Machine learning classification of functional neurological disorder using structural brain MRI features.

Area of Science:

• Neuroscience
• Radiology
• Machine Learning

Background:

• Previous functional neurological disorder (FND) research relied on univariate brain imaging, limiting individual patient classification.
• Group-level differences in grey matter between FND patients and healthy controls (HCs) have been reported but lack clinical translatability.
• Individual-level differentiation is crucial for accurate diagnosis and understanding FND pathophysiology.

Purpose of the Study:

• To investigate the potential of using machine learning classifiers with structural MRI grey matter features to differentiate individuals with FND from control groups.
• To assess the classification accuracy for mixed FND, FND-motor, and FND-seizure subtypes against healthy and psychiatric controls.
• To identify key brain regions contributing to successful classification.

Main Methods:

• 183 participants (61 FND-mixed, 61 HCs, 61 psychiatric controls) were recruited.
• Support vector machine classifiers analyzed 134 FreeSurfer-derived grey matter MRI features.
• Cross-validation was employed to distinguish individuals with FND from HCs and psychiatric controls.

Main Results:

• Classifiers differentiated FND-mixed from HCs with 66% accuracy (AUROC=0.74) and from psychiatric controls with 60% accuracy (AUROC=0.56).
• The FND-motor subtype was robustly differentiated from HCs with 72% accuracy (AUROC=0.80).
• Key differentiating brain regions included the cingulate gyrus, hippocampal subfields, and amygdalar nuclei.

Conclusions:

• Structural MRI, analyzed with machine learning, shows feasibility for classifying FND at an individual level.
• Findings highlight the link between brain structure and FND pathophysiology.
• Further validation with larger cohorts and diverse control groups is recommended.