Presentation Authors: Amir Baghdadi, Ahmed A. Hussein, Ahmed S. Elsayed, Naif A. Aldhaam*, Lora A. Cavuoto, Eric Kauffman, Khurshid A. Guru, Buffalo, NY
Introduction: Among the >65,000 new kidney cancer cases in the U.S., about 10% of these patients turn out to have benign renal cell oncocytomas (Onco). They face a 30% risk of complications due to unnecessary tumor resection. The biggest challenge in renal mass biopsy is the differentiation of Onco from chromophobe renal cell carcinoma (ChRCC). Our group recently developed and validated a manual computed tomography (CT)-based method that distinguishes CD117(+) Onco from ChRCC. This study aimed to develop an automated method for this differentiation, which results in saving time and avoiding potential subjective interpretations.
Methods: Retrospective analysis was performed on 20 patients from one institute with pathology proven disease, who underwent the surgical resection of renal tumors between 2009 and 2018. The tumor and kidney images were manually segmented from the CT scans. Peak early-phase enhancement ratio (PEER) was defined as the ratio of signal intensity differences between early and delayed contrast phases for the peak enhancing portion of the tumor compared to the renal cortex. Through a series of image processing operations, a fixed size circular or elliptic shape as a region-of-interest (ROI) within the kidney and an arbitrary elliptical shape ROI within the tumor were identified for measuring the image intensities. Based on the manual development and validation, Onco was associated with a PEER > 0.5. The performance of the automated diagnostic model was compared to the manual expert identification and based on the tumor pathology through accuracy, sensitivity, and specificity along with the root mean square error (RMSE) between the calculated PEERs.
Results: Our image processing algorithm was able to accurately identify and locate the ROI within the renal cortex and tumor (Figure 1). This automated method achieved the accuracy of 95% in tumor type classification (100% sensitivity and 89% specificity) compared to the final pathology results (RMSE of 0.15 for PEER ratio).
Conclusions: To the best of our knowledge, this study provides the first automated framework for differentiating benign and malignant renal tumors with a high accuracy.
Source of Funding: Roswell Park Alliance Foundation