Kidney cancer, specifically renal cell carcinoma (RCC), represents a major global health issue, with increasing incidence rates, highlighting the necessity for effective and accurate diagnostic systems. Traditional methods for RCC grading, such as manual histopathological analysis, are labor intensive, susceptible to human error, and lack the scalability required for modern clinical environments. To address these issues, we propose a novel multiphase classification framework that combines YOLOv8 for high-accuracy RCC grading and GradCAM for enhanced model interpretability. The framework progressively refines WHO/ISUP grades through a cascading approach: M₁ distinguishes between Grades 0/1 and 2/3/4, M₃ differentiates between Grades 0 and 1, and M₃ classifies Grades 2, 3, and 4. A unique "Avoid Error Propagation Layer" ensures that only high-confidence predictions are passed to subsequent phases, reducing cumulative errors. Our approach achieves remarkable accuracy (97. 51%), precision (93. 72%), recall (93. 28%), specificity (98. 32%), and F1 score (93. 35%) on a publicly available RCC dataset. By integrating cutting-edge object detection with visual explanations, our model not only provides superior diagnostic performance but also offers transparent decision-making, which is crucial for clinical adoption. This approach addresses a major shortcoming of existing RCC grading systems by simultaneously prioritizing accuracy and interpretability, promising for revolutionizing kidney cancer diagnosis by enhancing both diagnostic precision and clinician trust.
Bamaqa et al. (Thu,) studied this question.