Toward Zero-Calibration for Day-to-Day Use of Brain-Compute Interface

Abstract

To forward daily life out of the laboratory, the P300-based BCI system had several issues to solve. One of them is the time-consumed issues related to the need for re-calibration sessions. Without recalibration in using the P300-based BCI system, a deterioration in performance could occur. This problem would be more emphasized in routine life which is daily repeatedly performed the uses of the P300-based BCI system. Because a wave pattern of the P300 component, appearing during using the P300-based BCI system and being a useful marker for detecting users’ intent, seem to be changed by the physiological or habitual effect in daily uses of its system. Therefore, in this paper, we first verified appearing the deterioration of performance in using the P300-based BCI system without the calibration session on a different day. The task was performed by subjects for 5 days, and a break before doing the task on the last day. The performance comparison was conducted with or without the calibration, which is a session obtaining the information of brain signals to train the discriminator for the user's intent in the BCI systems. As a result of the performance decreasing without the calibration session significantly on days 13-15 (p < 0.05) in the P300-based BCI system, the factor causing the deterioration of the performance was investigated. Positively signed area amplitude in the time ranges from 275 to 450ms was obtained for comparing the amplitude of the P300 component between days. Regarding that, the characteristics of the P300 component decreased mainly in the central brain area on day 13-15 (p<0.05) relative to the characteristic of that on day 1. Thus, daily uses of the P300-based BCI could attribute to making a change in the signal amplitude of the P300 component. To solve the issues without re-calibration of the performance decreasing for the P300-based BCI system, an approach as a solution was applied, which approach is the separation of the common signals between days from the variable signals. The sparse dictionary learning methods were applied as one method for this approach, and it recovered the performance especially on day 13-15 without performing the calibration session for the new day. Our proposed methods in which performance less worsen expect to contribute to the use of the P300-based BCI system without an additional re calibration for a new day. In relation to that, we described how daily repetitive usage altered the P300-based BCI system and a task-related brain signal. In fact, it was given instructions on how to address problems with the amount of time re-calibration sessions took during the day. As a result, it is anticipated that this research will aid in the commercialization of the P300-based BCI system's everyday uses.