A Deep Contrastive Embedding Model for Neural Congruency-Based EEG Analysis in Reading Disorders

Abstract

Electroencephalography (EEG) provides valuable insight into the neural mechanisms underlying dyslexia, yet analysis is challenged by low signal to noise ratio (SNR), high inter-subject variability, and complex spatio-temporal dynamics. The Neural Congruency framework offers a promising way to identify consistent brain activity patterns among readers, but its integration with deep learning remains limited. This study introduces Neural Congruency Contrastive Learning (NCCL), a framework that combines spatial, frequency, and temporal convolutions to learn EEG embeddings aligned with neural congruency principles. Using synthetic EEG representing dyslexic and control participants across SNRs from -37 dB to -7 dB, the model was trained with a contrastive loss to maximize within group similarity and enhance between group separation. NCCL reliably distinguished dyslexic from control groups even at -25 dB, showing high stability across runs and maintaining discriminative performance under severe noise. These results highlight the framework’s robustness and its potential applicability to real EEG datasets, including tasks such as Rapid Automatized Naming and Phonological Awareness. Overall, this work establishes a noise resilient approach for modeling neural congruency with deep contrastive learning, advancing the use of artificial intelligence in dyslexia research and future clinical assessment.