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

Area of Science:

• Nephrology
• Oncology
• Artificial Intelligence

Background:

• Deciding between partial nephrectomy (PN) and radical nephrectomy (RN) for complex renal cell carcinoma (RCC) is challenging.
• Rapid glomerular filtration rate (GFR) decline post-RN indicates abnormal renal function and impacts treatment choices.

Purpose of the Study:

• To develop and validate a multimodal deep learning model to predict rapid GFR decline after RN.
• To assist urologists in treatment decisions for complex RCC patients.

Main Methods:

• Retrospective analysis of contrast-enhanced computed tomography images and clinical data from 1621 patients.
• Development of a multimodal deep learning model for predicting GFR decline.
• External validation of the model's predictive performance.

Main Results:

• The deep learning model achieved an area under the curve of 0.788-0.873 in external test sets.
• The model successfully stratified patients into high- and low-risk groups for chronic kidney disease progression.
• The model shows potential for aiding clinical decision-making in complex RCC cases.

Conclusions:

• A deep learning model can predict rapid GFR decline after RN with high accuracy.
• This predictive tool can support urologists in selecting nephrectomy strategies for complex RCC.
• The model may help preserve renal function by guiding the decision towards PN when feasible.