Artificial Intelligence–Based Classification of Renal Oncocytic Neoplasms: Advancing From a 2-Class Model of Renal Oncocytoma and Low-Grade Oncocytic Tumor to a 3-Class Model Including Chromophobe Renal Cell Carcinoma

• DOI: 10.5858/arpa.2024-0374-oa
• Publication Date: 2025-02-17T06:57:41Z
• Authors: Katrina Collins; Shubham Innani; Kingsley Ebare; Mohammed Saad; Stephanie E. Sieg...; Sean R. Williamson; Fiona Maclean; Andres Matoso; Ankur Sangoi; Michelle S. Hirsch; Dibson D. Gondim; Andres M. Acosta; Bhakti Baheti; Spyridon Bakas; Muhammad T. Idrees
• Keywords: Renal oncocytoma; Chromophobe cell

Distinguishing between renal oncocytic tumors, such as oncocytoma (RO), and a subset of tumors with overlapping characteristics, including the recently identified low-grade tumor (LOT), can present diagnostic challenge for pathologists owing to shared histopathologic features. To develop an automatic computational classifier stratifying whole slide images biopsy resection specimens into 2 distinct groups: RO LOT. A total 269 from 125 cases across 6 institutions were collected. weakly supervised attention-based multiple-instance-learning deep learning (DL) model was trained initially evaluated through 5-fold cross validation case-level stratification, followed by using independent holdout data set. Quantitative performance evaluation based on accuracy area under receiver operating characteristic curve (AUC). The developed set yielded generalizable performance, average testing 84% (AUC = 0.78), closely aligning 83% 0.92) proposed artificial intelligence approach contributes toward comprehensive solution addressing commonly encountered neoplasms, encompassing well-established entities like along challenging "gray zone" LOT, thereby proving applicable in clinical practice.