Machine Learning Diagnostic Modeling for Classifying Fibromyalgia Using B-mode Ultrasound Images

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning in Medicine

Background:

• Fibromyalgia (FM) diagnosis is challenging due to the absence of objective diagnostic markers.
• Quantitative ultrasound (US) techniques, including image texture analysis, show promise for diagnosing chronic pain conditions.
• Machine learning (ML) models can potentially differentiate between healthy and diseased tissues using US data.

Purpose of the Study:

• To develop and compare two ML models, Support Vector Machine (SVM) and logistic regression, for differentiating trapezius muscle in healthy individuals versus FM patients.
• To utilize image texture variables derived from quantitative US to build these diagnostic models.

Main Methods:

• Ultrasound videos of trapezius muscles were acquired from 51 healthy participants and 57 FM patients.
• Skeletal muscle regions of interest (ROIs) were extracted and filtered using complex wavelet structural similarity index (CW-SSIM).
• Thirty-one texture variables were extracted from ROIs to train and validate SVM and elastic net regularized logistic regression models via nested cross-validation.

Main Results:

• The SVM model achieved a predicted generalized performance accuracy of 83.9 ± 2.6%, validated at 84.1% on a holdout test set.
• The logistic regression model achieved a predicted generalized performance accuracy of 65.8 ± 1.7%, validated at 66.0% on the holdout test set.
• Both models differentiated between healthy and FM trapezius muscle, but only the SVM model demonstrated clinically relevant performance.

Conclusions:

• The SVM model, utilizing quantitative ultrasound image texture analysis, shows significant potential as an objective tool for fibromyalgia diagnosis.
• This approach offers a clinically relevant method for differentiating trapezius muscle characteristics between healthy individuals and FM patients.
• Further research is warranted to validate and implement this ML-based US technique in clinical settings for improved FM diagnosis.