Establishing fMRI Test-Retest Reliability: Implications for fMRI Biomarkers of Treatment Efficacy

Background: Little is known about the reproducibility of fMRI data at the level of the individual person.  Characterizing fMRI measurement reliability is important for studies that aim to use fMRI data to make individual level predictions, e.g., for studies that use fMRI to measure the effectiveness of an intervention, or to assess change across development.  Unreliable measurements cannot successfully document meaningful changes that actually occur.

Objectives: To estimate the reliability of fMRI signals across time during a face recognition experiment over a 9-week interval in typically developing controls (TDC) and youth with an autism spectrum disorder (ASD).  

Methods:  Sixty-five participants (63 male) aged 12-17 (M= 14.78, SD= 1.66) participated in a study comprised of fMRI and behavioral markers at two time points separated by an interval of 9 weeks. Thirty-five participants had a diagnosis of ASD based on expert clinical diagnosis, ADOS and ADI-R scores.  Thirty were recruited as TDCs. All participants were administered a block-design fMRI face-processing task consisting of 5 runs of both active (identity discrimination) and passive face viewing.  Each run alternated between blocks of faces and blocks of houses requiring discrimination. Data were analyzed in FSL and with matlab scripts developed in our lab. Data analyses for 43 (29 ASD) of the 65 participants are presented here, with the remainder to be complete for the conference presentation.

Results: Interclass correlation coefficients (ICC) were computed for specific regions of interest (ROI) that were significantly active across groups at the whole brain level. Collapsing across groups, mean z scores within functionally defined ROIs (constrained within an anatomical search space) showed significant test-retest reliability for the fusiform gyrus (right: ICC=0.72, p=0.000; left: ICC=0.77, p=0.000) and the parahippocampal gyrus (right: ICC=0.86, p= 0.000; left: ICC=0.79, p= 0.000).  However, mean signal from the amygdala did not show significant test-retest reliability. 

Conclusions: Cortical regions that are face and house selective show robust test-retest reliability across the typical length of a medication trial with the stimuli, task design and other procedures used here. Thus, interventions that might target functional regions like these would be expected to show fMRI related signal change, should real neuronal level improvement occur with treatment.  Notably, these preliminary data show better face-processing signal reliability for the medial fusiform gyrus, while group level contrasts of faces vs. houses show greater lateral fusiform activation, which is typical with the design of this study.  Similarly, the group showed robust bilateral amygdala activation for the face vs. house contrast, but amygdala reliability was low and not significant in this portion of the total sample.  Conference presentation will present the full data set and analyses that examine possible reasons for lower reliability in some regions (e.g., magnetic susceptibility artifacts at the interface of sinus and ear canals).  We will also examine possible group differences in reliability and describe implications for using fMRI biomarkers in clinical translational research.