Predicting Pain: An Analysis of fMRI Data through Machine Learning Techniques

Abstract

Recently, the increasing availability of fMRI data has allowed for better understanding and quantification of brain activity, with applications in predicting conditions such as neurological disease or pain. In this study, it is shown that classification of pain using support vector machine (SVM) yielded maximum accuracy rates of 92% and 87% when applied to two independent samples. Various feature elimination techniques were applied and assessed in order to reach optimal prediction accuracy. In addition, functional connectivity patterns in the two conditions were examined through correlation analysis. This analysis provides a step towards an accurate physiological determinant of pain, which is predominantly identified through self-report currently.