I. Introduction
Patients suffering from ALS, brainstem stroke, spinal cord injury, and other severe neuromuscular disorders have a very little or no means of communication due to their hard condition in which they have no control on their muscles [1], [2]. Self-paced BCI is a crucial trend towards automating mind reading. They offer more natural human-machine interaction compared to synchronous or cue-paced paradigms [3], [4]. In synchronous BCI, the user performs tasks for predefined periods of time. Those synchronous paradigms already have good performance. In self-paced BCI, the user performs motor imagery tasks at will, without specifying a predefined period for doing so. This poses the challenge of detecting the non-control periods in which the user switches his/her attention.