Chronic Kidney Disease Diagnosis

This retrospective study developed a deep learning radiomics model using 4,365 grayscale renal ultrasound images from 1,049 patients to screen for chronic kidney disease (CKD) and determine disease stage. By combining ResNet34-derived deep features with texture features, the model achieved superior performance compared with senior physicians for overall CKD detection (AUC 0.918 vs 0.869) and particularly for early-stage CKD (G1–G3). Diagnostic performance for advanced stages (G4–G5) was comparable to senior physicians, highlighting the model’s potential value for early CKD screening.