A machine learning pipeline for efficient differentiation between bipolar and major depressive disorder based on...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Major depressive disorder (MDD) and bipolar disorder (BD) share overlapping depressive symptoms, leading to frequent misdiagnosis.
• Approximately 60% of bipolar disorder patients are initially misdiagnosed as having MDD.
• Reliable biomarkers are needed to support accurate diagnosis and treatment of these conditions.

Purpose of the Study:

• To optimize a machine learning pipeline for differentiating between depressed patients with bipolar disorder (BD) and major depressive disorder (MDD).
• To investigate the utility of multimodal structural neuroimaging features as biomarkers for distinguishing BD from MDD.

Main Methods:

• Acquired Diffusion Tensor Imaging (DTI) and T1-weighted MRI data from 282 patients (180 BD, 102 MDD).
• Preprocessed images to extract DTI metrics (AD, RD, MD, FA) and Voxel-Based Morphometry (VBM) data.
• Utilized a 5-fold nested cross-validated pipeline with confound regression, standardization, PCA, and elastic-net regression for classification.

Main Results:

• DTI-based models achieved classification accuracies of 75-78%, significantly above chance.
• The VBM model showed a classification accuracy of 61%.
• Widespread structural differences, including white matter microstructure (higher AD, FA in BD) and grey matter volumes, drove the classification.

Conclusions:

• Structural neuroimaging features, particularly white matter integrity and grey matter volumes, can effectively differentiate between MDD and BD patients.
• The developed machine learning pipeline demonstrates good predictive accuracy for distinguishing these disorders.
• Neuroimaging biomarkers hold promise for improving the diagnostic accuracy of bipolar disorder and major depressive disorder.