nnU-Net for CT Segmentation

This study investigated whether deep learning could automate CT-based volumetric segmentation of Wilms’ tumor to reduce inter-radiologist variability and improve treatment guidance. Using intravenous-phase CT scans from 105 patients, an nnU-Net model achieved good segmentation performance, and further optimization incorporating three-dimensional tumor diameter significantly reduced tumor volume estimation errors. The combined AI and clinical feature approach outperformed AI segmentation alone, demonstrating potential to enhance volumetric assessment and clinical decision-making in pediatric Wilms’ tumor and other childhood diseases.