Real-Time Motor Imagery-Based Brain–Computer Interface System by Implementing a Frequency Band Selection

Motor imagery-based brain–computer interfaces (MI-BCIs) are a promise to revolutionize the way humans interact with machinery or software, performing actions by just thinking about them. Patients suffering from critical movement disabilities, such as amyotrophic lateral sclerosis (ALS) or tetraplegia, could use this technology to interact more independently with their surroundings. This paper aims to aid communities affected by these disorders with the development of a method that is capable of detecting the intention to execute movements in the upper extremities of the body. This will be done through signals acquired with an electroencephalogram (EEG), their conditioning and processing, and their subsequent classification with artificial intelligence models. In addition, a digital signal filter will be designed to keep the most characteristic frequency bands of each individual and increase accuracy significantly. After extracting discriminative statistical, frequential, and spatial features, it was possible to obtain an 88% accuracy on validation data with a random forest (RF) model when it came to detecting whether a participant was imagining a left-hand or a right-hand movement. Furthermore, a convolutional neural network (CNN) was used to distinguish if the participant was imagining a movement or not, which achieved 78% accuracy and 90% precision. These results will be verified by implementing a real-time simulation with the usage of a robotic arm.