Local Binary Patterning Approach for Movement Related Potentials based Brain Computer Interface

This paper proposes the design and the implementation of an innovative algorithm for a 2-choices synchronous Brain-Computer Interface (BCI).The proposed BCI operates on signals from eight EEG channels evenly distributed along the sensorimotor area.The acquired EEGs are then analyzed by using a symbolization method. Typically, the symbolization includes data-analysis algorithms that translate physical processes from experimental measurements into a series of discrete symbols (e.g., bit strings). For the BCI application, the chosen symbolization algorithm is the Local Binary Pattern (LBP).Since the selected LBP method uses binary operations for the whole processing chain (end-to-end), the complexity and the computing timing of the features extraction (FE) and real-time classification stages have been strongly reduced.Finally, a time-continuous Support Vector Machine (tcSVM) classifies the LBP-extracted features.The here proposed BCI algorithm has been validated on 3 subjects (aged 26±1), who underwent a stimulation protocol oriented to Movement-Related Potentials (MRPs) elicitation. The in-vivo validation showed how the system is able to reach an intention recognition accuracy of 85.61 ± 1.19 %. In addition, starting from the complete data storage, the whole implemented computing chain asks, on average, for just ~3ms to provide the classification.As a proof of concept, the tcSVM outcomes have been used to drive, via Bluetooth, a 3D printed robotic hand.