Stopping vs Resting state during motor imagery paradigm

Current non-invasive Brain Machine interfaces commonly rely on the decoding of sustained motor imagery activity (MI). This approach enables a user to control brain-actuated devices by triggering predetermined motor actions. One major drawback of such strategy is that users are not trained to stop their actions. Indeed, the termination process involved in BMI is poorly understood with most of the studies assuming that the end of an MI action is similar to the resting state. Here we hypothesize that the process of stopping MI (MI termination) and resting state are two different processes that should be decoded independently due to the exhibition of different neural pattens. We compared the detection of both states transitions of an imagined movement, i.e. rest-to-movement (onset) and movement-to-rest (offset). Our results shows that both decoders show significant differences in term of performances and latency (N=17 Subjects) with the offset decoder able to detect faster and better MI termination. While studying this difference, we found that the offset decoder is primarily based on the use of features in Beta band which appears earlier. Based on this finding, we also proposed a Random Forrest based decoder which enable to distinguish three classes (MI, MI termination and REST).