Motion Discrimination from EEG Using Logistic Regression and Schmitt-Trigger-Type Threshold

Abstract

In this study, a robot rehabilitation system is developed for motor paralysis using a brain -- machine interface (BMI) that estimates a patient's intention of motion from his electroencephalogram (EEG). Then, we consider the forcible movement of the patient's affected parts by an exoskeleton robot in synchronicity with his estimated intention. Further, we considered a motion discrimination method using an EEG that was measured when a healthy subject executed an upper-arm bending and stretching exercise. As a result, we used the10 -- 14 Hz band overall intensity of an EEG measured at the right parietal region as a feature of motion discrimination and proposed a method that employed logistic regression to obtain the likelihood function of the discriminated motion. Moreover, we used dual threshold processing, well-known as a "Schmitt-trigger gate" in the field of electronic engineering, to calculate the motion discrimination result using the obtained likelihood value. The effectiveness of our proposed method was confirmed through a motion discrimination experiment.