Transfer approach for the detection of missed task-relevant events in P300-based brain-computer interfaces

摘要

Detection of human cognitive states using biosignals such as the electroencephalogram (EEG) is gaining relevance in different application areas, e.g., teleoperation, human-robot collaboration, and rehabilitation. Especially, the P300, which is evoked as an event-related potential (ERP), when humans perceive task-relevant infrequent events among task-irrelevant frequent events, is widely used in brain-computer interfaces (BCIs). P300 detection has been used as an indicator that a human perceives task-relevant events or detects the occurrence of task-relevant or important events. In this paper, we focus on not only perceived task-relevant events but also not-perceived task-relevant events (i.e., missed events). In fact, a human can miss task-relevant events for different reasons, e.g., reduced attention level or increased workload level during parallel task-processing situations among others. Moreover, a human can also intentionally ignore task-relevant events to manage several simultaneous tasks. However, such missed events do not often occur in real-world applications. In this paper, we propose a transfer approach to handle insufficient number of events for training a classifier. For example, task-irrelevant infrequent events are used for training of classifier to detect missed task-relevant events. We evaluated our approach in different settings of training and testing a classifier with and without classifier transfer.