EDAPT: towards calibration-free BCIs with continual online adaptation

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) face accuracy challenges due to neural signal drift and user variability.
• Frequent recalibration hinders the practical deployment and real-world application of BCIs.

Purpose of the Study:

• To develop a framework for continual, real-time model adaptation in BCIs, eliminating the need for separate calibration phases.
• To enable BCIs to adapt to new users and changing signal characteristics seamlessly.

Main Methods:

• Proposed EDAPT, a task- and model-agnostic framework for continual online learning.
• Utilized population-level pretraining for a robust baseline decoder, followed by supervised continual finetuning for personalization.
• Integrated unsupervised domain adaptation techniques to address distribution shifts.

Main Results:

• Validated EDAPT across nine datasets, three BCI paradigms, and four deep learning architectures.
• Consistently improved decoding accuracy, raising mean balanced accuracy from 0.80 to 0.87.
• Demonstrated real-time feasibility with adaptation latencies under 200 milliseconds.

Conclusions:

• Continual online learning is a practical and effective strategy for high-performance, user-adaptive BCIs.
• EDAPT significantly reduces the recalibration bottleneck, a major barrier to BCI adoption.
• The framework advances neurotechnology towards robust and user-friendly real-world applications.