Abstract
Background: The Mayo imaging classification (MIC) for polycystic kidney disease (PKD) is a crucial basis for clinical treatment decisions; however, the volumetric assessment for its evaluation remains tedious and inaccurate. While the ellipsoid method for measuring the total kidney volume (TKV) in patients with PKD provides a practical TKV estimation using computed tomography (CT), its inconsistency and inaccuracy are limitations, highlighting the need for improved, accessible techniques in real-world clinics. Methods: We compared manual ellipsoid and artificial intelligence (AI)-based kidney volumetry methods using a convolutional neural network-based segmentation model (3D Dynamic U-Net) for measuring the TKV by assessing 32 patients with PKD in a single tertiary hospital. Results: The median age and average TKV were 56 years and 1200.24 mL, respectively. Most of the patients were allocated to Mayo Clinic classifications 1B and 1C using the ellipsoid method, similar to the AI volumetry classification. AI volumetry outperformed the ellipsoid method with highly correlated scores (AI vs. nephrology professor ICC: r = 0.991, 95% confidence interval (CI) = 0.9780-0.9948, p < 0.01; AI vs. trained clinician ICC: r = 0.983, 95% CI = 0.9608-0.9907, p < 0.01). The Bland-Altman plot also showed that the mean differences between professor and AI volumetry were statistically insignificant (mean difference 159.5 mL, 95% CI = 11.8368-330.7817, p = 0.07). Conclusions: AI-based kidney volumetry demonstrates strong agreement with expert manual measurements and offers a reliable, labor-efficient alternative for TKV assessment in clinical practice. It is helpful and essential for managing PKD and optimizing therapeutic outcomes.