., 2016. An integrated approach to Fischl, Bruce, 2012. FreeSurfer. NeuroImage 62 (2), 774–781 (Elsevier). correction for off-resonance effects and subject movement in diffusion mr imaging. Fischl, Bruce, Salat, David H., Busa, Evelina, Albert, Marilyn, Dieterich, Megan, NeuroImage 125, 1063–1078 (Elsevier). Haselgrove, Christian, van der Kouwe, Andre, et al., 2002. Whole brain segmentation: Bauer, Roman, Kaiser, Marcus, 2017. Nonlinear growth: an origin of hub organization in automated labeling of neuroanatomical structures in the human brain. Neuron 33 (3), complex networks. Royal Soc. Open Sc. 4 (3), 160691 (The Royal Society). 341–355 (Elsevier). Bennett, O., Cardoso, J., Duncan, J.S., Winston, G., Ourselin, S., 2017. Learning how to Fischl, Bruce, van der Kouwe, André, Destrieux, Christophe, Halgren, Eric, Ségonne, See the Invisible - Using Machine Learning to Find Underlying Abnormality Patterns Florent, Salat, David H., Busa, Evelina, et al., 2004. Automatically parcellating the in Reportedly Normal MR Brain Images from Patients with Epilepsy. Proceedings. human cerebral cortex. Cereb. Cortex 14 (1), 11–22 (Oxford Univ Press). ISMRM. Focke, Niels K., Yogarajah, Mahinda, Symms, Mark R., Gruber, Oliver, Paulus, Walter, Bernhardt, Boris C., Chen, Zhang, He, Yong, Evans, Alan C., Bernasconi, Neda, 2011. Duncan, John S., 2012. Automated Mr image classification in temporal lobe epilepsy. Graph-theoretical analysis reveals disrupted small-world organization of cortical NeuroImage 59 (1), 356–362 (Elsevier). thickness correlation networks in temporal lobe epilepsy. Cereb. Cortex 21 (9), Goñi, J., van den Heuvel, M.P., Avena-Koenigsberger, A., Velez de Mendizabal, N., Betzel, 2147–2157 (Oxford Univ Press). R.F., Griffa, A., ... Sporns, O., 2014. Resting-brain functional connectivity predicted Bernhardt, Boris C., Hong, Seok-Jun, Bernasconi, Andrea, Bernasconi, Neda, 2015. by analytic measures of network communication. Proc. Natl. Acad. Sci. U. S. A. 111 Magnetic resonance imaging pattern learning in temporal lobe epilepsy: classification (2), 833–838. http://dx.doi.org/10.1073/pnas.1315529111. and prognostics. Ann. Neurol. 77 (3), 436–446 (Wiley Online Library). Goodfellow, M., Rummel, C., Abela, E., Richardson, M.P., Schindler, K., Terry, J.R., 2016. Besson, Pierre, Dinkelacker, Vera, Valabregue, Romain, Thivard, Lionel, Leclerc, Xavier, Estimation of brain network ictogenicity predicts outcome from epilepsy surgery. Sci. Baulac, Michel, Sammler, Daniela, et al., 2014a. Structural connectivity differences in Rep. 6. left and right temporal lobe epilepsy. NeuroImage 100, 135–144 (Elsevier). Gooneratne, I.K., Mannan, S., de Tisi, J., Gonzalez, J.C., McEvoy, A.W., Miserocchi, A., Besson, Pierre, Lopes, Renaud, Leclerc, Xavier, Derambure, Philippe, Tyvaert, Louise, Diehl, B., Wehner, T., Bell, G.S., Sander, J.W., Duncan, J.S., 2017. Somatic compli- 2014b. Intra-subject reliability of the high-resolution whole-brain structural con- cations of epilepsy surgery over 25 years at a single center. Epilepsy Res. 132, 70–77. nectome. NeuroImage 102, 283–293 (Elsevier). Hardy, Steven G., Miller, John W., Holmes, Mark D., Born, Donald E., Ojemann, George Besson, Pierre, Kathleen Bandt, S., Proix, Timothée, Lagarde, Stanislas, Jirsa, Viktor K., A., Dodrill, Carl B., Hallam, Danial K., 2003. Factors predicting outcome of surgery Ranjeva, Jean-Philippe, Bartolomei, Fabrice, Guye, Maxime, 2017. Anatomic con- for intractable epilepsy with pathologically verified mesial temporal sclerosis. sistencies across epilepsies: a stereotactic-EEG informed high-resolution structural Epilepsia 44 (4), 565–568 (Wiley Online Library). connectivity study. Brain 140 (10), 2639–2652. He, X., Doucet, G.E., Pustina, D., Sperling, M.R., Sharan, A.D., Tracy, J.I., 2017. Betzel, Richard F., Avena-Koenigsberger, Andrea, Goñi, Joaquín, He, Ye, De Reus, Marcel Presurgical thalamic “hubness” predicts surgical outcome in temporal lobe epilepsy. A., Griffa, Alessandra, Vértes, Petra E., et al., 2016. Generative models of the human Neurology 10–1212. connectome. NeuroImage 124, 1054–1064 (Elsevier). van den Heuvel, M.P., Stam, C.J., Kahn, R.S., Pol, H.E.H., 2009. Efficiency of functional Bonilha, Leonardo, Nesland, Travis, Martz, Gabriel U., Joseph, Jane E., Spampinato, brain networks and intellectual performance. J. Neurosci. 29 (23), 7619–7624. Maria V., Edwards, Jonathan C., Tabesh, Ali, 2012. Medial temporal lobe epilepsy is Hoerl, A.E., Kennard, R.W., 2004. Ridge Regression. 9780471667193 John Wiley and associated with neuronal fibre loss and paradoxical increase in structural connectivity Sons, Inc.http://dx.doi.org/10.1002/0471667196.ess2280. of limbic structures. J. Neurol. Neurosurg. Psychiatry 83 (9), 903–909 (BMJ Irimia, Andrei, Van Horn, John Darrell, 2016. Scale-dependent variability and quantita- Publishing Group Ltd). tive regimes in graph-theoretic representations of human cortical networks. Brain Bonilha, Leonardo, Helpern, Joseph A., Sainju, Rup, Nesland, Travis, Edwards, Jonathan Connect. 6 (2), 152–163. C., Glazier, Steven S., Tabesh, Ali, 2013. Presurgical connectome and postsurgical Jehi, L., Yardi, R., Chagin, K., Tassi, L., Russo, G.L., Worrell, G., ... Chauvel, P., 2015. seizure control in temporal lobe epilepsy. Neurology 81 (19), 1704–1710 (AAN Development and validation of nomograms to provide individualised predictions of Enterprises). seizure outcomes after epilepsy surgery: a retrospective analysis. Lancet Neurol. 14 Bonilha, Leonardo, Jensen, Jens H., Baker, Nathaniel, Breedlove, Jesse, Nesland, Travis, (3), 283–290. Lin, Jack J., Drane, Daniel L., Saindane, Amit M., Binder, Jeffrey R., Kuzniecky, Jenkinson, Mark, Smith, Stephen, 2001. A global optimisation method for robust affine Ruben I., 2015. The brain connectome as a personalized biomarker of seizure out- registration of brain images. Med. Image Anal. 5 (2), 143–156 (Elsevier). comes after temporal lobectomy. Neurology 84 (18), 1846–1853 (AAN Enterprises). Jenkinson, Mark, Bannister, Peter, Brady, Michael, Smith, Stephen, 2002. Improved op- Breiman, L., Friedman, J.H., Alshen, R.A., Stone, C.J., 1984. CART: Classification and timization for the robust and accurate linear registration and motion correction of Regression Trees. Wadsworth, Belmont, CA. brain images. NeuroImage 17 (2), 825–841 (Elsevier). Cardoso, M. Jorge, Modat, Marc, Wolz, Robin, Melbourne, Andrew, Cash, David, Jenkinson, Mark, Beckmann, Christian F., Behrens, Timothy E.J., Woolrich, Mark W., Rueckert, Daniel, Ourselin, Sebastien, 2015. Geodesic information flows: spatially- Smith, Stephen M., 2012. Fsl. NeuroImage 62 (2), 782–790 (Elsevier). variant graphs and their application to segmentation and fusion. IEEE Trans. Med. Jeong, Jeong-Won, Asano, Eishi, Juhász, Csaba, Behen, Michael E., Chugani, Harry T., Imaging 34 (9), 1976–1988 (IEEE). 2016. Postoperative axonal changes in the contralateral hemisphere in children with Casanova, Ramon, Wagner, Benjamin, Whitlow, Christopher, Williamson, Jeff, Shumaker, medically refractory epilepsy: a longitudinal diffusion tensor imaging connectome Sally, Maldjian, Joseph, Espeland, Mark, 2011. High dimensional classification of analysis. Hum. Brain Mapp. 37 (11), 3946–3956 (Wiley Online Library). structural Mri Alzheimer's disease data based on large scale regularization. Front. Ji, Gong-Jun, Zhang, Zhiqiang, Xu, Qiang, Wei, Wei, Wang, Jue, Wang, Zhengge, Yang, Neuroinform. 5, 22. http://dx.doi.org/10.3389/fninf.2011.00022. Fang, et al., 2015. Connectome reorganization associated with surgical outcome in Chavez, Mario, Valencia, Miguel, Navarro, Vincent, Latora, Vito, Martinerie, Jacques, temporal lobe epilepsy. Medicine 94 (40), e1737 (LWW). 2010. Functional modularity of background activities in normal and epileptic brain Jones, D.K., 2010. Challenges and limitations of quantifying brain connectivity in vivo networks. Phys. Rev. Lett. 104 (11), 118701 (APS). with diffusion MRI. Quant. Imaging Med. Surg. 2, 341–355. Cook, Philip A., Symms, Mark, Boulby, Philip A., Alexander, Daniel C., 2007. Optimal Jutila, L., Immonen, A., Mervaala, E., Partanen, J., Partanen, K., Puranen, M., Kälviäinen, acquisition orders of diffusion-weighted mri measurements. J. Magn. Reson. Imaging R., et al., 2002. Long term outcome of temporal lobe epilepsy surgery: analyses of 140 25 (5), 1051–1058 (Wiley Online Library). consecutive patients. J. Neurol. Neurosurg. Psychiatry 73 (5), 486–494 (BMJ Crossley, Nicolas A., Mechelli, Andrea, Scott, Jessica, Carletti, Francesco, Fox, Peter T., Publishing Group Ltd). McGuire, Philip, Bullmore, Edward T., 2014. The hubs of the human connectome are Kamiya, Kouhei, Amemiya, Shiori, Suzuki, Yuichi, Kunii, Naoto, Kawai, Kensuke, Akira, generally implicated in the anatomy of brain disorders. Brain 137 (8), 2382–2395 Kunimatsu, Nobuhito, Saito, Shigeki, Aoki, Kuni, Ohtomo, 2016. Machine learning of (Oxford Univ Press). Dti structural brain connectomes for lateralization of temporal lobe epilepsy. Magn. De Tisi, Jane, Bell, Gail S., Peacock, Janet L., McEvoy, Andrew W., Harkness, William F.J., Reson. Med. Sci. 15 (1), 121–129 (Japanese Society for Magnetic Resonance in Sander, Josemir W., Duncan, John S., 2011. The long-term outcome of adult epilepsy Medicine). surgery, patterns of seizure remission, and relapse: a cohort study. Lancet 378 (9800), Keller, Simon S., Richardson, Mark P., Schoene-Bake, Jan-Christoph, O'muircheartaigh, 1388–1395 (Elsevier). Jonathan, Elkommos, Samia, Kreilkamp, Barbara, Goh, Yee Yen, Marson, Anthony G., Desikan, Rahul S., Ségonne, Florent, Fischl, Bruce, Quinn, Brian T., Dickerson, Bradford Elger, Christian, Weber, Bernd, 2015. Thalamotemporal alteration and postoperative C., Blacker, Deborah, Buckner, Randy L., et al., 2006. An automated labeling system seizures in temporal lobe epilepsy. Ann. Neurol. 77 (5), 760–774 (Wiley Online for subdividing the human cerebral cortex on Mri scans into gyral based regions of Library). interest. NeuroImage 31 (3), 968–980 (Elsevier). Kuceyeski, Amy, Maruta, Jun, Relkin, Norman, Raj, Ashish, 2013. The network mod- Douw, Linda, DeSalvo, Matthew N., Tanaka, Naoaki, Cole, Andrew J., Liu, Hesheng, ification (nemo) tool: elucidating the effect of white matter integrity changes on Reinsberger, Claus, Stufflebeam, Steven M., 2015. Dissociated multimodal hubs and cortical and subcortical structural connectivity. Brain Connect. 3 (5), 451–463 (Mary seizures in temporal lobe epilepsy. Ann. Clin. Transl. Neurol. 2 (4), 338–352 (Wiley Ann Liebert, Inc. 140 Huguenot Street, 3rd Floor New Rochelle, NY 10801 USA). Online Library). Kuceyeski, A.F., Vargas, W., Dayan, M., Monohan, E., Blackwell, C., Raj, A., Fujimoto, K., Duchi, John, Shalev-Shwartz, Shai, Singer, Yoram, Chandra, Tushar, 2008. Efficient Gauthier, S.A., 2015a. Modeling the relationship among gray matter atrophy, ab- projections onto the L1-ball for learning in high dimensions. In: Proceedings of the normalities in connecting white matter, and cognitive performance in early multiple 25th International Conference on Machine Learning. ACM, New York, NY, USA, pp. sclerosis. Am. J. Neuroradiol. 36 (4), 702–709 (Am Soc Neuroradiology). 272–279. (ICML ’08). https://doi.org/10.1145/1390156.1390191. Kuceyeski, Amy, Navi, Babak B., Kamel, Hooman, Relkin, Norman, Villanueva, Mark, Raj, 213 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Ashish, Toglia, Joan, O'dell, Michael, Iadecola, Costantino, 2015b. Exploring the Epilepsy Res. 98 (1), 1–13. brain's structural connectome: a quantitative stroke lesion-dysfunction mapping Stretton, J., Sidhu, M.K., Winston, G.P., Bartlett, P., McEvoy, A.W., Symms, M.R., Koepp, study. Hum. Brain Mapp. 36 (6), 2147–2160 (Wiley Online Library). M.J., Thompson, P.J., Duncan, J.S., 2014. Working memory network plasticity after Kuceyeski, Amy, Navi, Babak B., Kamel, Hooman, Raj, Ashish, Relkin, Norman, Toglia, anterior temporal lobe resection: a longitudinal functional magnetic resonance Joan, Iadecola, Costantino, O'dell, Michael, 2016. Structural connectome disruption imaging study. Brain 137 (5), 1439–1453. at baseline predicts 6‐months post‐stroke outcome. Hum. Brain Mapp. 37 (7), Taylor, Peter N., Han, Cheol E., Schoene-Bake, Jan-Christoph, Weber, Bernd, Kaiser, 2587–2601. Marcus, 2015. Structural connectivity changes in temporal lobe epilepsy: spatial Leemans, Alexander, Jones, Derek K., 2009. The B-matrix must be rotated when cor- features contribute more than topological measures. Neuroimaging Clin. N. Am. 8, recting for subject motion in DTI data. Magn. Reson. Med. 61 (6), 1336–1349. 322–328 (Elsevier). Lemkaddem, Alia, Daducci, Alessandro, Kunz, Nicolas, Lazeyras, François, Seeck, Taylor, Peter N., Wang, Yujiang, Kaiser, Marcus, 2017. Within brain area tractography Margitta, Thiran, Jean-Philippe, Vulliémoz, Serge, 2014. Connectivity and tissue suggests local modularity using high resolution connectomics. Sci. Rep. 7 (Nature microstructural alterations in right and left temporal lobe epilepsy revealed by dif- Publishing Group). fusion spectrum imaging. Neuroimaging Clin. N. Am. 5, 349–358 (Elsevier). Tibshirani, Robert, 1996. Regression shrinkage and selection via the lasso. J. R. Stat. Soc. Liao, Wei, Zhang, Zhiqiang, Pan, Zhengyong, Mantini, Dante, Ding, Jurong, Duan, Xujun, Ser. B Methodol. 267–288. Cheng, Luo, Lu, Guangming, Chen, Huafu, 2010. Altered functional connectivity and Tournier, J., Calamante, F., Connelly, A., 2012. MRtrix: diffusion tractography in crossing small-world in mesial temporal lobe epilepsy. PLoS One 5 (1), e8525 (Public Library fiber regions. Int. J. Imaging Syst. Technol. 22 (1), 53–66. of Science). Veropoulos, Konstantinos, Campbell, Colin, Cristianini, Nello, 1999. Controlling the Liao, Wei, Ji, Gong-Jun, Xu, Qiang, Wei, Wei, Wang, Jue, Wang, Zhengge, Yang, Fang, sensitivity of support vector machines. In: Proceedings of the International Joint et al., 2016. Functional connectome before and following temporal lobectomy in Conference on AI, pp. 55–60. mesial temporal lobe epilepsy. Sci. Rep. 6, 23153 (Nature Publishing Group). Vos, Sjoerd B., Tax, Chantal M.W., Luijten, Peter R., Ourselin, Sebastien, Leemans, Liu, Jun, Ye, Jieping, 2009. Efficient Euclidean projections in linear time. In: Proceedings Alexander, Froeling, Martijn, 2016. The importance of correcting for signal drift in of the 26th Annual International Conference on Machine Learning. ACM, New York, diffusion Mri. In: Magnetic Resonance in Medicine. Wiley Online Library. NY, USA, pp. 657–664. ICML ‘09. https://doi.org/10.1145/1553374.1553459. Wang, Zuoguan, Gunduz, Aysegul, Brunner, Peter, Ritaccio, Anthony L., Ji, Qiang, Schalk, Liu, J., Ji, S., Ye, J., 2009a. SLEP: Sparse Learning with Efficient Projections. Arizona Gerwin, 2012. Decoding onset and direction of movements using electrocortico- State University. http://www.public.asu.edu/~jye02/Software/SLEP. graphic (ECoG) signals in humans. Front. Neuroeng. 5. Liu, Jun, Chen, Jianhui, Ye, Jieping, 2009b. Large-scale sparse logistic regression. In: Wheeler-Kingshott, Claudia A.M., Hickman, Simon J., Parker, Geoffrey J.M., Ciccarelli, Proceedings of the 15th Acm Sigkdd International Conference on Knowledge Olga, Symms, Mark R., Miller, David H., Barker, Gareth J., 2002. Investigating cer- Discovery and Data Mining. ACM, New York, NY, USA, pp. 547–556. KDD ‘09. vical spinal cord structure using axial diffusion tensor imaging. NeuroImage 16 (1), https://doi.org/10.1145/1557019.1557082. 93–102 (Elsevier). Liu, Min, Chen, Zhang, Beaulieu, Christian, Gross, Donald W., 2014. Disrupted anatomic Wieser, H.G., Blume, W.T., Fish, D., Goldensohn, E., Hufnagel, A., King, D., Sperling, white matter network in left mesial temporal lobe epilepsy. Epilepsia 55 (5), 674–682 M.R., Lüders, H., Pedley, Timothy A., 2001. Proposal for a new classification of (Wiley Online Library). outcome with respect to epileptic seizures following epilepsy surgery. Epilepsia 42 Martin, R.C., Sawrie, S.M., Edwards, R., Roth, D.L., Faught, E., Kuzniecky, R.I., Morawetz, (2), 282–286 (Wiley Online Library). R.B., Gilliam, F.G., 2000. Investigation of executive function change following Winston, G.P., Daga, P., Stretton, J., Modat, M., Symms, M.R., McEvoy, A.W., Ourselin, S., anterior temporal lobectomy: selective normalization of verbal fluency. Duncan, J.S., 2012. Optic radiation tractography and vision in anterior temporal lobe Neuropsychology 14 (4), 501. resection. Ann. Neurol. 71 (3), 334–341. Maslov, Sergei, Sneppen, Kim, 2002. Specificity and stability in topology of protein Winston, Gavin P., Stretton, Jason, Sidhu, Meneka K., Symms, Mark R., Duncan, John S., networks. Science 296 (5569), 910–913 (American Association for the Advancement 2014. Progressive white matter changes following anterior temporal lobe resection of Science). for epilepsy. Neuroimaging Clin. N. Am. 4, 190–200 (Elsevier). Memarian, Negar, Kim, Sally, Dewar, Sandra, Engel, Jerome, Staba, Richard J., 2015. Wirsich, Jonathan, Perry, Alistair, Ridley, Ben, Proix, Timothée, Golos, Mathieu, Bénar, Multimodal data and machine learning for surgery outcome prediction in compli- Christian, Ranjeva, Jean-Philippe, et al., 2016. Whole-brain analytic measures of cated cases of mesial temporal lobe epilepsy. Comput. Biol. Med. 64, 67–78 network communication reveal increased structure-function correlation in right (Elsevier). temporal lobe epilepsy. Neuroimaging Clin. N. Am. 11, 707–718 (Elsevier). Morgan, Victoria L., Englot, Dario J., Rogers, Baxter P., Landman, Bennett A., Cakir, Wu, Xiaoyun, Rohini, Srihari, 2004. Incorporating prior knowledge with weighted margin Ahmet, Abou-Khalil, Bassel W., Anderson, Adam W., 2017. Magnetic resonance support vector machines. In: Proceedings of the Tenth ACM SIGKDD International imaging connectivity for the prediction of seizure outcome in temporal lobe epilepsy. Conference on Knowledge Discovery and Data Mining. ACM. Epilepsia 58 (7), 1251–1260. Wyler, Allen R., Hermann, Bruce P., Somes, Grant, 1995. Extent of medial temporal re- Munsell, Brent C., Wee, Chong-Yaw, Keller, Simon S., Weber, Bernd, Elger, Christian, da section on outcome from anterior temporal lobectomy: a randomized prospective Silva, Laura Angelica Tomaz, Nesland, Travis, Styner, Martin, Shen, Dinggang, study. Neurosurgery 37 (5), 982–991 (LWW). Bonilha, Leonardo, 2015. Evaluation of machine learning algorithms for treatment Xia, Mingrui, Wang, Jinhui, He, Yong, 2013. BrainNet viewer: a network visualization outcome prediction in patients with epilepsy based on structural connectome data. tool for human brain connectomics. PLoS One 8 (7), e68910 (Public Library of NeuroImage 118, 219–230. http://dx.doi.org/10.1016/j.neuroimage.2015.06.008. Science). Osuna, Edgar, Freund, Robert, Girosi, Federico, 1997. Support vector machines: training Yeh, Fang-Cheng, Wedeen, Van Jay, Tseng, W.-Y.I., 2010. Generalized-sampling imaging. and applications. Libr. Technol. Rep. AIM-1602. Med. Imaging IEEE Trans. 29 (9), 1626–1635 (IEEE). Pustina, Dorian, Doucet, Gaelle, Evans, James, Sharan, Ashwini, Sperling, Michael, Yeh, Fang-Cheng, Verstynen, Timothy D., Wang, Yibao, Fernández-Miranda, Juan C., Skidmore, Christopher, Tracy, Joseph, 2014. Distinct types of white matter changes Tseng, Wen-Yih Isaac, 2013. Deterministic diffusion fiber tracking improved by are observed after anterior temporal lobectomy in epilepsy. PLoS One 9 (8), e104211. quantitative anisotropy. PLoS One 8 (11), e80713 (Public Library of Science). Rodionov, Roman, Vollmar, Christian, Nowell, Mark, Miserocchi, Anna, Wehner, Tim, Yeo, B.T. Thomas, Krienen, Fenna M., Sepulcre, Jorge, Sabuncu, Mert R., Lashkari, Micallef, Caroline, Zombori, Gergely, et al., 2013. Feasibility of multimodal 3d Danial, Hollinshead, Marisa, Roffman, Joshua L., et al., 2011. The organization of the neuroimaging to guide implantation of intracranial eeg electrodes. Epilepsy Res. 107 human cerebral cortex estimated by intrinsic functional connectivity. J. (1), 91–100 (Elsevier). Neurophysiol. 106 (3), 1125–1165. Rubinov, Mikail, Sporns, Olaf, 2010. Complex network measures of brain connectivity: Yogarajah, M., Focke, N.K., Bonelli, S.B., Thompson, P., Vollmar, C., McEvoy, A.W., uses and interpretations. NeuroImage 52 (3), 1059–1069 (Elsevier). Alexander, D.C., Symms, M.R., Koepp, M.J., Duncan, J.S., 2010. The structural de Salvo, Matthew N., Douw, Linda, Tanaka, Naoaki, Reinsberger, Claus, Stufflebeam, plasticity of white matter networks following anterior temporal lobe resection. Brain Steven M., 2014. Altered structural connectome in temporal lobe epilepsy. Radiology 133 (8), 2348–2364. 270 (3), 842–848. http://dx.doi.org/10.1148/radiol.13131044. Zalesky, Andrew, Fornito, Alex, Cocchi, Luca, Gollo, Leonardo L., van den Heuvel, Martijn Shamim, Sadat, Wiggs, Edythe, Heiss, John, Sato, Susumu, Liew, Clarissa, Solomon, P., Breakspear, Michael, 2016. Connectome sensitivity or specificity: which is more Jeffrey, Theodore, William H., 2009. Temporal lobectomy: resection volume, neu- important? NeuroImage 142, 407–420. ropsychological effects, and seizure outcome. Epilepsy Behav. 16 (2), 311–314 Zhang, Z., Liao, W., Chen, H., Mantini, D., Ding, J.R., Xu, Q., Wang, Z., et al., 2011. (Elsevier). Altered functional–structural coupling of large-scale brain networks in idiopathic Sinha, N., Dauwels, J., Kaiser, M., Cash, S.S., Brandon Westover, M., Wang, Y., Taylor, generalized epilepsy. Brain 134 (10), 2912–2928 (Oxford Univ Press). P.N., 2017. Predicting neurosurgical outcomes in focal epilepsy patients using com- Zou, Hui, Hastie, Trevor, 2005. Regularization and variable selection via the elastic net. J. putational modelling. Brain 140 (2), 319–332. R. Stat. Soc. Ser. B (Stat Methodol.) 67 (2), 301–320. (Royal Statistical Society, Stretton, J., Thompson, P.J., 2012. Frontal lobe function in temporal lobe epilepsy. Wiley). http://www.jstor.org/stable/3647580. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Quantitative Biology arXiv (Cornell University)

The impact of epilepsy surgery on the structural connectome and its relation to outcome

Loading next page...
 
/lp/arxiv-cornell-university/the-impact-of-epilepsy-surgery-on-the-structural-connectome-and-its-0iI9fy87RU
ISSN
2213-1582
eISSN
ARCH-3345
DOI
10.1016/j.nicl.2018.01.028
Publisher site
See Article on Publisher Site

Abstract

NeuroImage: Clinical 18 (2018) 202–214 Contents lists available at ScienceDirect NeuroImage: Clinical journal homepage: www.elsevier.com/locate/ynicl The impact of epilepsy surgery on the structural connectome and its relation to outcome a,b,c, a,b a,b,c d,e c Peter N. Taylor , Nishant Sinha , Yujiang Wang , Sjoerd B. Vos , Jane de Tisi , c c c,e,1 c,e,1 Anna Miserocchi , Andrew W. McEvoy , Gavin P. Winston , John S. Duncan Interdisciplinary Computing and Complex BioSystems Group, School of Computing Science, Newcastle University, UK Institute of Neuroscience, Faculty of Medical Science, Newcastle University, UK NIHR University College London Hospitals Biomedical Research Centre, UCL Institute of Neurology, Queen Square, London WC1N 3BG, UK Translational Imaging Group, Centre for Medical Image Computing, University College London, UK Chalfont Centre for Epilepsy, Chalfont St Peter SL9 0LR, UK ARTICLE I NFO ABSTRACT Keywords: Background: Temporal lobe surgical resection brings seizure remission in up to 80% of patients, with long-term Connectome complete seizure freedom in 41%. However, it is unclear how surgery impacts on the structural white matter Network network, and how the network changes relate to seizure outcome. Temporal lobe epilepsy Methods: We used white matter fibre tractography on preoperative diffusion MRI to generate a structural white Surgery matter network, and postoperative T1-weighted MRI to retrospectively infer the impact of surgical resection on Machine learning this network. We then applied graph theory and machine learning to investigate the properties of change be- Support vector machine (SVM) tween the preoperative and predicted postoperative networks. Results: Temporal lobe surgery had a modest impact on global network efficiency, despite the disruption caused. This was due to alternative shortest paths in the network leading to widespread increases in betweenness centrality post-surgery. Measurements of network change could retrospectively predict seizure outcomes with 79% accuracy and 65% specificity, which is twice as high as the empirical distribution. Fifteen connections which changed due to surgery were identified as useful for prediction of outcome, eight of which connected to the ipsilateral temporal pole. Conclusion: Our results suggest that the use of network change metrics may have clinical value for predicting seizure outcome. This approach could be used to prospectively predict outcomes given a suggested resection mask using preoperative data only. 1. Introduction subjected to quantitative analysis techniques, which measure local and global properties in networks (see Bernhardt et al. (2015) for review). Epilepsy is a serious neurological disorder characterised by re- Network measures that have been found to be altered in temporal lobe current unprovoked seizures affecting 1% of the population. epilepsy (TLE) include the clustering coefficient of a region, which Neurosurgical resection can bring remission in up to 80% of those with captures the connectedness of neighbours of a region (Bernhardt et al., refractory focal epilepsy, with 41% remaining entirely seizure free for 2011). Furthermore, the strength of a connection (e.g. the number of years (De Tisi et al., 2011). The most common type of epilepsy surgery streamlines connecting two areas), or the strength of a region's con- is anterior temporal lobe resection, in which the amygdala, anterior nectivity (e.g. the number of streamlines connecting a region to all hippocampus, and anterior temporal neocortex are removed. The other regions) may also be altered in TLE (Besson et al., 2014a, 2014b; commonest neurological sequelae of temporal lobe surgery are memory Taylor et al., 2015). Another measure of a network is its efficiency, impairment, visual field deficits and word-finding difficulties (Jutila which is a measure of network integration - i.e. how easy it is to travel et al., 2002; Gooneratne et al., 2017). between one region to another via direct and indirect paths, and has Recent studies have investigated surgical outcome by considering been shown to be altered in patients with TLE (Liu et al., 2014). Finally, the brain as a network of connected regions. Such networks can then be regression analysis and machine learning approaches have also been Corresponding author at: Interdisciplinary Computing and Complex BioSystems Group, School of Computing Science, Newcastle University, UK. E-mail address: peter.taylor@newcastle.ac.uk (P.N. Taylor). Equal contribution. https://doi.org/10.1016/j.nicl.2018.01.028 Received 26 July 2017; Received in revised form 5 December 2017; Accepted 21 January 2018 2213-1582/ © 2018 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/). P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 applied to brain networks of TLE to relate them to surgical outcome Table 1 Patient demographics and relation to outcome group. (Bonilha et al., 2013; Munsell et al., 2015; Bonilha et al., 2015; Ji et al., 2015). ILAE 1 ILAE 2–6 Significance A challenge in comparing networks across subjects is the choice of an appropriate baseline or benchmark. There are two common ap- N 36 (68%) 17 (32%) proaches to this. One is to threshold the connectivity so that all subjects Male/female 16/20 4/13 p = 0.3597, have the same number of connections. A range of thresholds are then χ =0.839 checked, and the most significant results reported across thresholds Left/right TLE 22/14 8/9 p = 0.3353, (Zhang et al., 2011). This has the drawback of removing ‘weak’ but χ =0.923 Age (mean, S.D./median, 37, 11.6/ 41.5, 10.6/ p = 0.2374 potentially important connections. A second approach is to compare the I.Q.R.) 39.6, 19.25 42.3, 10.8 network to a random network with the same number of regions and Hippocampal sclerosis 25 (69%) 10 (59%) p = 0.4460, connections. Typically this is done by either rewiring the existing net- χ =0.5808 work (Maslov and Sneppen, 2002), or by generating a new network according to predefined rules (Betzel et al., 2016; Bauer and Kaiser, 2017). Many different types of baseline networks can be used and this Electric, Waukesha, Milwaukee, WI). Standard imaging gradients with a −1 −1 −1 will therefore influence results. maximum strength of 40mT m and slew rate 150 T m s were Recently Kuceyeski et al. (2013) introduced the network modifica- used. All data were acquired using a body coil for transmission, and 8- tion (NeMo) tool in the context of stroke (Kuceyeski et al., 2015a, channel phased array coil for reception. Standard clinical sequences 2015b) and multiple sclerosis (Kuceyeski et al., 2015a, 2015b). The were performed including a coronal T1-weighted volumetric acquisi- NeMo tool is a method to enable a direct comparison between networks tion with 170 contiguous 1.1 mm-thick slices (matrix, 256 × 256; in- that undergo change. For example, in their study of stroke, the authors plane resolution, 0.9375 × 0.9375 mm). drew masks over stroke affected areas and overlaid this mask with data Diffusion MRI data were acquired using a cardiac-triggered single- from healthy subjects. Normal connectivity from the healthy subjects, shot spin-echo planar imaging sequence (Wheeler-Kingshott et al., and altered connectivity (i.e. tracts which pass through the stroke 2002) with echo time = 73 ms. Sets of 60 contiguous 2.4 mm-thick mask) were calculated. This approach allowed the authors to calculate a axial slices were obtained covering the whole brain, with diffusion change in connectivity metric (ChaCo), which was shown to correlate sensitizing gradients applied in each of 52 noncollinear directions (b 2 −1 with outcomes. Since the authors use the pre-stroke network as a value of 1,200mm s [δ = 21 ms, Δ = 29 ms, using full gradient −1 baseline to investigate the implied post-stroke differences, the analysis strength of 40 mT m ]) along with 6 non-diffusion weighted scans. is possible without the need to generate random networks or threshold The gradient directions were calculated and ordered as described the connectivities. This obviates the need for arbitrarily chosen surro- elsewhere (Cook et al., 2007). The field of view was 24 cm, and the gate networks by effectively using the patient's own network as the acquisition matrix size was 96 × 96, zero filled to 128 × 128 during surrogate instead – a distinct advantage of the technique. A drawback of reconstruction, giving a reconstructed voxel size of that study is that the tractography was derived from a cohort of healthy 1.875 × 1.875 × 2.4 mm. The DTI acquisition time for a total of 3480 controls, rather than the stroke patients. Nonetheless, this framework is image slices was approximately 25 min (depending on subject heart ideally suited to investigate changes in networks, given a well-defined rate). alteration such as a stroke or surgery. In this study we used a ChaCo-like approach in the context of epi- 2.2. Image processing lepsy surgery and addressed the following questions: What is the impact of surgery on the patient's network? How does this impact graph the- 2.2.1. T1 processing oretic properties such as region strength, network efficiency? Do these Preoperative anatomical MRI was used to generate parcellated re- changes to patient networks correlate with surgical outcome? gions of interest (network nodes: ROIs). We used two different ap- Although the resection masks we use in this study are derived ret- proaches to do this, generating two different parcellation schemes. rospectively from postoperative data, our methods could in future be First, we used the FreeSurfer recon-all pipeline (https://surfer.nmr. applied preoperatively using a mask of the intended resection. mgh.harvard.edu/), which performs intensity normalization, skull stripping, subcortical volume generation, gray/white segmentation, 2. Materials and methods and parcellation (Fischl, 2012). The default parcellation scheme from FreeSurfer (the Desikan-Killiany atlas (Fischl et al., 2002; Desikan et al., 2.1. Patients & MRI acquisition 2006)) contains 82 cortical ROIs and subcortical ROIs and is widely used in the literature (e.g. Munsell et al., 2015; Taylor et al., 2015). The We retrospectively studied 53 patients who underwent temporal method FreeSurfer uses to generate its ROIs uses anatomical priors lobe epilepsy surgery at the National Hospital for Neurology and based on a manually annotated dataset from healthy controls. However, Neurosurgery, London, United Kingdom. Full patient details can be this may be suboptimal in the case of disease and therefore, we use a found in Table S11, a summary is given in Table 1. Patient outcomes second approach based on geodesic information flow (GIF) to generate were defined at 12 months postoperatively, according to the ILAE ROIs which has the advantage of performing well even in the presence classification of surgical outcomes (Wieser et al., 2001) and separated of neuropathology (Cardoso et al., 2015). Using the GIF approach, we into two groups. Group 1 includes patients who were completely sei- generate 114 cortical and subcortical ROIs (Table 2). A drawback of zure free (ILAE 1), and group 2 incorporates all other possibilities (ILAE using the GIF approach is comparison to previous studies is less 2–6). No patient had any prior history of neurosurgery. We used a χ straightforward since most previous work use alternative atlases. The test to check for differences between outcome groups in gender, side of results presented in the main manuscript use the GIF derived ROIs, surgery, and evidence of hippocampal sclerosis. We applied Kruskal- while we include results using FreeSurfer derived ROIs in supplemen- Wallis test to check for differences in age between outcome groups. tary materials to aid comparison to previous studies. All patients underwent preoperative anatomical T1-weighted MRI and preoperative diffusion MRI. Postoperative T1-weighted MRI was obtained within 12 months of surgery with the exception of one patient, 2.2.2. DWI processing who was rescanned later. Preoperative diffusion MRI data were first corrected for signal drift MRI studies were performed on a 3T GE Signa HDx scanner (General (Vos et al., 2016), then eddy current and movement artefacts were 203 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Table 2 MRI to the preoperative MRI using the FSL FLIRT tool (Jenkinson and Full names of the abbreviated regions of interest (ROIs). Smith, 2001; Jenkinson et al., 2002). Resection masks were then manually drawn using the FslView software by overlaying the post- Full name Abbreviation Full name Abbreviation operative MRI with the preoperative MRI starting at the most anterior Anterior cingulate ACgG Occipital pole OCP coronal slice, then proceeding posteriorly every three slices. Attention gyrus was given to ensure masks did not extend beyond the Sylvian fissure Anterior insula AIns Occipital fusiform gyrus OFuG into inferior frontal lobe since this is unrealistic for anterior temporal Anterior orbital AOrG Opercular part of the OpIFG resection and there was no evidence for this on post-operative MRI. gyrus inferior frontal gyrus Once complete, coronal slices were then joined by drawing in every Angular gyrus AnG Orbital part of the inferior OrIFG frontal gyrus sagittal slice. Masks were saved in the preoperative T1 space. Linear Calcarine cortex Calc Posterior cingulate gyrus PCgG registration is justified in this instance because nonlinear deformation Central operculum CO Precuneus PCu of the tissue around the resection site may lead to misleading bound- Cuneus Cun Parahippocampal gyrus PHG aries due to local deformations induced as part of the processing. All Entorhinal area Ent Posterior insula PIns Frontal operculum FO Parietal operculum PO registrations were visually inspected in detail during the drawing of the Frontal pole FRP Postcentral gyrus PoG resection mask. Fusiform gyrus FuG Posterior orbital gyrus POrG Gyrus rectus GRe Planum polare PP 2.3. Network generation Inferior occipital IOG Precentral gyrus PrG gyrus Inferior temporal ITG Planum temporale PT To align the tracts with the ROIs we linearly registered the pre- gyrus operative T1 image to the first non-diffusion-weighted image and saved Lingual gyrus LiG Subcallosal area SCA the transformation matrix using FSL FLIRT. We then multiplied every Lateral orbital gyrus LOrG Superior frontal gyrus SFG coordinate in every tract by the inverse of this transformation matrix to Middle cingulate MCgG Supplementary motor SMC gyrus cortex get the tracts in T1 space. The quality of the registration between tracts, Medial frontal cortex MFC Supramarginal gyrus SMG ROIs, and the resection mask was confirmed through visual inspection Middle frontal gyrus MFG Superior occipital gyrus SOG for all subjects. Since networks are constructed in native space, this Middle occipital MOG Superior parietal lobule SPL removes any mismatching of track types due to potential nonlinear gyrus registration issues which is advantageous compared to previous studies Medial orbital gyrus MOrG Superior temporal gyrus STG Postcentral gyrus MPoG Temporal pole TMP of network change (Kuceyeski et al., 2016). To generate preoperative medial segment connectivity matrices, we looped through all tracts and deemed two Precentral gyrus MPrG Triangular part of the TrIFG regions as connected if the two endpoints of the tract terminate in those medial segment inferior frontal gyrus regions. This generated a weighted connectivity (adjacency) matrix in Superior frontal MSFG Transverse temporal gyrus TTG which each entry in the matrix represents the number of streamlines gyrus medial segment connecting two regions. To generate predicted postoperative con- Middle temporal MTG nectivity matrices we performed the same process as above with one gyrus exception. Any tract that had any point within the resection mask was excluded from building the matrix. The inferred postoperative network therefore always had fewer streamlines than the preoperative network corrected using the FSL eddy_correct tool (Andersson and Sotiropoulos, and makes the assumption that remaining portions of removed tracts 2016). The b vectors were then rotated appropriately using the ‘fdt- subserve no functionality. This reduction in streamlines alone did not rotate-bvecs’ tool as part of FSL (Jenkinson et al., 2012; Leemans and explain outcome (Fig. S1). For analysis we applied a log transfor- Jones, 2009). The diffusion data were reconstructed using generalized mation to connection weights. Following this, right hemisphere regions q-sampling imaging (Yeh et al., 2010) with a diffusion sampling length (rows and columns in the matrix) were flipped in patients with a right ratio of 1.2. A deterministic fibre tracking algorithm (Yeh et al., 2013) sided resection as we investigated ipsi- and contralateral differences. All was then used, allowing for crossing fibres within voxels, with seeds subjects had 72 regions per hemisphere. placed at the whole brain. Probabilistic approaches to fibre tracking A summary of the processing pipeline to generate the GIF con- (e.g. Tournier et al., 2012) have been shown to perform less well with nectivity matrices is shown in Fig. 1. respect to accuracy and number of false positive connections inferred. The choice of tractography algorithm is therefore related to whether 2.4. Visualisation sensitivity or specificity is more important. It was recently shown (Zalesky et al., 2016) that the introduction of false positive connections To visualise the resection mask we first linearly registered the pre- is substantially more detrimental to the calculation of graph theoretic operative T1 to the MNI brain template using FSL FLIRT to generate a measures such as efficiency and clustering than the introduction of false transformation matrix. Next, we nonlinearly registered the preoperative negatives. We therefore use deterministic tractography since this has T1-weighted image to the MNI template using FSL FNIRT initialised been shown to have fewer false positive connections . Default tracto- with the aforementioned transformation to generate a nonlinear warp. graphy parameters from the 14 February 2017 build of DSI studio Finally, we applied this warp to both the preoperative T1-weighted software were used as follows. The angular threshold used was 60 de- image, and mask. All images were visually inspected to check for re- grees and the step size was set to 0.9375 mm. The anisotropy threshold gistration quality. We repeated this for all patients to generate a mask in was determined automatically by DSI Studio. Tracks with length < 10 MNI space. These masks were then loaded into MatLab, binarised mm and > 300 mm were discarded. A total of 1,000,000 tracts were (thresholded at 0.5), and summed across all subjects. This was then calculated per subject and saved in diffusion space. saved, and overlaid with the MNI brain using FSLView. Masks in MNI space were generated for visualisation purposes only and not used in 2.2.3. Resection mask the analysis. To draw resection masks we linearly registered the postoperative T1 Three-dimensional projections of brain regions were produced using the centre of mass for each region and visualised using BrainNet Viewer (Xia et al., 2013). We created the scatter plots using the UniVarScatter function http://www.tractometer.org/downloads/downloads/ismrm_presentation/Challenge_ ResultTractometer_updated_for_pdf_generation.pdf. in Matlab (https://github.com/GRousselet/matlab_visualisation). 204 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 1. Processing pipeline summary for GIF matrix generation. The pipeline is applied to each subject. 2.5. Network analysis ((−+yc xa )) i i min ∑we log(1++ ) xx+ λ i=1 1 (1) We investigated properties of the networks using well established graph theory measures that assessed the properties of a region, the where, y =(y ,y ,⋯,y ) is the n-dimensional vector representing the 1 2 n properties of a connection, and the properties of the network as a whole surgical outcomes (+1 for seizure-free outcome and −1 for nonseizure- (i.e. global). We excluded self-connections from all measures while n×m free outcome); a denotes the i− th row of feature vector A∈ ℝ ; w i i applying the weighted and undirected versions of the functions in the is the weight for the i− th sample (all samples were equally weighted Brain Connectivity Toolbox (Rubinov and Sporns, 2010). We in- in our implementation); c is the scalar intercept; λ∥ x∥ is the l reg- 1 1 vestigated measures of strength, regional communicability (sum of ularisation (sparsity) term; and x is the l regularisation (smooth- communicability matrix Estrada and Hatano, 2008), betweenness cen- ness) term. We optimised the cost function in Eq. 1 by applying the trality, clustering, and global efficiency, since several of these have LogisticR method in the sparse learning with efficient projections been suggested to be altered in previous studies (Bernhardt et al., 2011; (SLEP) software package (Liu et al., 2009a, 2009b; Liu and Ye, 2009; Besson et al., 2014a, 2014b; Taylor et al., 2015; Liu et al., 2014). Liu and Ye, 2009; Duchi et al., 2008). Region volume, region strength, connection strength changes were The weight vector x computed from Eq. 1 is a sparse representation computed by dividing the post-operative network property by the pre- of the training data set, i.e., the zero weights indicate the features operative network property. For example, if node strength reduces from which are not associated with surgical outcome, whereas the non-zero 10 pre-operatively to 2 in the inferred post-operative network the weights indicate the features associated with the surgical outcome. We change measurement will be 0.2. Other measures (betweenness, clus- derived a binary mask by setting all the non-zero weights in x to 1. tering) can increase or decrease after surgery and so the difference was Finally, we applied this mask to the high-dimensional feature space computed by subtracting the pre-operative value from the post-opera- while selecting only the features associated with surgical outcome, re- tive value. sulting in a low-dimensional feature space. In the second step of our classification framework, we incorporated the aforementioned reduced feature representation in a support vector 2.6. Elastic net regularisation and support vector machine classification machine (SVM) classifier with a linear kernel. We designed the SVM framework classifier in MATLAB using the ‘fitcsvm’ class. Specifically, we divided our data samples into test and training sets, incorporated a leave-one- We implemented a machine learning framework to investigate the out cross-validation scheme, and computed the average accuracy, sen- association of metrics indicating changes in the region strength, region sitivity, and specificity. To prevent the classifier from evaluating volumes, and connection strength with seizure free (ILAE 1) and non- skewed classification performance due to the between-class im- seizure free (ILAE 2–6) outcomes. We computed the aforementioned balance—36 in the seizure-free class as opposed to 17 in the non-sei- metrics and treated them as feature vectors. Specifically, the feature zure-free class—we implemented a class-weighted SVM by setting all space included 380, 85, and 14 features computed from proportional class prior probabilities as uniform similar to previous studies (e.g. reduction in the connection strength, region strength, and region vo- Wang et al., 2012). This sets a higher misclassification penalty for the lumes respectively for each subject. This resulted in a high-dimensional n×m minority class as compared to the majority class, thereby preventing the feature matrix, A ∈ ℝ , where m = 479 indicates the total number of classifier from learning simply from the underlying empirical data features and n = 53 denotes the total number of subjects in our study. distribution (Osuna et al., 1997; Veropoulos et al., 1999; Wu and Since the number of features, m, is much larger than the number of Rohini, 2004). subjects, n, we implemented a two-step classification paradigm Since our objective was to determine a minimum set of features that (Munsell et al., 2015; Casanova et al., 2011). In the first step, we per- would most accurately separate the seizure-free outcome and the formed feature selection by implementing logistic regression with nonseizure-free outcome classes, we incorporated a leave-one-out cross- elastic net regularisation (Zou and Hastie, 2005). This gives a sparse m- validated grid search on the l and l regularisation parameters (λ,ρ) dimensional weight vector, x ∈ ℝ , which minimises the following 1 2 (Munsell et al., 2015; Casanova et al., 2011). We varied the λ parameter regularised logistic regression problem: 205 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 2. Resection mask and region volume changes after surgery. (a) Region mask in MNI space for visualisation only (all other analyses were performed in preoperative T1 space) shows similar spatial profile between outcome groups. (b) Absolute amount of tissue resected does not relate with outcome. (c) Percentage of whole brain removed, does not relate to outcome. (d) Proportion of tissue remaining after surgery varies between subjects and regions. Here, 1 represents that the region remained intact, 0 represents that region was removed entirely. Full region names can be found in Table 2 Each dot per region is a single subject (53 per region in total). sequentially from 0.01 to 0.96 in steps of 0.05 and the ρ parameter from The sum of all streamlines directly connected to a region is termed the 0.01 to 1.96 in steps of 0.05. At each grid point, we incorporated the region strength - i.e. how strongly a region is directly connected to all machine learning framework detailed above, obtained a reduced fea- other regions. Fig. 3 shows the impact of removing all streamlines that ture space, and computed the performance metrics indicating the ac- pass into the resected tissue by plotting the resultant change in strength. curacy, sensitivity, and specificity. As expected, those regions that are partially removed (i.e. those showing a reduction in Fig. 2d) also have reduced strengths. The effect of surgery on other brain regions can be demonstrated with this ap- 3. Results proach. For example, the ipsilateral basal forebrain, although it is ex- tratemporal and not resected, has a substantial reduction in strength for Firstly, we investigate the inferred impact of surgery on network many patients. This suggests that preoperatively the basal forebrain has regions and secondly, on network connections. Finally, using machine many of its connections with the resected tissue, rather than other learning we investigate how network change metrics relate to outcomes areas. Several other areas beyond the resected regions also undergo from surgery. alterations in strength in many subjects including the ipsilateral lateral, medial and posterior, orbital gyri among others. These results are 3.1. Impact of surgery on network regions shown in detail in Table S12(a, b), and repeated for the FreeSurfer regions in Fig. S4. It is worth noting that region strength changes shown 3.1.1. Region volumes here may not be limited to only regions partly, or wholly resected, but All patients had similar surgical resections with the most variation may also change if a portion of a tract travels through the resection site, between subjects on the boundary of the resection site (Fig. 2a). The effectively disconnecting two non-resected areas. absolute (Fig. 2b), and relative (Fig. 2c) volume of the resected tissue Broadly speaking, the strengths of contralateral regions were not as did vary between subjects, but did not explain outcome (p = 0.56 and substantially affected by the surgery which suggests most connections p = 0.62 respectively – permutation test for difference between mean, are intrahemispheric. None of the changes in node strength met sig- 10,000 permutations used). Fig. 2d shows the proportion of tissue re- nificance for association with seizure-free outcome or not. maining after resection in different brain regions. Unsurprisingly the regions which were most reduced in volume were amygdala, hippo- campus, entorhinal cortex, parahippocampal gyrus, and the temporal 3.1.3. Region network measures pole. This suggests partial disruption to multiple regions, rather than The region volume change (Fig. 2) captures information solely complete disruption to specific regions. Although there is variation about individual regions, while the region strength change (Fig. 3) between subjects in the proportion of each region removed, this did not captures information about a region and its directly connected neigh- predict outcome (Fig. S2). These results are reproduced for the Free- bours. Other measures of network regions capture information re- Surfer atlas in Fig. S3. garding aspects of their role in the wider network. The betweenness centrality of a region represents how many of the 3.1.2. Region strength shortest paths between other regions traverse the region. Hub regions Each brain region is connected to other regions with streamlines. tend to occur in many of the shortest paths and have been implicated in 206 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 3. Changes in region strength are predominantly ipsilateral. (a) Proportion of strength remaining after surgery. A value of 1 (0) indicates all streamlines connecting to that region are kept (removed). (b) An example region shows ipsilateral hippocampal connection change in the ILAE > 1 group, and the ILAE 1 group. (c) Substantially affected regions visualised in 3D MNI space. Size and colour correspond to median reduction across all patients. Colour bars in panel (c) and (a) are consistent and correspond to the proportion of strength remaining between 0 and 1. a wide range of neurological disorders (Crossley et al., 2014). If a region measure beyond the resection site following surgery (Fig. S6), which is surgically removed, the shortest path between regions may be via did not relate to outcome. Reductions of regional network communic- other areas. The betweenness centrality of a region can therefore either ability were most pronounced in ipsilateral amygdala, entorhinal, and increase or decrease following surgery. temporal pole regions but also did not relate to outcome (p > 0.05, Fig. 4 shows the change in betweenness centrality after surgery. Kruskal-Wallis test, supplementary results S7). Regions involved in the resection have reductions in betweenness centrality suggesting reduced integration of those areas with the wider 3.2. Impact of surgery on network connections network. However, if those regions are no longer occurring in shortest paths, the following question remains: which ones are? Fig. 4a shows 3.2.1. Connection strength that many regions have substantial increase in betweenness centrality. Fig. 5a shows the connections which are affected by surgery in all The redirection of shortest paths in the network following surgery ap- patients. These connections typically involve those areas which are pears to be widely distributed among many other regions, rather than resected. Panel (b) shows the median reduction in connectivity fol- just the hubs - suggesting there are many alternative pathways (i.e. lowing surgery. Connections in panel (a) are a subset of those in panel redundancy). Panel (b) in Fig. 4 shows the spatial distribution of the (b). There are many connections which are only partly (as opposed to median change across all patients. This confirms that the increases entirely) reduced in strength. Partial reduction in strength can be (shown in blue) are widespread, even in contralateral regions. considered as removing some, but not all, of the connectivity between Because of the widespread redistribution of the shortest paths, we two areas. The greatest impact was seen in connections into the ipsi- also evaluated global efficiency, which is the inverse of the average lateral posterior temporal lobe and the inferior frontal lobe, with in shortest path length of the network. Fig. 4c demonstrates the reduction some patients, connections to the ipsilateral parietal lobe and con- in global efficiency following surgery. Given the redirection of shortest tralateral temporal lobe and medial frontal and parietal lobes as would paths shown in panel (a) of the same figure, it is now easy to see why be expected given Fig. 3. the reduction on global efficiency is only < 10% in most cases. This change does not relate to seizure outcome (p = 0.49 – permutation test 3.2.2. Connection network measures for difference between mean, 10,000 permutations used). We observed Similar to the regional betweenness centrality, connection (edge) similar results when using the FreeSurfer atlas (Fig. S5). betweenness centrality captures information regarding how often a We additionally investigated the change in regional clustering connection occurs in the shortest path between other regions – i.e. how coefficient, a measure of the interconnectedness of neighbours of a frequently that connection is traversed in the shortest paths. This value region, following surgery. We found very subtle differences in this can increase or decrease following surgery, as alternative paths is taken. 207 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 4. Widespread changes in region betweenness centrality following surgery. (a) Enhancements (positive values) and reductions (negative values) in betweenness centrality following surgery. A value of 0 indicates no change. (b) Median change of betweenness centrality shows that decreases are generally constrained to the resected regions (red) whereas increases are widespread across many other regions (blue). Colour bars in panel (b) and (a) are consistent and correspond to the change in betweenness centrality (y axis in panel a). Red (blue) indicates a decrease (increase) in betweenness centrality after surgery. (c) Global network efficiency is typically reduced by < 10% following surgery, and does not predict outcome. Fig. 6 shows the median change in betweenness centrality across all 3.3. Network change association with seizure outcome subjects. Decreases in connection betweenness centrality were broadly limited to ipsilateral temporal areas whereas the increases were more As an outlook to prospective applications, we used a machine widely distributed. learning strategy to retrospectively investigate the association between Fig. 5. Connectivity is disrupted by surgery. (a) Connections which are reduced in strength following surgery in all patients. (b) Median reduction in connectivity strength across all patients. A value of 1 (0) indicates that the median change to a connection following surgery is a 0% (100%) reduction in the number of streamlines. Only values < 1 are shown and therefore shows only connections which are reduced in strength in the majority of patients (as opposed to connections which are reduced in all patients - shown in panel a). Connections in panel (b) are a superset of those in panel (a). 208 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 6. Widespread changes in connection betweenness centrality following surgery. Median decreases (increases) in connection betweenness centrality across all subjects shown in red (blue). network changes – region volume, region strength, and connection incorporated all changes in volume, region strength, and connection strength – with seizure outcome. strength and applied an elastic net algorithm to select those which were Our machine learning approach minimises the number of features most useful. Fig. 7 shows those features selected by the algorithm. that give a high accuracy of outcome classification. As input features we Fifteen features were selected, all of which were connections. This Fig. 7. Features derived with the elastic net algorithm. (a) Fifteen features which were found to be most informative using our machine learning pipeline. All are connections which are reduced in strength following surgery. Colour coding shows the difference in mean reduction between groups. Connections in red are reduced by surgery more in seizure-free patients than in not-seizure-free patients. (b) Correspondence with resected tissue. 209 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 suggests that incorporating knowledge of regional properties (strength networks inferred from diffusion MRI to investigate network properties and volume change) does not improve the prediction substantially. All with respect to surgical outcomes (Bonilha et al., 2013). This includes a fifteen connections in Fig. 7 are coloured according to the difference in multi-centre study using machine learning which is perhaps most si- mean value between groups. In other words, a negative value shows milar to our work (Munsell et al., 2015). That study analysed pre- that the mean ILAE > 1 patient has a greater reduction in connectivity operative structural networks and compared several machine learning than the mean ILAE 1 patient (and vice-versa). The greatest reduction algorithms for feature selection and classification for prediction of any group can have is 1 (i.e. a 100% reduction). Values are therefore surgical outcomes. They found that the best performing algorithm uti- bound between ± 1, and consequently not normally distributed. We lised an elastic net/support vector machine combination, and achieved therefore show the difference in mean, rather than the effect size. Eight a cross-site accuracy of 66%, with a same-site accuracy of 70%. We of the features are connections involving the temporal pole. The list of therefore chose to use broadly the same technique for classification selected features is given in Table S13 and are visualised in movie S14. here. Instead of investigating preoperative networks, however, we in- Replication of these results using normalised data, preoperative data, or vestigate the difference networks - i.e. the inferred changes brought the alternative FreeSurfer parcellation did not lead to substantial im- about by patient-specific surgery. These difference networks have led us provements in accuracy (Supplementary results S10). Those who be- to achieve a same-site accuracy of 79% in our data. came seizure free had greater reductions of connectivity into the ipsi- We have performed a retrospective analysis of connectivity change lateral superior temporal lobe and frontal lobe. using a mask drawn from postoperative data – i.e. the tissue that was Table 3 shows our findings including accuracy, sensitivity and resected. However, we envisage that in future a prospective prediction specificity when using these features. Our specificity of 64.7% reflects could be performed. This should be possible using the preoperative data the finding that of all patients that the model predicted would have a and a mask of what the surgeon intends to resect (e.g. through the not-seizure-free outcome, 64.7% actually had a not-seizure-free out- manual adaptation of an ‘average mask’ in the case of a standardised come. This is substantially higher than the empirical distribution surgery such as anterior temporal lobe resection). Indeed, in theory a (32%). prediction is possible for any given presurgical resection plan. Fig. 8 suggests how this could be incorporated into such an evaluation, with the aim to address the question of, ‘will this proposed resection render 4. Discussion the patient seizure free?’ We do not address the separate question of, ‘what is the optimal resection for this patient?’, which has to integrate In this study we have used detailed neuroanatomical information the need to avoid damage to critical structures (e.g. Winston et al., from both pre- and postoperative MRI to infer the impact of surgery on 2012). This study has developed a pipeline for answering the first of patient-specific brain networks. Our three main contributions are as those two questions, and retrospectively validated it with 79% accu- follows. racy. Approaches to investigate the second of those questions have been First, we find that the impact of surgery leads to a reduction proposed recently (Sinha et al., 2017; Goodfellow et al., 2016). of < 10% in efficiency in the majority of patients, due to a redirection of shortest paths. Second, there is no single feature which fully accounts for outcome with 100% accuracy. This means that there is likely not a 4.1. Regional change following surgery single ‘target’ to aim for with surgery in this patient group that is captured by our data. Third, we have demonstrated that machine A potentially surprising result (Fig. 2), is that we found that the learning, in conjunction with connectivity change metrics, can produce amount of resected tissue did not appear to predict outcome. This is in predictions with high accuracy, and specificity which is twice as high as contrast to the work of (Wyler et al., 1995) and (Shamim et al., 2009) the empirical distribution. This means that around two thirds of the who have shown that larger resections do correlate with seizure out- patients we predict are not seizure free, are actually not seizure free. comes. However, other studies have found conflicting results which This could help preoperatively in identifying which patients might have accord with our own findings (Hardy et al., 2003; Keller et al., 2015; lower chances of becoming seizure free and put extra consideration in Liao et al., 2016). The variations in the literature may be due to dif- offering surgery or when consenting those patients. ferent populations of left/right TLE, evidence of hippocampal sclerosis, Most previous network-based studies in epilepsy typically perform and follow-up duration. Here we have controlled for these aspects group comparisons of patients relative to controls using either struc- where possible with no significant differences between outcome groups tural networks (Kamiya et al., 2016; Taylor et al., 2015; Lemkaddem (Table 1). et al., 2014; de Salvo et al., 2014; Bonilha et al., 2012), functional Our finding that there is variation between subjects, in terms of networks (Chavez et al., 2010; Liao et al., 2010), or both (Zhang et al., proportion of regions removed, can be explained by two potential fac- 2011; Douw et al., 2015; Wirsich et al., 2016). Few have investigated tors. First, there are differences in tissue removed between subjects, and alterations in global network properties with respect to surgical out- second, there are differences in regions between subjects. The latter was come (although see He et al., 2017 and Morgan et al., 2017 for recent one of our motivations for using two different parcellations to aid examples). Bonilha and colleagues utilised preoperative structural comparison (the second of which is shown in supplementary materials). Table 3 Confusion matrix indicating performance of machine learning framework for predicting surgical outcomes. Actual surgical outcome Seizure free = 36 Not seizure free = 17 Predicted Seizure free = 37 True positive = 31 False positive = 6 outcome (Type I Error) Not seizure free = 16 False negative = 5 True negative = 11 (Type II Error) True Positive Rate, False Positive Rate or Sensitivity = 0.861 or Fall out = 0.353 Accuracy = 0.792 False Negative Rate, True Negative Rate, or Miss rate = 0.139 or Specificity = 0.647 210 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Fig. 8. Potential incorporation of pipeline for prospective evaluation. Suggestion of how our approach could be used as part of the pre-surgical evaluation. From left to right: Multimodal data is acquired and evaluated at the multi-disciplinary team meeting. If surgery is recommended a proposed surgical resection mask is drawn, then fed into the connection change pipeline. The pipeline then gives a predicted post-operative network given the patient-specific mask and patient-specific presurgical DTI & T1w MRI. The pipeline uses these predicted connectivity changes, along with the pre-existing feature set to predict seizure outcome. In this example the predicted outcome would be seizure free. Furthermore, the resolution of the parcellation (i.e. the number of re- paths, leading to increased FA and thus avoiding major cognitive def- gions that the brain network is divided into) will also affect results. icits following surgery. Here we use fairly low resolution parcellations of ≈ 100 regions in The relationship between network efficiency, shortest path lengths, contrast to recently proposed higher resolutions (Besson et al., 2014a, and function is not entirely clear. However, path transitivity, which is 2014b; Taylor et al., 2017; Besson et al., 2017). Our reasoning for this is strongly related to path length has been shown to strongly correlate to that in this study we compare regional properties between subjects; functional connectivity (Goñi et al., 2014). Network efficiency itself has thus it is important that we are confident that one region in one subject also been shown to relate to intelligence in functional networks in is the same region in another, and can be reliably reproduced. One of healthy controls (van den Heuvel et al., 2009). It remains to be seen the reasons that the FreeSurfer software is so popular for this is that the how changes to network efficiency metrics following surgery relate to authors have demonstrated confidence maps at region boundaries patient changes in brain function beyond seizure outcome. (Fischl et al., 2004). One of the drawbacks of using such low resolution In addition to widespread network changes such as those described connectivity matrices however is the inability to investigate detailed above, the network change pipeline also allows the inference of more within-area architecture as in (Taylor et al., 2017). Indeed, this may be localised differences. As shown in Fig. 3, we found decreases in node an important and productive area of future research. strength in extratemporal areas, particularly the basal forebrain in the frontal lobe. This leads us to suggest that frontal lobe function may be altered following surgery. Several studies are in agreement with this. 4.2. Network change following surgery For example, Stretton et al. (2014) found improvements in frontal lobe working memory function after surgery, while Martin et al. (2000) We suggest that our finding of only a < 10% reduction in network showed significant increases in verbal fluency following surgery. efficiency following surgery (Fig. 4c) may explain why surgery usually However, other studies have also found conflicting results (see Stretton leads to relatively few severe side effects. The reason for only a subtle and Thompson, 2012 for review). We hypothesise these improvements change in efficiency is the redirection of shortest paths via other con- are related to post-surgical alterations in connectivity, and the varia- nections as shown in Fig. 6. We speculate that the decreases in edge bility in the results may be down to patient variation (i.e. the large betweenness correspond to those connections which undergo Wallerian spread in Fig. 3). This effect is typically referred to as a process of degeneration. Furthermore, we speculate that connections with in- functional improvement once nociferous cortex has been removed. creased betweenness centrality will undergo ‘strengthening’, potentially Future studies correlating these changes in connectivity with changes in reflected by increases in FA on structural imaging postoperatively. frontal lobe function will help to elucidate these relationships, poten- Qualitative evidence to support this has been shown by (Winston et al., tially predicting not only seizure outcome, but also functional out- 2014) who found decreases in FA in tracts involving the resected tissue comes. Indeed, this has already been demonstrated in stroke where and increased FA in more distant ipsilateral superior tracts within fairly connectivity change metrics have been found to relate to functional short timescales of three months. Further support for this hypothesis outcomes (Kuceyeski et al., 2015a, 2015b). comes from a study by (Jeong et al., 2016) who report increases in the FA of connections following frontal lobe epilepsy surgery far from the surgical site. Other studies have also made similar findings (Yogarajah 4.3. Machine learning for outcome prediction et al., 2010; Pustina et al., 2014; Faber et al., 2013). Taken together, this suggests a potential rerouting of connectivity, via new shortest Although very few have used brain network data, several studies in 211 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 recent years have begun to apply machine learning strategies for sur- giving confidence in the reproducibility of the masks drawn here. Our gical outcome prediction. The benefits of the approach appear pro- choice of tractography algorithm was influenced by the intention to mising. For example, (Bernhardt et al., 2015), using hippocampal limit false positive connections as suggested by Zalesky et al. (2016). measurements as features, made outcome predictions with 81% accu- Despite this, there are known limitations with all approaches and this racy. In another study of surgical outcomes in temporal lobe epilepsy study should also be considered with the known limitations of the (Memarian et al., 2015) demonstrated accuracy of 95% using various imaging technologies in mind. For example, the inability to resolve features derived from both imaging and demographic data. In addition, connection direction and potential bias for tractography to favour (Feis et al., 2013) achieved accuracy of outcome prediction of up to shorter, straighter connections (Jones, 2010). Our use of number of 96% using preoperative T1 measurements of white matter volumes. streamlines as a connectivity metric is based on the extensive prior Other studies have also used machine learning algorithms, along with literature of ChaCo analysis by Kuceyeski et al. (2013, 2015a, 2015b). imaging data, to classify lateralisation of seizure onset (Focke et al., However, other connection strength metrics, such as density (Taylor 2012; Bennett et al., 2017). Our prediction accuracy is in a broadly et al., 2015), or volume (Irimia and Van Horn, 2016) may also yield similar range to many of these and may be improved by including other informative results. features such as those used for generating predictive nomograms (Jehi Our network change pipeline allows to infer the immediate network et al., 2015). change brought about by surgery. It is worth highlighting that this does Multiple approaches exist in machine learning to attempt to im- not necessarily represent the long-term brain network changes in the prove the accuracy of the outcome prediction by selecting the most months/years that occur as a result of e.g. plasticity or degeneration. predictive features from a high-dimensional feature space (Tibshirani, This is both a strength and a weakness of our approach and is one 1996, Hoerl and Kennard, 2004, Breiman et al., 1984, Breiman et al., reason why we assess outcomes at only 12 months. The mechanisms 1984). Here we have adopted a fairly conservative approach in which underlying why some patients relapse to seizure recurrence only at we implement elastic net regularisation (Zou and Hastie, 2005) while several years after surgery (De Tisi et al., 2011) are unclear and would incorporating regularisation parameters which identify a minimum require longitudinal diffusion MRI data gathered over several years to number of features important for the prediction. Even using this con- investigate in this framework. The potential to use our approach pro- servative approach we obtain a prediction accuracy of 79.2%. It is spectively (Fig. 8) is a distinct advantage however. worth noting that we impose a prior probability expectation of 50%. If we did not impose this, and instead used the empirical distribution of 4.5. Outlook and wider implications outcomes (68%: 32% - row 1 in Table 1), it would be easy to get ac- curacies of at least 68% by simply suggesting all patients will be seizure- Although we have focussed exclusively on temporal lobe epilepsy free. The prediction accuracy of 79.2%, even using our highly con- surgery here, this method could be generalized to other types of neu- servative approach, is therefore encouraging. rosurgery such as those for other epilepsies such as neocortical epilepsy, tumours or movement disorders. Additionally, other outcome criteria 4.4. Strengths & weaknesses could also be predicted. For example, instead of predicting seizure outcome, functional deficits arising due to damage to known functional There are several strengths to this work. For example, here we have subnetworks (e.g. somatomotor, attention, visual) could be predicted a fairly large cohort of 53 patients which were all scanned on the same using existing atlases (Yeo et al., 2011). Our approach would be pos- scanner using the same scanning protocol. We also have good con- sible if preoperative imaging data is available and a proposed resection sistency in terms of operation type (all anterior temporal lobe resection mask drawn before the surgery. Deficits for the given mask (surgery) performed by the same 2 surgeons using the same technique), and could then be predicted for individual subjects. Additionally, multiple consistency in follow-up duration for outcome - 12 months. The con- proposed masks (surgeries) could be tested against each other in-silico, sistency in follow-up duration is particularly important because out- the results for each compared, and the optimal surgery predicted. Al- comes can change over time (De Tisi et al., 2011). A further strength of gorithms for a similar approach are already being developed for in- our study, in comparison to some other network connectivity change vasive electrode recordings (Rodionov et al., 2013; Sinha et al., 2017). studies, is that our analysis is performed in subject (native) space. This means nonlinear deformations of the scans to a common space, which 4.6. Conclusion may decrease the accuracy of the region mapping, are not necessary. Our investigation of changes in structural connectivity following surgery It is to be expected that outcome from surgery will depend not only using a ChaCo-like approach is, to our knowledge, novel in epilepsy. on the network connectivity of the brain before surgery, but the dis- Moreover, the ChaCo approach has the advantage of not requiring ar- ruption to this network caused by the surgery itself. This can be mea- bitrary thresholding of the network, or comparison to arbitrarily chosen sured as the change in connectivity, which can be predicted for a given random networks as a baseline – instead the patient's own pre-operative resection preoperatively. Our analysis shows that such connectivity network is used. change metrics can be used to retrospectively predict seizure outcome A possible weakness of this study is that we have flipped the con- with high accuracy, and may relate to functional outcomes. nectivity matrices in right TLE patients, simplifying our analysis into Supplementary data to this article can be found online at https:// ipsi- and contralateral hemispheres. Although there is precedence for doi.org/10.1016/j.nicl.2018.01.028. this approach (Bonilha et al., 2013; Munsell et al., 2015; Ji et al., 2015) it has been suggested that left and right TLE are not simply the mirror Acknowledgments image of each other (Besson et al., 2014a, 2014b), especially with re- spect to post-surgical cognitive deficits and resection size which is ty- We are grateful to the Wolfson Foundation and the Epilepsy Society pically larger in the non-dominant hemisphere. We did try to mitigate for supporting the Epilepsy Society MRI scanner. this however by ensuring the proportion of left:right patients in each PNT was funded by Wellcome Trust (105617/Z/14/Z). PNT thanks outcome group are similar (p = 0.3353; Table 1). Rob Forsyth and Helen Marley for discussion. GPW was supported by Another potential bias is the resection mask, which was manually the MRC (G0802012, MR/M00841X/1). SBV is funded by the National drawn for all subjects by the same rater, with a majority subset drawn Institute for Health Research University College London Hospitals by a second rater (results S9). Variations between raters exist, but are Biomedical Research Centre (NIHR BRC UCLH/UCL High Impact small. The correlation of network features between raters is > 0.98, Initiative). 212 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 This study was carried out in part at UCLH and the support of the Estrada, E., Hatano, N., 2008. Communicability in complex networks. Phys. Rev. E 77 (3), NIHR-funded UCLH/UCL BRC is gratefully acknowledged. JD is sup- Faber, Jennifer, Schoene-Bake, Jan-Christoph, Trautner, Peter, Lehe, Marec, Elger, ported by the NIHR Senior investigator scheme. Christian E., Weber, Bernd, 2013. Progressive fiber tract affections after temporal lobe surgery. Epilepsia 54 (4), e53-e57. Feis, Delia-Lisa, Schoene-Bake, Jan-Christoph, Elger, Christian, Wagner, Jan, References Tittgemeyer, Marc, Weber, Bernd, 2013. Prediction of post-surgical seizure outcome in left mesial temporal lobe epilepsy. Neurol. Clin. 2, 903–911 (Elsevier). Andersson, Jesper L.R., Sotiropoulos, Stamatios N., 2016. An integrated approach to Fischl, Bruce, 2012. FreeSurfer. NeuroImage 62 (2), 774–781 (Elsevier). correction for off-resonance effects and subject movement in diffusion mr imaging. Fischl, Bruce, Salat, David H., Busa, Evelina, Albert, Marilyn, Dieterich, Megan, NeuroImage 125, 1063–1078 (Elsevier). Haselgrove, Christian, van der Kouwe, Andre, et al., 2002. Whole brain segmentation: Bauer, Roman, Kaiser, Marcus, 2017. Nonlinear growth: an origin of hub organization in automated labeling of neuroanatomical structures in the human brain. Neuron 33 (3), complex networks. Royal Soc. Open Sc. 4 (3), 160691 (The Royal Society). 341–355 (Elsevier). Bennett, O., Cardoso, J., Duncan, J.S., Winston, G., Ourselin, S., 2017. Learning how to Fischl, Bruce, van der Kouwe, André, Destrieux, Christophe, Halgren, Eric, Ségonne, See the Invisible - Using Machine Learning to Find Underlying Abnormality Patterns Florent, Salat, David H., Busa, Evelina, et al., 2004. Automatically parcellating the in Reportedly Normal MR Brain Images from Patients with Epilepsy. Proceedings. human cerebral cortex. Cereb. Cortex 14 (1), 11–22 (Oxford Univ Press). ISMRM. Focke, Niels K., Yogarajah, Mahinda, Symms, Mark R., Gruber, Oliver, Paulus, Walter, Bernhardt, Boris C., Chen, Zhang, He, Yong, Evans, Alan C., Bernasconi, Neda, 2011. Duncan, John S., 2012. Automated Mr image classification in temporal lobe epilepsy. Graph-theoretical analysis reveals disrupted small-world organization of cortical NeuroImage 59 (1), 356–362 (Elsevier). thickness correlation networks in temporal lobe epilepsy. Cereb. Cortex 21 (9), Goñi, J., van den Heuvel, M.P., Avena-Koenigsberger, A., Velez de Mendizabal, N., Betzel, 2147–2157 (Oxford Univ Press). R.F., Griffa, A., ... Sporns, O., 2014. Resting-brain functional connectivity predicted Bernhardt, Boris C., Hong, Seok-Jun, Bernasconi, Andrea, Bernasconi, Neda, 2015. by analytic measures of network communication. Proc. Natl. Acad. Sci. U. S. A. 111 Magnetic resonance imaging pattern learning in temporal lobe epilepsy: classification (2), 833–838. http://dx.doi.org/10.1073/pnas.1315529111. and prognostics. Ann. Neurol. 77 (3), 436–446 (Wiley Online Library). Goodfellow, M., Rummel, C., Abela, E., Richardson, M.P., Schindler, K., Terry, J.R., 2016. Besson, Pierre, Dinkelacker, Vera, Valabregue, Romain, Thivard, Lionel, Leclerc, Xavier, Estimation of brain network ictogenicity predicts outcome from epilepsy surgery. Sci. Baulac, Michel, Sammler, Daniela, et al., 2014a. Structural connectivity differences in Rep. 6. left and right temporal lobe epilepsy. NeuroImage 100, 135–144 (Elsevier). Gooneratne, I.K., Mannan, S., de Tisi, J., Gonzalez, J.C., McEvoy, A.W., Miserocchi, A., Besson, Pierre, Lopes, Renaud, Leclerc, Xavier, Derambure, Philippe, Tyvaert, Louise, Diehl, B., Wehner, T., Bell, G.S., Sander, J.W., Duncan, J.S., 2017. Somatic compli- 2014b. Intra-subject reliability of the high-resolution whole-brain structural con- cations of epilepsy surgery over 25 years at a single center. Epilepsy Res. 132, 70–77. nectome. NeuroImage 102, 283–293 (Elsevier). Hardy, Steven G., Miller, John W., Holmes, Mark D., Born, Donald E., Ojemann, George Besson, Pierre, Kathleen Bandt, S., Proix, Timothée, Lagarde, Stanislas, Jirsa, Viktor K., A., Dodrill, Carl B., Hallam, Danial K., 2003. Factors predicting outcome of surgery Ranjeva, Jean-Philippe, Bartolomei, Fabrice, Guye, Maxime, 2017. Anatomic con- for intractable epilepsy with pathologically verified mesial temporal sclerosis. sistencies across epilepsies: a stereotactic-EEG informed high-resolution structural Epilepsia 44 (4), 565–568 (Wiley Online Library). connectivity study. Brain 140 (10), 2639–2652. He, X., Doucet, G.E., Pustina, D., Sperling, M.R., Sharan, A.D., Tracy, J.I., 2017. Betzel, Richard F., Avena-Koenigsberger, Andrea, Goñi, Joaquín, He, Ye, De Reus, Marcel Presurgical thalamic “hubness” predicts surgical outcome in temporal lobe epilepsy. A., Griffa, Alessandra, Vértes, Petra E., et al., 2016. Generative models of the human Neurology 10–1212. connectome. NeuroImage 124, 1054–1064 (Elsevier). van den Heuvel, M.P., Stam, C.J., Kahn, R.S., Pol, H.E.H., 2009. Efficiency of functional Bonilha, Leonardo, Nesland, Travis, Martz, Gabriel U., Joseph, Jane E., Spampinato, brain networks and intellectual performance. J. Neurosci. 29 (23), 7619–7624. Maria V., Edwards, Jonathan C., Tabesh, Ali, 2012. Medial temporal lobe epilepsy is Hoerl, A.E., Kennard, R.W., 2004. Ridge Regression. 9780471667193 John Wiley and associated with neuronal fibre loss and paradoxical increase in structural connectivity Sons, Inc.http://dx.doi.org/10.1002/0471667196.ess2280. of limbic structures. J. Neurol. Neurosurg. Psychiatry 83 (9), 903–909 (BMJ Irimia, Andrei, Van Horn, John Darrell, 2016. Scale-dependent variability and quantita- Publishing Group Ltd). tive regimes in graph-theoretic representations of human cortical networks. Brain Bonilha, Leonardo, Helpern, Joseph A., Sainju, Rup, Nesland, Travis, Edwards, Jonathan Connect. 6 (2), 152–163. C., Glazier, Steven S., Tabesh, Ali, 2013. Presurgical connectome and postsurgical Jehi, L., Yardi, R., Chagin, K., Tassi, L., Russo, G.L., Worrell, G., ... Chauvel, P., 2015. seizure control in temporal lobe epilepsy. Neurology 81 (19), 1704–1710 (AAN Development and validation of nomograms to provide individualised predictions of Enterprises). seizure outcomes after epilepsy surgery: a retrospective analysis. Lancet Neurol. 14 Bonilha, Leonardo, Jensen, Jens H., Baker, Nathaniel, Breedlove, Jesse, Nesland, Travis, (3), 283–290. Lin, Jack J., Drane, Daniel L., Saindane, Amit M., Binder, Jeffrey R., Kuzniecky, Jenkinson, Mark, Smith, Stephen, 2001. A global optimisation method for robust affine Ruben I., 2015. The brain connectome as a personalized biomarker of seizure out- registration of brain images. Med. Image Anal. 5 (2), 143–156 (Elsevier). comes after temporal lobectomy. Neurology 84 (18), 1846–1853 (AAN Enterprises). Jenkinson, Mark, Bannister, Peter, Brady, Michael, Smith, Stephen, 2002. Improved op- Breiman, L., Friedman, J.H., Alshen, R.A., Stone, C.J., 1984. CART: Classification and timization for the robust and accurate linear registration and motion correction of Regression Trees. Wadsworth, Belmont, CA. brain images. NeuroImage 17 (2), 825–841 (Elsevier). Cardoso, M. Jorge, Modat, Marc, Wolz, Robin, Melbourne, Andrew, Cash, David, Jenkinson, Mark, Beckmann, Christian F., Behrens, Timothy E.J., Woolrich, Mark W., Rueckert, Daniel, Ourselin, Sebastien, 2015. Geodesic information flows: spatially- Smith, Stephen M., 2012. Fsl. NeuroImage 62 (2), 782–790 (Elsevier). variant graphs and their application to segmentation and fusion. IEEE Trans. Med. Jeong, Jeong-Won, Asano, Eishi, Juhász, Csaba, Behen, Michael E., Chugani, Harry T., Imaging 34 (9), 1976–1988 (IEEE). 2016. Postoperative axonal changes in the contralateral hemisphere in children with Casanova, Ramon, Wagner, Benjamin, Whitlow, Christopher, Williamson, Jeff, Shumaker, medically refractory epilepsy: a longitudinal diffusion tensor imaging connectome Sally, Maldjian, Joseph, Espeland, Mark, 2011. High dimensional classification of analysis. Hum. Brain Mapp. 37 (11), 3946–3956 (Wiley Online Library). structural Mri Alzheimer's disease data based on large scale regularization. Front. Ji, Gong-Jun, Zhang, Zhiqiang, Xu, Qiang, Wei, Wei, Wang, Jue, Wang, Zhengge, Yang, Neuroinform. 5, 22. http://dx.doi.org/10.3389/fninf.2011.00022. Fang, et al., 2015. Connectome reorganization associated with surgical outcome in Chavez, Mario, Valencia, Miguel, Navarro, Vincent, Latora, Vito, Martinerie, Jacques, temporal lobe epilepsy. Medicine 94 (40), e1737 (LWW). 2010. Functional modularity of background activities in normal and epileptic brain Jones, D.K., 2010. Challenges and limitations of quantifying brain connectivity in vivo networks. Phys. Rev. Lett. 104 (11), 118701 (APS). with diffusion MRI. Quant. Imaging Med. Surg. 2, 341–355. Cook, Philip A., Symms, Mark, Boulby, Philip A., Alexander, Daniel C., 2007. Optimal Jutila, L., Immonen, A., Mervaala, E., Partanen, J., Partanen, K., Puranen, M., Kälviäinen, acquisition orders of diffusion-weighted mri measurements. J. Magn. Reson. Imaging R., et al., 2002. Long term outcome of temporal lobe epilepsy surgery: analyses of 140 25 (5), 1051–1058 (Wiley Online Library). consecutive patients. J. Neurol. Neurosurg. Psychiatry 73 (5), 486–494 (BMJ Crossley, Nicolas A., Mechelli, Andrea, Scott, Jessica, Carletti, Francesco, Fox, Peter T., Publishing Group Ltd). McGuire, Philip, Bullmore, Edward T., 2014. The hubs of the human connectome are Kamiya, Kouhei, Amemiya, Shiori, Suzuki, Yuichi, Kunii, Naoto, Kawai, Kensuke, Akira, generally implicated in the anatomy of brain disorders. Brain 137 (8), 2382–2395 Kunimatsu, Nobuhito, Saito, Shigeki, Aoki, Kuni, Ohtomo, 2016. Machine learning of (Oxford Univ Press). Dti structural brain connectomes for lateralization of temporal lobe epilepsy. Magn. De Tisi, Jane, Bell, Gail S., Peacock, Janet L., McEvoy, Andrew W., Harkness, William F.J., Reson. Med. Sci. 15 (1), 121–129 (Japanese Society for Magnetic Resonance in Sander, Josemir W., Duncan, John S., 2011. The long-term outcome of adult epilepsy Medicine). surgery, patterns of seizure remission, and relapse: a cohort study. Lancet 378 (9800), Keller, Simon S., Richardson, Mark P., Schoene-Bake, Jan-Christoph, O'muircheartaigh, 1388–1395 (Elsevier). Jonathan, Elkommos, Samia, Kreilkamp, Barbara, Goh, Yee Yen, Marson, Anthony G., Desikan, Rahul S., Ségonne, Florent, Fischl, Bruce, Quinn, Brian T., Dickerson, Bradford Elger, Christian, Weber, Bernd, 2015. Thalamotemporal alteration and postoperative C., Blacker, Deborah, Buckner, Randy L., et al., 2006. An automated labeling system seizures in temporal lobe epilepsy. Ann. Neurol. 77 (5), 760–774 (Wiley Online for subdividing the human cerebral cortex on Mri scans into gyral based regions of Library). interest. NeuroImage 31 (3), 968–980 (Elsevier). Kuceyeski, Amy, Maruta, Jun, Relkin, Norman, Raj, Ashish, 2013. The network mod- Douw, Linda, DeSalvo, Matthew N., Tanaka, Naoaki, Cole, Andrew J., Liu, Hesheng, ification (nemo) tool: elucidating the effect of white matter integrity changes on Reinsberger, Claus, Stufflebeam, Steven M., 2015. Dissociated multimodal hubs and cortical and subcortical structural connectivity. Brain Connect. 3 (5), 451–463 (Mary seizures in temporal lobe epilepsy. Ann. Clin. Transl. Neurol. 2 (4), 338–352 (Wiley Ann Liebert, Inc. 140 Huguenot Street, 3rd Floor New Rochelle, NY 10801 USA). Online Library). Kuceyeski, A.F., Vargas, W., Dayan, M., Monohan, E., Blackwell, C., Raj, A., Fujimoto, K., Duchi, John, Shalev-Shwartz, Shai, Singer, Yoram, Chandra, Tushar, 2008. Efficient Gauthier, S.A., 2015a. Modeling the relationship among gray matter atrophy, ab- projections onto the L1-ball for learning in high dimensions. In: Proceedings of the normalities in connecting white matter, and cognitive performance in early multiple 25th International Conference on Machine Learning. ACM, New York, NY, USA, pp. sclerosis. Am. J. Neuroradiol. 36 (4), 702–709 (Am Soc Neuroradiology). 272–279. (ICML ’08). https://doi.org/10.1145/1390156.1390191. Kuceyeski, Amy, Navi, Babak B., Kamel, Hooman, Relkin, Norman, Villanueva, Mark, Raj, 213 P.N. Taylor et al. NeuroImage: Clinical 18 (2018) 202–214 Ashish, Toglia, Joan, O'dell, Michael, Iadecola, Costantino, 2015b. Exploring the Epilepsy Res. 98 (1), 1–13. brain's structural connectome: a quantitative stroke lesion-dysfunction mapping Stretton, J., Sidhu, M.K., Winston, G.P., Bartlett, P., McEvoy, A.W., Symms, M.R., Koepp, study. Hum. Brain Mapp. 36 (6), 2147–2160 (Wiley Online Library). M.J., Thompson, P.J., Duncan, J.S., 2014. Working memory network plasticity after Kuceyeski, Amy, Navi, Babak B., Kamel, Hooman, Raj, Ashish, Relkin, Norman, Toglia, anterior temporal lobe resection: a longitudinal functional magnetic resonance Joan, Iadecola, Costantino, O'dell, Michael, 2016. Structural connectome disruption imaging study. Brain 137 (5), 1439–1453. at baseline predicts 6‐months post‐stroke outcome. Hum. Brain Mapp. 37 (7), Taylor, Peter N., Han, Cheol E., Schoene-Bake, Jan-Christoph, Weber, Bernd, Kaiser, 2587–2601. Marcus, 2015. Structural connectivity changes in temporal lobe epilepsy: spatial Leemans, Alexander, Jones, Derek K., 2009. The B-matrix must be rotated when cor- features contribute more than topological measures. Neuroimaging Clin. N. Am. 8, recting for subject motion in DTI data. Magn. Reson. Med. 61 (6), 1336–1349. 322–328 (Elsevier). Lemkaddem, Alia, Daducci, Alessandro, Kunz, Nicolas, Lazeyras, François, Seeck, Taylor, Peter N., Wang, Yujiang, Kaiser, Marcus, 2017. Within brain area tractography Margitta, Thiran, Jean-Philippe, Vulliémoz, Serge, 2014. Connectivity and tissue suggests local modularity using high resolution connectomics. Sci. Rep. 7 (Nature microstructural alterations in right and left temporal lobe epilepsy revealed by dif- Publishing Group). fusion spectrum imaging. Neuroimaging Clin. N. Am. 5, 349–358 (Elsevier). Tibshirani, Robert, 1996. Regression shrinkage and selection via the lasso. J. R. Stat. Soc. Liao, Wei, Zhang, Zhiqiang, Pan, Zhengyong, Mantini, Dante, Ding, Jurong, Duan, Xujun, Ser. B Methodol. 267–288. Cheng, Luo, Lu, Guangming, Chen, Huafu, 2010. Altered functional connectivity and Tournier, J., Calamante, F., Connelly, A., 2012. MRtrix: diffusion tractography in crossing small-world in mesial temporal lobe epilepsy. PLoS One 5 (1), e8525 (Public Library fiber regions. Int. J. Imaging Syst. Technol. 22 (1), 53–66. of Science). Veropoulos, Konstantinos, Campbell, Colin, Cristianini, Nello, 1999. Controlling the Liao, Wei, Ji, Gong-Jun, Xu, Qiang, Wei, Wei, Wang, Jue, Wang, Zhengge, Yang, Fang, sensitivity of support vector machines. In: Proceedings of the International Joint et al., 2016. Functional connectome before and following temporal lobectomy in Conference on AI, pp. 55–60. mesial temporal lobe epilepsy. Sci. Rep. 6, 23153 (Nature Publishing Group). Vos, Sjoerd B., Tax, Chantal M.W., Luijten, Peter R., Ourselin, Sebastien, Leemans, Liu, Jun, Ye, Jieping, 2009. Efficient Euclidean projections in linear time. In: Proceedings Alexander, Froeling, Martijn, 2016. The importance of correcting for signal drift in of the 26th Annual International Conference on Machine Learning. ACM, New York, diffusion Mri. In: Magnetic Resonance in Medicine. Wiley Online Library. NY, USA, pp. 657–664. ICML ‘09. https://doi.org/10.1145/1553374.1553459. Wang, Zuoguan, Gunduz, Aysegul, Brunner, Peter, Ritaccio, Anthony L., Ji, Qiang, Schalk, Liu, J., Ji, S., Ye, J., 2009a. SLEP: Sparse Learning with Efficient Projections. Arizona Gerwin, 2012. Decoding onset and direction of movements using electrocortico- State University. http://www.public.asu.edu/~jye02/Software/SLEP. graphic (ECoG) signals in humans. Front. Neuroeng. 5. Liu, Jun, Chen, Jianhui, Ye, Jieping, 2009b. Large-scale sparse logistic regression. In: Wheeler-Kingshott, Claudia A.M., Hickman, Simon J., Parker, Geoffrey J.M., Ciccarelli, Proceedings of the 15th Acm Sigkdd International Conference on Knowledge Olga, Symms, Mark R., Miller, David H., Barker, Gareth J., 2002. Investigating cer- Discovery and Data Mining. ACM, New York, NY, USA, pp. 547–556. KDD ‘09. vical spinal cord structure using axial diffusion tensor imaging. NeuroImage 16 (1), https://doi.org/10.1145/1557019.1557082. 93–102 (Elsevier). Liu, Min, Chen, Zhang, Beaulieu, Christian, Gross, Donald W., 2014. Disrupted anatomic Wieser, H.G., Blume, W.T., Fish, D., Goldensohn, E., Hufnagel, A., King, D., Sperling, white matter network in left mesial temporal lobe epilepsy. Epilepsia 55 (5), 674–682 M.R., Lüders, H., Pedley, Timothy A., 2001. Proposal for a new classification of (Wiley Online Library). outcome with respect to epileptic seizures following epilepsy surgery. Epilepsia 42 Martin, R.C., Sawrie, S.M., Edwards, R., Roth, D.L., Faught, E., Kuzniecky, R.I., Morawetz, (2), 282–286 (Wiley Online Library). R.B., Gilliam, F.G., 2000. Investigation of executive function change following Winston, G.P., Daga, P., Stretton, J., Modat, M., Symms, M.R., McEvoy, A.W., Ourselin, S., anterior temporal lobectomy: selective normalization of verbal fluency. Duncan, J.S., 2012. Optic radiation tractography and vision in anterior temporal lobe Neuropsychology 14 (4), 501. resection. Ann. Neurol. 71 (3), 334–341. Maslov, Sergei, Sneppen, Kim, 2002. Specificity and stability in topology of protein Winston, Gavin P., Stretton, Jason, Sidhu, Meneka K., Symms, Mark R., Duncan, John S., networks. Science 296 (5569), 910–913 (American Association for the Advancement 2014. Progressive white matter changes following anterior temporal lobe resection of Science). for epilepsy. Neuroimaging Clin. N. Am. 4, 190–200 (Elsevier). Memarian, Negar, Kim, Sally, Dewar, Sandra, Engel, Jerome, Staba, Richard J., 2015. Wirsich, Jonathan, Perry, Alistair, Ridley, Ben, Proix, Timothée, Golos, Mathieu, Bénar, Multimodal data and machine learning for surgery outcome prediction in compli- Christian, Ranjeva, Jean-Philippe, et al., 2016. Whole-brain analytic measures of cated cases of mesial temporal lobe epilepsy. Comput. Biol. Med. 64, 67–78 network communication reveal increased structure-function correlation in right (Elsevier). temporal lobe epilepsy. Neuroimaging Clin. N. Am. 11, 707–718 (Elsevier). Morgan, Victoria L., Englot, Dario J., Rogers, Baxter P., Landman, Bennett A., Cakir, Wu, Xiaoyun, Rohini, Srihari, 2004. Incorporating prior knowledge with weighted margin Ahmet, Abou-Khalil, Bassel W., Anderson, Adam W., 2017. Magnetic resonance support vector machines. In: Proceedings of the Tenth ACM SIGKDD International imaging connectivity for the prediction of seizure outcome in temporal lobe epilepsy. Conference on Knowledge Discovery and Data Mining. ACM. Epilepsia 58 (7), 1251–1260. Wyler, Allen R., Hermann, Bruce P., Somes, Grant, 1995. Extent of medial temporal re- Munsell, Brent C., Wee, Chong-Yaw, Keller, Simon S., Weber, Bernd, Elger, Christian, da section on outcome from anterior temporal lobectomy: a randomized prospective Silva, Laura Angelica Tomaz, Nesland, Travis, Styner, Martin, Shen, Dinggang, study. Neurosurgery 37 (5), 982–991 (LWW). Bonilha, Leonardo, 2015. Evaluation of machine learning algorithms for treatment Xia, Mingrui, Wang, Jinhui, He, Yong, 2013. BrainNet viewer: a network visualization outcome prediction in patients with epilepsy based on structural connectome data. tool for human brain connectomics. PLoS One 8 (7), e68910 (Public Library of NeuroImage 118, 219–230. http://dx.doi.org/10.1016/j.neuroimage.2015.06.008. Science). Osuna, Edgar, Freund, Robert, Girosi, Federico, 1997. Support vector machines: training Yeh, Fang-Cheng, Wedeen, Van Jay, Tseng, W.-Y.I., 2010. Generalized-sampling imaging. and applications. Libr. Technol. Rep. AIM-1602. Med. Imaging IEEE Trans. 29 (9), 1626–1635 (IEEE). Pustina, Dorian, Doucet, Gaelle, Evans, James, Sharan, Ashwini, Sperling, Michael, Yeh, Fang-Cheng, Verstynen, Timothy D., Wang, Yibao, Fernández-Miranda, Juan C., Skidmore, Christopher, Tracy, Joseph, 2014. Distinct types of white matter changes Tseng, Wen-Yih Isaac, 2013. Deterministic diffusion fiber tracking improved by are observed after anterior temporal lobectomy in epilepsy. PLoS One 9 (8), e104211. quantitative anisotropy. PLoS One 8 (11), e80713 (Public Library of Science). Rodionov, Roman, Vollmar, Christian, Nowell, Mark, Miserocchi, Anna, Wehner, Tim, Yeo, B.T. Thomas, Krienen, Fenna M., Sepulcre, Jorge, Sabuncu, Mert R., Lashkari, Micallef, Caroline, Zombori, Gergely, et al., 2013. Feasibility of multimodal 3d Danial, Hollinshead, Marisa, Roffman, Joshua L., et al., 2011. The organization of the neuroimaging to guide implantation of intracranial eeg electrodes. Epilepsy Res. 107 human cerebral cortex estimated by intrinsic functional connectivity. J. (1), 91–100 (Elsevier). Neurophysiol. 106 (3), 1125–1165. Rubinov, Mikail, Sporns, Olaf, 2010. Complex network measures of brain connectivity: Yogarajah, M., Focke, N.K., Bonelli, S.B., Thompson, P., Vollmar, C., McEvoy, A.W., uses and interpretations. NeuroImage 52 (3), 1059–1069 (Elsevier). Alexander, D.C., Symms, M.R., Koepp, M.J., Duncan, J.S., 2010. The structural de Salvo, Matthew N., Douw, Linda, Tanaka, Naoaki, Reinsberger, Claus, Stufflebeam, plasticity of white matter networks following anterior temporal lobe resection. Brain Steven M., 2014. Altered structural connectome in temporal lobe epilepsy. Radiology 133 (8), 2348–2364. 270 (3), 842–848. http://dx.doi.org/10.1148/radiol.13131044. Zalesky, Andrew, Fornito, Alex, Cocchi, Luca, Gollo, Leonardo L., van den Heuvel, Martijn Shamim, Sadat, Wiggs, Edythe, Heiss, John, Sato, Susumu, Liew, Clarissa, Solomon, P., Breakspear, Michael, 2016. Connectome sensitivity or specificity: which is more Jeffrey, Theodore, William H., 2009. Temporal lobectomy: resection volume, neu- important? NeuroImage 142, 407–420. ropsychological effects, and seizure outcome. Epilepsy Behav. 16 (2), 311–314 Zhang, Z., Liao, W., Chen, H., Mantini, D., Ding, J.R., Xu, Q., Wang, Z., et al., 2011. (Elsevier). Altered functional–structural coupling of large-scale brain networks in idiopathic Sinha, N., Dauwels, J., Kaiser, M., Cash, S.S., Brandon Westover, M., Wang, Y., Taylor, generalized epilepsy. Brain 134 (10), 2912–2928 (Oxford Univ Press). P.N., 2017. Predicting neurosurgical outcomes in focal epilepsy patients using com- Zou, Hui, Hastie, Trevor, 2005. Regularization and variable selection via the elastic net. J. putational modelling. Brain 140 (2), 319–332. R. Stat. Soc. Ser. B (Stat Methodol.) 67 (2), 301–320. (Royal Statistical Society, Stretton, J., Thompson, P.J., 2012. Frontal lobe function in temporal lobe epilepsy. Wiley). http://www.jstor.org/stable/3647580.

Journal

Quantitative BiologyarXiv (Cornell University)

Published: Jul 25, 2017

References