Predicting differential diagnosis between bipolar and unipolar depression with multiple kernel learning on multimodal...

Area of Science:

• Neuroscience
• Psychiatry
• Medical Imaging

Background:

• Differentiating major depression (MDD) and bipolar disorder (BD) is a significant clinical challenge.
• Reliable biomarkers are needed for early and accurate diagnosis of mood disorders.
• Current diagnostic methods can be time-consuming and may lead to misdiagnosis, impacting treatment efficacy.

Purpose of the Study:

• To develop a predictive model for distinguishing between bipolar disorder (BD) and major depressive disorder (MDD) using neuroimaging data.
• To identify structural brain differences that can serve as biomarkers for MDD and BD.
• To evaluate the efficacy of combining multiple neuroimaging techniques with machine learning for differential diagnosis.

Main Methods:

• Integration of structural neuroimaging techniques: Tract-based Spatial Statistics (TBSS) and Voxel-based morphometry.
• Application of a multiple kernel learning (MKL) procedure for predictive modeling.
• Analysis of a sample of 148 patients diagnosed with MDD or BD, and 74 healthy controls.

Main Results:

• Achieved a balanced accuracy of 73.65% in differentiating BD from MDD.
• Identified reduced grey matter volume in specific brain regions (hippocampus, amygdala, insula, etc.) in BD patients compared to MDD patients and healthy controls.
• TBSS revealed widespread white matter microstructure disruptions in both MDD and BD patients compared to controls, with a more pronounced pattern in MDD.

Conclusions:

• The combination of structural neuroimaging and multiple kernel learning shows promise for the differential diagnosis of mood disorders.
• Quantitative brain biomarkers derived from neuroimaging can aid in personalized treatment strategies for MDD and BD.
• This study highlights the potential of advanced computational methods in psychiatric diagnostics.