Diagnostic classification based on functional connectivity in chronic pain: model optimization in fibromyalgia and rheumatoid arthritis.

RATIONALE AND

OBJECTIVES

The combination of functional magnetic resonance imaging (fMRI) of the brain with multivariate pattern analysis (MVPA) has been proposed as a possible diagnostic tool. Goal of this investigation was to identify potential functional connectivity (FC) differences in the salience network (SN) and default mode network (DMN) between fibromyalgia syndrome (FMS), rheumatoid arthritis (RA), and controls (HC) and to evaluate the diagnostic applicability of derived pattern classification approaches.

MATERIALS AND

METHODS

The resting period during an fMRI examination was retrospectively analyzed in women with FMS (n = 17), RA (n = 16), and HC (n = 17). FC was calculated for SN and DMN subregions. Classification accuracies of discriminative MVPA models were evaluated with cross-validation: (1) inferential test of a single method, (2) explorative model optimization.

RESULTS

No inferentially tested model was able to classify subjects with statistically significant accuracy. However, the diagnostic ability for the differential diagnostic problem exhibited a trend to significance (accuracy: 69.7%, P = .086). Optimized models in the explorative analysis reached accuracies up to 73.5% (FMS vs. HC), 78.8% (RA vs. HC), and 78.8% (FMS vs. RA) whereas other models performed at or below chance level. Comparable support vector machine approaches performed above average for all three problems.

CONCLUSIONS

Observed accuracies are not sufficient to reliably differentiate between FMS and RA for diagnostic purposes. However, some indirect evidence in support of the feasibility of this approach is provided. This exploratory analysis constitutes a fundamental model optimization effort to be based on in further investigations.