Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Utilizing Resting- and Task-State EEG with Convolutional Neural Networks to Screen for Developmental Dyslexia in Hong Kong Chinese Children
Poster Session C, Friday, October 25, 4:30 - 6:00 pm, Great Hall 3 and 4
Yaqi YANG1, Zhaoyu Liu2, Brian W.L. Wong1,3, Shuting Huo1,4, Jie Wang5, Tan Lee6, Fumiko Hoeft7, Urs Maurer1,8,9; 1Department of Psychology, The Chinese University of Hong Kong, Hong Kong, China, 2Department of Biomedical Engineering, The Chinese University of Hong Kong, Hong Kong, China, 3Basque Center on Brain, Language and Cognition (BCBL), Donostia-San Sebastián, Spain, 4Department of Applied Social Sciences, The Hong Kong Polytechnic University, Hong Kong, China, 5Department of Psychology, The Education University of Hong Kong, Hong Kong, China, 6Department of Electronic Engineering, The Chinese University of Hong Kong, Hong Kong, China, 7Department of Psychological Sciences, University of Connecticut, Waterbury, Connecticut, United States of America, 8Centre for Developmental Psychology, The Chinese University of Hong Kong, Hong Kong, China, 9Brain and Mind Institute, The Chinese University of Hong Kong, Hong Kong, China
Developmental dyslexia (DD) is a prevalent learning disorder with presumed neurological origins. Despite numerous attempts to integrate machine learning algorithms with electroencephalogram (EEG) techniques for efficient and cost-effective screening, a reliable EEG-based classification framework for screening Chinese children with DD and identifying potential neural biomarkers has yet to be discovered. Our study addresses this gap by combining EEG data with Convolutional Neural Networks (CNNs) to screen for DD in 130 Chinese children (7-11 years old), including both DD and typically developing (TD) children. We transformed pre-processed EEG signals into functional connectivity (FC) matrices using three distinct FC measures—Pearson Correlation Coefficients (PCC), Phase Locking Value (PLV), and Rho Index (RHO)—across four frequency bands (delta, theta, alpha, and beta) under four experimental conditions (resting-state eye-open, eye-closed, one-back, and two-back working memory tasks). After pre-processing the FC matrices, we divided them into two independent samples. We trained and validated the CNNs model using one sample through 5-fold cross-validation and conducted cross-sample validation with a permutation test on the other sample. The eye-open beta-band-based RHO index achieved the highest 5-fold average classification accuracy, reaching approximately 98%. In cross-sample validation, the accuracy remained around 70%, significantly above the chance level. We also identified discriminative FCs for DD classification. The TD group showed stronger temporal-parietal, temporal-frontocentral, and central-centroparietal FCs compared to the DD group. Conversely, the DD group exhibited stronger frontocentral-frontal, prefrontal-anterior frontal, and frontal-centroparietal FCs than the TD group. Additionally, we found a negative correlation between Chinese reading abilities and discriminative FCs in the DD group. Overall, our study presents a robust deep-learning framework using EEG-based methods for DD screening in Chinese children, validated in an independent sample. The findings may help uncover potential behavior-related neural biomarkers. This innovative approach enhances the practicality of EEG-based deep learning in DD screening and expands our understanding of the neural substrates of Chinese DD.
Topic Areas: Disorders: Developmental, Reading