Single-trial classification of NIRS data from prefrontal cortex during working memory tasks

This study presents single-trial classification performance on high density Near Infrared Spectroscopy (NIRS) data collected from the prefrontal cortex of 11 healthy subjects while performing working memory tasks and idle condition. The NIRS data collected comprised a total of 40 trials of n-back tasks for 2 difficulty levels: n=1 for easy and n=3 for hard. The single-trial classification was performed on features extracted using common average reference spatial filtering and single-trial baseline reference. The single-trial classification was performed using 5×5-fold cross-validations on the NIRS data collected by using mutual information-based feature selection and the support vector machine classifier. The results yielded average accuracies of 72.7%, 68.0% and 84.0% in classifying hard versus easy tasks, easy versus idle tasks and hard versus idle tasks respectively. Hence the results demonstrated a potential feasibility of using high density NIRS-based BCI for assessing working memory load.