Shawn Haji-Momenian1, Zixian Lin2, Bhumi Patel3, Nicole Law3, Adam Michalak4, Anishsanjay Nayak2, James Earls3, Murray Loew2. 1. Department of Radiology, George Washington University Hospital, 900 23rd St NW, Washington, DC, 20037, USA. shajimomenian@mfa.gwu.edu. 2. Department of Biomedical Engineering, George Washington University, 800 22nd Street NW, 5000 Science & Engineering Hall, Washington, DC, 20052, USA. 3. Department of Radiology, George Washington University Hospital, 900 23rd St NW, Washington, DC, 20037, USA. 4. Department of Family Medicine, University of Pittsburgh Medical Center (UPMC) Altoona, 501 Howard Avenue, Suite F2, Altoona, PA, 16601, USA.
Abstract
PURPOSE: To predict the histologic grade of small clear cell renal cell carcinomas (ccRCCs) using texture analysis and machine learning algorithms. METHODS: Fifty-two noncontrast (NC), 26 corticomedullary (CM) phase, and 35 nephrographic (NG) phase CTs of small (< 4 cm) surgically resected ccRCCs were retrospectively identified. Surgical pathology classified the tumors as low- or high-Fuhrman histologic grade. The axial image with the largest cross-sectional tumor area was exported and segmented. Six histogram and 31 texture (gray-level co-occurrences (GLC) and gray-level run-lengths (GLRL)) features were calculated for each tumor in each phase. T testing compared feature values in low- and high-grade ccRCCs, with a (Benjamini-Hochberg) false discovery rate of 10%. Area under the receiver operating curve (AUC) was calculated for each feature to assess prediction of low- and high-grade ccRCCs in each phase. Histogram, texture, and combined histogram and texture data sets were used to train and test four algorithms (k-nearest neighbor (KNN), support vector machine (SVM), random forests, and decision tree) with tenfold cross-validation; AUCs were calculated for each algorithm in each phase to assess prediction of low- and high-grade ccRCCs. RESULTS: Zero, 23, and 0 features in the NC, CM, and NG phases had statistically significant differences between low and high-grade ccRCCs. CM histogram skewness and GLRL short run emphasis had the highest AUCs (0.82) in predicting histologic grade. All four algorithms had the highest AUCs (0.97) predicting histologic grade using CM histogram features. The algorithms' AUCs decreased using histogram or texture features from NC or NG phases. CONCLUSION: The histologic grade of small ccRCCs can be accurately predicted with machine learning algorithms using CM histogram features, which outperform NC and NG phase image data.
PURPOSE: To predict the histologic grade of small clear cell renal cell carcinomas (ccRCCs) using texture analysis and machine learning algorithms. METHODS: Fifty-two noncontrast (NC), 26 corticomedullary (CM) phase, and 35 nephrographic (NG) phase CTs of small (< 4 cm) surgically resected ccRCCs were retrospectively identified. Surgical pathology classified the tumors as low- or high-Fuhrman histologic grade. The axial image with the largest cross-sectional tumor area was exported and segmented. Six histogram and 31 texture (gray-level co-occurrences (GLC) and gray-level run-lengths (GLRL)) features were calculated for each tumor in each phase. T testing compared feature values in low- and high-grade ccRCCs, with a (Benjamini-Hochberg) false discovery rate of 10%. Area under the receiver operating curve (AUC) was calculated for each feature to assess prediction of low- and high-grade ccRCCs in each phase. Histogram, texture, and combined histogram and texture data sets were used to train and test four algorithms (k-nearest neighbor (KNN), support vector machine (SVM), random forests, and decision tree) with tenfold cross-validation; AUCs were calculated for each algorithm in each phase to assess prediction of low- and high-grade ccRCCs. RESULTS: Zero, 23, and 0 features in the NC, CM, and NG phases had statistically significant differences between low and high-grade ccRCCs. CM histogram skewness and GLRL short run emphasis had the highest AUCs (0.82) in predicting histologic grade. All four algorithms had the highest AUCs (0.97) predicting histologic grade using CM histogram features. The algorithms' AUCs decreased using histogram or texture features from NC or NG phases. CONCLUSION: The histologic grade of small ccRCCs can be accurately predicted with machine learning algorithms using CM histogram features, which outperform NC and NG phase image data.
Authors: Nathalie Rioux-Leclercq; Pierre I Karakiewicz; Quoc-Dien Trinh; Vincenzo Ficarra; Luca Cindolo; Alexandre de la Taille; Jacques Tostain; Richard Zigeuner; Arnaud Mejean; Jean-Jacques Patard Journal: Cancer Date: 2007-03-01 Impact factor: 6.860
Authors: Steven Campbell; Robert G Uzzo; Mohamad E Allaf; Eric B Bass; Jeffrey A Cadeddu; Anthony Chang; Peter E Clark; Brian J Davis; Ithaar H Derweesh; Leo Giambarresi; Debra A Gervais; Susie L Hu; Brian R Lane; Bradley C Leibovich; Philip M Pierorazio Journal: J Urol Date: 2017-05-04 Impact factor: 7.450
Authors: A Zisman; A J Pantuck; F Dorey; J W Said; O Shvarts; D Quintana; B J Gitlitz; J B deKernion; R A Figlin; A S Belldegrun Journal: J Clin Oncol Date: 2001-03-15 Impact factor: 44.544
Authors: John M Hollingsworth; David C Miller; Stephanie Daignault; Brent K Hollenbeck Journal: J Natl Cancer Inst Date: 2006-09-20 Impact factor: 13.506
Authors: V Ficarra; R Righetti; G Martignoni; A D'Amico; S Pilloni; E Rubilotta; G Malossini; G Mobilio Journal: Urol Int Date: 2001 Impact factor: 2.089
Authors: Matthew R Cooperberg; Katherine Mallin; Jamie Ritchey; Jacqueline D Villalta; Peter R Carroll; Christopher J Kane Journal: J Urol Date: 2008-06 Impact factor: 7.450