Predicting Renal Tumor Pathology From Intraoperative Gross Appearance: An AI-Based Pilot Study

Area of Science:

• Urology
• Artificial Intelligence
• Surgical Pathology

Background:

• Intraoperative differentiation of renal tumors is crucial for surgical planning and patient outcomes.
• Current methods rely on surgeon experience and may lack definitive accuracy.
• AI offers potential for objective, real-time pathological assessment.

Purpose of the Study:

• To assess the feasibility of using deep learning models, specifically CNNs and Vision Transformers (ViTs), for intraoperative renal tumor classification based on gross appearance.
• To compare the performance of a CNN (ResNet50) and a ViT (GSViT) in distinguishing between common renal tumor subtypes.

Main Methods:

• Retrospective analysis of 443 intraoperative images from partial nephrectomies (2008-2024).
• Training CNN and ViT models to classify six renal tumor types: clear cell RCC, papillary RCC, chromophobe RCC, hybrid oncocytic tumors, oncocytoma, and angiomyolipoma.
• Models were evaluated using accuracy, AUC-ROC, and confusion matrices on held-out test data.

Main Results:

• The CNN achieved moderate performance in binary classification tasks, with AUCs ranging from 0.70 to 0.74 for specific tumor types.
• Multi-class CNN performance showed variability, with notable AUCs for papillary RCC (0.70) and oncocytoma (0.71).
• The GSViT model underperformed compared to the CNN and exhibited prediction bias towards clear cell RCC; overfitting occurred when unfreezing pretrained backbones.

Conclusions:

• CNN models show a moderate capability for intraoperative renal tumor classification using gross appearance, supporting the concept of AI-assisted surgical decision-making.
• The study highlights the need for larger datasets and external validation to enable clinical application of these AI tools.