Multimodal deep learning model for AI-based functional prognostic risk stratification in patients undergoing radical nephrectomy

Making the decision between technically challenging partial nephrectomy (PN) and radical nephrectomy (RN) in patients with complex renal cell carcinoma (RCC) remains a significant challenge for urologists. Rapid glomerular filtration rate (GFR) decline (annual decline >3 mL/min/1.73 m²) after RN is considered an abnormal renal function state, and if this risk can be predicted preoperatively, PN may be pursued even when technically demanding. We retrospectively analyze contrast-enhanced computed tomography images and clinical data from 1621 patients across multiple centers. A multimodal deep learning model is developed to predict rapid GFR decline after RN. The model achieves an area under the curve of 0.788–0.873 in external test sets. It stratifies patients into high- and low-risk groups with significantly different risks of chronic kidney disease progression. Here we show that the model demonstrates potential for assisting treatment decisions in patients with complex RCC for whom PN is challenging but feasible. Deciding between partial nephrectomy (PN) or radical nephrectomy (RN) in patients with complex renal cell carcinoma (RCC) remains a significant clinical challenge. Here, the authors develop an AI model for functional risk stratification in complex RCC using preoperative computed tomography images and clinical data, aiming to assist treatment choice for complex RCC patients, for whom PN is challenging but remains feasible.