Abstract

Renal Cell Carcinoma (RCC) is the most common form of kidney cancer found in adults. Due to varying types of RCC such as oncocytoma and papillary RCC, distinguishing between these sub-types becomes vital for identifying and treating patients. A ResNet18 and EfficientNetV2 Small model was trained and evaluated on various whole slide images using a patch-based approach. These images consisted of both surgical and biopsy resection slides and were labelled with their corresponding diagnoses. In our study, the two models were combined in a feature fusion approach to predict a diagnosis on a whole slide image level. With the combined model reaching an accuracy of nearly 87%, there is potential for an accurate and usable method of diagnosing and distinguishing between different sub-types for RCC with deep learning and feature fusion.