Related Experiment Video
Updated: Jan 9, 2026
11:28
Concurrent EEG and Functional MRI Recording and Integration Analysis for Dynamic Cortical Activity Imaging
Published on: June 30, 2018
12.2K
FEMBA: Efficient and Scalable EEG Analysis with a Bidirectional Mamba Foundation Model
Summary
We developed FEMBA, a novel framework for efficient electroencephalography (EEG) analysis. This bidirectional state-space model offers linear scalability, outperforming Transformers for long-term EEG monitoring in clinical and wearable applications.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Biomedical Engineering
Background:
- Accurate electroencephalography (EEG) analysis is crucial for long-term patient monitoring and diagnostics.
- Traditional deep learning models, like Transformers, face computational challenges due to quadratic complexity, limiting their use in resource-constrained settings.
- There is a need for efficient and scalable EEG analysis methods for applications ranging from clinical diagnostics to wearable health devices.
Purpose of the Study:
- To introduce FEMBA (Foundational EEG Mamba + Bidirectional Architecture), a novel self-supervised framework for efficient EEG analysis.
- To establish new benchmarks in EEG analysis efficiency using bidirectional state-space modeling.
- To demonstrate the viability of FEMBA for both clinical applications and resource-constrained wearable devices.
Main Methods:
- Developed FEMBA, a self-supervised framework utilizing bidirectional state-space modeling for EEG analysis.
- FEMBA exhibits linear time and memory complexity, contrasting with the quadratic complexity of Transformer models.
- Trained FEMBA on over 21,000 hours of unlabeled EEG data and fine-tuned it on three downstream tasks.
Main Results:
- FEMBA achieves competitive performance compared to Transformer models with significantly lower computational costs.
- Achieved 81.82% balanced accuracy (0.8921 AUROC) on TUAB and 0.949 AUROC on TUAR.
- A 7.8M-parameter variant of FEMBA demonstrates effectiveness for resource-constrained devices, enabling on-device EEG monitoring.
Conclusions:
- FEMBA offers a scalable and efficient solution for EEG analysis, addressing the limitations of traditional deep learning models.
- The framework's efficiency and performance make it suitable for widespread adoption in clinical settings and wearable technology.
- FEMBA paves the way for improved patient care through continuous, on-device EEG monitoring for seizure detection and artifact reduction.
