Related Experiment Video
Updated: Aug 5, 2026
05:41
Synchronous Triplanar Reconstruction Integrated with Color Doppler Mapping for Precise and Rapid Localization of Thyroid Lesions
Published on: February 9, 2024
HFG-Net: high-frequency guided multi-view graph convolution and dynamic spatio-temporal fusion for ASD diagnosis
Jiannan Kang1, Zhiyuan Fan1, Zongbing Xiao1
1College of Electronic & Information Engineering, Hebei University, Baoding, People's Republic of China.
Journal of Neural Engineering
|July 28, 2026
Summary
A new deep learning model, HFG-Net, effectively diagnoses Autism Spectrum Disorder (ASD) using electroencephalogram (EEG) data by analyzing high-frequency brainwave patterns and adapting to individual patient variability.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Biomedical Engineering
Background:
- Autism Spectrum Disorder (ASD) presents significant heterogeneity in electroencephalogram (EEG) features, challenging current deep learning models.
- Existing methods like CNNs and GCNs struggle with complex spatio-temporal dependencies and non-Euclidean abnormalities in EEG data.
- Static feature fusion in current models limits their ability to capture dynamic brain activity patterns.
Purpose of the Study:
- To develop a novel deep learning framework, HFG-Net, for robust and accurate EEG-based diagnosis of ASD.
- To address the limitations of existing models in capturing spatio-temporal dependencies and handling data heterogeneity.
- To improve the interpretability and adaptability of deep learning models for neurodevelopmental disorder diagnosis.
Main Methods:
- Proposed HFG-Net, a High-Frequency Guided Spatio-Temporal Synergistic Network incorporating a Channel-Temporal Multi-scale Attention (CTMA) mechanism.
- Implemented a High-Frequency Guided Multi-View Graph Construction strategy using sparse skeletons from Beta/Gamma bands to create noise-resistant topologies.
- Utilized a Dynamic Synergistic Fusion module with a gating network for adaptive feature integration based on sample-level confidence weights.
Main Results:
- HFG-Net achieved 95.79% classification accuracy and 92.89% F1-score on an independent test set, outperforming state-of-the-art models.
- The model demonstrated high recognition rates for mild ASD cases (98.43%).
- Interpretability analysis confirmed alignment with neuropathological findings and adaptive capability for heterogeneous samples.
Conclusions:
- HFG-Net offers an efficient, robust, and interpretable paradigm for EEG-based ASD diagnosis.
- The framework successfully addresses challenges in synergistic spatio-temporal modeling and heterogeneity adaptation in neurodevelopmental disorders.
- High-frequency guided analysis and dynamic fusion enhance diagnostic performance and clinical applicability.