Purpose or Case Report: Pediatric kidney abnormalities, such as urinary tract dilation (UTD), are common, and current reporting methods vary significantly. There's a need for more precise evaluation methods. Using automated 3D kidney volume measurements from ultrasound (US) offers an alterantive objective approach. Our aim is to evaluate the diagnostic potential of automated 3D kidney volume measurements for detecting and categorizing UTD, comparing their accuracy with clinical assessment.
Methods & Materials: Renal US images from 37 patients were analyzed using a novel 3D segmentation tool to measure various kidney volumes (total renal volume, pelvicalyces volume, intrarenal pelvicalyces, extrarenal pelvis, and parenchyma) and computed HydroIndex values for the collecting system and inside the capsule. Kidneys were categorized as normal or having UTD. Each measurement was compared between the two groups using Mann Whitney test. A logistic regression model assessed the diagnostic performance of 3D kidney measurements and compared it to 2D US assessment.
Results: Six out of seven measurements showed significant differences (p<0.05) between UTD and normal. Key disparities included increased total renal volume (110.32 ± 18.9 vs. 55.04 ± 8.28), pelvicalyces volume (33.76 ± 6.76 vs. 1.49 ± 0.4), intrarenal pelvicalyces volume (26.98 ± 5.65 vs. 1.49 ± 0.4), extrarenal pelvis volume (8.03 ± 2.55 vs. 0.0 ± 0), and, reduced HydroIndex in the collecting system (0.73 ± 0.02 vs. 0.97 ± 0.01) and collecting system inside the capsule (0.77 ± 0.02 vs. 0.97 ± 0.01) ,for UTD cases compared to normal.

The model based on 3D kidney measurements exhibited strong diagnostic performance, with an AUC of 0.96, sensitivity of 87.2%, and specificity of 92% in distinguishing normal from UTD cases. Notably, individual features within the model exhibited varying levels of accuracy, with AUC values ranging from 0.62 to 0.96. Specifically, parenchymal volume had the lowest AUC (0.62), while, pelvicalyces volume showed the highest AUC (0.97) among the features.

When comparing the performance of the 3D kidney volume model to clinical assessment, a strong correlation (correlation coefficient r=0.85, p<0.001) was observed.
Conclusions: 3D kidney volume measurements significantly differentiated individuals with UTD from those without, suggesting diagnostic potential. The diagnostic model using these measurements effectively distinguishes between the two groups, supporting the integration of 3D kidney measurements into clinical practice for UTD identification.