Automated body composition quantification from non-contrast CT for urolithiasis classification and exploratory...

Area of Science:

• Radiology
• Artificial Intelligence
• Urology

Background:

• Urolithiasis classification and incident risk stratification are crucial for patient management.
• Non-contrast CT is widely used, but extracting detailed body composition data for stone risk is challenging.

Purpose of the Study:

• To develop and validate a deep learning framework for body composition quantification from non-contrast CT.
• To utilize this framework for urolithiasis classification and incident stone risk stratification.

Main Methods:

• Retrospective multicenter study with 781 participants (classification and longitudinal cohorts).
• Automated segmentation of L1/L3 muscle and fat from CT images.
• Development and evaluation of clinical, radiomics, and combined deep learning models.

Main Results:

• The combined muscle-clinical model achieved high external classification performance (AUC: 0.90).
• In the longitudinal cohort, the model identified 75% of incident urolithiasis cases with 89% specificity.
• Preliminary association demonstrated between model-derived risk status and subsequent stone events.

Conclusions:

• Automated body composition analysis from non-contrast CT is effective for urolithiasis classification.
• The model shows preliminary potential for stratifying incident stone formation risk.
• Prospective validation is warranted to confirm these findings.