PSD-CNN Approach for Subject Independent Dementia Recognition from EEG Signals

• Published in: Proceedings of the ... International Joint Conference on Computer Science and Software Engineering (Online) pp. 588 - 594
• Main Authors: Kongwudhikunakorn, Supavit, Kiatthaveephong, Suktipol, Thanontip, Kamonwan, Leelaarporn, Pitshaporn, Dujada, Pathitta, Yagi, Tohru, Senanarong, Vorapun, Saengmolee, Wanumaidah, Wilaiprasitporn, Theerawit
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.06.2024
• Subjects: Accuracy; Computer science; Convolutional neural networks; Deep learning; Dementia; Electroencephalography; Medical services; Mild Cognitive Impairment; Power Spec-tral Density; Predictive models
• ISSN: 2642-6579
• DOI: 10.1109/JCSSE61278.2024.10613730

Several studies of EEG-based dementia classification were conducted on the subject-dependent scenarios, which could not be a common ground truth for the new subjects, as EEG has high variability across different subjects. Thus, this work aims to propose our method based on power spectral features and convolutional neural network compared to the state-of-the-art deep learning and machine learning techniques for subject-independent dementia recognition with a leave-two-out cross-validation approach. All of the methods tested their binary classification performance on two datasets: (i) normal vs. dementia and (ii) normal vs. abnormal groups across three tasks (i.e., eyes-closed, eyes-opened, and mental imagery). The proposed method accomplished the highest performance against the state-of-the-art methods on both datasets. The eyes-closed provided the best classification result at an accuracy of 0.89 ± 0.03. Our result presents a promising future to apply the PSD-CNN method for dementia screening application.