A comparison of region-of-interest measures for extracting whole brain data using survival analysis in alcoholism as an example.

Abstract

Aggregation of functional magnetic resonance imaging (fMRI) data in regions-of-interest (ROIs) is required for complex statistical analyses not implemented in standard fMRI software. Different data-aggregation measures assess various aspects of neural activation, including spatial extent and intensity. In this study, conducted within the framework of the PREDICT study, we compared different aggregation measures for voxel-wise fMRI activations to be used as prognostic factors for relapse in 49 abstinent alcohol-dependent individuals in an outpatient setting using a cue-reactivity task. We compared the importance of the data-aggregation measures as prognostic factors for treatment outcomes by calculating the proportion of explained variation. Relapse risk was associated with cue-induced brain activation during abstinence in the ventral striatum (VS) and in the orbitofrontal cortex (OFC). While various ROI measures proved appropriate for using fMRI cue-reactivity to predict relapse, on the descriptive level the most "important" prognostic factor was a measure defined as the sum of t-values exceeding an individually defined threshold. Data collected in the VS was superior to that from other regions. In conclusion, it seems that fMRI cue-reactivity, especially in the VS, can be used as prognostic factor for relapse in abstinent alcohol-dependent patients. Our findings suggest that data-aggregation measures that take both spatial extent and intensity of cue-induced brain activation into account make better biomarkers for predicting relapse than measures that consider an activation's spatial extent or intensity alone.