A Site-Wise Reliability Analysis of the ABCD Diffusion Fractional Anisotropy and Cortical Thickness: Impact of Scanner Platforms

Authors

Yezhi Pan
L Elliot Hong
Ashley Acheson
Paul M Thompson
Neda Jahanshad
Alyssa H Zhu
Jiaao Yu
Chixiang Chen
Tianzhou Ma
Ho-Ling Liu
Jelle Veraart
Els Fieremans
Nicole R Karcher
Peter Kochunov
Shuo Chen

Publication Date

11-1-2024

Journal

Human Brain Mapping

Abstract

The Adolescent Brain and Cognitive Development (ABCD) project is the largest study of adolescent brain development. ABCD longitudinally tracks 11,868 participants aged 9-10 years from 21 sites using standardized protocols for multi-site MRI data collection and analysis. While the multi-site and multi-scanner study design enhances the robustness and generalizability of analysis results, it may also introduce nonbiological variances including scanner-related variations, subject motion, and deviations from protocols. ABCD imaging data were collected biennially within a period of ongoing maturation in cortical thickness and integrity of cerebral white matter. These changes can bias the classical test-retest methodologies, such as intraclass correlation coefficients (ICC). We developed a site-wise adaptive ICC (AICC) to evaluate the reliability of imaging-derived phenotypes while accounting for ongoing brain development. AICC iteratively estimates the population-level age-related brain development trajectory using a weighted mixed model and updates age-corrected site-wise reliability until convergence. We evaluated the test-retest reliability of regional fractional anisotropy (FA) measures from diffusion tensor imaging and cortical thickness (CT) from structural MRI data for each site. The mean AICC for 20 FA tracts across sites was 0.61 ± 0.19, lower than the mean AICC for CT in 34 regions across sites, 0.76 ± 0.12. Remarkably, sites using Siemens scanners consistently showed significantly higher AICC values compared with those using GE/Philips scanners for both FA (AICC = 0.71 ± 0.12 vs. 0.46 ± 0.17, p <  0.001) and CT (AICC = 0.80 ± 0.10 vs. 0.69 ± 0.11, p <  0.001). These findings demonstrate site-and-scanner related variations in data quality and underscore the necessity for meticulous data curation in subsequent association analyses.

Keywords

Humans, Child, Male, Female, Reproducibility of Results, Diffusion Tensor Imaging, Anisotropy, Longitudinal Studies, Cerebral Cortex, Brain Cortical Thickness, Adolescent, brain development, diffusion tensor imaging, longitudinal, quality control, structural MRI, test–retest reliability

DOI

10.1002/hbm.70070

PMID

39526419

PMCID

PMC11551787

PubMedCentral® Posted Date

11-11-2024

PubMedCentral® Full Text Version

Post-print

Published Open-Access

yes