Deep learning-based EEG analysis to classify normal, mild cognitive impairment, and dementia: Algorithms and dataset.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Automatic electroencephalogram (EEG) analysis is crucial for diagnosing brain disorders.
• Existing methods often require significant human intervention and lack comprehensive datasets.
• There is a need for well-annotated EEG datasets and efficient deep learning models for accurate, non-invasive diagnosis.

Purpose of the Study:

• To introduce the Chung-Ang University Hospital EEG (CAUEEG) dataset with detailed clinical annotations.
• To develop and evaluate a fully end-to-end deep learning model, CEEDNet, for automatic EEG diagnosis.
• To establish reliable evaluation tasks for detecting dementia and general abnormalities using EEG data.

Main Methods:

• Creation of the CAUEEG dataset, including patient age, event history, and diagnosis labels.
• Development of the CEEDNet (CAUEEG End-to-end Deep neural Network) model for seamless EEG analysis.
• Design of two evaluation tasks: CAUEEG-Dementia (normal, MCI, dementia) and CAUEEG-Abnormal (normal, abnormal).

Main Results:

• CEEDNet achieved high accuracy in EEG-based diagnosis, outperforming existing methods.
• The model recorded ROC-AUC scores of 0.9 on CAUEEG-Dementia and 0.86 on CAUEEG-Abnormal.
• The proposed method effectively leverages end-to-end learning for improved diagnostic performance.

Conclusions:

• The CAUEEG dataset and CEEDNet model offer a robust framework for automatic EEG diagnosis.
• The developed deep learning approach facilitates early detection of brain disorders through automated screening.
• This work advances the potential for low-cost, non-invasive diagnostic tools in neurology.