Model-based Classification of Pain fMRI Data

Purpose: Classification of Somatoform Pain Disorder with multivariate pattern recognition methods applied to fMRI data.

Materials and Methods: 12 subjects with somatoform pain disorder and 13 age-matched healthy controls obtained task-fMRI. The task involved noxious heat stimulation on the inner side of the left forearm. Imaging has been performed with a 1.5 T scanner with EPI sequences (TE: 50ms, TR: 2.51s, voxel size 3×3 x 5mm). After standard pre-processing and automatic parcellation into 90 ROIs by AAL labelling, the data has been analyzed with a novel model-based classifier especially designed for multivariate time series.

Results: Somatoform pain disorder has been identified with high accuracy. With leave-one-out validation only one of the subjects has been incorrectly classified. Our model-based classifier found differences in the interaction patterns among the ROIs in subjects with somatoform pain disorder. Most evident, the right amygdale shows increased connectivity to frontal control areas (superior frontal) in patients.

Conclusion: Our results indicate that multivariate pattern recognition methods applied to fMRI data provide the potential to contribute to a better understanding of somatoform pain disorder.