Evaluation of motor learning by fNIRS signal analysis using a general linear model

People can learn how to use a new tool with repeated practice. It has been suggested that such motor learning is closely related to specific changes in brain activity in humans. Here, we used functional near-infrared spectroscopy (fNIRS), a non-invasive neuroimaging technology, and examined whether it can detect learning-related activity. Subjects performed a cursor-tracking movement task with rotational transformation. Although fNIRS is easy to use, has little restraint, and can be employed during motor tasks, artifacts such as scalp blood flow contaminate the signal. Therefore, to analyze fNIRS signals (oxygenated hemoglobin), we applied a model-based approach in which scalp blood flow and learning-related behavioral changes were integrated into the design matrix of a general linear model (GLM). By doing so, we were able to examine the validity of the analysis. Group analysis indicated that a decrease in behavioral errors was accompanied by a decrease in inferior frontal gyrus activity. In contrast, significantly negative t-values were observed in superior frontal areas. For the scalp blood-flow component, significant activity was observed in almost all fNIRS channels. Since it is expected that scalp blood flow distributes uniformly and widely, this result indicates that the scalp blood-flow component was factored out correctly. These results show the potential of model-based GLM analysis for fNIRS to evaluate brain activity related to motor learning.