cohort·epidemiology, research methods, observational study, preclinical research·PMC10090309
Better Together
Investigative Radiology · 25 authors, 26 centres
AI SUMMARY
FIDELITY 100%
This summary was generated by AI from a single paper. It has not been reviewed by a clinician and is not clinical advice. Verify against the source before acting on it.
After using causal discovery to identify data-source biases and ComBat harmonization to correct them, the authors derived quantitative reference models for phenotypic variation of abdominal organs across the general adult population.
Full summary
1,010 CHARS
Image-derived features of abdominal organs (liver, spleen, kidney, and pancreas volumes; kidney hilum adipose tissue volume; liver and pancreas fat fractions) were extracted from T1-weighted Dixon MRI data of UKBB and NAKO participants using deep learning segmentations. Causal discovery methods revealed direct effects of age, sex, height, weight, and data source on image-derived features. ComBat harmonization was applied with UKBB as the reference dataset to correct source-related biases. Quantile regression on harmonized data provided models stratified by sex and dependent on age, height, and weight. Harmonization markedly improved alignment of image-derived features between studies. Limitations include that feature extraction could be improved using dedicated multiecho sequences for PDFF estimation, that ComBat reference selection (UKBB) was based on an assumption about scanner robustness, and that unobserved confounders (e.g., ethnicity, lifestyle) between cohorts could not be fully excluded.
