# Towards Asynchronous Motor Imagery-Based Brain-Computer Interfaces: a joint training scheme using deep learning

**Authors:** Patcharin Cheng, Phairot Autthasan, Boriwat Pijarana, Ekapol Chuangsuwanich, Theerawit Wilaiprasitporn  
**arXiv:** [1808.10852](https://arxiv.org/abs/1808.10852)  
**Date:** 2018-08-31  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1808.10852v1/

## Key points

- Authors explore deep learning (DL) in joint training scheme for asynchronous motor imagery-based Brain-Computer Interface (BCI)
- Proposed DL approach involves cascade of one-dimensional convolutional neural networks and fully-connected neural networks (CNN-FC)
- Study focuses on three types of brain responses: non-imagery EEG, pure imagery EEG, and transitional imagery EEG
- Pure and transitional imagery signals are treated as same class under joint training scheme
- Performance evaluated using ten-fold cross-validation with mean accuracy reaching 71.52% for CNN-FC and 70.27% for CSP-SVM using sparse selection of electrode channels ($C_{z}$, $C_{3}$ and $C_{4}$)
- Joint training scheme significantly enhances classification performance compared to approaches without it

## Summary

In this paper, titled "Towards Asynchronous Motor Imagery-Based Brain-Computer Interfaces: a joint training scheme using deep learning," authors Patcharin Cheng, Phairot Autthasan, Boriwat Pijarana, Ekapol Chuangsuwanich, and Theerawit Wilaiprasitporn explore the application of deep learning (DL) in a joint training scheme for asynchronous motor imagery-based Brain-Computer Interface (BCI). The proposed DL approach involves a cascade of one-dimensional convolutional neural networks and fully-connected neural networks (CNN-FC). The study focuses on three types of brain responses: non-imagery EEG (background EEG), pure imagery EEG, and transitional imagery EEG during the period between background EEG and pure imagery. The research aims to provide greater insight into real-world scenarios by studying transitional imagery signals. It is suggested that pure imagery and transitional EEG represent high and low power EEG imagery, respectively. The results obtained from the CNN-FC model are compared to the conventional approach for motor imagery-BCI, which involves using common spatial pattern (CSP) for feature extraction and support vector machine (SVM) for classification (CSP-SVM). Under the joint training scheme proposed in this study, pure and transitional imagery signals are treated as the same class, while background EEG is considered another class. The performance of classifying pure and transitional imagery signals from background EEG is evaluated using ten-fold cross-validation. Remarkably, when utilizing only a sparse selection of electrode channels ($C_{z}$, $C_{3}$ and $C_{4}$), the mean accuracy achieved reaches 71.52% for CNN-FC and 70.27% for CSP-SVM. Furthermore, it is demonstrated that implementing the joint training scheme significantly enhances classification performance compared to approaches without it. Specifically, mean accuracy without the joint training scheme only reaches 62.68% for CNN-FC and 52.41% for CSP-SVM. This research contributes valuable insights into utilizing deep learning techniques in BCI systems for improved classification of motor imagery signals based on different brain responses.