ETRC-net: Efficient transformer for grading renal cell carcinoma in histopathological images

Abstract

Renal cell carcinoma (RCC), the most prevalent form of kidney cancer, accounts for nearly 85 % of kidney cancer-related deaths. Manual diagnosis of RCC from histopathology images relies heavily on the expertise of pathologists, often leading to variability in results. Although deep learning methods have been explored for disease diagnosis, research on RCC remains limited, and existing approaches are insufficient for accurate grading. Since each RCC stage requires a distinct treatment plan, reliable grading is crucial, as errors can result in inappropriate therapies and poor patient outcomes. To address this challenge, we propose the Efficient Transformer for Renal Classification Network (ETRCNet), a novel deep learning framework specifically designed for accurate RCC classification from histopathology images. ETRCNet combines EfficientNet with Squeeze-and-Excitation (SE) blocks for enhanced feature representation and a customized Vision Transformer encoder to capture global context and long-range dependencies. The SE blocks adaptively recalibrate channel-wise responses, enabling the model to focus on relevant features while suppressing less informative ones. We evaluate ETRCNet on the Kasturba Medical College (KMC) dataset, achieving 94.37 % accuracy, 94.54 % precision, 94.37 % recall, and an F1-score of 94.37 %. On the Lung and Colon dataset, it further demonstrates superior generalization with 99.92 % accuracy, 99.64 % precision, 99.71 % recall, and a 99.80 % F1-score. Compared to state-of-the-art methods, ETRCNet delivers higher accuracy with fewer trainable parameters and lower computational cost. Its efficiency and scalability make itfor resource constrained clinical environments, offering a robust and intelligent solution for early RCC diagnosis.