CYSTSCAN–PKD: a comprehensive pipeline for automatic cyst segmentation and counting on µCT scans from PKD animal models

Autosomal dominant polycystic kidney disease (ADPKD) is a genetic disorder causing progressive renal cyst formation, increased kidney volume, and impaired function. The PCK rat is a preclinical model for investigating new treatments, requiring accurate quantification of total kidney volume (TKV), total cyst volume (TCV), and cyst count. We propose an automated segmentation pipeline of kidneys and cysts on µCT scans of excised rat kidneys, followed by automated cyst counting. For segmentation, a 3D U-Net ensemble was implemented using the nnU-Net framework with dual-channel input (raw µCT volumes, Sobel-filtered images). Models were trained on Dataset 1 subsets (D1, n = 5, 10, 15, 20), and evaluated on internal test set (n = 5) and independent external Dataset 2 (D2, n = 5). Segmentation achieved Dice Similarity Coefficients > 0.99 for kidney and > 0.98 for cysts on D1, with comparable performance on D2, exploring cross-scanner generalizability. For cyst counting, a morphological algorithm based on 3D distance transform and peak detection was optimized via genetic algorithm on 10 samples and evaluated on independent 10-sample set. The automated method achieved low variability, with operator-algorithm agreement exceeding inter-operator agreement across all metrics, drastically reducing processing time. The pipeline provides fast, accurate, and reproducible quantification of morphological biomarkers required for preclinical PKD research.