A Convolutional Autoencoder-based Explainable Clustering Approach for Resting-State EEG Analysis.

Area of Science:

• Neuroscience
• Machine Learning
• Computational Psychiatry

Background:

• Supervised machine learning for electroencephalography (EEG) is common, but EEG clustering is underexplored.
• Clustering EEG data could reveal novel subtypes of neuropsychiatric disorders.
• Existing EEG clustering methods often rely on manual feature extraction or separate deep learning and clustering steps, potentially limiting cluster quality.

Purpose of the Study:

• To develop an integrated, explainable deep learning approach for high-quality EEG clustering.
• To apply this method to resting-state EEG data in individuals with schizophrenia (SZ).
• To investigate the relationship between identified EEG states and clinical symptoms in SZ.

Main Methods:

• Proposed an explainable convolutional autoencoder-based approach combining model training and clustering.
• Applied the method to resting-state EEG data from individuals with schizophrenia.
• Analyzed the identified EEG states and their correlation with clinical symptom severity.

Main Results:

• Identified 8 distinct EEG states characterized by varying levels of delta (δ) activity.
• Found a correlation between time spent outside the dominant EEG state and increased negative symptom severity in SZ.
• Demonstrated the effectiveness of the integrated deep learning approach for EEG clustering.

Conclusions:

• The proposed explainable autoencoder-based clustering method significantly advances EEG data analysis.
• This approach can identify meaningful EEG states and their clinical relevance in disorders like schizophrenia.
• The method holds potential for future discoveries in various neurological and neuropsychological conditions.