POMDP-BCI: A Benchmark of (Re)Active BCI Using POMDP to Issue Commands.

Area of Science:

• Neuroscience and Artificial Intelligence
• Brain-Computer Interface (BCI) Systems

Background:

• Existing Brain-Computer Interface (BCI) research often features diverse decoding algorithms and decision-making processes tailored to specific modalities.
• Some BCI modalities lack a dedicated high-level decision-making component within their standard pipeline, potentially limiting performance and flexibility.

Purpose of the Study:

• To propose and evaluate a unified decision-making framework for multiple active/reactive BCI modalities using a Partially Observable Markov Decision Process (POMDP).
• To enhance BCI systems by providing a flexible, high-level decision-making component adaptable to various modalities.

Main Methods:

• Developed a general POMDP model to serve as a sequential decision-making framework for BCI.
• Tested the POMDP model on three distinct BCI modalities using publicly available datasets.
• Compared the POMDP model's performance against state-of-the-art methods for each BCI modality, measuring accuracy, false positive trials, no-action trials, and decision time.

Main Results:

• The proposed POMDP models achieved performance comparable to or exceeding existing state-of-the-art methods across the three tested BCI modalities.
• The POMDP framework demonstrated generalizability, requiring no significant modifications for application to different BCI experiments.
• A key advantage of the POMDP approach is its ability to implement a 'no-action' decision when decoding performance is suboptimal.

Conclusions:

• The implemented POMDP model provides a flexible and effective sequential decision framework for BCI systems, particularly those lacking an integrated decision-making component.
• The framework achieves performance on par with existing specialized decision-making processes.
• The POMDP framework's generalizability across BCI modalities and potential for integrating diverse data streams offer significant promise for future BCI advancements.