A Convolutional Neural Network for the Detection of Asynchronous Steady State Motion Visual Evoked Potential

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Detecting intentional control (IC) and non-intentional control (NC) states is crucial for asynchronous brain-computer interface (BCI) operation.
• Steady-state visual evoked potential (SSVEP) BCI systems face challenges in accurately distinguishing multiple IC sub-states and achieving high true positive rates with low false positive rates using traditional threshold methods.

Purpose of the Study:

• To propose and evaluate a novel convolutional neural network (CNN) for the first time to detect IC and NC states in SSVEP-BCI systems.
• To investigate two distinct CNN-based processing pipelines for improved asynchronous state detection.

Main Methods:

• Utilized the steady-state motion visual evoked potentials (SSMVEP) paradigm for reduced visual discomfort.
• Developed two pipelines: FFT-CNN (CNN as a multi-class classifier) and FFT-CNN-CCA (CNN for IC/NC discrimination, CCA for IC sub-state classification).

Main Results:

• Both CNN pipelines demonstrated significant accuracy improvements for healthy participants with low traditional algorithm performance.
• The FFT-CNN-CCA pipeline outperformed the FFT-CNN pipeline when tested with stroke patient data.

Conclusions:

• Convolutional neural networks (CNNs) offer a robust method for detecting states in asynchronous SSMVEP-BCI systems.
• The proposed CNN approaches show great potential for enhancing the reliability and applicability of out-of-lab BCI systems.