55 巻 (2017) 1 号 p. 9-16
Partaking in physical exercise while listening to one's favorite music ameliorates the perception of dyspnea and fatigue and increases one's enjoyment of the exercise. Hence, the use of specific music in a physical training class for elderly people could increase the endurance of the participants. Additionally, this could benefit elderly individuals by reducing frailty and improving cognitive function. However, it may be difficult to elicit information pertaining to preferred music as it may be difficult to communicate with elderly individuals owing to cognitive decline. Thus, in this study, an objective classification method of music preferences was developed using the changes in prefrontal hemodynamic signals while listening to music, measured by functional near-infrared spectroscopy (fNIRS). The experiment consisted of 17 young adults and 17 elderly individuals. While listening to 6 popular songs during different time periods, fNIRS was used to scan the prefrontal brain activity of the subjects. For each song, seventeen features of the fNIRS waveform were extracted, which included statistics of temporal distribution and laterality of changes in oxy-hemoglobin, deoxy-hemoglobin, and total hemoglobin concentrations. From these features, three that exhibited the highest correlation to score of a subjective preference scale obtained from a questionnaire completed by each participant were selected. The extracted features were used to train a 2-class linear classifier that determined whether the listener preferred each song. Mean classification accuracy was calculated by a leave-one-out cross-validation method. The proposed algorithm had mean classification accuracy of 86.3 ± 11.8% and 88.9 ± 14.5% (mean ± standard deviation) in detecting individual favorite songs in young adults and elderly individuals, respectively. The mean classification accuracy was significantly higher when features selected for individual participants were used compared to when fixed and common features for all participants were used. This finding suggested that listening to favorite songs evoked individual changes in emotional responses.