Enhancement in classification accuracy of motor imagery signals with visual aid: An fNIRS-BCI Study

One of the most promising brain activity utilized in brain-computer interface (BCI) is motor imagery (MI). Due to weak hemodynamic response (HR) signal, the achieved classification accuracies using MI are not sufficiently high. In this study, the enhancement in HR was investigated during motor imagery tasks of ball squeezing with the right hand. Brain signals in the form of concentration changes in oxy-hemoglobin (ΔHbO) and deoxy-hemoglobin (ΔHbR) from the left sensorimotor cortex were obtained using functional near-infrared spectroscopy (fNIRS). The experiment was separated in two sessions: In the first session the MI task was performed without a visual aid, and in the second session of the same task, the visual aid was provided: A video was played on a screen that showed a person continuously squeezing the ball, which can help in enhancing the imagination, thus improvement in HR. Later the features of averaged ΔHbO were used for classification. The active channels were selected on the basis of t-values and trials of those channels were mean to obtain averaged ΔHbO. Consistent with literature, imagery task with visual aid, showed increased activation in ΔHbO. Moreover, linear discriminant analysis was used to classify signals by taking the mean and peak of the averaged ΔHbO resulting in average classification accuracies of approximately 66% and 77% for MI task, with and without visual aid, respectively. These results are convincing that showed improvement in MI ability which will be useful for fNIRS-based BCI applications.