Automatic depression diagnosis through hybrid EEG and near-infrared spectroscopy features using support vector...

Area of Science:

• Neuroscience
• Medical Technology
• Psychiatry

Background:

• Depression diagnosis is challenging, impacting social function and daily life.
• Accurate diagnostic tools are crucial for effective depression treatment.
• Current diagnostic methods may lack objective biological markers.

Purpose of the Study:

• To investigate the efficacy of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) for depression classification.
• To identify potential neuroimaging biomarkers for depression.
• To develop an automated machine learning model for individual-level depression diagnosis.

Main Methods:

• Recorded resting-state EEG and fNIRS signals from 25 depression patients and 30 healthy controls.
• Analyzed EEG brain functional network properties (clustering coefficient, local efficiency) in delta and theta bands.
• Extracted features including brain network properties, asymmetry, and brain oxygen entropy.
• Employed a data-driven approach for feature selection and a support vector machine for classification.

Main Results:

• EEG analysis revealed significantly higher clustering coefficient and local efficiency in delta and theta bands for depression patients.
• The automated model achieved 81.8% accuracy using EEG features alone.
• Hybrid EEG and fNIRS features increased classification accuracy to 92.7%.
• Key distinguishing features included delta band local efficiency, theta band asymmetry, and brain oxygen sample entropy (p < 0.05).

Conclusions:

• Hybrid EEG and fNIRS combined with machine learning provide a highly accurate method for individual depression classification.
• Specific EEG and fNIRS-derived features show potential as effective biological markers for depression.
• This multimodal neuroimaging approach offers a promising avenue for objective depression diagnosis.