What question did this study set out to answer?

This framework addresses accuracy degradation in brain-computer interfaces (BCIs) when real-time classifiers are used after offline training.

May 8, 2026Open Access

Stage-Wise Divergence Detection: A Diagnostic Framework for Closing Real-Time vs. Offline Accuracy Gaps in ERP-Based Brain-Computer Interfaces

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial detects accuracy gaps in ERP-based brain-computer interfaces, suggesting a diagnostic protocol for improvement.

This framework addresses accuracy degradation in brain-computer interfaces (BCIs) when real-time classifiers are used after offline training.
Developed a stage-wise divergence detection framework for ERP preprocessing consisting of six stages (raw epoch slicing, bandpass filtering, baseline correction, Euclidean Alignment, channel reordering, trial-level scoring).
Used Pearson correlation and L2-ratio metrics to assess divergence between offline and real-time data at each stage on 12 subjects.
Validated with a compact convolutional ERP classifier using the BigP3BCI dataset.
Identified a filter phase mismatch at Stage B with a correlation of r = 0.724, impacting accuracy.
Found a trial-ordering error at Stage F with a mean posterior correlation of r = 0.023 across 25 affected trials.
Improved real-time accuracy from 0.5% to 68.1%, achieving a final scoring of 70.9%, within 2.8 pp of offline performance.

Khanh-Dung Tran (2026) studied this question.

View Full Paper

Ask AI

Bookmark

May 8, 2026Open Access

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial detects accuracy gaps in ERP-based brain-computer interfaces, suggesting a diagnostic protocol for improvement.

This framework addresses accuracy degradation in brain-computer interfaces (BCIs) when real-time classifiers are used after offline training.
Developed a stage-wise divergence detection framework for ERP preprocessing consisting of six stages (raw epoch slicing, bandpass filtering, baseline correction, Euclidean Alignment, channel reordering, trial-level scoring).
Used Pearson correlation and L2-ratio metrics to assess divergence between offline and real-time data at each stage on 12 subjects.
Validated with a compact convolutional ERP classifier using the BigP3BCI dataset.
Identified a filter phase mismatch at Stage B with a correlation of r = 0.724, impacting accuracy.
Found a trial-ordering error at Stage F with a mean posterior correlation of r = 0.023 across 25 affected trials.
Improved real-time accuracy from 0.5% to 68.1%, achieving a final scoring of 70.9%, within 2.8 pp of offline performance.

Khanh-Dung Tran (2026) studied this question.

View Full Paper

Ask AI

Bookmark