MyoVision-US: an Artificial Intelligence-Powered Software for Automated Analysis of Skeletal Muscle Ultrasonography

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Manual analysis of muscle ultrasound images for morphometry is time-consuming and requires extensive training.
• Objective quantification of muscle morphometry is crucial for clinical practice and research.
• Existing methods face challenges in efficiency and consistency.

Purpose of the Study:

• To develop and validate an AI-powered software tool (MyoVision-US) for automated lower limb muscle ultrasound analysis.
• To assess the consistency and predictability of the AI tool against expert manual analysis.
• To evaluate the AI tool's performance across diverse patient populations (healthy, acute, and chronic illness).

Main Methods:

• Deep learning model (MyoVision-US) developed for automated muscle morphometry analysis.
• Quadriceps complex (QC) and tibialis anterior (TA) muscle ultrasound images captured from healthy, ICU, and lung cancer subjects.
• Intraclass correlation coefficients (ICC) used to determine consistency between manual and automated analyses.
• Adjusted linear regression (adj.R ²) used to calculate predictability.

Main Results:

• AI analysis reduced image processing time by approximately 99.8% (247 seconds vs. 24 hours).
• Good to excellent consistency (ICC: 0.85-0.99) observed between AI and manual analyses for QC and TA muscles.
• Strong predictability (adj.R ²: 0.56-0.97) of AI measurements compared to manual analysis across all subject groups.
• AI demonstrated high consistency even for critically ill and lung cancer patient images.

Conclusions:

• AI-driven automation of lower limb muscle ultrasound analysis offers excellent consistency and strong predictability.
• The MyoVision-US tool significantly enhances efficiency in muscle morphometry quantification.
• Future research should extend AI models to other skeletal muscle groups.