Access the full text.
Sign up today, get DeepDyve free for 14 days.
Background: We investigated the correlation between texture features extracted from apparent diffusion coefficient (ADC) maps or diffusion-weighted images (DWIs), and grade group (GG) in the prostate peripheral zone (PZ) and transition zone (TZ), and assessed reliability in repeated examinations. Methods: Patients underwent 3-T pelvic magnetic resonance imaging (MRI) before radical prostatectomy with repeated DWI using b-values of 0, 100, 1,000, and 1,500 s/mm . Region of interest (ROI) for cancer was assigned to the first and second DWI acquisition separately. Texture features of ROIs were extracted from comma-separated values (CSV) data of ADC maps generated from several sets of two b-value combinations and DWIs, and correlation with GG, discrimination ability between GG of 1–2 versus 3–5, and data repeatability were evaluated in PZ and TZ. Results: Forty-four patients with 49 prostate cancers met the eligibility criteria. In PZ, ADC 10% and 25% based on ADC map of two b-value combinations of 100 and 1,500 s/mm and 10% based on ADC map with b-value of 0 and 1,500 s/mm showed significant correlation with GG, acceptable discrimination ability, and good repeatability. In TZ, higher-order texture feature of busyness extracted from ADC map of 100 and 1,500 s/mm , and high gray-level run emphasis, short-run high gray-level emphasis, and high gray-level zone emphasis from DWI with b-value of 100 s/ mm demonstrated significant correlation, excellent discrimination ability, but moderate repeatability. Conclusions: Some DWI-related features showed significant correlation with GG, acceptable to excellent discrimination ability, and moderate to good data repeatability in prostate cancer, and differed between PZ and TZ. Keywords: Diffusion magnetic resonance imaging, Image interpretation (computer-assisted), Neoplasm grading, Prostate neoplasms, Reproducibility of results * Correspondence: email@example.com Department of Diagnostic Radiology, School of Medicine, Sapporo Medical University, South 1, West 17, Chuo-ku, Sapporo 060-8556, Japan Full list of author information is available at the end of the article © The Author(s). 2022 Open Access under exclusive licence to European Society of Radiology. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 2 of 12 Key points significantly correlated with GS in PZ cancer, but not in � Some diffusion-weighted imaging (DWI)-related tex- TZ cancer. These results raise the possibility that DWI- ture features significantly correlated with histological ag- related features may demonstrate a different relationship gressiveness in prostate cancer. with tumor aggressiveness between the PZ and TZ. In a � Some DWI-related texture features show clinically recent systematic review, Surov A et al.  reported acceptable data repeatability in prostate cancer. that in PZ cancer, ADC moderately correlates with GS, � Texture features showing correlation with histo- but it weakly correlated with in TZ cancer. logical aggressiveness and repeatability differ between This study aimed to analyze the correlation between zones. texture features extracted from ADC maps generated � DWI with b-values of 100 and 1,500 s/mm may be from several sets of two b-value combinations or DWIs relevant. with several b-values, and GG in the PZ and TZ, separ- ately, and to evaluate the reliability of texture features in Background repeated examinations. Texture analysis of clinical imaging has been increas- ingly carried out to determine its correlation with histo- Methods logical findings, such as lesion aggressiveness and Population, inclusion, and exclusion criteria clinical outcome [1–3]. Texture features extracted from This study was compliant with Helsinki Declaration. magnetic resonance diffusion-weighted imaging (DWI), The following inclusion and exclusion criteria were con- including apparent diffusion coefficient (ADC) maps, sidered: Inclusion criteria: patients who underwent 3-T have shown promising results. However, there is no con- multiparametric MRI (mpMRI) at our institute, includ- sensus regarding the method to calculate DWI-related ing two sets of repeated DWI acquisitions for evaluating metrics such as monoexponential fitting, intravoxel inco- prostate lesions with informed consent from July 2016 herent motion, and diffusion kurtosis imaging. From a to May 2020. Exclusion criteria: treatment except radical clinical perspective, ADC maps calculated from two dif- prostatectomy; lesions with a longitudinal diameter < 10 ferent b-values can be simple and easy to use, but there mm; lesions not detected on DWI; lesions with a voxel is no consensus regarding the use of a combination of number within the region of interest (ROI) < 50; lesions two b-values. Furthermore, there are concerns regarding containing voxel with ADC value < 0; poor image qual- the reliability of texture features which are sensitive to ity. Figure 1 shows the flowchart of patient inclusion imaging characteristics, possibly having coincidental sig- and exclusion. nificance due to a larger number of parameters [4–6]. Magnetic resonance imaging (MRI) is a primary im- MRI aging modality used for prostate cancer. Many studies MRI was performed using a 3-T system (Ingenia, Philips have been reported regarding the correlation between Healthcar, Eindhoven, The Netherlands) with a pelvic DWI-related parameters and lesion aggressiveness, such phased-array coil. No endorectal coil was used. Either as the Gleason score (GS) and grade group (GG), with 20-mg hyoscine-N-butyl-bromide or 1-mg glucagon was inconsistent results. It was reported that ADC entropy injected intramuscularly before examination to minimize showed significant difference between GS of 3 + 4 and 4 bowel peristalsis. + 3 but not in ADC mean . Alessandrino F et al.  A routine mpMRI protocol was applied to all patients, reported similar results, with no significance in ADC including sagittal, coronal, and axial T2-weighted im- mean. In contrast, Itou Y et al.  reported that ADC aging; axial DWIs; and axial dynamic contrast-enhanced median showed a significant correlation with GS and a imaging before and after gadolinium chelate injection of significant difference between GS of 3+4 and 4+3. Shan 0.1 mmol/kg gadoterate meglumine, Magnescope, Y et al.  reported that ADC mean showed a signifi- Dotarem (Guerbet, Villepinte, France). For DWI, two se- cant correlation with GS and a significant difference be- quential free-breathing DWI single-shot spin-echo echo- tween GS of 3 + 4 and 4 + 3. Though some studies planar images were acquired. The patient remained in evaluated data reliability focusing on intraobserver and the same position between the two DWI acquisitions. interobserver agreement for the same images (ADC Four b-values (0, 100, 1,000, and 1,500 s/mm ) with maps) [11, 12], few studies have been performed with re- three orthogonal diffusion probing gradients were gener- spect to image data reliability itself. Furthermore, though ated. ADC maps were generated using DWIs with b- the above studies dealt with cancers in the peripheral values of 100 and 1,000 s/mm , ADC map (100, 1,000) zone (PZ) and transition zone (TZ) together, Hambrock in line with the Prostate Imaging–Reporting and Data T et al.  reported that ADC median showed a signifi- System (PI-RADS) version 2.1 (https://www.acr.org/-/ cant correlation with Gleason grade in PZ cancer. Jyoti media/ACR/Files/RADS/PI-RADS/PIRADS-V2-1.pdf) R et al.  also reported that ADC minimum was for the first and second DWIs, respectively. The DWI Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 3 of 12 Fig. 1 Flowchart of study showing inclusion and exclusion criteria, and patient and lesion numbers sequence parameters are summarized in Supplemental evaluation of prostatectomy specimen. After this, the Table S1. ROI was placed on the first DWI datasets of DWI 0, DWI 100, and DWI 1,000 through intravoxel incoherent Image analysis motion (IVIM) application of the Synapse Vincent 3D Image analysis including ROI assignment was performed Image Analysis System. The same procedures were re- by a consensus decision of two observers (C.T. and M.H. peated for the second DWI datasets. Voxel data distribu- with 4 and over 30 years of experience in diagnostic tions within the ROI were rendered in comma-separated radiology, respectively) using a Synapse Vincent 3D values (CSV) format (Supplemental Figs. S1 and S2) Image Analysis System (Fujifilm Corporation, Tokyo, using a Synapse Vincent 3D Image Analysis System. Japan). For PZ cancer, the polygonal two-dimensional Then, the ADC of each voxel was calculated by fitting ROI was manually determined on the lesion in the cen- signal intensity decay between four patterns of b-value ter slice showing hyperintensity on the first DWI with a combinations using a monoexponential curve fit: 0 and 2 2 2 b-value of 1,500 s/mm (DWI 1,500) and hypointensity 1,000 s/mm , ADC (0, 1,000); 0 and 1,500 s/mm , ADC on the first ADC map (100, 1,000), referring to T2- (0, 1,500); 100 and 1,000 s/mm , ADC (100, 1,000); and weighted imaging, dynamic contrast-enhanced imaging, 100 and 1,500 s/mm , ADC (100, 1,500). Representative and whole-mount, step-sectioned histological evaluation cases are shown in Figs. 2 and 3. of prostatectomy specimen. Then, the ROI was placed We assigned a two-dimensional ROI in the center slice on the first DWI datasets of DWI 0, DWI 100, and DWI of the lesion because more than half lesions were not 1,000 through IVIM application of a Synapse Vincent large enough to place a three-dimensional ROI. Only 21 3D Image Analysis System. For non-peripheral transition lesions, 43% of total lesions, showed longitudinal diam- zone (TZ) cancers, the polygonal two-dimensional ROI eter > 12 mm on images and could be determined on was manually determined on the lesion in the center equal to or more than four slices (DWI slice thickness of slice showing hypointensity on T2-weighted images and 3 mm/ gap of 0 mm, Supplemental Table 1) that would hyperintensity on the first DWI 1,500, referring to the have satisfied assigning ROIs on two or more slices first ADC map (100, 1,000), dynamic contrast-enhanced avoiding peripheral images, possibly being affected by imaging, and whole-mount, step-sectioned histological partial volume effect. Texture analysis calculates the Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 4 of 12 Fig. 2 Multiparametric magnetic resonance imaging of the case (70 years, right peripheral zone cancer, GG of 3, PI-RADS of 4, T2aN0M0). a Axial T2-weighted image (repetition time of 4,000 ms and echo time of 80 ms). b First axial apparent diffusion coefficient (ADC) map (100, 1,000). c Second axial ADC map (100, 1,000). d Dynamic contrast-enhanced T1-weighted image. e First DWI 1,500. f Second DWI 1,500. Arrows indicate polygonal areas of the region of interests on e and f relationship between adjacent voxels, and thus, we as- matrix (GLZSM), and neighborhood gray-level differ- sumed that appropriate texture analysis required at least ence matrix (NGLDM). Homogeneity, energy, correl- four voxels along each direction. ation, contrast, entropy, and dissimilarity were All voxels within the ROI were extracted from the calculated from the GLCM. Short-run emphasis CSV data of ADC (0, 1,000), (0, 1,500), (100, 1,000), (SRE), long-run emphasis (LRE), low gray-level run and (100, 1,500), and DWI 0, 100, 1,000, and 1,500. emphasis (LGRE), high gray-level run emphasis First-order statistical variables (minimum, 10%, 25%, (HGRE), short-run low gray-level emphasis (SRLGE), median, 75%, 90%, maximum, mean, sum, standard short-run high gray-level emphasis (SRHGE), long-run deviation, skewness, kurtosis, energy, and entropy) low gray-level emphasis (LRLGE), long-run high gray- were calculated. After discretization of voxel values level emphasis (LRHGE), gray-level non-uniformity (bin number 32), higher-order texture analysis was for run (GLNUr), run-length non-uniformity (RLNU), performed in a two-dimensional manner to generate a and run percentage (RP) were calculated from GLRL gray-level co-occurrence matrix (GLCM), gray-level M. Short-zone emphasis (SZE), long-zone emphasis run-length matrix (GLRLM), gray-level zone-size (LZE), low gray-level zone emphasis (LGZE), high Fig. 3 Multiparametric magnetic resonance imaging of the case (61 years, transition zone cancer, GG of 2, PI-RADS of 5, T2cN0M0). a Axial T2- weighted image (repetition time of 4,000 ms and echo time of 80 ms). b First axial apparent diffusion coefficient (ADC) map (100, 1,000). c Second axial ADC map (100, 1,000). d Dynamic contrast-enhanced T1-weighted image. e First DWI 1,500. f Second DWI 1,500. Arrows indicate polygonal areas of the region of interests on e and f Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 5 of 12 gray-level zone emphasis (HGZE), short-zone low Finally, 44 patients with 49 cancers were analyzed. The gray-level emphasis (SZLGE), short-zone high gray- characteristics of patients and lesions are summarized in level emphasis (SZHGE), long-zone low gray-level em- Table 1. Of them, 11 patients underwent prostate biopsy phasis (LZLGE), long-zone high gray-level emphasis within 6 weeks (17−36 days) before mpMRI, but no clear (LZHGE), gray-level non-uniformity for zone adverse effects were included for analysis. The duration (GLNUz), zone length non-uniformity (ZLNU), and between MRI and radical prostatectomy was from 8 to zone percentage (ZP) were calculated from GLZSM. 191 days (median 69 days). Coarseness, contrast, and busyness were calculated As a representative of ADC histograms, ADC 10% cal- from NGLDM. Texture features were computed using culated from ADC (0, 1,000), (0, 1,500), (100, 1,000), and the PTexture package (www.github.com/metavol/ (100, 1,500) are summarized in Table 2 and classified ac- ptexture) written in Python language. The detailed cording to the PZ and TZ as well as GG. methods are described elsewhere . In PZ cancer, the following metrics showed significant correlation with GG at both examinations: ADC 10% Statistical analysis and 25% based on ADC (0, 1,000); ADC 5%, 10%, and Statistical analyses were carried out separately for PZ 25% based on ADC (0, 1,500); ADC 10%, 25%, and 50% and TZ cancers. First, the correlation between texture based on ADC (100, 1,000); and ADC 5%, 10%, and 25% features and GG was evaluated using Spearman's rank based on ADC (100, 1,500). Other metrics including correlation test. For the features showing significance at higher-order texture features did not show significance. both the first and second examinations, receiver operat- The results, including Spearman's ρ and its 95% confi- ing characteristic (ROC) curves for differentiating be- dence interval, the AUC of ROC for differentiation be- tween GG of 1 and 2 versus GG of 3, 4, and 5 were tween GG of 1 and 2 versus GG of 3, 4, and 5, ICC and drawn, and the area under the curve (AUC) was calcu- its 95% confidence interval, and Bland-Altman analysis lated because there was a difference in prognosis be- (%) (bias, standard deviation of bias, 95% limit of agree- tween GG of 1 and 2, versus GG of 3, 4, and 5 . To ment), are summarized in Table 3. Among them, as check test-retest data repeatability, intraclass correlation ADC 10%-based on ADC (0, 1,500) and ADC (100, coefficient (ICC) and Bland-Altman plot (%) (% differ- 1,500) as well as ADC 25% based on ADC (100, 1,500) ence was used to normalize differences in original data showed moderate correlation coefficient with GG (|ρ|> magnitude) were used. Statistical analyses were per- 0.4, p < 0.05), acceptable discrimination ability (AUC > formed using GraphPad Prism ver. 7.05 (GraphPad Soft- Table 1 Summary of patient and lesion characteristics ware, San Diego, USA) and SPSS statistics ver. 25 Parameter Value (International Business Machines Corporation, Armonk, No. of patients 44 USA); p-values < 0.05 were considered statistically significant. Median age (year), (range) 68 (49–79) We considered the following values for classifying the No. of cancers 49 strength of correlation: moderate (|ρ|: 0.4-0.7), strong PI-RADS 2.1 score (|ρ|: 0.7-0.9), and very strong (|ρ|: 0.9-1) , discrimin- ation ability: acceptable (AUC: 0.7–0.8), excellent (AUC: 0.8–0.9), and outstanding (AUC > 0.9) , and data re- peatability: moderate (ICC: 0.5–0.75), good: (ICC: 0.75– 0.9), and excellent (ICC: 0.9–1) . Grade group Results From July 2016 to May 2020, a total of 296 patients with suspected prostate cancer underwent mpMRI including two sets of repeated DWI acquisitions for evaluating prostate lesions with informed consent. Among them, 52 patients underwent mpMRI before prostatectomy and Location of cancer were histologically diagnosed as prostate cancer by the Peripheral zone 30 institutional pathologists. There were 62 cancers with a Not peripheral zone (Transition zone) 19 longitudinal diameter ≥ 10 mm. Furthermore, one lesion Number of voxels included in the ROI undetected on DWI, six lesions with a voxel number First, median, (range) 104 (58–231) within the ROI < 50, three lesions with poor image qual- Second, median, (range) 101 (51–271) ity either in the first or second DWI, and three lesions containing voxel with ADC value < 0 were excluded. PI-RADS Prostate Imaging–Reporting and Data System, ROI Region of interest Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 6 of 12 Table 2 Summary of the breakdown of ADC 10% according to 1,500), and their XY plot in TZ cancer are also shown in grade group and the b-value combination in peripheral and Supplemental Fig. S3. transition zones For TZ cancer, SRHGE and busyness based on ADC Zone/grade Combination of two b-values × s/mm (100, 1,500), and skewness, HGRE, SRHGE, LRHGE, group (0, 1,000) (0, 1,500) (100, 1,000) (100, 1,500) HGZE, SZHGE, and busyness based on DWI 100, and -3 2 skewness, HGRE, SRHGE, and HGZE based on DWI 0 Peripheral zone First/second of ADC 10% × 10 mm /s showed significant correlations with GG. As opposed to Grade group 1 0.763/0.869 0.663/0.743 0.760/0.857 0.648/0.694 PZ, the first-order statistical metrics did not show sig- 2 0.767/0.782 0.660/0.678 0.651/0.694 0.584/0.612 nificance. The results, including Spearman's ρ and its 3 0.595/0.633 0.525/0.572 0.505/0.541 0.452/0.490 95% confidence interval, the ROC-AUC of for differenti- 4 0.609/0.654 0.537/0.599 0.557/0.558 0.482/0.532 ating GG of 1 and 2 from GG of 3, 4, and 5, ICC and its 5 0.529/0.517 0.431/0.451 0.454/0.438 0.386/0.397 95% confidence interval, and Bland-Altman analysis (%) -3 2 (bias, standard deviation of bias, 95% limit of agreement) Transition zone First/second of ADC 10% ×10 mm /s are summarized in Table 4. Among them, busyness Grade group 1––– – based on ADC (100, 1,500), and HGRE, SRHGE, and 2 0.708/0.676 0.626/0.598 0.636/0.584 0.566/0.536 HGZE based on DWI 100 demonstrated moderate cor- 3 0.745/0.744 0.659/0.658 0.700/0.637 0.610/0.555 relation coefficients with GG (|ρ| > 0.5, p < 0.05), excel- 4 0.748/0.733 0.682/0.653 0.645/0.728 0.656/0.573 lent discrimination ability (AUC > 0.8) at both 5 0.741/0.773 0.636/0.634 0.638/0.592 0.456/0.563 examinations, and moderate data repeatability (ICC > 0.5; skewness based on DWI 100 or 0 was excluded due 0.7) at both examinations, and good data repeatability to large standard deviation [> 300%] in Bland-Altman (ICC > 0.8). The correlation between GG and ADC 10% analysis [%]). The correlation between GG, and busyness mean of the first and second examinations based on mean of the first and second examinations based on ADC (100, 1,500), and their XY plot are shown in Fig. 4. ADC (100, 1,500) and HGRE mean of the first and sec- To demonstrate the difference between the PZ and TZ, ond examinations based on DWI 100, and their XY plot the correlation between GG and ADC 10% mean of the are shown in Figs. 5 and 6, respectively. first and second examinations based on ADC (100, Fig. 4 Peripheral zone cancer. a Correlation between grade group and mean of the first and second apparent diffusion coefficient (ADC) 10% based on ADC (100, 1,500). b XY plot of the first and second ADC 10% based on ADC (100, 1,500). Open circle, closed square, and line indicate grade group (GG) of 1 and 2, GG of 3, 4, and 5, and Y = X line, respectively Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 7 of 12 Fig. 5 Transition zone cancer. a Correlation between grade group and mean of the first and second busyness of neighborhood gray-level difference matrix based on apparent diffusion coefficient (ADC) (100, 1,500). b XY plot of the first and second busyness of neighborhood gray- level difference matrix based on ADC (100, 1,500). Open circle, closed square, and line indicate grade group (GG) of 1 and 2, GG of 3, 4, and 5, and Y = X line, respectively Discussion together and the distribution of lesion aggressiveness To our knowledge, this is the first study showing a dif- (16out of total65lesions were GS of 6),including ference in DWI-related texture features that demon- 19 biopsy-proven lesions, might explain the discrep- strate not only significant correlations with GG and ancy. When analyzed by combining PZ and TZ can- discrimination ability between GG of 1 and 2, versus GG cers, the entropy of GLCM based on ADC (0, 1,000) of 3, 4, and 5, but also practical data repeatability be- did not show significance either in bin of 8, 16, or 32 tween the PZ and TZ in prostate cancer. setting (Supplemental Table S2). In PZ cancer, ADC 10% based on ADC (0, 1,500) In TZ cancer, busyness based on ADC (100, 1,500), and (100, 1,500) as well as ADC 25% based on ADC and HGRE, SRHGE, and HGZE based on DWI 100 (100, 1,500) satisfied moderate correlation and had demonstrated moderate correlation coefficients, excel- acceptable discrimination and good repeatability. lent discrimination, and moderate data repeatability. To These results were in accordance with a systematic evaluate the effect of bin number, texture features using review reporting that ADC correlated moderately with bin 8 and 16 were also analyzed. Similar results were ob- GS (correlation coefficient of -0.48, 95% confidence tained (Supplemental Tables S3 and S4). In general, tex- interval of -0.54 to -0.42) . However, Hectors SJ ture features for TZ cancer tend to show higher et al. reported thatSRE and LRE usingbin 16 correlation and discrimination but lower data repeatabil- extracted from ADC map showed significance with ity than those for PZ cancer. GS. Several differences, such as analyzing the PZ and Another important finding is that ADC histogram TZ together, calculating ADC with four b-values (0, metrics such as 10%, which showed significance in PZ 1,000, 1,600, and 2,000 s/mm ), and measuring tex- cancer, showed no significance in TZ cancer (Supple- ture feature using different methods, could explain mental Fig. S3). This result was not inconsistent with the differences. Baek T et al. reported that the en- the results of a systematic review, which reported that tropy of GLCM from ADC map generated from b- ADC correlated weakly (correlation coefficient of -0.22, values of 0 and 1,000 s/mm showed significance with 95% confidence interval of -0.47 to + 0.03) with GS in GS. The differences in analyzing the PZ and TZ TZ cancer . Furthermore, ADC 10% did not show Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 8 of 12 Fig. 6 Transition zone cancer. a Correlation between grade group and mean of the first and second high gray-level run emphasis of gray-level run-length matrix based on diffusion-weighted (DWI) 100. b XY plot of the first and second high gray-level run emphasis of gray-level run-length matrix based on DWI 100. Open circle, closed square, and line indicate grade group (GG) of 1 and 2, GG of 3, 4, and 5, and Y = X line, respectively significance and some features from DWI 100 and 0 method reported to show the highest AUC among the demonstrating significance in TZ cancer may indirectly IVIM, kurtosis, and IVIM-kurtosis methods , which support that PI-RADS 2.1 puts emphasis on the findings is consistent with ours. of T2-weighted imaging for TZ cancer, because DWI Another focus of the present study is data repeat- with low b-value looks similar to fat-saturated T2- ability. DWI-related features with significance for PZ weighted imaging. However, it is unclear why DWI- cancer demonstrated good repeatability, but those for related features showing significance with GG differ be- TZ cancer remained moderate. However, moderate tween PZ and TZ. One possible explanation might be repeatability may be acceptable in clinical practice. In that while the volume of the lumen and stroma is posi- apreviousstudy,the κ valuefor thereproducibilityof tively correlated with ADC, that of the epithelium is the PI-RADS 2 score in TZ was 0.525 . In an- negatively correlated , and the degree of each com- other study, ICCs of lesion size in the TZ were 0.80 position differs between the PZ and TZ . This may and 0.58 for intra-reader and inter-reader analyses, re- explain the results. However, the detailed mechanism spectively . underlying this is unknown. Texture features themselves have high potential with Regarding which two b-value combination is appropri- respect to correlation with lesion aggressiveness and ate for calculating ADC, ADC generated from DWI 100 clinical outcome. However, those have a tendency prone and 1,500 would be relevant in terms of a correlation to be affected by a mild difference of the imaging data with GG (Tables 3 and 4). We cannot interpret these re- including artifacts. Therefore, reliability studies not only sults with reasonable model and/or relevant hypothesis for observers but also for imaging data themselves at this time but image quality improvement of DWI should be verified sufficiently before being applied to 1,500 due to performance advance of MRI-system would clinical practice. contribute to these results. In a study comparing diag- This study has some limitations. First, we analyzed pa- nostic ability of prostate cancer based on DWI-related tients who underwent radical prostatectomy because of features, ADC value calculated from DWIs with b-values the clear correlation between histology and mpMRI, but of 50 and 1,500 s/mm using a mono-exponential this concept would have reduced the number of the Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 9 of 12 Table 3 Summary of the correlation between features and grade group as well as the repeatability in peripheral zone cancer Base First/ Grade group Grade group 1, 2 vs.3, ICC 95% CI Bland-Altman plot (%) image/ second 4, 5 features Spearman's 95% CI p AUC of ROC Bias SD of 95% LOA ρ bias ADC (0, 1,000) ADC First -0.4027 -0.6725 to 0.0273 0.732 0.773 0.580 to -0.2016 21.15 -41.65 to 10% -0.03854 0.885 41.25 Second -0.4241 -0.6864 to 0.0195 0.737 -0.06427 ADC First -0.3638 -0.6467 to 0.0481 0.705 0.825 0.667 to 0.06553 15.54 -30.40 to 25% 0.007035 0.912 30.53 Second -0.411 -0.6779 to 0.0241 0.728 -0.04843 ADC (0, 1,500) ADC 5% First -0.3738 -0.6534 to 0.0419 0.705 0.817 0.654 to -8.692 19.58 -47.08 to -0.004499 0.908 29.69 Second -0.4389 -0.6959 to 0.0152 0.759 -0.08239 ADC First -0.418 -0.6824 to 0.0215 0.737 0.829 0.674 to -7.875 15.9 -39.04 to 10% -0.05691 0.915 23.29 Second -0.463 -0.7111 to -0.1123 0.01 0.763 ADC First -0.4035 -0.673 to -0.03941 0.027 0.714 0.865 0.738 to -5.203 11.99 -28.71 to 25% 0.933 18.31 Second -0.3736 -0.6532 to 0.042 0.71 -0.004216 ADC (100, 1,000) ADC First -0.4083 -0.6761 to 0.0251 0.732 0.75 0.542 to -6.843 21.59 -49.16 to 10% -0.04522 0.872 35.47 Second -0.5043 -0.7368 to -0.1652 0.0045 0.786 ADC First -0.4103 -0.6774 to 0.0243 0.723 0.786 0.601 to -4.467 15.91 -35.66 to 25% -0.04755 0.892 26.73 Second -0.4555 -0.7064 to -0.1029 0.0114 0.754 ADC First -0.3794 -0.6571 to 0.0387 0.692 0.82 0.658 to -3.982 11.12 -25.77 to 50% -0.01101 0.910 17.81 Second -0.4013 -0.6715 to -0.0368 0.028 0.714 ADC (100, 1,500) ADC 5% First -0.3903 -0.6643 to 0.033 0.719 0.816 0.651 to -4.456 18.31 -40.35 to -0.02385 0.907 31.44 Second -0.4142 -0.6799 to 0.0229 0.746 -0.05222 ADC First -0.4275 -0.6886 to 0.0184 0.741 0.839 0.692 to -6.85 12.82 -31.97 to 10% -0.06841 0.920 18.27 Second -0.4591 -0.7087 to -0.1074 0.0107 0.763 ADC First -0.4528 -0.7047 to 0.012 0.746 0.843 0.699 to - 5.629 10.96 -27.10 to 25% -0.09954 0.922 15.85 Second -0.4039 -0.6733 to 0.0268 0.732 -0.03999 95% CI 95% confidence interval, 95% LOA 95% Limit of agreement, AUC Area under the curve, ICC Intraclass correlation coefficient, ROC Receiver operating characteristic, SD Standard deviation cases and lesions included in the study. Second, texture features were extracted from two-dimensional ROI be- features of T2-weighted imaging were not evaluated be- cause the lesion size was not large enough to extract fea- cause matrix size was different from DWI and the voxel tures from three-dimensional ROI. Fourth, ROI number in the ROI differed greatly. Third, texture assignment was performed by consensus between two Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 10 of 12 Table 4 Summary of the correlation between features and grade group as well as the repeatability in transition zone cancer Base First/ Grade group Grade group 1, 2 vs.3, 4, ICC 95% CI Bland-Altman plot (%) image/ second 5 features Spearman's 95% CI p AUC of ROC Bias SD of 95% LOA ρ bias ADC (100, 1,500), bin 32 SRHGE First 0.488 0.02898 to 0.7771 0.034 0.814 0.427 -0.011 to 0.73 -21.4 27.77 -75.83 to 33.04 Second 0.6912 0.3326 to 0.8752 0.001 0.943 First -0.5039 -0.7853 to 0.0278 0.814 0.576 0.188 to 21.1 77.36 -130.5 to busyness -0.05002 0.811 172.7 Second -0.6742 -0.8675 to -0.3039 0.0015 0.929 DWI 100, bin 32 First -0.5232 -0.7951 to 0.0215 0.829 0.629 0.266 to -28.03 310.3 -636.2 to skewness -0.07612 0.837 580.1 Second -0.4699 -0.7676 to 0.0424 0.786 -0.005422 HGRE First 0.5663 0.1368 to 0.8166 0.0115 0.857 0.658 0.312 to -14.68 35.99 -85.22 to 0.852 55.85 Second 0.5516 0.1157 to 0.8093 0.0144 0.843 SRHGE First 0.5663 0.1368 to 0.8166 0.0115 0.857 0.619 0.251 to -15.33 35.93 -85.76 to 55.1 0.833 Second 0.5902 0.1717 to 0.8282 0.0078 0.871 LRHGE First 0.4892 0.03047 to 0.7777 0.0335 0.8 0.802 0.563 to -12.04 37.52 -85.58 to 61.5 0.918 Second 0.4744 0.01126 to 0.77 0.0401 0.786 HGZE First 0.5277 0.08236 to 0.7974 0.0202 0.829 0.558 0.161 to -13.03 32.89 -77.5 to 51.44 0.802 Second 0.5277 0.08236 to 0.7974 0.0202 0.829 SZHGE First 0.5516 0.1157 to 0.8093 0.0144 0.843 0.373 -0.075 to -14.48 33.09 -79.34 to 0.699 50.38 Second 0.6072 0.1974 to 0.8364 0.0058 0.886 First -0.6151 -0.8401 to -0.2095 0.0051 0.900 0.548 0.148 to 9.866 44.39 -77.14 to busyness 0.797 96.87 Second -0.4642 -0.7646 to 0.0453 0.771 0.001836 DWI 0, bin 32 First -0.5743 -0.8205 to -0.1483 0.0101 0.871 0.575 0.185 to 0.81 -55.02 303.2 -649.3 to skewness 539.2 Second -0.4982 -0.7824 to 0.0299 0.800 -0.04246 HGRE First 0.5947 0.1785 to 0.8304 0.0072 0.871 0.579 0.192 to -14.98 34.32 -82.24 to 0.813 52.28 Second 0.4721 0.008339 to 0.7688 0.0412 0.786 SRHGE First 0.614 0.2078 to 0.8396 0.0052 0.886 0.537 0.132 to -15.22 34.93 -83.67 to 0.791 53.24 Second 0.4892 0.03047 to 0.7777 0.0335 0.8 HGZE First 0.5947 0.1785 to 0.8304 0.0072 0.871 0.524 0.115 to -10.79 32.12 -73.75 to 0.784 52.18 Second 0.4937 0.03645 to 0.78 0.0317 0.8 95% CI 95% confidence interval, 95% LOA 95% Limit of agreement, AUC Area under the curve, HGRE High gray-level run emphasis, HGZE High gray-level zone emphasis, ICC Intraclass correlation coefficient, LRHGE Long-run high gray-level emphasis, ROC Receiver operating characteristic, SD Standard deviation, SRHGE Short-run high gray-level emphasis, SZHGE Short-zone high gray-level emphasis observers, not carried out independently. We consider In conclusion, some DWI-related features showed sig- consensus reading would be acceptable because one of nificant correlation with GG and clinically acceptable the main purposes of the present study was to assess re- data repeatability in histologically confirmed prostate liability of the imaging data themselves. Finally, in both cancer, and they differed between the PZ and TZ. The PZ and TZ cancer, the number of lesions was not suffi- texture features for TZ cancer tended to show higher ciently large; therefore, further analyses by combining correlation with GG and higher discrimination ability features through logistic regression and/or discriminant between GG of 1 and 2 versus GG of 3, 4, and 5, but analyses were not performed. lower data repeatability than those for PZ cancer. Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 11 of 12 Regarding a correlation with GG, DWI 100 and 1,500 s/ Received: 29 June 2021 Accepted: 25 October 2021 mm , and ADC generated from these two images would be relevant. References Abbreviations 1. Lubner MG, Smith AD, Sandrasegaran K, Sahani DV, Pickhardt PJ (2017) CT AUC: Area under the curve; GG: Grade group; GLCM: Gray-level co- texture analysis: definitions, applications, biologic correlates, and challenges. occurrence matrix; GLZSM: Gray-level zone-size matrix; GS: Gleason score; Radiographics 37:1483–1503 https://pubs.rsna.org/doi/10.1148/rg.20171 HGRE: High gray-level run emphasis; HGZE: High gray-level zone emphasis; ICC: Intraclass correlation coefficient; LRE: Long-run emphasis; LRHGE: Long- 2. Schieda N, Lim CS, Zabihollahy F, et al (2021) Quantitative prostate MRI. J run high gray-level emphasis; mpMRI: Multiparametric MRI; MRI: Magnetic Magn Reson Imaging 53:1632─1645. https://doi.org/10.1002/jmri.27191 resonance imaging; NGLDM: Neighborhood gray-level difference matrix; PI- 3. Lovinfosse P, Polus M, Van Daele D, et al (2018) FDG PET/CT radiomics for RADS: Prostate Imaging–Reporting and Data System; PZ: Peripheral zone; predicting the outcome of locally advanced rectal cancer. Eur J Nucl Med ROC: Receiver operating characteristic; ROI: Region of interest; SRE: Short-run Mol Imaging 45:365─375. https://doi.org/10.1007/s00259-017-3855-5 emphasis; SRHGE: Short-run high gray-level emphasis; SZE: Short-zone 4. Traverso A, Wee L, Dekker A, Gillies R (2018) Repeatability and emphasis; SZHGE: Short-zone high gray-level emphasis; TZ: Transition zone Reproducibility of Radiomic Features: A Systematic Review. Int J Radiat Oncol Biol Phys 102:1143─1158. https://doi.org/10.1016/j.ijrobp.2018.05.053 5. Varghese BA, Cen SY, Hwang DH, Duddalwar VA (2019) Texture Analysis of Imaging: What Radiologists Need to Know. AJR Am J Roentgenol 212: Supplementary Information 520─528. https://www.ajronline.org/doi/10.2214/AJR.18.20624 The online version contains supplementary material available at https://doi. 6. Han S, Woo S, Suh CH, Kim YJ, Oh JS, Lee JJ (2018) A systematic review of org/10.1186/s41747-021-00252-y. the prognostic value of texture analysis in 18F─FDG PET in lung cancer. Ann Nucl Med 32:602─610. https://doi.org/10.1007/s12149-018-1281-9 Additional file 1: Supplemental Table 1 Acquisition parameters for 7. Rosenkrantz AB, Triolo MJ, Melamed J, Rusinek H, Taneja SS, Deng FM diffusion-weighted imaging (2015) Whole-lesion apparent diffusion coefficient metrics as a marker of percentage Gleason 4 component within gleason 7 prostate cancer at radical prostatectomy. J Magn Reson Imaging 41:708─714. https://doi.org/1 Authors’ contributions 0.1002/jmri.24598 All authors read, edited, and approved the final version of the manuscript. CT 8. Alessandrino F, Taghipour M, Hassanzadeh E, et al (2019) Predictive role of contributed to data acquisition, data analysis, and manuscript preparation. PI-RADSv2 and ADC parameters indifferentiating Gleason pattern 3 + 4 and KH contributed to software creation, data analysis, interpretation, and 4 + 3 prostate cancer. Abdom Radiol (NY) 44:279─285. https://doi.org/10.1 manuscript editing. KK contributed to study concept, design, data 007/s00261-018-1718-6. interpretation, and manuscript editing. NM contributed to study design, MRI 9. Itou Y, Nakanishi K, Narumi Y, Nishizawa Y, Tsukuma H (2011) Clinical utility examination, data interpretation, and manuscript editing. MH contributed to of apparent diffusion coefficient (ADC) values in patients with prostate study concept, design, data acquisition, data analysis, interpretation, and cancer: Can ADC values contribute to assess the aggressiveness of prostate manuscript editing. cancer? J Magn Reson Imaging 33:167─172. https://doi.org/10.1002/ jmri.22317 10. Itou Y, Nakanishi K, Narumi Y, Nishizawa Y, Tsukuma HItou Y, Nakanishi K, Funding Narumi Y, Nishizawa Y, Tsukuma HItou Y, Nakanishi K, Narumi Y, Nishizawa This work has not received any funding. Y, Tsukuma H (2019) Prostate cancer aggressive prediction: preponderant diagnostic performances of intravoxel incoherent motion (IVIM) imaging and diffusion kurtosis imaging (DKI) beyond ADC at 3.0 T scanner with Availability of data and materials gleason score at final pathology. Abdom Radiol (NY) 44:3441─3452. https:// The datasets used and/or analyzed during the current study are available doi.org/10.1007/s00261-019-02075-3 from the corresponding author on reasonable request. 11. Diaz de Leon A, Leyendecker JR, Otero-Muinelo S et al (2018) Reproducibility of Index Lesion Size and Mean Apparent Diffusion Declarations Coefficient Values Measured by Prostate Multiparametric MRI: Correlation With Whole-Mount Sectioning of Specimens. AJR Am J Roentgenol 211: Ethics approval and consent to participate 783─788. https://www.ajronline.org/doi/10.2214/AJR.17.19172 This prospective observational study was approved by the Sapporo 12. Nguyen TB, Ushinsky A, Yang A et al (2018) Utility of quantitative apparent University Hospital review board. All participants gave informed consent to diffusion coefficient measurements and normalized apparent diffusion this study. coefficient ratios in the diagnosis of clinically significant peripheral zone prostate cancer. Br J Radiol 91:20180091. https://www.birpublications.org/ doi/10.1259/bjr.20180091 Consent for publication 13. Hambrock T, Somford DM, Huisman HJ, et al (2011) Relationship between As part of the hospital treatment contract, informed written consent is apparent diffusion coefficients at 3 . 0-T MR imaging and Gleason Grade in obtained from all patients and/or their legal guardians for the use of their peripheral zone prostate cancer. Radiology 259:453─461. https://pubs.rsna. anonymized data for research purposes. No personal data is included in this org/doi/10.1148/radiol.11091409 manuscript. 14. Jyoti R, Jain TP, Haxhimolla H, Liddell H, Barrett SE (2018) Correlation of apparent diffusion coefficient ratio on 3.0 T MRI with prostate cancer Gleason score. Eur J Radiol Open 5:58─63. https://doi.org/10.1016/j.ejro.201 Competing interests 8.03.002 The authors declare that they have no competing interests. 15. Surov A, Meyer HJ, Wienke A (2020) Correlations between Apparent Diffusion Coefficient and Gleason Score in Prostate Cancer: A Systematic Author details Review. Eur Urol Oncol 3:489─497. https://doi.org/10.1016/j.euo.2018.12.006 Department of Diagnostic Radiology, School of Medicine, Sapporo Medical University, South 1, West 17, Chuo-ku, Sapporo 060-8556, Japan. 16. Manabe O, Ohira H, Hirata K, et al (2019) Use of 18 F-FDG PET/CT texture Department of Diagnostic Imaging, Graduate School of Medicine, Hokkaido analysis to diagnose cardiac sarcoidosis. Eur J Nucl Med Mol Imaging 46: University, Sapporo, Japan. Global Center for Biomedical Science and 1240─1247. https://doi.org/10.1007/s00259-018-4195-9 Engineering, Faculty of Medicine, Hokkaido University, Sapporo, Japan. 17. Stark JR, Perner S, Stampfer MJ, et al (2009) Gleason score and lethal Department of Urology, School of Medicine, Sapporo Medical University, prostate cancer: Does 3 + 4 = 4 + 3? J Clin Oncol 27:3459─3464. https:// Sapporo, Japan. doi.org/10.1200/JCO.2008.20.4669 Tsuruta et al. European Radiology Experimental (2022) 6:1 Page 12 of 12 18. Hosmer DW Jr, Lemeshow S, Sturdivant RX (eds) (2013) Applied Logistic Regression, 3rd Edition. Wiley, New York 19. Koo TK, Li MY (2016) A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med 15:155─163. https://doi.org/10.1016/j.jcm.2016.02.012 20. Hectors SJ, Cherny M, Yadav KK, et al (2019) Radiomics features measured with multiparametric magnetic resonance imaging predict prostate cancer aggressiveness. J Urol 202:498─504. https://doi.org/10.1097/JU. 21. Baek TW, Kim SH, Park SJ, Park EJ (2020) Texture analysis on bi-parametric MRI for evaluation of aggressiveness in patients with prostate cancer. Abdom Radiol (NY) 45:4214─4222. https://doi.org/10.1007/s00261-020-02 683-4 22. Kwak JT, Sankineni S, Xu S, et al (2017) Prostate cancer: A correlative study of Multiparametric MR imaging and digital histopathology. Radiology 285: 147─156. https://doi.org/10.1148/radiol.2017160906 23. Chatterjee A, Bourne RM, Wang S, et al (2018) Diagnosis of prostate cancer with noninvasive estimation of prostate tissue composition by using hybrid multidimensional MR imaging: A feasibility study. Radiology 287:863─873. https://doi.org/10.1148/radiol.2018171130 24. Liu Y, Wang X, Cui Y, et al (2020) Comparative Study of Monoexponential, Intravoxel Incoherent Motion, Kurtosis, and IVIM-Kurtosis Models for the Diagnosis and Aggressiveness Assessment of Prostate Cancer. Front Oncol 10:1763. https://doi.org/10.3389/fonc.2020.01763 25. Hwang DH, Duddalwar VA (2018) Clinical Application of Biparametric MRI Texture Analysis for Detection and Evaluation of High-Grade Prostate Cancer in Zone-Specific Regions. AJR Am J Roentgenol 210:549–556 https:// www.ajronline.org/doi/10.2214/AJR.17.18494. 26. Smith CP, Harmon SA, Barrett T, et al (2019) Intra- and interreader reproducibility of PI-RADSv2: A multireader study. J Magn Reson Imaging 49:1694─1703. https://doi.org/10.1002/jmri.26555. Publisher’sNote Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
European Radiology Experimental – Springer Journals
Published: Jan 12, 2022
Keywords: Diffusion magnetic resonance imaging; Image interpretation (computer-assisted); Neoplasm grading; Prostate neoplasms; Reproducibility of results
Access the full text.
Sign up today, get DeepDyve free for 14 days.