A site-wise reliability analysis of the ABCD diffusion fractional anisotropy data, impact of the scanner and analytical pipeline

The Adolescent Brain and Cognitive Development (ABCD) project is the largest longitudinal study of brain development that tracts 11,820 subjects from 21 sites using standardized protocols for multi-site data collection and analysis. Adolescence is a critical period of brain development associated with white matter myelination and requires reliable measures to detect these changes. We assessed confounding non-biological variances in diffusion tensor imaging (DTI) data that may be present due to technological variations, participant compliance and data analysis protocols. ABCD imaging data were collected biannually, and thus ongoing maturation may artificially introduce bias to classical test-retest approaches such as the interclass correlation coefficients (ICC). To address this, we developed a site-wise adaptive ICC (AICC) to systematically evaluate the quality of imaging-derived phenotypes while accounting for ongoing brain development. We measured the age-related brain development trajectory and estimated site-wise AICC iteratively, adjusting the weight for each site based on the ICC scores. We evaluated longitudinal reliability of diffusion fractional anisotropy (FA) data for each site, considering the impact of MRI scanner platform and standard ABCD versus ENIGMA-DTI data extraction pipelines, and comparing longitudinal stability of FA measurements to these of the cortical thickness. The ENIGMA structural and diffusion pipeline with QA/QC improved the average reliability for cortical FA to AICC=0.70±0.19, compared to 0.61±0.19 for the standard ABCD pipeline (Wilcoxon test p<0.001). Furthermore, we showed that the AICC for sites that used Siemens scanners significantly outperformed those using GE/Phillips scanners (AICC=0.78±0.11 vs 0.62±0.21, p<0.001). In conclusion, variations in data quality among study sites and preprocessing pipelines underscore the necessity for meticulous data curation in subsequent association analyses. ### Competing Interest Statement The authors have declared no competing interest.