An Analysis of Deep Learning Models in SSVEP-Based BCI: A Survey

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) offer direct human-robot communication, bypassing the peripheral nervous system.
• Steady-state visual evoked potentials (SSVEPs) provide high information transfer rates and short training times in BCI paradigms.
• Deep learning models are increasingly applied to classify complex SSVEP signals.

Purpose of the Study:

• To survey and analyze deep learning models used for SSVEP signal classification.
• To provide an up-to-date design guide for researchers in this field.
• To identify trends and common practices in deep learning for SSVEP classification.

Main Methods:

• Systematic literature review of 31 deep learning models from 2011-2023.
• Analysis of model design aspects: input data, network architecture, and performance evaluation.
• Focus on studies published primarily in 2021 and 2022.

Main Results:

• Deep learning models for SSVEP classification show diverse designs.
• Hyperparameter choices significantly impact model performance unpredictably.
• Recent research (2021-2022) dominates the surveyed literature.

Conclusions:

• A comprehensive understanding of current deep learning approaches for SSVEP classification is presented.
• The survey serves as a valuable resource for optimizing BCI system design.
• Further research can benefit from standardized methodologies and comparative analyses of deep learning models.