Related Experiment Video
Updated: Sep 18, 2025
17:06
Co-analysis of Brain Structure and Function using fMRI and Diffusion-weighted Imaging
Published on: November 8, 2012
26.4K
Big Data, Small Bias: Harmonizing Diffusion MRI-Based Structural Connectomes to Mitigate Site-Related Bias in Data
Rui Sherry Shen1,2, Drew Parker2, Andrew An Chen3
1Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania, USA.
Human Brain Mapping
|June 26, 2025
Summary
This study introduces MATCH, a novel harmonization framework for multi-site diffusion MRI connectome data. MATCH effectively removes site-specific biases, preserving biological information for more reliable brain connectivity research.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Data Science
Background:
- Diffusion MRI-based structural connectomes are vital for studying brain connectivity in disorders.
- Pooling multi-site data is crucial due to small sample sizes and heterogeneous manifestations.
- Site-related differences from scanners and protocols necessitate data harmonization.
Purpose of the Study:
- To develop and evaluate a statistical harmonization framework tailored for structural connectomes.
- To address the challenge of site-related biases in multi-site neuroimaging datasets.
- To preserve biological variability while mitigating technical artifacts.
Main Methods:
- Investigated various statistical harmonization methods adapted for connectome data.
- Developed the MATCH algorithm based on a gamma-distributed model.
- Evaluated harmonization performance on edge-based and graph analyses.
Main Results:
- The MATCH algorithm effectively models structural connectomes and removes site-related biases.
- MATCH outperforms existing methods in harmonizing connectome data.
- Harmonization enhanced machine learning predictor generalizability and group-level difference detection.
Conclusions:
- MATCH provides a robust framework for harmonizing multi-site structural connectomes.
- This approach is essential for reliable discoveries in collaborative, big data neuroimaging research.
- Guidelines are provided for effective multi-site connectome analysis.
