Artificial intelligence–assisted rectus femoris ultrasound vs. L3 computed tomography for sarcopenia assessment in oncology patients: establishing diagnostic cut-offs for muscle mass and quality

Introduction Sarcopenia is prevalent among oncological patients; its diagnosis remains challenging due to technical limitations. This study aims to compare an AI-assisted ultrasound (US) evaluation of the rectus femoris with the gold-standard L3 computed tomography (CT) scan and to establish diagnostic cut-off values for sarcopenia in oncologic patients. Methods This cross-sectional observational study assessed body composition in oncology patients undergoing treatment, comparing two AI-assisted imaging modalities: L3 vertebral level CT and US of the rectus femoris. Muscle mass was quantified using Skeletal Muscle Area (SMA) and Lean Muscle Area (LMA) from CT, and Rectus Femoris Area (RFA) and Thickness (RFT) from US. Muscle quality was evaluated via skeletal muscle Hounsfield units (SM-HU) on CT, and muscle (Mi) and fat (FATi) percentages within the region of interest on US. CT served as the reference standard for diagnosing malnutrition, sarcopenia, and myosteatosis. Results A total of 337 patients (58.8% male; mean age 69.7 (10.9) years) were analyzed, with malnutrition identified in 78.3%, including 8% with sarcopenia. Ultrasound measurements correlated with CT-derived muscle metrics, including RFA with SMA ( r = 0.44; p 0.01) and LMA ( r = 0.47; p 0.01). Muscle quality parameters from US showed a positive correlation between Mi and SM-HU ( r = 0.27; p 0.01), while FATi negatively correlated with SM-HU ( r = −0.19; p 0.01). Ultrasound-based cut-off points for diagnosing sarcopenia were established at 3.425 cm 2 RFA (AUC: 0.628) and 1.180 cm RFT (AUC: 0.636) for men, and 2.845 cm 2 RFA (AUC: 0.628) and 0.868 cm RFT (AUC: 0.636) for women; and ultrasound-based cut-off points for diagnosing myosteatosis were 46.77% for MiT (AUC: 0.640) and 41.20% for FATi (AUC: 0.623). Derived cut-off values demonstrated high negative predictive value for low muscle mass and high positive predictive value for myosteatosis. Conclusion AI-enhanced rectus femoris ultrasonography is a feasible, non-invasive, and clinically relevant approach for sarcopenia assessment in oncology. AI-assisted ultrasonography showed moderate correlations with CT-derived muscle mass indices and reliably capture qualitative changes in muscle quality and fat infiltration. The AI heralds a new era of rapid, radiation-free body composition assessment.