LSTA-CNN: A Lightweight Spatiotemporal Attention-Based Convolutional Neural Network for ASD Diagnosis Using EEG

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is a cost-effective tool for identifying autism spectrum disorders (ASD).
• Deep learning methods are increasingly used for analyzing complex EEG signals.
• EEG signals contain rich temporal and spatial information crucial for accurate diagnosis.

Purpose of the Study:

• To propose a lightweight spatio-temporal attention-based convolutional neural network (LSTA-CNN) for ASD diagnosis using EEG.
• To effectively extract and integrate spatio-temporal features from EEG recordings.

Main Methods:

• Developed a lightweight spatio-temporal attention-based convolutional neural network (LSTA-CNN).
• Utilized multi-scale temporal and spatial convolution layers for diverse feature representation.
• Introduced a novel spatio-temporal attention mechanism for joint feature integration.

Main Results:

• The LSTA-CNN achieved superior classification performance on a self-collected EEG dataset compared to existing deep learning models.
• The proposed model demonstrated a significantly lower number of parameters.
• The LSTA-CNN exhibited reduced inference time, indicating its lightweight nature.

Conclusions:

• The LSTA-CNN is a highly effective and efficient deep learning model for diagnosing ASD from EEG data.
• Its lightweight architecture holds significant potential for practical clinical applications.
• The model's ability to integrate spatio-temporal features enhances diagnostic accuracy in EEG analysis.