On the Classification of SSVEP-Based Dry-EEG Signals via Convolutional Neural Networks (1805.04157v2)

Abstract: In this paper, we propose a novel Convolutional Neural Network (CNN) approach for the classification of raw dry-EEG signals without any data pre-processing. To illustrate the effectiveness of our approach, we utilise the Steady State Visual Evoked Potential (SSVEP) paradigm as our use case. SSVEP can be utilised to allow people with severe physical disabilities such as Complete Locked-In Syndrome or Amyotrophic Lateral Sclerosis to be aided via BCI applications, as it requires only the subject to fixate upon the sensory stimuli of interest. Here we utilise SSVEP flicker frequencies between 10 to 30 Hz, which we record as subject cortical waveforms via the dry-EEG headset. Our proposed end-to-end CNN allows us to automatically and accurately classify SSVEP stimulation directly from the dry-EEG waveforms. Our CNN architecture utilises a common SSVEP Convolutional Unit (SCU), comprising of a 1D convolutional layer, batch normalization and max pooling. Furthermore, we compare several deep learning neural network variants with our primary CNN architecture, in addition to traditional machine learning classification approaches. Experimental evaluation shows our CNN architecture to be significantly better than competing approaches, achieving a classification accuracy of 96% whilst demonstrating superior cross-subject performance and even being able to generalise well to unseen subjects whose data is entirely absent from the training process.

• The paper proposes an end-to-end CNN model that bypasses extensive pre-processing and achieves high accuracy for SSVEP signal classification.
• The CNN architecture, featuring a novel SSVEP Convolutional Unit, attains 96% accuracy on single-subject data and 89% mean accuracy across multiple subjects.
• The study reveals that deeper CNN designs enhance generalization on unseen subjects, paving the way for practical, subject-independent BCI applications.

On the Classification of SSVEP-Based Dry-EEG Signals via Convolutional Neural Networks

This paper presents an investigation into the viability of using Convolutional Neural Networks (CNNs) for classifying Steady State Visual Evoked Potential (SSVEP) signals in dry-EEG data, a notoriously challenging task due to the inherent noise and high impedance levels associated with dry-EEG measurements. The research highlights the advantages of an end-to-end CNN approach that negates the need for cumbersome signal pre-processing, which is typically required with traditional EEG classification methods.

Methodology and Experimental Setup

The work employs a commercially available Quick-20 dry-EEG headset, capable of streaming data from 20 sensors without skin preparation, thus offering convenience and practicality in Brain-Computer Interface (BCI) applications. Experimental data focuses on SSVEP stimuli with frequencies ranging from 10 Hz to 30 Hz, recorded across various scalp regions associated with visual processing. The authors present a CNN architecture that leverages a novel SSVEP Convolutional Unit (SCU), which includes multiple convolutional, batch normalization, and pooling layers aimed at extracting pertinent features directly from raw EEG signals.

Results Summary

The CNN model demonstrates a classification accuracy of 96% for a single subject's data, outperforming traditional classifiers like Support Vector Machines (SVMs), Linear Discriminant Analysis (LDA), and Recurrent Neural Networks (RNNs). When evaluated across multiple subjects, the CNN maintains a mean accuracy of 89%, further highlighting its robustness and reliability in multi-subject scenarios. Notably, the CNN architecture also shows the potential to generalize well, achieving an accuracy of 78% when classifying data from subjects not included in the training set.

The paper discusses the generalization capabilities of the CNN against unseen subjects, revealing that while classification accuracy initially decreases, adjustments to the depth of the CNN architecture can improve performance significantly, suggesting that deeper architectures are conducive to handling the variability inherent in subject-independent EEG tasks.

Implications and Future Directions

The implications of this paper are significant, especially considering the potential for subject-independent BCI applications where training with the specific individual's data is unnecessary. The use of CNNs represents a step forward in EEG signal processing, offering a streamlined approach that sidesteps traditional data pre-processing pitfalls and potentially reducing the computational burden associated with feature extraction.

Looking ahead, one could envision scaling the paper to include larger datasets and additional EEG paradigms to further explore the versatility of CNNs in neurocomputational applications. The integration of hybrid models combining CNNs with recurrent architectures like LSTM or GRU could also be explored to capitalize on temporal dependencies in EEG data. This work provides a pathway toward more practical and user-friendly neuroimaging applications where dry-EEG systems could be employed effectively across varied environments and settings, enhancing the accessibility and utility of BCI systems.