Artificial intelligence-based solutions in renal scoring and nephrectomy decision-making processes

Objective: The aim of this study is to develop a novel deep learning model capable of automatically classifying renal tumor complexity based on R.E.N.A.L. nephrometry scores using computed tomography (CT) data, thereby supporting clinical decision-making in partial and total nephrectomy procedures. Materials and Methods: Contrast-enhanced portal phase CT data from 197 patients diagnosed with malignant renal tumors and who underwent nephrectomy at Afyonkarahisar Health Sciences University Urology Department were used. All CT images were resampled to 128×128×64 voxel dimensions, and tumor-containing kidney were manually delineated. A custom-built three-dimensional convolutional neural network (3D CNN) model was developed to classify tumors as "low" or "high" complexity without requiring segmentation preprocessing. Results: The overall classification accuracy of the model was calculated as 90%, with an area under the ROC curve (AUC) of 0.96. All cases in the low-complexity group were correctly classified, achieving a sensitivity of 100%. In the high-complexity group, a sensitivity of 76.9% was obtained. A statistically significant and strong positive correlation was observed between tumor maximum diameter and nephrometry score (r = 0.718; p < 0.001). The model demonstrated high reliability in identifying cases suitable for partial nephrectomy, particularly within the low-complexity group. Conclusion: The developed 3D CNN model offers a rapid and reproducible alternative to manual scoring systems by predicting tumor complexity directly from CT images. This capability has the potential to reduce subjectivity in surgical planning and support urologists in determining whether partial or total nephrectomy is more appropriate. The model's 100% sensitivity in the low-complexity group highlights its promising role as an artificial intelligence-based clinical decision support tool for preoperative planning.