Specification of Neck Muscle Dysfunction through Digital Image Analysis Using Machine Learning.

Area of Science:

• Biomedical Engineering
• Musculoskeletal Imaging
• Rehabilitation Science

Background:

• Neck pain is a prevalent condition often perceived as muscle stiffness.
• Objective evidence of increased neck muscle stiffness in individuals with chronic neck pain remains unclear.
• Ultrasound elastography offers a non-invasive method to assess muscle mechanical properties.

Purpose of the Study:

• To objectively investigate differences in neck muscle stiffness between women with chronic neck pain and asymptomatic controls.
• To develop and validate machine learning models for classifying neck pain based on ultrasound elastography data.
• To identify potential ultrasound elastography-based biomarkers for neck muscle dysfunction.

Main Methods:

• Acquisition of 1099 ultrasound elastography images from 38 adult women (20 with chronic neck pain, 18 asymptomatic).
• Extraction of 323 numerical features from shear wave velocity images and segments for machine learning.
• Supervised binary classification using six machine learning algorithms, with Random Forest being the primary model.

Main Results:

• The Random Forest model achieved an Area Under the Curve (AUC) of 0.898 in distinguishing between groups.
• Features derived from the M. multifidus (deepest neck muscles) were most effective in classifying elastograms.
• Summary images and Hotelling's T2 maps visualized and statistically confirmed group differences.

Conclusions:

• Ultrasound elastography, combined with machine learning, can objectively differentiate neck muscles of individuals with and without chronic neck pain.
• Deep neck muscles, particularly the M. multifidus, show distinct elastographic properties associated with neck pain.
• This approach holds potential for identifying objective biomarkers of muscle dysfunction in neck pain.