Multichannel temporal data classification of motor imagination using fNIRS

We describe classification of functional near-infrared spectroscopy (fNIRS) data acquired during finger tapping imagery tasks performed by a human subject, using an artificial neural network designed for image sequence recognition. Our goal is to develop a brain-computer interface that can handle various intentions of users. We used an fNIRS system to collect neural information from brain activity. For discrimination of the fNIRS data, we used our previously proposed neural network model called the Neocognitron-type Image Sequence Recognition Model (Neo-ISRM), which is suitable for analyzing multichannel temporal patterns. Finger tapping imagery of both left and right hands was used as the mental tasks to be discriminated with Neo-ISRM. The model gave good discrimination results for each category of tasks from data for the motor area, the prefrontal area, and the frontal lobe. In all experiments, we confirmed that the discrimination results for the frontal lobe had fewer discrimination errors compared with the results for both the motor and prefrontal areas.