arxiv:2507.20254

MIRepNet: A Pipeline and Foundation Model for EEG-Based Motor Imagery Classification

Published on Jul 27

· Submitted by

DiMaria0817 on Jul 30

Upvote

Authors:

Dingkun Liu ,

Zhu Chen ,

Jingwei Luo ,

Abstract

MIRepNet, an EEG foundation model tailored for motor imagery, achieves state-of-the-art performance across multiple datasets using a hybrid pretraining strategy that combines self-supervised and supervised learning.

AI-generated summary

Brain-computer interfaces (BCIs) enable direct communication between the brain and external devices. Recent EEG foundation models aim to learn generalized representations across diverse BCI paradigms. However, these approaches overlook fundamental paradigm-specific neurophysiological distinctions, limiting their generalization ability. Importantly, in practical BCI deployments, the specific paradigm such as motor imagery (MI) for stroke rehabilitation or assistive robotics, is generally determined prior to data acquisition. This paper proposes MIRepNet, the first EEG foundation model tailored for the MI paradigm. MIRepNet comprises a high-quality EEG preprocessing pipeline incorporating a neurophysiologically-informed channel template, adaptable to EEG headsets with arbitrary electrode configurations. Furthermore, we introduce a hybrid pretraining strategy that combines self-supervised masked token reconstruction and supervised MI classification, facilitating rapid adaptation and accurate decoding on novel downstream MI tasks with fewer than 30 trials per class. Extensive evaluations across five public MI datasets demonstrated that MIRepNet consistently achieved state-of-the-art performance, significantly outperforming both specialized and generalized EEG models. Our code will be available on GitHubhttps://github.com/staraink/MIRepNet.

View arXiv page View PDF GitHub 45 Add to collection

Community

DiMaria0817

Paper author Paper submitter 2 days ago

First Foundation Model for EEG-based MI Classification:

This paper introduces the first foundation model specifically designed for motor imagery (MI)-based EEG classification, named MIRepNet. It tackles a crucial gap in the BCI field by addressing the challenge of generalizing across diverse subjects, electrode configurations, and new task categories, all with minimal calibration. The authors present a robust pipeline for high-quality MI data construction that unifies heterogeneous EEG signals, enhancing the adaptability of the model. This is particularly important as EEG headsets often vary in terms of electrode count and placement, and the data quality can be inconsistent across different sources.

One of the paper's key contributions is its innovative pretraining strategy, combining self-supervised masked token reconstruction with supervised MI classification. This hybrid approach allows MIRepNet to rapidly adapt to new subjects and tasks, requiring only a small amount of new data for fine-tuning (less than 30 trials per class). Extensive experiments on five MI datasets demonstrate the efficacy of the proposed model, consistently outperforming state-of-the-art EEG models.

The paper successfully highlights the importance of paradigm-specific foundation models for EEG classification. The results strongly suggest that such tailored models can significantly enhance the accuracy and practicality of BCIs, making them more feasible for real-world applications, such as assistive technologies and rehabilitation systems.