Date Published: June 18, 2019
Publisher: Public Library of Science
Author(s): Asier Salazar-Ramirez, Jose I. Martin, Raquel Martinez, Andoni Arruti, Javier Muguerza, Basilio Sierra, Paweł Pławiak.
A brain-computer interface (BCI), based on motor imagery EEG, uses information extracted from the electroencephalography signals generated by a person who intends to perform any action. One of the most important issues of current research is how to detect automatically whether the user intends to send some message to a certain device. This study presents a proposal, based on a hierarchical structured system, for recognising intentional and non-intentional mental tasks on a BCI system by applying machine learning techniques to the EEG signals. First-level clustering is performed to distinguish between intentional control (IC) and non-intentional control (NC) state patterns. Then, the patterns recognised as IC are passed on to a second stage where supervised learning techniques are used to classify them. In BCI applications, it is critical to correctly classify NC states with a low false positive rate (FPR) to avoid undesirable effects. According to the literature, we selected a maximum FPR of 10%. Under these conditions, our proposal achieved an average test accuracy of 66.6%, with an 8.2% FPR, for the BCI competition IIIa dataset. The main contribution of this paper is the hierarchical approach, based on machine learning paradigms, which performs intentional and non-intentional discrimination and, depending on the case, classifies the intended command selected by the user.
It was in 1924 when Hans Berger achieved to record the first human electroencephalogram . Since then, the study of the brain has been a matter of interest for scientists, researchers and medical professionals. Related to the study of the activity of the brain, in the 1970s, researchers of the field of engineering began to show interest on brain activity and started producing the first Brain-Computer Interface (BCI) applications. In the early days of BCI, these applications had mainly a medical background and were focused on restoring either lost audition/visibility capabilities or mobility by allowing users to interact with a computer using their thoughts. However, the span covered by BCI nowadays has vastly increased and it is not only focused on medical applications but also to other applications from different fields, such as assistive technologies for elder people, videogame and entertainment, smart home control or even in for military applications.
As previously mentioned, the final step of the design process is to verify that the system is able to differentiate between classes for new EEG signal patterns. To do so, new data that the system has not previously used is needed, i.e., the test datasets (TestSet) that had previously been obtained from the 5 repetitions of the random dataset construction process. Therefore, the average results of applying the first level classification to TestSet over 5 runs are presented in Table 5, where the following information is given: the average confusion matrix, the FPR and the accuracy of the primary classification.
Despite the experimental setup of this work being slightly different from that of other research, it can be considered that the results obtained are similar to those presented in other studies in the literature. Table 10 summarises the information related to similar works found in the bibliography in the context of BCI systems. For each system, Table 10 shows the number of classes or motor imagery movements (and if the system considers the NC state, NC class), the false positive rate (FPR), the accuracy of the system and the algorithms used for building the system.
This paper presents a BCI system that is capable of distinguishing between intentional and non-intentional control (IC and NC) mental states. In addition, the system is capable of determining the specific imaginary movement of those mental states considered to be IC. The system divides the problem hierarchically: first, the mental patterns are classified with an unsupervised clustering algorithm that determines whether the pattern belongs to an IC or an NC state. Second, a supervised learning algorithm decides the specific imaginary movement class of the patterns that were classified as IC states (left hand, right hand, tongue and foot).