Functional Near Infrared Spectroscopy (fNIRS) based Odor Detection and Classification using Functional Data Analysis

An objective approach for odor detection is to analyze the brain activity using imaging techniques during the odor stimulation. In this study, Functional Near Infrared Spectroscopy (fNIRS) is used to record hemodynamic response from the frontal region of the brain by using a 4-channel fNIRS system. The fNIRs data is collected during the odor detection task in which the subjects were asked to press a button when they detect the given odor. Functional Data Analysis (FDA) was applied on fNIRs data to convert discrete measured samples of data to continuous smooth curves. The FDA method enables us to use the bases coefficients of fNIRS smoothed curves for features that represent the shape of the raw fNIRS signal. With the learning algorithm that we proposed, these features were used to train the support vector machine classifier. We evaluated the odor detection problem, in two binary classification cases: odorant vs. non-odorant and odorant vs. fingertapping. The model achieved a classification accuracy of 94.12% and 97.06% over the stimulus condition in the two cases, respectively. Moreover to find the actual predictors we used the extracted defined features (slope, standard deviation, and delta) to train our classifier. We achieved an average accuracy of 91.18 % on classifying odorant vs. non-odorant and an accuracy of 94.12% for odorant vs. fingertapping on the stimulus condition. The results determined that fNIRs signals of odorant and non-odorant are distinguishable without being affected by the motor activity during the experiment. These findings suggest that fNIRs measurement on the forehead could be potentially used for objective and comparably inexpensive assessment of odor detection in cases that the subjective report is unreliable.