Towards real time efficient and robust ECoG decoding for mobile brain-computer interface | JoVE Visualize

Area of Science:

Neuroscience and Biomedical Engineering
Brain-Computer Interfaces (BCIs)

Background:

Decoding locomotion-related brain activity from electrocorticographic (ECoG) signals is vital for brain-computer interfaces (BCIs).
Existing ECoG decoders often require significant computational resources and are susceptible to noise and outliers, limiting their use in mobile applications for physically disabled patients.

Purpose of the Study:

To explore and identify the optimal decoding pipeline for mobile BCIs, focusing on computational efficiency, precision, and robustness.
To evaluate various decoding algorithms and feature optimization techniques for real-time ECoG signal processing in resource-constrained environments.

Main Methods:

Comprehensive evaluation of diverse decoding algorithms (e.g., Partial Least Squares, Bayesian Ridge Regression, Support Vector Regression, Neural Networks, Random Forest) using a combined ECoG dataset from 12 subjects performing individual finger movements.
Exploration of feature optimization using model explainability for selected algorithms.
Comparison of decoding performance for updatable algorithms using sequential data batches.

Main Results:

The Random Forest (RF) model demonstrated the best trade-off between decoding precision (average Pearson's correlation coefficient, r = 0.466) and computational efficiency (0.5 K FLOPs/inference, 900 KiB model size).
RF exhibited superior robustness to noisy ECoG signals, achieving over twice the decoding precision compared to state-of-the-art deep neural networks.
An optimized RF pipeline was successfully deployed on an STM32-based embedded platform, achieving a low computation delay of 15.2 ms.

Conclusions:

The Random Forest algorithm presents a highly effective and efficient solution for decoding finger movements from ECoG signals, outperforming traditional methods in terms of precision, efficiency, and robustness.
The developed decoding pipeline, implemented on a compact embedded platform, enables low-latency, power-efficient real-time decoding, paving the way for practical mobile BCIs.
This research significantly advances the development of mobile BCIs for real-life applications, particularly for restoring motor function in individuals with physical disabilities.