Abdominal CT Body Composition Thresholds Using Automated AI Tools for Predicting 10-year Adverse Outcomes

Area of Science:

• Radiology and Imaging Science
• Artificial Intelligence in Medicine
• Preventive Cardiology and Geriatrics

Background:

• Fully automated artificial intelligence (AI) tools for CT-based body composition analysis show promise for opportunistic screening.
• A significant gap exists in established body composition thresholds linked to adverse clinical outcomes.

Purpose of the Study:

• To establish population and sex-specific thresholds for muscle, abdominal fat, and abdominal aortic calcium using abdominal CT.
• To evaluate the predictive capability of these CT-derived measures for mortality, adverse cardiovascular events, and fragility fractures.

Main Methods:

• Retrospective analysis of noncontrast abdominal CT scans from 9223 asymptomatic adults (2004-2016).
• Application of fully automated AI algorithms to quantify skeletal muscle, abdominal fat (L3 level), and abdominal aortic calcium.
• Longitudinal follow-up for death, cardiovascular events, and fragility fractures; Receiver Operating Characteristic (ROC) curve analysis for threshold derivation (90% specificity).

Main Results:

• Muscle attenuation and aortic calcium demonstrated high diagnostic performance for predicting death (AUC 0.76 men, 0.72 women for muscle attenuation).
• Sex-specific thresholds were significantly higher in men than women for muscle attenuation.
• Key 90% specificity thresholds identified: muscle attenuation (23 HU men, 13 HU women for death/fractures), aortic calcium (1475 Agatston men, 735 Agatston women for death/CVD events).

Conclusions:

• Automated abdominal CT-derived body composition measures, with sex-specific thresholds, can effectively predict the risk of death, adverse cardiovascular events, and fragility fractures.
• These findings support the use of AI-driven body composition analysis in asymptomatic populations for enhanced risk stratification.