Taek Min Kim1,2, Hyungwoo Ahn2,3, Hyo Jeong Lee1, Min Gwan Kim1, Jeong Yeon Cho1,2,4, Sung Il Hwang2,3, Sang Youn Kim5,6. 1. Department of Radiology, Seoul National University Hospital, Seoul, Korea. 2. Department of Radiology, Seoul National University College of Medicine, Seoul, Korea. 3. Department of Radiology, Seoul National University Bundang Hospital, Seongnam-si, Gyeonggi-do, Korea. 4. Institute of Radiation Medicine and Kidney Research Institute, Seoul National University Medical Research Center, Seoul, 03080, Korea. 5. Department of Radiology, Seoul National University Hospital, Seoul, Korea. iwishluv@empas.com. 6. Department of Radiology, Seoul National University College of Medicine, Seoul, Korea. iwishluv@empas.com.
Abstract
PURPOSE: This study aims to assess the computed tomography (CT) findings of renal epithelioid angiomyolipoma (EAML) and develop a radiomics-based model for differentiating EAMLs and clear cell renal cell carcinomas (RCCs). METHOD: This two-center retrospective study included 28 histologically confirmed EAMLs and 56 size-matched clear cell RCCs with preoperative three-phase kidney CTs. We conducted subjective image analysis to determine the CT parameters that can distinguish EAMLs from clear cell RCCs. Training and test sets were divided by chronological order of CT scans, and radiomics model was built using ten selected features among radiomics and CT features. The diagnostic performance of the radiomics model was compared with that of the three radiologists using the area under the receiver-operating characteristic curve (AUC). RESULTS: The mean size of the EAMLs was 6.2 ± 5.0 cm. On multivariate analysis, a snowman or ice cream cone tumor shape (OR 16.3; 95% CI 1.7-156.9, P = 0.02) and lower tumor-to-cortex (TOC) enhancement ratio in the corticomedullary phase (OR 33.4; 95% CI 5.7-197, P < 0.001) were significant independent factors for identifying EAMLs. The diagnostic performance of the radiomics model (AUC 0.89) was similar to those of genitourinary radiologists (AUC 0.78 and 0.81, P > 0.05) and superior to that of a third-year resident (AUC 0.63, P = 0.04). CONCLUSIONS: A snowman or ice cream cone shape and lower TOC ratio were more closely associated with EAMLs than with clear cell RCCs. A CT radiomics model was useful for differentiating EAMLs from clear cell RCCs with better diagnostic performance than an inexperienced radiologist.
PURPOSE: This study aims to assess the computed tomography (CT) findings of renal epithelioid angiomyolipoma (EAML) and develop a radiomics-based model for differentiating EAMLs and clear cell renal cell carcinomas (RCCs). METHOD: This two-center retrospective study included 28 histologically confirmed EAMLs and 56 size-matched clear cell RCCs with preoperative three-phase kidney CTs. We conducted subjective image analysis to determine the CT parameters that can distinguish EAMLs from clear cell RCCs. Training and test sets were divided by chronological order of CT scans, and radiomics model was built using ten selected features among radiomics and CT features. The diagnostic performance of the radiomics model was compared with that of the three radiologists using the area under the receiver-operating characteristic curve (AUC). RESULTS: The mean size of the EAMLs was 6.2 ± 5.0 cm. On multivariate analysis, a snowman or ice cream cone tumor shape (OR 16.3; 95% CI 1.7-156.9, P = 0.02) and lower tumor-to-cortex (TOC) enhancement ratio in the corticomedullary phase (OR 33.4; 95% CI 5.7-197, P < 0.001) were significant independent factors for identifying EAMLs. The diagnostic performance of the radiomics model (AUC 0.89) was similar to those of genitourinary radiologists (AUC 0.78 and 0.81, P > 0.05) and superior to that of a third-year resident (AUC 0.63, P = 0.04). CONCLUSIONS: A snowman or ice cream cone shape and lower TOC ratio were more closely associated with EAMLs than with clear cell RCCs. A CT radiomics model was useful for differentiating EAMLs from clear cell RCCs with better diagnostic performance than an inexperienced radiologist.
Authors: Jun H Lei; Liang R Liu; Qiang Wei; Tu R Song; Lu Yang; Hai C Yuan; Yong Jiang; Huan Xu; Sheng H Xiong; Ping Han Journal: Sci Rep Date: 2015-05-05 Impact factor: 4.379