Related Experiment Video
Updated: Jul 31, 2025
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.2K
DeepComBat: A Statistically Motivated, Hyperparameter-Robust, Deep Learning Approach to Harmonization of Neuroimaging
Fengling Hu1, Alfredo Lucas2, Andrew A Chen1
1Penn Statistics in Imaging and Visualization Endeavor (PennSIVE), Department of Biostatistics, Epidemiology, and Informatics, Perelman School of Medicine, University of Pennsylvania.
Biorxiv : the Preprint Server for Biology
|May 10, 2023
Summary
DeepCombat, a novel deep learning method, effectively removes batch effects in neuroimaging data. This harmonization enhances the generalizability and reproducibility of brain imaging studies using machine learning.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Biostatistics
Background:
- Multi-batch neuroimaging data are crucial for brain research but contain technical artifacts (batch effects).
- Batch effects confound data, obscuring biological signals and reducing generalizability and reproducibility, especially with machine learning.
- Current harmonization methods, including statistical and deep learning approaches, often fail to remove significant multivariate batch effects.
Conclusions:
- DeepCombat offers an advanced deep learning approach for harmonizing neuroimaging data, outperforming current methods.
- The technique improves the reliability of multi-batch neuroimaging studies, particularly those employing machine learning.
- DeepCombat provides a new perspective for statistically-motivated deep learning harmonization strategies.
