Radiomics-Based Machine Learning for Sarcopenia Detection in Abdominal and Low-Dose CT

Background: Sarcopenia, characterized by progressive loss of skeletal muscle mass and function, is becoming increasingly prevalent with the global population aging. Computed tomography (CT) is widely used for muscle assessment; however, concerns regarding radiation exposure have prompted interest in lower-dose imaging protocols. This study investigated the performance of radiomics-based machine learning (ML) models for sarcopenia detection using abdominal CT (APCT) and low-dose CT (LDCT). Methods: Radiomics features were extracted from CT images following skeletal muscle segmentation, and ML models were developed using logistic regression, support vector machine, and random forest. Model performance was evaluated using fivefold cross-validation with out-of-fold predictions. Results: The random forest model demonstrated the best performance among the evaluated models, achieving an area under the receiver operating characteristic curve of 0.720 (95% CI: 0.532–0.881) for APCT and 0.692 (95% CI: 0.573–0.801) for LDCT. Model interpretation using SHapley Additive exPlanations analysis identified several intensity-based radiomics features, including TotalEnergy, as important contributors to sarcopenia prediction. Conclusions: These findings suggest that radiomics features derived from LDCT images may provide useful information for sarcopenia detection. Because LDCT is widely used in clinical settings such as lung cancer screening, radiomics analysis of LDCT images may offer an additional opportunity for opportunistic sarcopenia assessment.