Cascade and Parallel Convolutional Recurrent Neural Networks on EEG-based Intention Recognition for Brain Computer Interface (1708.06578v2)

Abstract: Brain-Computer Interface (BCI) is a system empowering humans to communicate with or control the outside world with exclusively brain intentions. Electroencephalography (EEG) based BCIs are promising solutions due to their convenient and portable instruments. Motor imagery EEG (MI-EEG) is a kind of most widely focused EEG signals, which reveals a subjects movement intentions without actual actions. Despite the extensive research of MI-EEG in recent years, it is still challenging to interpret EEG signals effectively due to the massive noises in EEG signals (e.g., low signal noise ratio and incomplete EEG signals), and difficulties in capturing the inconspicuous relationships between EEG signals and certain brain activities. Most existing works either only consider EEG as chain-like sequences neglecting complex dependencies between adjacent signals or performing simple temporal averaging over EEG sequences. In this paper, we introduce both cascade and parallel convolutional recurrent neural network models for precisely identifying human intended movements by effectively learning compositional spatio-temporal representations of raw EEG streams. The proposed models grasp the spatial correlations between physically neighboring EEG signals by converting the chain like EEG sequences into a 2D mesh like hierarchy. An LSTM based recurrent network is able to extract the subtle temporal dependencies of EEG data streams. Extensive experiments on a large-scale MI-EEG dataset (108 subjects, 3,145,160 EEG records) have demonstrated that both models achieve high accuracy near 98.3% and outperform a set of baseline methods and most recent deep learning based EEG recognition models, yielding a significant accuracy increase of 18% in the cross-subject validation scenario.

• The paper proposes Cascade and Parallel Convolutional Recurrent Neural Networks (C-RNNs and P-RNNs) for recognizing human movement intentions directly from raw EEG signals.
• Using the PhysioNet dataset, the models achieved approximately 98.3% accuracy in cross-subject tests, an 18% improvement over previous state-of-the-art methods.
• This approach enhances EEG-based BCI reliability by reducing reliance on preprocessing and shows potential for real-world applications like mind-controlled devices.

EEG-Based Intention Recognition for Brain-Computer Interface Utilizing Convolutional Recurrent Neural Networks

The paper entitled "Cascade and Parallel Convolutional Recurrent Neural Networks on EEG-based Intention Recognition for Brain Computer Interface" proposes two convolutional recurrent network architectures—cascade and parallel models—for the task of EEG-based intention recognition within Brain-Computer Interface (BCI) applications. This work focuses on enhancing the capability to identify human intended movements and instructions through learning the compositional spatio-temporal representations from raw EEG streams.

Methodology

The authors present two advanced neural network models, the Cascade and Parallel Convolutional Recurrent Neural Networks (C-RNNs and P-RNNs respectively), designed to analyze raw EEG signal data. The unique feature of these models lies in their ability to effectively capture spatio-temporal dynamics without the need for preprocessing steps common in other EEG signal analysis methodologies. The C-RNN model sequentially extracts spatial features of each EEG data mesh using CNNs, which are then input into RNNs to decipher temporal features. Conversely, the P-RNN model employs CNNs and RNNs concurrently to process spatial and temporal features in parallel, later fusing these to yield final intention predictions.

The core operation begins by converting one-dimensional EEG sequences to two-dimensional EEG meshes, aggregating spatial correlations between electrode readings. Two-dimensional convolutional layers then extract spatial features from these meshes, while recurrent layers identify temporal dependencies within the data mesh sequences. The key innovation here is the alignment of EEG signals with spatial electrode information to provide comprehensive representations of neural activity.

Experimental Results

The models were evaluated using the PhysioNet EEG Dataset, one of the largest movement intention datasets, containing data from 108 subjects. Both the cascade and parallel models demonstrated superior performance, achieving remarkable accuracy levels of approximately 98.3% in cross-subject, multi-class scenarios. This signifies an immense improvement, a 18% increase in accuracy compared to current state-of-the-art EEG recognition methods, which typically demonstrate performance levels around 80%. In a real-world BCI system evaluation, the models achieved a 93% accuracy in recognizing five different instruction intentions across limited EEG channels.

Implications and Future Directions

This research enhances the reliability and applicability of EEG-based BCI systems by reducing dependency on handcrafted features and preprocessing steps, thereby minimizing the impact of noisy data—a prevalent challenge within EEG signal processing. The achievement of high cross-subject and multi-class accuracy levels suggests significant potential for these models to be implemented in diverse real-world applications, such as mind-controlled devices, prosthetics, and exoskeletons.

The theoretical implications of this approach lie in its endorsement of end-to-end learnable models that can robustly classify intentions by leveraging raw EEG data. The demonstration of robust cross-subject generalization suggests that further research may explore expanding these applications to other cognitive and neurological tasks, thereby enhancing human-computer interaction. Future work could explore the adaptation of these models to online systems, creating real-time intention recognition frameworks for dynamic BCI applications.

Overall, this paper substantially contributes to the EEG-based intention recognition field by presenting scalable, accurate, and adaptable solutions using advanced neural network architectures.