A confidence-gated source selection strategy for cross-session transfer in brain-computer interfaces.

Cross-session variability remains a major obstacle to the reliable operation of motor imagery (MI)-based brain-computer interfaces (BCI), particularly when systems are reused across multiple days. When multiple prior sessions from the same subject are available, two key questions arise before domain transfer: which source sessions to select and how to effectively utilize them. We address these questions by developing confidence-gated, selective-transfer pipelines: a Minimum-Distance Multi-Source Pipeline (MMP) that only uses source sessions close to the target, and a Bridge Domain Pipeline (BDP) that exploits both near and far sources to improve robustness. We evaluated these novel pipelines against uniform-pooling (MAP) and distance-weighted-pooling (DWP) baseline methods on two public motor imagery EEG datasets under matched experimental configurations of feature extraction, classification, and domain adaptation algorithms. The benchmark results support an endpoint-specific interpretation rather than a single accuracy ranking. Specifically, MAP achieved the highest maximum-configuration accuracy, DWP demonstrated the highest average accuracy across configurations, and on the primary dataset MAP, DWP, and BDP exhibited no statistically significant differences as the top-performing pipelines for data-driven configuration selection. In contrast, MMP