Related Experiment Video
Updated: May 5, 2026
17:06
Co-analysis of Brain Structure and Function using fMRI and Diffusion-weighted Imaging
Published on: November 8, 2012
26.4K
Diffusion based multi-domain neuroimaging harmonization method with preservation of anatomical details
Haoyu Lan1, Bino A Varghese2, Nasim Sheikh-Bahaei2
1Laboratory of Neuro Imaging, USC Mark and Mary Stevens Neuroimaging and Informatics Institute, USC Keck School of Medicine, University of Southern California, Los Angeles, CA, USA.
Neuroimage
|May 28, 2025
Summary
This study introduces a diffusion model for neuroimaging harmonization, effectively reducing batch effects in MRI scans. The diffusion model preserves anatomical details better than Generative Adversarial Networks (GANs), improving data reliability for multi-center studies.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Image Analysis
Background:
- Multi-center neuroimaging studies face challenges with technical variability due to batch differences, impacting data aggregation and reliability.
- Existing harmonization methods, like Generative Adversarial Networks (GANs), can introduce artifacts or anatomical distortions.
Purpose of the Study:
- To assess the efficacy of denoising diffusion probabilistic models for neuroimaging harmonization.
- To demonstrate a superior method for harmonizing images across multiple domains using a single model.
Main Methods:
- Utilized a denoising diffusion probabilistic model for neuroimaging harmonization.
- Developed a method incorporating learned domain invariant anatomical conditions to preserve anatomy while differentiating batch effects.
- Tested the model on T1-weighted MRI images from the ADNI1 and ABIDE II datasets.
Main Results:
- The diffusion model achieved superior harmonization results compared to GAN-based methods, evidenced by consistent anatomy preservation and better Fréchet Inception Distance (FID) scores.
- Demonstrated the model's capability to harmonize images across multiple domains with a single model.
- Showcased improved consistency in perivascular spaces segmentation and volumetric analysis post-harmonization.
Conclusions:
- Denoising diffusion probabilistic models offer a promising approach for neuroimaging harmonization, outperforming GANs in anatomical detail preservation and cross-domain harmonization.
- The proposed method effectively reduces technical variability in multi-center studies, enhancing the reliability of neuroimaging data analysis.
