Analysis of classification performance of fNIRS signals from prefrontal cortex using various temporal windows

In this paper we investigate the role of different temporal windows in classification of functional near-infrared spectroscopy (fNIRS) signals corresponding to mental arithmetic and mental counting for development of a brain-computer interface. Signals are acquired from the prefrontal cortex of four healthy subjects during mental arithmetic and mental counting tasks using a continuous-wave fNIRS system, DYNOT: Dynamic Near-Infrared Optical Tomography. Support vector machine is used to classify the mean values of the change in concentration of oxygenated and deoxygenated hemoglobin during different temporal windows. The highest average classification accuracy of 82.4% is achieved during the 2–7 s time window within the total 10 s task period. The averaged classification accuracies achieved using 0–5 s, 1–6 s and 5–10 s temporal windows are 61.6%, 67.4% and 72.5% respectively. These results indicate that using signal mean, calculated during 2–7 s time window, as the features results in higher classification accuracies.