Movement related events classification from functional near infrared spectroscopic signal

Abstract

This study investigates left hand (LH) and right hand (RH) movements related events with respect to resting state (RS) on the basis of hemodynamic response of dorsolateral prefrontal cortex (DLPFC). The signals of hemodynamic responses are acquired by 16 channels functional near infrared (fNIR) spectroscopy. From these multiple channel data, it is difficult to classify the events. To solve this difficulty, statistically the most effective channels are identified. For identifying most effective channels, at first the raw fNIR signal is filtered and separated into three classes (RS, LH, and RH movements) based on the events. The most effective channels are found out by t-test hypothesis and effect size (ES) statistics. Furthermore, for classifying purpose, the time domain features are extracted from oxygenated hemoglobin (HbO2) signal of the most effective channels. From these features, artificial neural network (ANN) is used to classify the events. The classifying accuracy is achieved 79.5% in average. This study is helpful to estimate the voluntary movement from frontal cortex neural activity.