A Machine Learning Algorithm to Discriminating Between Bipolar and Major Depressive Disorders Based on Resting EEG...

Area of Science:

• Neuroscience
• Psychiatry
• Biomedical Engineering

Background:

• Distinguishing major depressive disorder (MDD) from bipolar disorder (BD) is a significant clinical challenge due to overlapping symptoms and lack of biomarkers.
• Misdiagnosis of bipolar depression as MDD can lead to inappropriate treatment.
• Electroencephalography (EEG) shows potential for biomarker discovery, but previous machine learning (ML) approaches have been limited by small datasets and methodology.

Purpose of the Study:

• To develop and validate a robust machine learning (ML) technique for differentiating MDD from bipolar disorder (BD) using resting-state EEG.
• To overcome limitations of previous studies, including small sample sizes and inadequate ML methodologies.

Main Methods:

• Utilized a training dataset of resting-state EEG from 71 MDD and 71 BD patients.
• Implemented a three-step ML technique: multi-step EEG signal preprocessing, application of symbolic transfer entropy (STE) for connectivity analysis, and feature extraction for classification.
• Employed STE, an effective connectivity measure, on preprocessed EEG signals.

Main Results:

• The proposed ML method achieved a total accuracy of 84.9%.
• Sensitivity and specificity for distinguishing MDD from BD were 83.4% and 87.1%, respectively.
• The algorithm demonstrated high performance on a substantial patient sample.

Conclusions:

• The developed ML technique shows significant promise as a clinical tool for differentiating MDD from BD.
• The method's high accuracy, sensitivity, and specificity suggest its potential utility in improving diagnostic accuracy.
• This approach may help reduce misdiagnosis and guide more appropriate therapeutic interventions.