Pseudo-online framework for BCI evaluation: a MOABB perspective using various MI and SSVEP datasets.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-Computer Interfaces (BCIs) are crucial for assistive technologies, operating in online, offline, and pseudo-online modes.
• Offline BCI analysis, while accurate, introduces biases by processing entire datasets, unlike real-time applications.
• Existing BCI frameworks primarily support offline analysis, limiting the evaluation of algorithms in more realistic scenarios.

Purpose of the Study:

• To extend the MOABB framework for pseudo-online BCI analysis, enabling direct comparison of algorithms.
• To introduce an 'idle state' event for comprehensive data processing in pseudo-online mode.
• To bridge the gap between offline BCI research and real-world performance assessment.

Main Methods:

• Implementation of overlapping sliding windows for pseudo-online data processing within the MOABB framework.
• Inclusion of an 'idle state' event to capture non-task-related data variations.
• Validation of algorithm performance using the normalized Matthews correlation coefficient and information transfer rate.

Main Results:

• Analysis of state-of-the-art BCI algorithms over 15 years across motor imagery and steady-state visually evoked potential datasets.
• Statistical demonstration of performance differences between offline and pseudo-online BCI analysis approaches.
• Identification of biases introduced by traditional offline BCI processing methods.

Conclusions:

• The developed pseudo-online BCI analysis method provides a more realistic evaluation of classification algorithms.
• Enabling comparative analysis in both offline and pseudo-online modes enhances the reliability of BCI research findings.
• This advancement will lead to more accurate and comprehensive reporting of BCI algorithm performance in the scientific community.