Multi-centre classification of functional neurological disorders based on resting-state functional connectivity.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Functional Neurological Disorder (FND) presents disabling symptoms without clear organic causes, complicating diagnosis.
• Current diagnostic methods for FND often rely on clinical signs but may require extensive, costly tests.
• Resting-state functional connectivity (RS FC) shows promise as an adjunctive imaging biomarker for FND diagnosis.

Purpose of the Study:

• To assess the robustness of a machine learning classification approach using RS FC in a multi-center setting.
• To evaluate the applicability of RS FC for distinguishing FND patients from healthy controls across different scanners and locations.
• To test the generalizability of RS FC-based classification in a real-world, multi-center clinical context.

Main Methods:

• A multivariate machine learning approach was applied to whole-brain RS FC data from 86 FND patients and 86 healthy controls across four centers.
• Intra-center cross-validation replicated previous findings, while pooled cross-validation assessed robustness against inter-scanner variability.
• Inter-center cross-validation evaluated the method's generalizability by training on data from some centers and testing on data from others.

Main Results:

• The RS FC classification approach successfully distinguished FND patients from controls with accuracies ranging from 70% to 74% in individual centers and 72% in pooled data.
• The classifier's performance remained robust despite variations in scanners and centers, surviving adjustments for clinical factors like anxiety and depression.
• Discriminant features included specific brain regions such as the angular/supramarginal gyri, sensorimotor, cingulate, insular cortices, and hippocampus.
• Inter-center validation, however, did not exceed chance levels, indicating a need for further generalization strategies.

Conclusions:

• RS FC demonstrates applicability and robustness in distinguishing FND patients from healthy controls across different centers and scanners.
• The findings support the potential of RS FC as an adjunctive diagnostic tool for FND in multi-center settings.
• Future research should focus on optimizing acquisition parameters and including diverse control groups to enhance clinical generalizability.