Related Experiment Video
Updated: Aug 6, 2025
13:08
Measurement of Fronto-limbic Activity Using an Emotional Oddball Task in Children with Familial High Risk for Schizophrenia
Published on: December 2, 2015
9.0K
ABCD_Harmonizer: An Open-source Tool for Mapping and Controlling for Scanner Induced Variance in the Adolescent Brain
Jonathan A Dudley1, Thomas C Maloney2, John O Simon3
1Department of Radiology, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, USA. Jonathan.Dudley@cchmc.org.
Neuroinformatics
|March 20, 2023
Summary
Scanner effects in brain imaging data introduce significant variance, often exceeding biological factors like age and sex. Data harmonization using ComBat effectively removes this scanner-induced variance, improving effect size estimation in large studies like the ABCD dataset.
Area of Science:
- Neuroimaging
- Data Science
- Biostatistics
Background:
- Multisite magnetic resonance imaging (MRI) studies, including the large-scale Adolescent Cognitive Brain Development (ABCD) study, face challenges with scanner-induced variance.
- This variance can reduce statistical power and bias results in structural MRI (sMRI) and diffusion MRI (dMRI) data.
- Scanner effects can be substantial, often outweighing biological variability from factors like age and sex.
Purpose of the Study:
- Quantify scanner-induced variance in sMRI and dMRI data from the ABCD study.
- Demonstrate the effectiveness of the ComBat harmonization method in addressing scanner effects.
- Present an open-source tool for harmonizing ABCD study image features.
Main Methods:
- Analysis of sMRI (e.g., cortical thickness) and dMRI (e.g., fractional anisotropy) data from the ABCD study.
- Quantification of variance attributable to 29 different scanners across 5 models and 3 vendors.
- Application and evaluation of the ComBat data harmonization technique.
- Comparison of ComBat harmonization with ordinary least squares regression for controlling scanner effects.
Main Results:
- Scanner-induced variance was detected in all analyzed image features, varying by feature type and brain location.
- For most features, scanner variance was greater than variability attributed to age and sex.
- ComBat harmonization successfully removed scanner-induced variance while preserving biological variability.
- ComBat harmonized data yielded more accurate effect size estimates than traditional regression methods in subsample analyses.
Conclusions:
- Scanner effects represent a significant source of noise in multisite neuroimaging datasets like the ABCD study.
- ComBat is an effective method for harmonizing neuroimaging data, mitigating scanner-induced variance.
- Harmonized data improves the reliability and accuracy of statistical analyses, particularly for smaller subsamples.
