Discriminative Analysis of Audio (Speech) and Ball Squeezing Tasks Using fNIRS Data for BCI-Based Control Applications

Functional near-infrared spectroscopy (fNIRS) has been a new technique in designing non-invasive brain imaging methods to measure cerebral hemodynamics in response to some cognitive and motor tasks. Hence, this study aimed to yield some insights into the possibilities of fNIRS to separate auditory versus motor ball-squeezing tasks. Particularly, it aimed to test the classification accuracy of fNIRS signals using statistical features for the possible use as a Brain Computer Interface (BCI) application targeted at motor-impaired people. The dataset included trials of healthy subjects performing the two tasks. A comprehensive feature extraction procedure was applied to individual epochs. Statistical analyses and supervised classification were performed. Among the features extracted, some demonstrated statistically significant differences between task conditions. A Random Forest classifier has been able to attain high classification accuracies of 83.3%, which confirms that the cognitive states can be separated. The results thus suggest that fNIRS can be solidly exploited in BCI systems for neurorehabilitation for identifying tasks.