Polycystic Kidney Disease Segmentation

This study developed and deployed a U-Net–based deep learning model with an EfficientNet encoder to automatically measure total kidney volume (TKV) from T2-weighted MRI in patients with autosomal dominant polycystic kidney disease (ADPKD). Trained on 213 MRI studies, the model demonstrated excellent segmentation accuracy in both external (DSC 0.98) and prospective validation cohorts (DSC 0.97), with minimal bias on Bland-Altman analysis. In clinical deployment, model-assisted annotation reduced expert contouring time by 51%, supporting its accuracy, efficiency, and real-world applicability for automated TKV estimation.