Diagnosis of Sarcopenia Using Convolutional Neural Network Models Based on Muscle Ultrasound Images: Prospective...

Area of Science:

• Gerontology
• Medical Imaging
• Artificial Intelligence

Background:

• Sarcopenia detection is crucial but complex and inaccessible.
• Current screening methods for sarcopenia involve muscle mass, strength, and function assessments.
• Simplified and accessible diagnostic tools are needed for early sarcopenia detection.

Purpose of the Study:

• To develop a convolutional neural network (CNN) model using ultrasound images for sarcopenia diagnosis.
• To simplify and enhance the accessibility of the sarcopenia diagnostic process.
• To identify the most effective data type and model architecture for accurate sarcopenia detection.

Main Methods:

• A prospective study evaluated 357 participants, collecting muscle ultrasound images, clinical, and laboratory data.
• Monomodal, bimodal, and multimodal models were developed and compared for diagnostic performance.
• The optimal model (SARCO model) was identified and validated externally and through proof-of-concept testing.

Main Results:

• The ultrasound image-based monomodal model showed the highest performance (AUC: 0.827, F1-score: 0.738).
• Ultrasound images were the dominant contributor in multimodal models for diagnosing sarcopenia.
• The final SARCO model achieved satisfactory performance in external (AUC: 0.801) and proof-of-concept (AUC: 0.757) validations.

Conclusions:

• Ultrasound images play a dominant role in sarcopenia diagnosis via AI models.
• The SARCO model, utilizing ultrasound images, offers a convenient and accurate solution for sarcopenia diagnosis.
• AI-based ultrasound analysis holds promise for improving early sarcopenia detection and management.