Due to advantages such as robustness with respect to motion artifact, suitability for special populations like infants, and being able to be measured in wearable settings, the functional near-infrared spectroscopy (fNIRS) is an emerging and more and more important brain functional imaging modality in many research fields e.g. the brain computer interface, the psychology and the cognitive science. Motor imagination is an important paradigm in the rehabilitation trainings for disabled people. With the development of wearable fNIRS systems, these systems may assist the disabled people in long-term brain rehabilitation trainings at home or in community. However, the classification accuracies of the current fNIRS-based motor imaginary tasks are generally low. This paper aims to improve the classification accuracy of the fNIRS-based motor imaginary task by the individualized parameter optimization using the Pearson correlation based R² method. In this experiment, the concentration variation data of hemoglobin species during the left and right hand motor imaginary tasks of 17 subjects were collected using the fNIRS method, and the support vector machine (SVM) classifier was then adopted for classification. Experimental results show that the classification accuracy is significantly improved by the parameter optimization using the R² method. With the R2 method, the percentage of the subjects with classification accuracies above 60% is turned from 58.8% to 94% and that with classification accuracies above 65% is turned from 41.2% to 64.7% in the whole subject pool.

functional near-infrared spectroscopy  /  motor imagery  /  support vector machine  /  correlation index analysis