An automatic classifier of pain scores in chronic pain patients from local field potentials recordings

Abstract

This paper investigates measures to assess automatically the level of pain in a group of chronic pain patients implanted with electrodes for deep brain stimulation. Electrical activity in local field potentials in particular frequency bands has been shown to be associated with changes in the perception of pain. In this paper we develop a method to classify pain intensity with two groups of patients, one with electrodes implanted in the thalamus (VPL) and the other with implants in periaqueductal grey (PAG/PVG), using wavelet analysis to process the local field potential data from the deep brain electrodes. A fuzzy network classifier is used to relate sections of the data to the pain intensity, as recorded by patients using a visual analogue scale (VAS) scale. Our results suggest that in the PAG implanted patients alpha activity is a good measure of pain in a single patient, whereas correlation with beta activity is more appropriate in thalamus implanted patients. The relation between such activity and pain level shows some consistency within a session. This suggests that a closed loop form of DBS may be possible for these patients to optimize their treatment. However it was not possible to train a classifier consistently across the groups of patients, possibly because of differences in pain perception across individuals.