A novel three dimensional probability-based classifier for improving motor imagery-based BCI

Objective: Motor imagery BCI based assistive robotics solution has the potential to empower the upper mobility independence of a disabled person. The objective of this work was to compare the classification performance of well-established classifiers with a novel prototype classifier.

Approach: We developed an adaptive decision surface ADS classifier with the future objective to augment an assistive robotic prosthetic hand to open and close to grasp an object in cooperation with LIDAR sensors. The ADS was trained with a training data set from the BCI competition IV dataset 2a from Graz University of Technology.

Main results: The classification accuracy in the offline tests reached 76.06 % class 1 and 81.50 % class 2 using a non-adaptive ADS and 79.55 % class 1 and 99.69 % class 2 using an adaptive ADS classifiers. We show a prototype adaptive decision classifier used with motor imagery datasets.