ORIGINAL RESEARCH article
Front. Neurol.
Sec. Pediatric Neurology
Volume 16 - 2025 | doi: 10.3389/fneur.2025.1700161
This article is part of the Research TopicThe Convergence of Cognitive Neuroscience and Artificial Intelligence: Unraveling the Mysteries of Emotion, Perception, and Human CognitionView all 12 articles
CMTS-GNN: A Cross-Modal Temporal-Spectral Graph Neural Network with Cognitive Network Explainability
Provisionally accepted
- 1Student Guidance Service Center, Northeastern University, Shenyang, China
- 2China Medical University, Shenyang, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Infantile spasms (IS) are a severe epileptic encephalopathy of early infancy. Delayed or inaccurate detection is linked to adverse neurodevelopmental trajectories that can compromise perception, cognition, and affective development. Traditional EEG analysis struggles with the complexity, heterogeneity, and volume of IS data, making manual interpretation time-consuming and vulnerable to inter-rater variability. We present CMTS-GNN, a Cross-Modal Temporal—Spectral Graph Neural Network that integrates complementary temporal and spectral EEG representations via bidirectional cross-modal attention and gated fusion, while explicitly modeling brain-region connectivity. By capturing functional interactions supporting perceptual processing, cognitive control, and affective dynamics, CMTS-GNN enhances detection accuracy and interpretability. We evaluate the model on an in-house infantile spasms dataset and the public CHB-MIT epilepsy dataset using five-fold cross-validation and leave-one-subject-out/leave-one-patient-out protocols. On our in-house dataset, five-fold cross-validation yields 99.02% accuracy, 98.96% precision, 97.47% recall, 98.20% F1-score, and 99.27% AUC. On CHB-MIT, five-fold cross-validation achieves 98.54% accuracy, 98.31% precision, 98.71% recall, 98.47% F1-score, and 98.87% AUC, surpassing recent methods on most metrics. Subject-independent evaluations further demonstrate consistent robustness and generalizability across patients. Modeling connectivity across brain regions also provides clinically meaningful explanations of model decisions. Overall, CMTS-GNN offers an accurate, generalizable, and interpretable solution for automated IS detection from EEG and has the potential to facilitate earlier intervention—helping mitigate long-term perceptual, cognitive, and affective morbidity in affected infants.
Keywords: infantile spasms, cognitive control, Explainability analysis, cross-modal, Brain regions, long-term perceptual
Received: 12 Sep 2025; Accepted: 08 Oct 2025.
Copyright: © 2025 Yi, Lu and ying. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Meng Lu, menglu@ise.neu.edu.cn
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.