Evaluating a Trial-Wise Motor Imagery Modulation Quality Metric in a Cross-Subject Adaptation Scenario

Despite their potential for rehabilitation therapies based on brain-computer interfaces (BCIs), motor imagery (MI) tasks are inherently challenging to evoke. Meaningful feedback should follow each MI trial to promote modulation strategies refinement in a closed loop way. In a previous work, we showed that the cost of aligning a trial to match the training session distribution in a within-subject scenario provides a real-time measure of MI modulation quality. Real-time experiments confirmed that using such a metric for feedback delivery can enhance MI modulation. While promising results, this setup requires a full calibration session. More recently, we extended such adaptation framework to a cross-subject transfer learning scenario, showing that it enables accurate MI decoding without subject-specific calibration. In this work, we investigate if the adaptation cost derived from a model trained on a cross-subject cross-dataset setting can also provide an effective MI modulation quality metric. Specifically, we assess if it reflects MI modulation quality and study its relationship with the intra-subject analogous metric. By demonstrating that a cross-subject adaptation cost metric provides informative trial-wise feedback, we could guide users to control the MI-BCI system from the first day, without requiring a subject-specific calibration. This approach is particularly relevant in rehabilitation contexts, where minimizing setup time directly translates to higher usability.