Using Minimal-Redundant and Maximal-Relevant Whole-Brain Functional Connectivity to Classify Bipolar Disorder

Area of Science:

• Neuroscience
• Psychiatry
• Medical Imaging

Background:

• Bipolar disorder (BD) is often misdiagnosed, highlighting the need for objective diagnostic markers.
• Dysconnectivity in prefronto-limbic-striatal regions is a key feature of BD.
• Functional connectivity may serve as a reliable neurobiological marker for BD diagnosis.

Purpose of the Study:

• To investigate the utility of functional connectivity in differentiating bipolar disorder patients from healthy controls.
• To identify specific functional connectivity patterns associated with bipolar disorder.

Main Methods:

• A machine learning approach using support vector machines (SVM) and feature selection (minimum redundancy/maximum relevance) was employed.
• Data from 192 BD patients and 173 healthy controls (HC) were analyzed using a 10-fold cross-validation strategy.
• Functional connectivity matrices were analyzed for within-group and between-group similarities.

Main Results:

• The classification model achieved a mean accuracy of 76.25% and an area under the curve (AUC) of 0.840.
• Key differentiating features involved functional connectivity within the subcortical/cerebellar regions and the frontoparietal network.
• Functional connectivity exhibited greater similarity within the BD group after feature selection compared to between groups.

Conclusions:

• The developed approach successfully distinguishes BD patients from HCs using functional connectivity.
• The findings suggest that functional connectivity patterns, particularly those showing increased within-group similarity in BD patients, are relevant to the neuropathology of the disorder.
• This method offers a potential objective tool to aid in the diagnosis of bipolar disorder.