EEG-ITNet: An Explainable Inception Temporal Convolutional Network for Motor Imagery Classification

Abstract

In recent years, neural networks and especially deep architectures have received substantial attention for EEG signal analysis in the field of brain-computer interfaces (BCIs). In this ongoing research area, the end-to-end models are more favoured than traditional approaches requiring signal transformation pre-classification. They can eliminate the need for prior information from experts and the extraction of handcrafted features. However, although several deep learning algorithms have been already proposed in the literature, achieving high accuracies for classifying motor movements or mental tasks, they often face a lack of interpretability and therefore are not quite favoured by the neuroscience community. The reasons behind this issue can be the high number of parameters and the sensitivity of deep neural networks to capture tiny yet unrelated discriminative features. We propose an end-to-end deep learning architecture called EEG-ITNet and a more comprehensible method to visualise the network learned patterns. Using inception modules and causal convolutions with dilation, our model can extract rich spectral, spatial, and temporal information from multi-channel EEG signals with less complexity (in terms of the number of trainable parameters) than other existing end-to-end architectures, such as EEG-Inception and EEG-TCNet. By an exhaustive evaluation on dataset 2a from BCI competition Ⅳ and OpenBMI motor imagery dataset, EEG-ITNet shows up to 5.9% improvement in the classification accuracy in different scenarios with statistical significance compared to its competitors. We also comprehensively explain and support the validity of network illustration from a neuroscientific perspective. We have also made our code freely accessible at https://github.com/AbbasSalami/EEG-ITNet.

EEG-CDILNet: a lightweight and accurate CNN network using circular dilated convolution for motor imagery classification

“…We compared the proposed method with the existing classification methods of MI-EEG, mainly including: FBCSP (Ang et al 2012), ShallowNet (Schirrmeister et al 2017), EEGNet (Lawhern et al 2018), MI-EEGNET (Riyad et al 2021a), EEG-TCNet (Ingolfsson et al 2020), AMSI-EEGNET (Riyad et al 2021b), EEG-Inception (Santamaria-Vazquez et al 2020), and EEG-ITNet (Salami et al 2022). To perform a fair comparison of the performance of the network on the dataset, we selected the best results of each method, when trained with the fixed network parameters for comparison.…”

“…EEG-TCNet inherited the advantages of EEGNet and achieved excellent classification accuracy with fewer parameters. Similarly, Salami et al (2022) combined Inception and TCN and proposed EEG-ITNet, which achieved a higher performance with lower complexity. However, TCN is a skewed network that uses the output structure at the last sequential position, resulting in a greater impact on the output further back.…”

Brain-Computer Interface using neural network and temporal-spectral features

“…(4) Gated Recurrent Unit Recurrent Neural Network Long-Short Term Memory-Recurrent Neural Network (Luo et al, 2018) (5) (Zhang D. et al, 2020) (17) Temporal-Spatial Convolutional Neural Network (Chen et al, 2020) (18) Temporal-Spectral-based Squeeze-and-Excitation Feature Fusion Network (Li Y. et al, 2021) (19) Shallow Convolution Neural Network and Bidirectional Long-Short Term Memory (Lian et al, 2021) (20) Temporal Convolutional Networks-Fusion (Musallam et al, 2021) (21) EEG-Inception-Temporal Network (Salami et al, 2022)…”

Optimal Channel Selection of Multiclass Motor Imagery Classification Based on Fusion Convolutional Neural Network with Attention Blocks

“…The model was validated using two strategies, within-subject and cross-subject, achieving accuracy rates of 83.92% and 63.34%, respectively. Following the same validation strategies, the authors in [27] introduced the EEG-ITNet model, which consists of four blocks: three layers of EE-GNet, temporal convolution, dimension reduction, and classification. As a result of the validation process, the model achieved an accuracy of 76.74% with a within-subject strategy and 69.44% with a cross-subject strategy.…”

“…Both models exhibit high accuracy in one strategy but not in the other, indicating that neither model is universally effective across different validation scenarios. In a similar manner to [27], the authors in [28,29] modified EEGNet to improve its performance and adapted it for general use. In [28], the MBSTCNN-ECA-LightGBM model combines EEGNet layers with a channel attention module and a LightGBM classifier to achieve up to 74% accuracy for four MI tasks of different classes.…”