Decoding four different sound-categories in the auditory cortex using functional near-infrared spectroscopy

• Asymmetry in functional hemodynamic responses upon 4 sound-categories using fNIRS is investigated.
• As features for classification, the mean, slope, and skewness of oxy-hemoglobin signals are used.
• For BCI based on auditory responses, the use of left-hemispheric signals is recommended.

The ability of the auditory cortex in the brain to distinguish different sounds is important in daily life. This study investigated whether activations in the auditory cortex caused by different sounds can be distinguished using functional near-infrared spectroscopy (fNIRS). The hemodynamic responses (HRs) in both hemispheres using fNIRS were measured in 18 subjects while exposing them to four sound categories (English-speech, non-English-speech, annoying sounds, and nature sounds). As features for classifying the different signals, the mean, slope, and skewness of the oxy-hemoglobin (HbO) signal were used. With regard to the language-related stimuli, the HRs evoked by understandable speech (English) were observed in a broader brain region than were those evoked by non-English speech. Also, the magnitudes of the HbO signals evoked by English-speech were higher than those of non-English speech. The ratio of the peak values of non-English and English speech was 72.5%. Also, the brain region evoked by annoying sounds was wider than that by nature sounds. However, the signal strength for nature sounds was stronger than that for annoying sounds. Finally, for brain-computer interface (BCI) purposes, the linear discriminant analysis (LDA) and support vector machine (SVM) classifiers were applied to the four sound categories. The overall classification performance for the left hemisphere was higher than that for the right hemisphere. Therefore, for decoding of auditory commands, the left hemisphere is recommended. Also, in two-class classification, the annoying vs. nature sounds comparison provides a higher classification accuracy than the English vs. non-English speech comparison. Finally, LDA performs better than SVM.