Discriminating between bipolar and major depressive disorder using a machine learning approach and resting-state EEG...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomarker Discovery

Background:

• Distinguishing major depressive disorder (MDD) from bipolar disorder (BD) is clinically challenging due to overlapping symptoms and differing treatment needs.
• Objective biomarkers are needed to improve diagnostic accuracy and guide treatment selection for these mood disorders.

Purpose of the Study:

• To explore electroencephalography (EEG) as an objective biomarker for differentiating MDD from BD.
• To develop and validate an efficient machine learning algorithm (MLA) for this diagnostic task using a large, balanced dataset.

Main Methods:

• A 3-step MLA was employed, involving multi-step EEG signal preprocessing.
• Symbolic transfer entropy (STE), a measure of effective connectivity, was calculated from the preprocessed EEG data.
• The MLA utilized extracted STE features to classify subjects with MDD (N=71) versus BD (N=71).

Main Results:

• The algorithm selected 14 key connectivity features, primarily associated with frontal and parietal lobe electrodes.
• Involved brain regions included Broca's area (language processing) and the somatosensory association cortex (sensory information processing).
• The classifier achieved an evaluation accuracy of 88.5% and a test accuracy of 89.3%.

Conclusions:

• The high accuracy of the MLA, trained on a substantial, balanced dataset, indicates its potential as a clinical diagnostic tool.
• This approach may offer an inexpensive, accessible method to enhance diagnostic precision and expedite effective treatment for MDD and BD.