A rapid automated segmentation method for total kidney volume measurement on unenhanced computed tomography in autosomal dominant polycystic kidney disease.

PURPOSE: To develop and validate a fully automated deep learning-based method for total kidney volume (TKV) measurement from unenhanced computed tomography (CT) images in patients with autosomal dominant polycystic kidney disease (ADPKD). METHODS: A fully automated kidney segmentation model based on the nnU-Net framework was developed using unenhanced CT images from 236 patients, including 111 patients with ADPKD. The development dataset was divided into a training cohort and an internal testing cohort. An independent cohort of 70 patients with ADPKD was used for validation, including longitudinal scans from 20 patients. The reference standard consisted of semi-automated kidney segmentation followed by slice-by-slice manual correction performed independently by two experienced radiologists. Automated and reference segmentations were compared using the Dice similarity coefficient (DSC), intraclass correlation coefficient (ICC), linear regression and Bland-Altman analysis. RESULTS: Automated segmentation exhibited excellent spatial agreement with the reference standard, with a mean DSC of 0.96 ± 0.01 for TKV. Volumetric agreement was near perfect (ICC = 0.999; 95% confidence interval, 0.998-0.999), with minimal bias (-0.14%) and high precision (±2.09%). Longitudinal analysis demonstrated robust agreement for kidney growth assessment (ICC = 0.99; bias, - 0.64%; precision, ± 4.81%). The automated method required approximately 5 s per case, compared with 30 min for expert-assisted reference segmentation. CONCLUSION: This automated segmentation method enables fast, accurate, and reproducible TKV quantification from unenhanced CT images in patients with ADPKD, offering a clinically feasible alternative to manual volumetric analysis.