Multiscale Temporal Self-Attention and Dynamical Graph Convolution Hybrid Network for EEG-Based Stereogram...

Area of Science:

• Neuroscience
• Computer Science
• Ophthalmology

Background:

• Conventional stereopsis measurement relies on subjective interpretation of stereograms, which can be influenced by individual bias.
• Objective and reliable methods are needed for diagnosing visual conditions like strabismus.

Purpose of the Study:

• To develop an objective method for stereopsis assessment using electroencephalography (EEG) signals.
• To propose a novel hybrid deep learning network for classifying EEG signals evoked by dynamic random dot stereograms (DRDS).
• To facilitate the diagnosis of strabismus patients, even without direct communication.

Main Methods:

• Collected EEG signals evoked by DRDS for stereogram recognition.
• Proposed a multi-scale temporal self-attention and dynamical graph convolution hybrid network (MTS-DGCHN).
• Employed multi-scale temporal self-attention for temporal feature extraction and dynamical graph convolution for spatial relationship analysis.

Main Results:

• The proposed MTS-DGCHN demonstrated outstanding classification performance on the SRDA and SRDB datasets.
• The method achieved superior results compared to existing approaches in EEG signal classification for stereopsis.

Conclusions:

• The MTS-DGCHN offers a promising, objective approach for stereopsis evaluation.
• This technique can assist ophthalmologists in diagnosing strabismus more effectively.