Unsupervised adaptation of an ECoG based brain-computer interface using neural correlates of task performance

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Brain-computer interfaces (BCIs) enable communication and control for individuals with severe motor disabilities.
• Current BCIs require extensive training for decoder calibration and updates, limiting their practical usability.
• The need for continuous, real-time adaptation of BCIs is critical for enhancing user experience and independence.

Purpose of the Study:

• To develop and validate an auto-adaptive BCI (aaBCI) system capable of online training and updating neural decoders during free use.
• To integrate a motor task performance (MTP) classifier to detect user intention and facilitate decoder adaptation.
• To demonstrate the efficacy of the aaBCI across discrete and continuous control paradigms.

Main Methods:

• Proposed a novel aaBCI architecture incorporating a control decoder and an MTP decoder.
• Utilized MTP decoder outputs to generate training data for online, real-time updates of the control decoder.
• Validated the aaBCI concept using an ECoG dataset from a tetraplegic participant in an online simulation study.
• Tested aaBCI performance on discrete (4-class exoskeleton control) and continuous (2D cursor control) BCI paradigms.

Main Results:

• The aaBCI achieved a multiclass area under the ROC curve of 0.7404 for discrete control, outperforming chance levels.
• For continuous control, the aaBCI achieved a cosine similarity of 0.1211, significantly above chance.
• While slightly below supervised training performance (0.8187 multiclass AUC, 0.2002 continuous cosine similarity), aaBCI demonstrated effective online adaptation.

Conclusions:

• The proposed aaBCI system enables real-time training and updating of neural decoders during BCI use, significantly enhancing usability.
• The MTP decoder effectively identifies task performance, facilitating seamless online adaptation of control decoders.
• aaBCI represents a promising advancement for more intuitive and accessible brain-computer interfaces, particularly for individuals with severe motor impairments.