AI-Enhanced Lower Extremity X-Ray Segmentation: A Promising Tool for Sarcopenia Diagnosis

Background/Objectives: Sarcopenia, characterized by progressive loss of skeletal muscle mass and strength, significantly impacts physical function and quality of life in older adults. Traditional measurement methods like Dual-energy X-ray absorptiometry (DEXA) are often inaccessible in primary care. This study aimed to develop and validate an AI-driven auto-segmentation model for muscle mass assessment using long X-rays as a more accessible alternative to DEXA. Methods: This was a retrospective validation study using data from the Real Hip Cohort at Inha University Hospital in South Korea. 351 lower extremity X-ray images from 157 patients were collected and analyzed. AI-based semantic segmentation models, including U-Net, V-Net, and U-Net++, were trained and validated on this dataset to automatically segment muscle regions. Model performance was assessed using Intersection over Union (IoU) and Dice Similarity Coefficient (DC) metrics. The correlation between AI-derived muscle measurements and the DEXA-derived skeletal muscle index was evaluated using Pearson correlation analysis and Bland-Altman analysis. Results: The study analyzed data from 157 patients (mean age 77.1 years). The U-Net++ architecture achieved the best segmentation performance with an IoU of 0.93 and DC of 0.95. Pearson correlation demonstrated a moderate to strong positive correlation between the AI model's muscle estimates and DEXA results (r = 0.72, *** p < 0.0001). Regression analysis showed a coefficient of 0.74, indicating good agreement with reference measurements. Conclusions: This study successfully developed and validated an AI-driven auto-segmentation model for estimating muscle mass from long X-rays. The model provides an accessible alternative to DEXA, with potential to improve sarcopenia diagnosis and management in community and primary care settings. Future work will refine the model and explore its application to additional muscle groups.