Performance of the Sparse Logistic Regression Classifier Chronic Pain versus Normal Group.

*Significant at p<0.05 utilizing 1000 permutation tests of sparse logistic regression SLR classification over randomly shuffled labels of the subjects in the training set. Percent correct was determined by the mean performance of the 26 classifiers trained using the leave one out cross validation method. Measures of sensitivity, specificity, positive predictive value, negative predictive value, posterior mean accuracy, posterior probability interval (p<0.05) and D Prime are also given. Posterior mean accuracy and posterior probability intervals were computed using methods given in Brodersen et al. (2010). The mask of voxels included in the analysis consisted of brain regions composing pain related areas: Primary Somatosensory Cortex, Secondary Somatosensory Cortex, Inferior Parietal Cortex, Insula, and Anterior Cingulate Cortex. The mask consisted of 6686 voxels. After training, 22 of the classifiers selected 3 features and 4 or the classifiers selected 2 features. The mean number of extracted features was 2.85.