Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion based multi-domain neuroimaging harmonization method with preservation of anatomical details

Haoyu Lan, Bino A. Varghese, Nasim Sheikh-Bahaei, Farshid Sepehrband, Arthur W Toga, Jeiran Choupan

TL;DR

This work has demonstrated the diffusion model's superior capability in harmonizing images across multiple domains with single model, and tested the efficacy of the diffusion model for neuroimaging harmonization using T1-weighted MRI images from two public neuroimaging datasets of ADNI1 and ABIDE II.

Abstract

Multi-center neuroimaging studies face technical variability due to batch differences across sites, which potentially hinders data aggregation and impacts study reliability.Recent efforts in neuroimaging harmonization have aimed to minimize these technical gaps and reduce technical variability across batches. While Generative Adversarial Networks (GAN) has been a prominent method for addressing image harmonization tasks, GAN-harmonized images suffer from artifacts or anatomical distortions. Given the advancements of denoising diffusion probabilistic model which produces high-fidelity images, we have assessed the efficacy of the diffusion model for neuroimaging harmonization. we have demonstrated the diffusion model's superior capability in harmonizing images from multiple domains, while GAN-based methods are limited to harmonizing images between two domains per model. Our experiments highlight that the learned domain invariant anatomical condition reinforces the model to accurately preserve the anatomical details while differentiating batch differences at each diffusion step. Our proposed method has been tested on two public neuroimaging dataset ADNI1 and ABIDE II, yielding harmonization results with consistent anatomy preservation and superior FID score compared to the GAN-based methods. We have conducted multiple analysis including extensive quantitative and qualitative evaluations against the baseline models, ablation study showcasing the benefits of the learned conditions, and improvements in the consistency of perivascular spaces (PVS) segmentation through harmonization.

Diffusion based multi-domain neuroimaging harmonization method with preservation of anatomical details

TL;DR

This work has demonstrated the diffusion model's superior capability in harmonizing images across multiple domains with single model, and tested the efficacy of the diffusion model for neuroimaging harmonization using T1-weighted MRI images from two public neuroimaging datasets of ADNI1 and ABIDE II.

Abstract

Multi-center neuroimaging studies face technical variability due to batch differences across sites, which potentially hinders data aggregation and impacts study reliability.Recent efforts in neuroimaging harmonization have aimed to minimize these technical gaps and reduce technical variability across batches. While Generative Adversarial Networks (GAN) has been a prominent method for addressing image harmonization tasks, GAN-harmonized images suffer from artifacts or anatomical distortions. Given the advancements of denoising diffusion probabilistic model which produces high-fidelity images, we have assessed the efficacy of the diffusion model for neuroimaging harmonization. we have demonstrated the diffusion model's superior capability in harmonizing images from multiple domains, while GAN-based methods are limited to harmonizing images between two domains per model. Our experiments highlight that the learned domain invariant anatomical condition reinforces the model to accurately preserve the anatomical details while differentiating batch differences at each diffusion step. Our proposed method has been tested on two public neuroimaging dataset ADNI1 and ABIDE II, yielding harmonization results with consistent anatomy preservation and superior FID score compared to the GAN-based methods. We have conducted multiple analysis including extensive quantitative and qualitative evaluations against the baseline models, ablation study showcasing the benefits of the learned conditions, and improvements in the consistency of perivascular spaces (PVS) segmentation through harmonization.
Paper Structure (21 sections, 1 equation, 7 figures, 2 tables, 1 algorithm)

This paper contains 21 sections, 1 equation, 7 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Proposed Method
  4. Multi-domain neuroimaging harmonization as a controllable domain adaption problem
  5. Learned domain invariant conditioned controllable denoising diffusion model
  6. Skip sampling
  7. Experiments and Results
  8. Dtaset
  9. Perivascular spaces
  10. Experimental setup
  11. Domain embedding controls DDPM to learn multiple diffusion trajectories
  12. Evaluation of harmonization on ABIDE II
  13. Harmonization improves consistency of PVS count ratio across MRI scanners on ADNI 1
  14. Ablation study (learned condition improves the generated image quality)
  15. Discussion and Conclusion
  16. ...and 6 more sections

Figures (7)

  • Figure 1: The overall framework of the proposed method with training and sampling details.
  • Figure 2: Visual demonstration of the learned trajectories. Domain embedding Z steers the sampling trajectories to achieve the multi-domain harmonization. Comparison between left and right side of the figure demonstrates the effectiveness of harmonization at each sampling step.
  • Figure 3: Qualitative results of the harmonization on ABIDE II. The far-left column shows samples of target image domains (reference of imaging texture), and the top row shows samples of source image domain (reference of anatomical details). The red dash squares highlight the harmonization case where target image domain equals to the source image domain.
  • Figure 4: Qualitative comparison with baseline models. Red arrows indicate the areas with generated artifacts and mild stripe patterns which are not present in both source and target image domain.
  • Figure 5: PVS segmentation quantification.A. Harmonized images show reduced heterogeneity regarding PVS count ratio across scanners as shown in orange, green and red boxplots compared to the blue boxplots. B. Visualization of the increased PVS count ratio derived from the harmonized image (blue PVS) in comparison with the original image (green PVS)
  • ...and 2 more figures