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

Area of Science:

• Radiology
• Artificial Intelligence
• Gerontology

Background:

• Sarcopenia, a progressive loss of muscle mass and function, is a growing concern due to global population aging.
• Computed tomography (CT) is a standard for muscle assessment, but radiation exposure is a limitation.
• Lower-dose imaging protocols are being explored to mitigate radiation risks.

Purpose of the Study:

• To evaluate the efficacy of radiomics-based machine learning (ML) models for sarcopenia detection.
• To compare model performance using standard-dose abdominal CT (APCT) and low-dose CT (LDCT).

Main Methods:

• Radiomics features were extracted from segmented skeletal muscle on CT images.
• Machine learning models, including logistic regression, support vector machine, and random forest, were developed.
• Model performance was assessed via fivefold cross-validation.

Main Results:

• The random forest model showed the highest performance, with an AUC of 0.720 for APCT and 0.692 for LDCT.
• SHapley Additive exPlanations identified intensity-based radiomics features, such as TotalEnergy, as key predictors.
• Radiomics features from LDCT demonstrated potential for sarcopenia detection.

Conclusions:

• Radiomics analysis of LDCT images can provide valuable insights for sarcopenia detection.
• LDCT is frequently used in clinical settings like lung cancer screening, enabling opportunistic sarcopenia assessment.
• This approach may facilitate early identification and management of sarcopenia without additional radiation dose.