Automatic Tracking of the Muscle Tendon Junction in Healthy and Impaired Subjects using Deep Learning^.

Area of Science:

• Biomechanics
• Medical Imaging
• Artificial Intelligence

Background:

• Understanding muscle and tendon mechanics is crucial for diagnosing movement disorders.
• Current methods for tracking muscle tendon junction (MTJ) displacements are often manual or semi-automatic, limiting efficiency and reproducibility.
• Ultrasound imaging offers a non-invasive window into musculoskeletal dynamics, but precise MTJ localization remains challenging.

Purpose of the Study:

• To develop a fully-automatic deep learning-based method for detecting the muscle tendon junction (MTJ) position in ultrasound images.
• To enable precise and objective quantification of MTJ displacements during various movements.
• To provide a robust tool for analyzing muscle and tendon behavior separately.

Main Methods:

• A novel deep learning approach utilizing an attention mechanism was employed for MTJ detection in ultrasound images.
• The dataset comprised ultrasound images from 79 healthy subjects and 28 subjects with movement limitations.
• Subjects performed passive full range of motion and maximum contraction movements to capture diverse scenarios.

Main Results:

• The trained deep learning network demonstrated robust MTJ detection across a diverse dataset with varying image quality.
• The method achieved a mean absolute error of 2.55 ± 1 mm for MTJ position detection.
• The approach proved effective for various subjects and capable of real-time operation.

Conclusions:

• The developed deep learning method provides an accurate and automated solution for tracking MTJ displacements in ultrasound.
• This technique facilitates objective analysis of muscle and tendon biomechanics, aiding in the assessment of movement capabilities.
• The open-source availability of the software promotes further research and clinical application in musculoskeletal diagnostics.