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

Learn More →

Segmentation of Circle of Willis from 7T TOF-MRI data and immersive exploration using VR

Segmentation of Circle of Willis from 7T TOF-MRI data and immersive exploration using VR DE GRUYTER Current Directions in Biomedical Engineering 2022;8(1): 12 9 -132 Lena Spitz*, Mareen Allgaier, Anastasios Mpotsaris, Daniel Behme, Bernhard Preim, and Sylvia Saalfeld Segmentation of Circle of Willis from 7T TOF-MRI data and immersive exploration using VR https://doi.org/10.1515/cdbme-2022-0033 Small vessels in the brain are complex structures which are highly relevant for anatomy education and disease diag- Abstract: 7T TOF MRI scans provide high resolution im- nosis. The structure of the intracranial vasculature, the Circle ages of intracranial vasculature. When segmented, the Circle of Willis (CoW) is taught in medical education, but is always of Willis is detailed and thus opens up new possibilities, in depicted as a 2D diagram in schooling books, and even most research but also in education. We propose a segmentation technologically enhanced materials are limited to 2D [7]. This pipeline for the Circle of Willis, and introduce a prototype that limits the understanding of the CoW’s natural alignment and enables exploration of not just the entire Circle of Willis, but its connections in a real brain. Additionally, a standard CoW also of its centerline, in an immersive VR enviroment. In our does not depict pathologies or prepares for the particularities prototype, the model can be freely rotated, placed and scaled. of patient-specific anatomy. Given that understanding anatomy A qualitative evaluation was performed with two experienced spatially is one of the most challenging areas for medical stu- neuroradiologists, who rated the prototype and its potential dents [3], exploring a 3D CoW in VR could thus enhance the positively. learning experience. Keywords: Circle of Willis, segmentation, VR, anatomy ed- ucation. 2 Materials & Methods 1 Introduction We worked with volunteer data. To obtain the MRI data, a 7T As technology advances, imaging scans get more and more whole-body MRI system from Siemens Healthineers (Erlan- detailed, revealing a better view inside the human body and gen, Germany) with a 32-channel head coil (Nova Medical, its functions. However, with more details, the segmentation of Wilmington, MA, USA) was used. For high resolution an- relevant structures from those images becomes more sophis- giograms the parameters were set accordingly, yielding a final ticated as well. 7 Tesla (T) Time-of-Flight (TOF) Magnetic resolution voxel size ranging from 0.26 𝑚𝑚 to 0.39 𝑚𝑚. Resonance Imaging (MRI) is a state of the art imaging method, though it is not part of clinical routine yet. It yields high resolu- tion scans that can show even very small vessels with a diame- 2.1 Preprocessing ter of 40 𝜇𝑚 [12]. Such small vessels are important for the ex- ploration of various pathologies, including ones that still need To prepare for segmentation and get rid of surrounding tissue, more research to be understood, such as cerebral small vessel the MR DICOM images were loaded into MeVisLab 3.4.2 [9]. disease (CSVD) or arteriovenous malformations (AVM) [12]. After adjusting page size for faster handling, a vesselness filter Therefore our contribution is to employ such high-resolution was applied multiple times to highlight both larger and smaller scans for deriving geometric models to be used in anatomy vessels. To account for the extreme variations of vessel size, education. we tested different segmentation strategies. Although a ves- selness filter with six scales ( 𝜎 = 1− 6) yields best results, we combined the result image again with a filtered image focus- *Corresponding author: Lena Spitz, Mareen Allgaier, Bernhard Preim, Sylvia Saalfeld, Institute for Simulation and ing on small vessels (𝜎 = 1 − 1.5, 2 scales) and with a filtered Graphics, Otto-von-Guericke University Magdeburg, Germany, image focusing on large vessels (𝜎 = 6, 1 scale). e-mail: lena.spitz@isg.cs.ovgu.de Next, a mask was created. The highlighted vessels were *Corresponding author: Lena Spitz, Mareen Allgaier, segmented with region growing, and a convex hull and dilation Anastasios Mpotsaris, Daniel Behme, Bernhard Preim, Sylvia was put on the result. This mask was then saved and applied to Saalfeld, Forschungscampus STIMULATE, Magdeburg, Germany the vessel image again. Anastasios Mpotsaris, Daniel Behme, University Clinic for Neu- roradiology, Otto-von-Guericke University Magdeburg, Germany Open Access. © 2022 The Author(s), published by De Gruyter. This work is licensed under the Creative Commons Attribution 4.0 International License. 129 L. Spitz et al., Segmentation of 7T TOF-MRI Circle of Willis and immersive exploration using VR MeVisLab Blender Matlab Centerline post- Vesselness Artefact processing Filter reduc�on Hierarchy Region genera�on Growing 7T TOF-MRI scans High-resolu�on VMTK DICOM data Masking 3D Circle of Willis model Centerline extrac�on Threshold Segmenta�on Fig. 1: The segmentation pipeline from the DICOM images to the segmented CoW model, including centerline. 2.2 Segmentation the centerline going from outlet to inlet, meaning each seg- ment ends in the inlet. We algorithmically adjusted number The background filtered image was then loaded into our in- and length of segments so that there is only one segment be- house software MERCIA, where mesh extraction is carried out tween two furcations or inlet/outlet. Further post-processing as described in [5, 10]. There, a threshold-based segmentation included cleaning up the way segments meet at furcations, is carried out on the background-filtered image in combina- deleting wrongly detected segments, and deleting points with tion with region growing and connected component analysis a distance of less than 0.1 between them as well as unused to extract the CoW. Special attention was paid to segment the points. Unused points refers to points that are not part of any posterior communicating artery (PCOM) in its entirety, as that segment. is a region of interest, as elaborated in Section 3. The hierarchy between the segments was determined, The resulting segmentation was then manually edited in meaning each segment (except the root segment) was assigned Blender 2.93.4 (The Blender Foundation, Amsterdam, Nether- a parent vessel. Child vessels were assigned for all non-leaf lands) to account for fusion, staircase and noise bleeding arti- segments similarly to Saalfeld et al.’s [11] flow splitting ap- facts from the region growing [5]. Additionally, the periphery proach. Hierarchy direction goes from inlet to outlets, thus was trimmed and inlets and outlets were cut to be approxi- following the vessel tree from root to leaves. This included mately perpendicular to the vessel centerline. Finally, the mesh recombining the three subtrees into one tree again. Here, the was smoothed manually in Blender’s sculpting workspace. A hierarchy direction from root to leaves was followed. pipeline of the entire segmentation is illustrated in Figure 1. Lastly, vessels with a maximum inscribed sphere radius smaller than 1 were marked as small vessels. 2.3 Centerline extraction 2.4 VR Prototype To prepare for centerline extraction, the CoW’s complex 3D model was split at the anterior and posterior communicating The VR environment developed by Allgaier et al. [1] in the arteries into three vessel trees to eliminate cycles. The cen- Unity game engine (Unity Technologies: https://unity.com, terline was then extracted for each of the three subtrees with San Francisco U.S.) and the XR Interaction Toolkit was used the Vascular Modelling Toolkit (VMTK) 1.4.0 [2]. The result- as base for the VR prototype. Their aneurysm selection scene ing .vtp file was converted into .vtk in Paraview v4.2 (Kitware was used as base for a medical environment for the CoW ex- Inc. and Los Alamos National Laboratory). The .vtk file saves ploration. The CoW model was placed in the middle of the oth- the centerline points, as well as which point is part of which erwise empty VR OR environment with a menu panel to its left segment, and the maximum inscribed sphere radius for each (see Figure 2). The menu contained a slider for transparency point. and a slider for scale. Transparency refers to the opaqueness Matlab R2021a (The MathWorks Inc., Natick, MA, USA) of the vessel walls, beneath which the centerline becomes vis- was used to further work with the centerline. VMTK extracts ible (see Figure 3). This can help to show the base structure of 130 L. Spitz et al., Segmentation of 7T TOF-MRI Circle of Willis and immersive exploration using VR Fig. 2: CoW model in the VR prototype with maximum Opacity Fig. 3: CoW model in the VR prototype with lowered opacity, and default scale. showing the centerline. the CoW without particularities of the vessels and their texture the arteria coroidea anterior, Heubner’s artery, the artery of and material in the way. The scale slider can be used to make Percheron, and the PCOM, as they could all be important and the CoW and centerline bigger or smaller. occlusion could cause severe consequences for a patient. The user can interact with the CoW by grabbing it with the Inclusion of the centerline was confirmed by the radiolo- controller. Once grabbed, it can be rotated and placed freely gists, as they find the centerline to be the important base struc- around the room. This way the CoW can be viewed from dif- ture of the CoW that makes it recognizeable based on just a ferent angles that would not otherwise be possible. few lines. They can also further help with access planning for interventions, acting as guidewires through the vessels. For anatomical teaching, they could imagine a VR CoW model not only helpful for general anatomical education, but 3 Results & Discussion also for demonstration of specific pathologies. This is in line with the positive effects noted in surveys on virtual anatomy The prototype was evaluated with two experienced neuroradi- systems [6, 8]. ologists and the think-aloud method [4], where the participants Additional inclusions would be adding a head and brain were asked to describe their usage and reasonings while test- to toggle to see orientation of the CoW, and according to the ing out the prototype. radiologists the eyes in particular to see the distance to them. They were fascinated by being able to view a CoW at such In the future, a study to compare how teaching CoW a big scale. They grabbed and rotated it without having to be anatomy to medical students with a VR application performs prompted, and noted that they had never been able to view it in contrast to the standard education with 2D diagrams would from such angles before. While grabbing and rotating it was be feasible. While the CoW is a very specialized structure, it their first reaction, they also placed it in the virtual room and is argued that introducing imaging scans much earlier in the then moved in the real space to walk around it. They did not medical curriculum than it is currently might help with general perceive any latency during interaction. anatomical and spatial understanding of anatomy and could An early comment was how remarkably clear the 3D have positive consequences [7]. This can help with the mental model made the connection between anterior and posterior cir- translation between 2D and 3D, 2D being provided by imag- culation, which is usually hard to see. The VR view demon- ing, 3D by a VR application like our tool. strated the connection and highlighted that it existed and how Another functionality of the application could be a puzzle the two supply each other. In this context the radiologists re- of complex anatomical structures that can be assembled and peatedly emphasized how valuable the prototype could be for disassembled to examine and understand the connections be- anatomical education, which corresponds with studies finding tween the parts. This could be done for the CoW, but for other that VR enviroments can help convey spatial cues and thus un- organs too, like the inner ear. derstanding and learning [6, 8]. Apart from the VR application, a high resolution CoW They also noted that the high resolution of the CoW was that includes centerlines and hierarchies can be useful, for ex- helpful, and that an even higher one could be appreciated too, ample for computational fluid dynamics, possibly in combi- as small vessels that would not usually be shown in most CoW nation with a prior phase-contrast MRI scan registration. This models and diagrams would be visible. Here they spoke of 131 L. Spitz et al., Segmentation of 7T TOF-MRI Circle of Willis and immersive exploration using VR can help explore hemodynamic parameters and blood flow in References patient-specific intracranial vasculature. In terms of drawbacks, our pipeline relies on four different [1] Allgaier M, Amini A, Neyazi B, Sandalcioglu IE, Preim B, tools that a user has to be familiar with. Especially the step in Saalfeld S, VR-based training of craniotomy for intracra- Blender requires time-intensive manual work. A streamlined nial aneurysm surgery. International Journal of Computer Assisted Radiology and Surgery 2022, 17:449–456. doi: and more automatic approach to integrate these programs into 10.1007/s11548-021-02538-3 a single framework would be desirable in the future. [2] Antiga L, Piccinelli M, Botti L, Ene-Iordache B, Remuzzi A, Steinman D, An image-based modeling framework for patient-specific computational hemodynamics, Medical & Bi- ological Engineering & Computing 2008, 46.11:1097–112. 4 Conclusion doi: 10.1007/s11517-008-0420-1. [3] Ben Awadh A, Lindsay S, Clark J, Clowry G, Keenan I, We segmented a high resolution 7T TOF-MRI scan of the Student perceptions of challenging topics and concepts in CoW vasculature, including pre-processing and subsequent anatomy education, In: Anatomical Society summer meet- ing, University of Oxford 2018. J Anat 3:392–418. doi = centerline extraction and post-processing. With the segmented 10.1111/joa.12923 model we developed a prototype VR application that enables [4] Fonteyn ME, Kuipers B, Grobe SJ, A Description of exploration of the CoW model, including centerline, in a free Think Aloud Method and Protocol Analysis, Qual- 3D environment with a medical OR backdrop. As VR systems itative Health Research 1993,3.4:430-441. doi: have become increasingly affordable in recent years, they have 10.1177/104973239300300403 [5] Glaßer S, Berg P, Neugebauer M, Preim B, Reconstruction of become more widespread. Our prototype includes options for 3D surface meshes for blood flow simulations of intracranial transparancy of the vessel walls and scale of the entire CoW aneurysms. In: Proceedings of the computer- and robot- model, as well as freedom to rotate the CoW around all axes assisted surgery (CURAC) 2015, 163–168 and place it within the room. [6] Huettl F, Saalfeld P, Hansen C, Preim B, Poplawski A, Kneist The prototype was rated very positively by neuroradiolo- W, Lang H, Huber T. Virtual reality and 3D printing improve gists, who commended the educational possibilities of such a preoperative visualization of 3D liver reconstructions-results from a preclinical comparison of presentation modalities and VR application and saw further potential for other use cases. user’s preference. Ann Transl Med. 2021 9.13:1074. doi: In the future, we aim at an automatization of the segmen- 10.21037/atm-21-512 tation pipeline including a quantitative study comparing ed- [7] Keenan ID, Powell M, Interdimensional Travel: Visualisation ucation with a VR model in comparison to standard 2D dia- of 3D-2D Transitions in Anatomy Learning, In: Rea P, edi- grams. tors. Biomedical Visualisation, Advances in Experimental Medicine and Biology, Springer, Cham. 2020:1235. doi = 10.1007/978-3-030-37639-0_6 Acknowledgment: We thank Dr.-Ing. Hendrik Mattern [8] Preim B, Saalfeld P, A survey of virtual human anatomy ed- (Department Biomedical Magnetic Resonance at the Otto- ucation systems, Computers & Graphics 2018, 71:132-153, von-Guericke University Magdeburg) for the 7T MRI data. doi: 10.1016/j.cag.2018.01.005. [9] Ritter F, Boskamp T, Homeyer A, Laue H, Schwier M, Link Author Statement F, Peitgen H-O, Medical image analysis, IEEE pulse 2011, 2.6:60–70. doi: 10.1109/MPUL.2011.942929. This work is partly funded by the Federal Ministry of Educa- [10] Saalfeld S, Berg P, Niemann A, Luz M, Preim B, Beuing tion and Research within the Forschungscampus STIMULATE O, Semiautomatic neck curve reconstruction for intracra- (grant no. 13GW0473A) and the German Research Founda- nial aneurysm rupture risk assessment based on morpho- tion (SA 3461/3-1). Authors state no conflict of interest. In- logical parameters, International Journal of Computer As- formed consent has been obtained from all individuals in- sisted Radiology and Surgery 2018, 13.11:1781–1793. doi: 10.1007/s11548-018-1848-x cluded in this study. The research related to human use com- [11] Saalfeld S, Voß S, Beuing O, Berg P, Flow-splitting-based plies with all the relevant national regulations, institutional computation of outlet boundary conditions for improved cere- policies and was performed in accordance with the tenets of brovascular simulation in multiple intracranial aneurysms, the Helsinki Declaration, and has been approved by the au- International Journal of Computer Assisted Radiology and thors’ institutional review board or equivalent committee. Surgery 2019, 14.10:1805–1813. doi: 10.1007/s11548-019- 02036-7 [12] Wardlaw JM, Smith C, and Dichgans M, Small Vessel Dis- ease: Mechanisms and Clinical Implications. The Lancet Neurology 2019, 18.7:684–696. doi: 10.1016/S1474- 4422(19)30079-1. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Current Directions in Biomedical Engineering de Gruyter

Segmentation of Circle of Willis from 7T TOF-MRI data and immersive exploration using VR

Loading next page...
 
/lp/de-gruyter/segmentation-of-circle-of-willis-from-7t-tof-mri-data-and-immersive-SSMoIbSPkV

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
de Gruyter
Copyright
© 2022 by Walter de Gruyter Berlin/Boston
eISSN
2364-5504
DOI
10.1515/cdbme-2022-0033
Publisher site
See Article on Publisher Site

Abstract

DE GRUYTER Current Directions in Biomedical Engineering 2022;8(1): 12 9 -132 Lena Spitz*, Mareen Allgaier, Anastasios Mpotsaris, Daniel Behme, Bernhard Preim, and Sylvia Saalfeld Segmentation of Circle of Willis from 7T TOF-MRI data and immersive exploration using VR https://doi.org/10.1515/cdbme-2022-0033 Small vessels in the brain are complex structures which are highly relevant for anatomy education and disease diag- Abstract: 7T TOF MRI scans provide high resolution im- nosis. The structure of the intracranial vasculature, the Circle ages of intracranial vasculature. When segmented, the Circle of Willis (CoW) is taught in medical education, but is always of Willis is detailed and thus opens up new possibilities, in depicted as a 2D diagram in schooling books, and even most research but also in education. We propose a segmentation technologically enhanced materials are limited to 2D [7]. This pipeline for the Circle of Willis, and introduce a prototype that limits the understanding of the CoW’s natural alignment and enables exploration of not just the entire Circle of Willis, but its connections in a real brain. Additionally, a standard CoW also of its centerline, in an immersive VR enviroment. In our does not depict pathologies or prepares for the particularities prototype, the model can be freely rotated, placed and scaled. of patient-specific anatomy. Given that understanding anatomy A qualitative evaluation was performed with two experienced spatially is one of the most challenging areas for medical stu- neuroradiologists, who rated the prototype and its potential dents [3], exploring a 3D CoW in VR could thus enhance the positively. learning experience. Keywords: Circle of Willis, segmentation, VR, anatomy ed- ucation. 2 Materials & Methods 1 Introduction We worked with volunteer data. To obtain the MRI data, a 7T As technology advances, imaging scans get more and more whole-body MRI system from Siemens Healthineers (Erlan- detailed, revealing a better view inside the human body and gen, Germany) with a 32-channel head coil (Nova Medical, its functions. However, with more details, the segmentation of Wilmington, MA, USA) was used. For high resolution an- relevant structures from those images becomes more sophis- giograms the parameters were set accordingly, yielding a final ticated as well. 7 Tesla (T) Time-of-Flight (TOF) Magnetic resolution voxel size ranging from 0.26 𝑚𝑚 to 0.39 𝑚𝑚. Resonance Imaging (MRI) is a state of the art imaging method, though it is not part of clinical routine yet. It yields high resolu- tion scans that can show even very small vessels with a diame- 2.1 Preprocessing ter of 40 𝜇𝑚 [12]. Such small vessels are important for the ex- ploration of various pathologies, including ones that still need To prepare for segmentation and get rid of surrounding tissue, more research to be understood, such as cerebral small vessel the MR DICOM images were loaded into MeVisLab 3.4.2 [9]. disease (CSVD) or arteriovenous malformations (AVM) [12]. After adjusting page size for faster handling, a vesselness filter Therefore our contribution is to employ such high-resolution was applied multiple times to highlight both larger and smaller scans for deriving geometric models to be used in anatomy vessels. To account for the extreme variations of vessel size, education. we tested different segmentation strategies. Although a ves- selness filter with six scales ( 𝜎 = 1− 6) yields best results, we combined the result image again with a filtered image focus- *Corresponding author: Lena Spitz, Mareen Allgaier, Bernhard Preim, Sylvia Saalfeld, Institute for Simulation and ing on small vessels (𝜎 = 1 − 1.5, 2 scales) and with a filtered Graphics, Otto-von-Guericke University Magdeburg, Germany, image focusing on large vessels (𝜎 = 6, 1 scale). e-mail: lena.spitz@isg.cs.ovgu.de Next, a mask was created. The highlighted vessels were *Corresponding author: Lena Spitz, Mareen Allgaier, segmented with region growing, and a convex hull and dilation Anastasios Mpotsaris, Daniel Behme, Bernhard Preim, Sylvia was put on the result. This mask was then saved and applied to Saalfeld, Forschungscampus STIMULATE, Magdeburg, Germany the vessel image again. Anastasios Mpotsaris, Daniel Behme, University Clinic for Neu- roradiology, Otto-von-Guericke University Magdeburg, Germany Open Access. © 2022 The Author(s), published by De Gruyter. This work is licensed under the Creative Commons Attribution 4.0 International License. 129 L. Spitz et al., Segmentation of 7T TOF-MRI Circle of Willis and immersive exploration using VR MeVisLab Blender Matlab Centerline post- Vesselness Artefact processing Filter reduc�on Hierarchy Region genera�on Growing 7T TOF-MRI scans High-resolu�on VMTK DICOM data Masking 3D Circle of Willis model Centerline extrac�on Threshold Segmenta�on Fig. 1: The segmentation pipeline from the DICOM images to the segmented CoW model, including centerline. 2.2 Segmentation the centerline going from outlet to inlet, meaning each seg- ment ends in the inlet. We algorithmically adjusted number The background filtered image was then loaded into our in- and length of segments so that there is only one segment be- house software MERCIA, where mesh extraction is carried out tween two furcations or inlet/outlet. Further post-processing as described in [5, 10]. There, a threshold-based segmentation included cleaning up the way segments meet at furcations, is carried out on the background-filtered image in combina- deleting wrongly detected segments, and deleting points with tion with region growing and connected component analysis a distance of less than 0.1 between them as well as unused to extract the CoW. Special attention was paid to segment the points. Unused points refers to points that are not part of any posterior communicating artery (PCOM) in its entirety, as that segment. is a region of interest, as elaborated in Section 3. The hierarchy between the segments was determined, The resulting segmentation was then manually edited in meaning each segment (except the root segment) was assigned Blender 2.93.4 (The Blender Foundation, Amsterdam, Nether- a parent vessel. Child vessels were assigned for all non-leaf lands) to account for fusion, staircase and noise bleeding arti- segments similarly to Saalfeld et al.’s [11] flow splitting ap- facts from the region growing [5]. Additionally, the periphery proach. Hierarchy direction goes from inlet to outlets, thus was trimmed and inlets and outlets were cut to be approxi- following the vessel tree from root to leaves. This included mately perpendicular to the vessel centerline. Finally, the mesh recombining the three subtrees into one tree again. Here, the was smoothed manually in Blender’s sculpting workspace. A hierarchy direction from root to leaves was followed. pipeline of the entire segmentation is illustrated in Figure 1. Lastly, vessels with a maximum inscribed sphere radius smaller than 1 were marked as small vessels. 2.3 Centerline extraction 2.4 VR Prototype To prepare for centerline extraction, the CoW’s complex 3D model was split at the anterior and posterior communicating The VR environment developed by Allgaier et al. [1] in the arteries into three vessel trees to eliminate cycles. The cen- Unity game engine (Unity Technologies: https://unity.com, terline was then extracted for each of the three subtrees with San Francisco U.S.) and the XR Interaction Toolkit was used the Vascular Modelling Toolkit (VMTK) 1.4.0 [2]. The result- as base for the VR prototype. Their aneurysm selection scene ing .vtp file was converted into .vtk in Paraview v4.2 (Kitware was used as base for a medical environment for the CoW ex- Inc. and Los Alamos National Laboratory). The .vtk file saves ploration. The CoW model was placed in the middle of the oth- the centerline points, as well as which point is part of which erwise empty VR OR environment with a menu panel to its left segment, and the maximum inscribed sphere radius for each (see Figure 2). The menu contained a slider for transparency point. and a slider for scale. Transparency refers to the opaqueness Matlab R2021a (The MathWorks Inc., Natick, MA, USA) of the vessel walls, beneath which the centerline becomes vis- was used to further work with the centerline. VMTK extracts ible (see Figure 3). This can help to show the base structure of 130 L. Spitz et al., Segmentation of 7T TOF-MRI Circle of Willis and immersive exploration using VR Fig. 2: CoW model in the VR prototype with maximum Opacity Fig. 3: CoW model in the VR prototype with lowered opacity, and default scale. showing the centerline. the CoW without particularities of the vessels and their texture the arteria coroidea anterior, Heubner’s artery, the artery of and material in the way. The scale slider can be used to make Percheron, and the PCOM, as they could all be important and the CoW and centerline bigger or smaller. occlusion could cause severe consequences for a patient. The user can interact with the CoW by grabbing it with the Inclusion of the centerline was confirmed by the radiolo- controller. Once grabbed, it can be rotated and placed freely gists, as they find the centerline to be the important base struc- around the room. This way the CoW can be viewed from dif- ture of the CoW that makes it recognizeable based on just a ferent angles that would not otherwise be possible. few lines. They can also further help with access planning for interventions, acting as guidewires through the vessels. For anatomical teaching, they could imagine a VR CoW model not only helpful for general anatomical education, but 3 Results & Discussion also for demonstration of specific pathologies. This is in line with the positive effects noted in surveys on virtual anatomy The prototype was evaluated with two experienced neuroradi- systems [6, 8]. ologists and the think-aloud method [4], where the participants Additional inclusions would be adding a head and brain were asked to describe their usage and reasonings while test- to toggle to see orientation of the CoW, and according to the ing out the prototype. radiologists the eyes in particular to see the distance to them. They were fascinated by being able to view a CoW at such In the future, a study to compare how teaching CoW a big scale. They grabbed and rotated it without having to be anatomy to medical students with a VR application performs prompted, and noted that they had never been able to view it in contrast to the standard education with 2D diagrams would from such angles before. While grabbing and rotating it was be feasible. While the CoW is a very specialized structure, it their first reaction, they also placed it in the virtual room and is argued that introducing imaging scans much earlier in the then moved in the real space to walk around it. They did not medical curriculum than it is currently might help with general perceive any latency during interaction. anatomical and spatial understanding of anatomy and could An early comment was how remarkably clear the 3D have positive consequences [7]. This can help with the mental model made the connection between anterior and posterior cir- translation between 2D and 3D, 2D being provided by imag- culation, which is usually hard to see. The VR view demon- ing, 3D by a VR application like our tool. strated the connection and highlighted that it existed and how Another functionality of the application could be a puzzle the two supply each other. In this context the radiologists re- of complex anatomical structures that can be assembled and peatedly emphasized how valuable the prototype could be for disassembled to examine and understand the connections be- anatomical education, which corresponds with studies finding tween the parts. This could be done for the CoW, but for other that VR enviroments can help convey spatial cues and thus un- organs too, like the inner ear. derstanding and learning [6, 8]. Apart from the VR application, a high resolution CoW They also noted that the high resolution of the CoW was that includes centerlines and hierarchies can be useful, for ex- helpful, and that an even higher one could be appreciated too, ample for computational fluid dynamics, possibly in combi- as small vessels that would not usually be shown in most CoW nation with a prior phase-contrast MRI scan registration. This models and diagrams would be visible. Here they spoke of 131 L. Spitz et al., Segmentation of 7T TOF-MRI Circle of Willis and immersive exploration using VR can help explore hemodynamic parameters and blood flow in References patient-specific intracranial vasculature. In terms of drawbacks, our pipeline relies on four different [1] Allgaier M, Amini A, Neyazi B, Sandalcioglu IE, Preim B, tools that a user has to be familiar with. Especially the step in Saalfeld S, VR-based training of craniotomy for intracra- Blender requires time-intensive manual work. A streamlined nial aneurysm surgery. International Journal of Computer Assisted Radiology and Surgery 2022, 17:449–456. doi: and more automatic approach to integrate these programs into 10.1007/s11548-021-02538-3 a single framework would be desirable in the future. [2] Antiga L, Piccinelli M, Botti L, Ene-Iordache B, Remuzzi A, Steinman D, An image-based modeling framework for patient-specific computational hemodynamics, Medical & Bi- ological Engineering & Computing 2008, 46.11:1097–112. 4 Conclusion doi: 10.1007/s11517-008-0420-1. [3] Ben Awadh A, Lindsay S, Clark J, Clowry G, Keenan I, We segmented a high resolution 7T TOF-MRI scan of the Student perceptions of challenging topics and concepts in CoW vasculature, including pre-processing and subsequent anatomy education, In: Anatomical Society summer meet- ing, University of Oxford 2018. J Anat 3:392–418. doi = centerline extraction and post-processing. With the segmented 10.1111/joa.12923 model we developed a prototype VR application that enables [4] Fonteyn ME, Kuipers B, Grobe SJ, A Description of exploration of the CoW model, including centerline, in a free Think Aloud Method and Protocol Analysis, Qual- 3D environment with a medical OR backdrop. As VR systems itative Health Research 1993,3.4:430-441. doi: have become increasingly affordable in recent years, they have 10.1177/104973239300300403 [5] Glaßer S, Berg P, Neugebauer M, Preim B, Reconstruction of become more widespread. Our prototype includes options for 3D surface meshes for blood flow simulations of intracranial transparancy of the vessel walls and scale of the entire CoW aneurysms. In: Proceedings of the computer- and robot- model, as well as freedom to rotate the CoW around all axes assisted surgery (CURAC) 2015, 163–168 and place it within the room. [6] Huettl F, Saalfeld P, Hansen C, Preim B, Poplawski A, Kneist The prototype was rated very positively by neuroradiolo- W, Lang H, Huber T. Virtual reality and 3D printing improve gists, who commended the educational possibilities of such a preoperative visualization of 3D liver reconstructions-results from a preclinical comparison of presentation modalities and VR application and saw further potential for other use cases. user’s preference. Ann Transl Med. 2021 9.13:1074. doi: In the future, we aim at an automatization of the segmen- 10.21037/atm-21-512 tation pipeline including a quantitative study comparing ed- [7] Keenan ID, Powell M, Interdimensional Travel: Visualisation ucation with a VR model in comparison to standard 2D dia- of 3D-2D Transitions in Anatomy Learning, In: Rea P, edi- grams. tors. Biomedical Visualisation, Advances in Experimental Medicine and Biology, Springer, Cham. 2020:1235. doi = 10.1007/978-3-030-37639-0_6 Acknowledgment: We thank Dr.-Ing. Hendrik Mattern [8] Preim B, Saalfeld P, A survey of virtual human anatomy ed- (Department Biomedical Magnetic Resonance at the Otto- ucation systems, Computers & Graphics 2018, 71:132-153, von-Guericke University Magdeburg) for the 7T MRI data. doi: 10.1016/j.cag.2018.01.005. [9] Ritter F, Boskamp T, Homeyer A, Laue H, Schwier M, Link Author Statement F, Peitgen H-O, Medical image analysis, IEEE pulse 2011, 2.6:60–70. doi: 10.1109/MPUL.2011.942929. This work is partly funded by the Federal Ministry of Educa- [10] Saalfeld S, Berg P, Niemann A, Luz M, Preim B, Beuing tion and Research within the Forschungscampus STIMULATE O, Semiautomatic neck curve reconstruction for intracra- (grant no. 13GW0473A) and the German Research Founda- nial aneurysm rupture risk assessment based on morpho- tion (SA 3461/3-1). Authors state no conflict of interest. In- logical parameters, International Journal of Computer As- formed consent has been obtained from all individuals in- sisted Radiology and Surgery 2018, 13.11:1781–1793. doi: 10.1007/s11548-018-1848-x cluded in this study. The research related to human use com- [11] Saalfeld S, Voß S, Beuing O, Berg P, Flow-splitting-based plies with all the relevant national regulations, institutional computation of outlet boundary conditions for improved cere- policies and was performed in accordance with the tenets of brovascular simulation in multiple intracranial aneurysms, the Helsinki Declaration, and has been approved by the au- International Journal of Computer Assisted Radiology and thors’ institutional review board or equivalent committee. Surgery 2019, 14.10:1805–1813. doi: 10.1007/s11548-019- 02036-7 [12] Wardlaw JM, Smith C, and Dichgans M, Small Vessel Dis- ease: Mechanisms and Clinical Implications. The Lancet Neurology 2019, 18.7:684–696. doi: 10.1016/S1474- 4422(19)30079-1.

Journal

Current Directions in Biomedical Engineeringde Gruyter

Published: Jul 1, 2022

Keywords: Circle of Willis; segmentation; VR; anatomy education

There are no references for this article.