TFSNet: A Time-Frequency Synergy Network Based on EEG Signals for Autism Spectrum Disorder Classification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Autism Spectrum Disorder (ASD) diagnosis relies on subjective methods, necessitating objective and accurate diagnostic tools.
• Existing machine learning and deep learning methods struggle with efficient feature extraction and capturing complex time-frequency EEG signal characteristics.

Purpose of the Study:

• To develop an advanced deep learning model, the time-frequency synergy network (TFSNet), for enhanced classification accuracy of Autism Spectrum Disorder (ASD) using electroencephalogram (EEG) signals.
• To address limitations in existing methods by effectively extracting and integrating time-domain and frequency-domain EEG features.

Main Methods:

• Proposed a novel time-frequency synergy network (TFSNet) incorporating a Dynamic Residual Block (TDRB) for time-domain feature enhancement.
• Utilized Short-Time Fourier Transform (STFT), a convolutional attention mechanism, and transformation technology for frequency-domain analysis.
• Designed an adaptive cross-domain attention mechanism (ACDA) for efficient fusion of time-frequency features.

Main Results:

• TFSNet achieved high average accuracies of 98.68% on the University of Sheffield dataset and 97.14% on the KAU dataset.
• Demonstrated superior performance compared to existing machine learning and deep learning methods for ASD EEG signal classification.
• Interpretability analysis confirmed the model's transparency and reliability in decision-making.

Conclusions:

• The TFSNet model offers a significant advancement in the accuracy and reliability of ASD diagnosis through EEG signal analysis.
• The proposed method effectively captures and integrates complex time-frequency features, outperforming previous approaches.
• This AI-driven tool holds promise for improving early detection and prognosis of Autism Spectrum Disorder.