The Analysis of the Brain State Measuring by NIRS-based BMI in Answering yes-no Questions

Abstract

Abstract In recent years, Brain-Machine Interface (BMI) has been improved with the rapid development of the cerebral function measurement technologies. BMI is used for measuring the brain activity of the subjects and inferring his/her intention. These measuring results can control the devices such as the electric wheelchair or the electric arm directly. Previous studies have such problems that BMI device is not portable, and it takes much time to attach a lot of sensor nodes on the head of the subject. Therefore, we use portable NIRS, because it is easy for the subjects to mount it and their burden is low. This paper describes the analytical method for cerebral blood flow during imagining affirmative or negative answers to the questions. It was verified whether it is possible to use the NIRS as BMI to discriminate the yes answer or no without voice and gesture. In our study, a subject keeps watching the display which shows a yes-no question, he/she imagines affirmative or negative answer to the questions. In the experiment, one test trial is in 30 seconds and it includes 10 seconds task between 10 seconds rests. Each test set consists of 10 trials. One subject has five test set. In our study, we used Wearable Optical Topography WOT-100 as measurement device of NIRS which has 10 channels (ch7-16) of prefrontal cortex. The NIRS data analysis procedure is as follows; in the first step, we used a band-pass filter to select the data of frequencies from 0.02 Hz up to 0.1 Hz. In the second step, measured NIRS data of each task set is divided into 10 blocks which are included 5 seconds data before the task and 10 seconds data after the task. In the third step, we calculated baseline of measured data from 5 seconds of the beginning and the end of the task blocks, and this baseline fitting is applied to the original data. In the last step, the neural network learned the important elements of the training data and classified the test data. Our method can discriminate between imagining affirmative or negative answers with 70% accuracy. At result, NIRS is useful to discriminate the yes-no answer of the questions.