Anomaly Detection of Moderate Traumatic Brain Injury Using Auto-Regularized Multi-Instance One-Class SVM

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Accurate detection of functional deficits from moderate traumatic brain injury (mTBI) is critical for effective clinical management and rehabilitation.
• Magnetoencephalography (MEG) offers a non-invasive method to assess brain activity and functional connectivity.

Purpose of the Study:

• To develop and validate a machine learning model using MEG-based functional connectivity features for detecting mTBI.
• To quantify synchronized brain activity patterns indicative of functional deficits in mTBI patients.

Main Methods:

• Utilized magnetoencephalography (MEG) to capture brain activity, focusing on magnitude squared coherence (MSC) and phase lag index (PLI) for functional connectivity analysis.
• Developed a multi-instance one-class support vector machine (SVM) model trained on healthy control data.
• Identified mTBI cases as anomalies deviating from the SVM decision boundary, optimized using data from patients with Glasgow Coma Scale (GCS) scores between 9 and 13.

Main Results:

• The proposed multi-instance one-class SVM model achieved high accuracy (94.19%) and sensitivity (90.00%) in detecting mTBI at the high beta band.
• Model performance was validated against magnetic resonance imaging (MRI) data.
• The model effectively identified deviations from normal brain activity patterns characteristic of mTBI.

Conclusions:

• The study demonstrates the efficacy of a multi-instance one-class SVM approach utilizing MEG functional connectivity for mTBI detection.
• This method provides a promising tool for objective quantification of functional deficits in mTBI.
• The findings support the integration of advanced machine learning techniques with neuroimaging for improved mTBI diagnostics.