Artificial intelligence for body composition assessment focusing on sarcopenia.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Gerontology and Geriatric Medicine

Background:

• Conventional methods for measuring skeletal muscle mass to diagnose sarcopenia have limitations in simplicity, reproducibility, and convenience.
• Accurate and efficient sarcopenia diagnosis is crucial for managing age-related muscle loss and its associated health consequences.

Purpose of the Study:

• To introduce and evaluate an artificial intelligence (AI) system for direct computed tomography (CT) analysis to measure skeletal muscle mass for sarcopenia diagnosis.
• To compare the accuracy, speed, reproducibility, and convenience of the AI system against conventional measurement methods.

Main Methods:

• Developed AI models (Deeplabv3 for body region detection, EfficientNetV2-XL for sarcopenia diagnosis) trained on CT scans from 3096 cases.
• Calculated Skeletal Muscle Index (SMI) using AI and conventional methods, comparing results for agreement and diagnostic changes.
• Evaluated AI performance using metrics including intersection over Union (IoU), sensitivity, specificity, and positive predictive value.

Main Results:

• Conventional methods showed low agreement (κ coefficients 0.478-0.236) and diagnostic changes in 43% of cases.
• The AI system demonstrated robust body region detection (IoU=0.93) and consistently produced identical results across measurements.
• AI sarcopenia diagnosis achieved high accuracy: 82.3% sensitivity, 98.1% specificity, and 89.5% positive predictive value.

Conclusions:

• The developed AI system offers superior reproducibility and convenience compared to conventional sarcopenia diagnostic methods.
• The AI system exhibits high diagnostic accuracy, presenting a promising alternative to overcome the limitations of current approaches.
• AI-driven CT analysis facilitates efficient and reliable sarcopenia diagnosis, potentially improving patient management.