Compact Convolutional Neural Networks for Classification of Asynchronous Steady-state Visual Evoked Potentials (1803.04566v2)

Abstract: Steady-State Visual Evoked Potentials (SSVEPs) are neural oscillations from the parietal and occipital regions of the brain that are evoked from flickering visual stimuli. SSVEPs are robust signals measurable in the electroencephalogram (EEG) and are commonly used in brain-computer interfaces (BCIs). However, methods for high-accuracy decoding of SSVEPs usually require hand-crafted approaches that leverage domain-specific knowledge of the stimulus signals, such as specific temporal frequencies in the visual stimuli and their relative spatial arrangement. When this knowledge is unavailable, such as when SSVEP signals are acquired asynchronously, such approaches tend to fail. In this paper, we show how a compact convolutional neural network (Compact-CNN), which only requires raw EEG signals for automatic feature extraction, can be used to decode signals from a 12-class SSVEP dataset without the need for any domain-specific knowledge or calibration data. We report across subject mean accuracy of approximately 80% (chance being 8.3%) and show this is substantially better than current state-of-the-art hand-crafted approaches using canonical correlation analysis (CCA) and Combined-CCA. Furthermore, we analyze our Compact-CNN to examine the underlying feature representation, discovering that the deep learner extracts additional phase and amplitude related features associated with the structure of the dataset. We discuss how our Compact-CNN shows promise for BCI applications that allow users to freely gaze/attend to any stimulus at any time (e.g., asynchronous BCI) as well as provides a method for analyzing SSVEP signals in a way that might augment our understanding about the basic processing in the visual cortex.

• The paper introduces a Compact-CNN that classifies asynchronous SSVEPs directly from raw EEG data, achieving approximately 80% accuracy across subjects, significantly outperforming traditional methods.
• Unlike traditional methods, Compact-CNN learns features directly from EEG, including phase and frequency information, allowing it to work effectively on smaller datasets and in asynchronous scenarios.
• This deep learning approach shows potential for practical, calibration-free BCI applications and offers insights into visual processing by identifying distinct signal variations.

Overview of Compact Convolutional Neural Networks for SSVEP Classification

This paper presents a novel approach to classifying Steady-state Visual Evoked Potentials (SSVEPs) from electroencephalographic (EEG) data using a Compact Convolutional Neural Network (Compact-CNN). SSVEPs are neural oscillations induced by visual stimuli flickering at specific frequencies, commonly employed in Brain-Computer Interface (BCI) systems. While conventional models require synchronous stimulus presentation and user-specific calibration, the proposed Compact-CNN only requires raw EEG signals without such constraints. The research assesses this CNN-based method against traditional techniques like Canonical Correlation Analysis (CCA) and Combined-CCA, demonstrating notable improvements in classification accuracy.

Main Findings and Implications

The Compact-CNN outperformed established methods, achieving an across-subject mean accuracy of approximately 80%, whereas traditional CCA and state-of-the-art Combined-CCA approaches lagged significantly behind. This achievement highlights the potential of deep learning for SSVEP classification without user-specific calibration, and suggests that feature extraction directly from raw EEG data can enhance performance in asynchronous scenarios.

Technical Specifics and Novel Contributions

Unlike prior techniques that rely heavily on handcrafted features and domain-specific knowledge, Compact-CNN learns patterns directly from EEG data. This is facilitated by employing convolutional layers capable of extracting phase and frequency information, which are critical in deciphering SSVEP signals. Importantly, the architecture's efficacy arises from its compactness, which allows operation on smaller datasets - a typical constraint in BCI applications.

Moreover, the paper introduces a visualization of the learned representations using t-SNE projection, which confirms that Compact-CNN identifies distinct class-level clusters in the neural data, as well as variations related to different signal epochs. The ability of Compact-CNN to distinguish these variations suggests its promise in contexts where precise temporal alignment of SSVEPs is not feasible, such as in asynchronous BCI paradigms.

Theoretical and Practical Implications

The findings underscore the efficacy of deep learning in neuroimaging contexts, suggesting they can overcome limitations of conventional methods in asynchronous and calibration-free setups. By revealing additional phase-related information often ignored by earlier methods, the Compact-CNN could be leveraged not only for practical BCI applications - such as spellers or control systems for assistive technologies - but also in probing fundamental aspects of visual processing.

Future Directions and Speculations

Expanding upon the success demonstrated here, future research might explore optimizing Compact-CNN architectures for varied BCI applications, including real-time adaptive systems where user freedom and flexible interaction are paramount. Additionally, further exploration into the landscape of phase and amplitude extraction could enhance understanding of neural mechanisms, potentially enhancing neurotechnological applications and neuroscience studies.