Abstract
Background
High-resolution structural MRI has been widely used in clinical research to detect and quantify subtle brain changes in patient populations. Findings from prospective, longitudinal studies show structural brain abnormalities as well as progressive gray matter changes over time in individuals at clinical high risk for psychosis compared to healthy subjects. In recent years, research in this field has seen an increase in multicentre neuroimaging projects, such as EU-GEI, PSYSCAN, PRONIA and NAPLS. Additional sources of variance, alongside known technological and biological factors, may be introduced when MRI images are acquired and combined from different sites. It is imperative for longitudinal multicentre studies to determine the accuracy of quantitative MRI measurements and account for systematic differences both between scanners and across scanning sessions. This is particularly true within psychosis research where morphometric changes as small as 3% or less are expected.
Methods
Six healthy participants were scanned on four separate occasions over a two-month period at King’s College London; twice on a GE SIGNA HDx 3T scanner used locally in the EU-GEI High Risk Study and twice on a GE MR750 3T scanner used locally in the PSYSCAN study. Both scanners implemented the ADNI-2 T1 protocol which is used globally across the EU-GEI and PSYSCAN consortia. Structural imaging data was segmented using the FreeSurfer 6.0 longitudinal pipeline. Intraclass correlation coefficients (ICCs) with a two-way mixed effects model of absolute agreement were calculated to assess intra- and inter-scanner reliability of brain morphometry. For volumetric studies, ICC values greater than 0.9 indicate ‘excellent’ reliability. Reliability analyses of key regions implicated in psychosis included gray matter volume estimates of the hippocampus, insula, lateral ventricle, orbitofrontal cortex and anterior cingulate cortex, and average cortical thickness measurements of the whole brain, parahippocampus and superior frontal cortex.
Results
Gray matter volume estimates of all structures yielded ‘excellent’ reliability for both intra-scanner (ICCs of 0.979 – 0.998) and inter-scanner analyses (ICCs of 0.976 – 0.999). Intra-scanner reliability for mean cortical thickness measurements was ‘excellent’ for right total cortex, resulting in an ICC of 0.901, but otherwise ‘good’ for left and total cortex, parahippocampus, superior frontal cortex (ICCs of 0.754 – 0.875). Inter-scanner reliability for mean cortical thickness estimates were most variable across the brain structures. Here, results demonstrated ‘excellent’ reliability for the parahippocampus and left total cortex (ICCs of 0.907 – 0.965), ‘good’ for total cortex (ICC of 0.835), ‘moderate’ for right total cortex, right and total superior frontal cortex (ICCs of 0.520 – 0.676), and ‘poor’ for the left superior frontal cortex which produced an ICC of 0.470. Overall, mean cortical thickness estimates of the superior frontal cortex from two different MR scanners showed the least reliability.
Discussion
Results confirmed highly reliable estimates for gray matter volumes in all brain structures, both from images acquired within the same scanner and across two different scanners. However, the findings indicated increased variability of mean cortical thickness estimates, particularly between scanners, which should be considered when interpreting study findings. Multicentre structural neuroimaging within the field of psychosis is becoming more common and it must be acknowledged that combining MRI data in multicentre studies will contribute additional sources of variance and potential bias with certain brain regions affected more than others.
Genes Encoding Heterotrimeric G-proteins Are Associated with Gray Matter Volume Variations in the Medial Frontal Cortex
