Improving single-trial detection of error-related potentials by considering the effect of heartbeat-evoked potentials in a motor imagery-based brain-computer interface

Abstract

Objective This study aimed to determine the effect of heartbeat-evoked potentials (HEPs) on changes in the error-related potential (ErrP) epoch and classification performance in single trials, specifically distinguishing between correct and error conditions in a three-class motor imagery-based brain-computer interface. Methods Eleven individuals participated in this study, with 10 participants assigned to the offline group and 10 to the real-time group. The experiment consisted of 360 motor imagery trials, involving both correct and erroneous feedback. The ErrP trial was categorized into three conditions based on whether the heartbeat was within the ErrP epoch time window or not: (1) including heartbeat trials (ErrPIHB), (2) excluding heartbeat trials (ErrPEHB), and (3) total trials (ErrPT). Results A small negativity was observed at approximately 200 ms, followed by a subsequent positivity at approximately 300 ms. The prominent amplitudes at approximately 200 and 300 ms in the ErrPEHB condition notably differed from those in the ErrPT and ErrPIHB conditions, showing the highest classification accuracy. In the offline experiment dataset of 10 participants, the ErrPEHB condition demonstrated the highest classification accuracy (0.89). This was higher by 0.07 and 0.11 compared to the ErrPT (0.82) and ErrPIHB (0.78) conditions, respectively. For real-time analysis, the novel ErrP paradigm proposed in this study achieved a classification accuracy of 0.89 for 10 participants, a 0.05 increase compared with that of the conventional ErrP paradigm. Conclusion and significance These findings can contribute to obtaining pure or clear ErrP epochs associated with the target response and significantly improve classification performance.