Biomedic Signal Processing and Analysis of Neuroimaging from fNIRS for Human Pain

One of major biomedical signals, pain, and its diagnosis has been critical but hard in clinical practice, in particularly for nonverbal patients. However, as we know that neuroimaging methods, such as functional near-infrared spectroscopy (fNIRS), have shown some great encouraging assessing neuronal function corresponding to nociception and pain. Specially some research results strongly suggest that neuroimaging, together with supports from machine learning, may be practically used to not only facilitate but also can predict different cognitive tasks over this challenge. The aim of this current research is to expand our previous studies by exploring the classification of fNIRS signals (oxyhaemoglobin) according to temperature level (we define cold and hot) and corresponding pain intensity (say low and high) by means of machine learning models. In order to find out the relations between temperatures and pain intensity, we defined and used the quantitative sensory testing to determine pain threshold and pain tolerance for the cold and heat in all eighteen-healthy people. The classification algorithm is based on a bag-of-words approach, a histogram representation was used in document classification based on the frequencies of extracted words and adapted for time series. Two machine learning algorithms were used separately, namely, K-nearest neighbor (K-NN) and support vector machines (SVM). A comparison between two sets of fNIRS channels was made in our classification task. The results showed that K-NN obtained slightly better results (92.1%) than SVM (91.3%) with all the 24 channels; however, the performances slightly dropped if using only channels from the region of interest with K-NN (91.5%) and SVM (90.8%). These research results encourage potential applications of fNIRS in the development of a physiologically based diagnosis of human pain, including in clinical parties.