An Interpretable Machine Learning Method for the Detection of Schizophrenia Using EEG Signals.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Schizophrenia diagnosis relies on clinical observation, lacking objective biomarkers.
• Electroencephalograms (EEGs) offer a non-invasive window into brain activity.
• Developing computational tools for EEG analysis can improve diagnostic accuracy.

Purpose of the Study:

• To develop and validate a machine learning (ML) method for schizophrenia diagnosis using EEG data.
• To identify specific EEG connectivity metrics and frequency bands relevant to schizophrenia.
• To provide clinically interpretable insights from the ML model's diagnostic process.

Main Methods:

• Utilized a random forest machine learning model.
• Extracted connectivity metrics, including generalized partial directed coherence (GPDC) and direct directed transfer function (dDTF), from EEG signals.
• Employed feature selection to identify the most relevant EEG characteristics for diagnosis.

Main Results:

• The ML model demonstrated potential in aiding schizophrenia diagnosis.
• Connectivity metrics derived from EEG signals served as effective input features.
• The occipital region and specific frequency bands (beta and theta) were identified as significant for diagnosis.

Conclusions:

• Machine learning analysis of EEG connectivity metrics offers a promising avenue for schizophrenia diagnosis.
• The study highlights the importance of the occipital region and beta/theta frequency bands in schizophrenia.
• This approach provides clinically interpretable information beyond a simple diagnostic classification.