Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generating healthy counterfactuals with denoising diffusion bridge models

Ana Lawry Aguila, Peirong Liu, Marina Crespo Aguirre, Juan Eugenio Iglesias

TL;DR

This work tackles the challenge of applying healthy-image analysis to pathology by generating realistic healthy counterfactuals that preserve subject-specific anatomy. It introduces Denoising Diffusion Bridge Models (DDBMs), which condition diffusion on both the initial healthy image and a synthetic pathological target, using a bridge-score matching objective and diffusion-bridge implicit model (DBIM) sampling to map between healthy and pathological domains. Training on healthy–pseudo-pathology pairs generated via UNA, the approach yields state-of-the-art performance in brain tissue segmentation and anomaly detection compared with DDPMs and supervised baselines, across multiple MR datasets. By enabling accurate pseudo-healthy references, DDBMs facilitate the application of healthy-brain analysis tools to diseased scans and improve lesion localization, offering a practical impact for medical image analysis and clinical decision support.

Abstract

Generating healthy counterfactuals from pathological images holds significant promise in medical imaging, e.g., in anomaly detection or for application of analysis tools that are designed for healthy scans. These counterfactuals should represent what a patient's scan would plausibly look like in the absence of pathology, preserving individual anatomical characteristics while modifying only the pathological regions. Denoising diffusion probabilistic models (DDPMs) have become popular methods for generating healthy counterfactuals of pathology data. Typically, this involves training on solely healthy data with the assumption that a partial denoising process will be unable to model disease regions and will instead reconstruct a closely matched healthy counterpart. More recent methods have incorporated synthetic pathological images to better guide the diffusion process. However, it remains challenging to guide the generative process in a way that effectively balances the removal of anomalies with the retention of subject-specific features. To solve this problem, we propose a novel application of denoising diffusion bridge models (DDBMs) - which, unlike DDPMs, condition the diffusion process not only on the initial point (i.e., the healthy image), but also on the final point (i.e., a corresponding synthetically generated pathological image). Treating the pathological image as a structurally informative prior enables us to generate counterfactuals that closely match the patient's anatomy while selectively removing pathology. The results show that our DDBM outperforms previously proposed diffusion models and fully supervised approaches at segmentation and anomaly detection tasks.

Generating healthy counterfactuals with denoising diffusion bridge models

TL;DR

This work tackles the challenge of applying healthy-image analysis to pathology by generating realistic healthy counterfactuals that preserve subject-specific anatomy. It introduces Denoising Diffusion Bridge Models (DDBMs), which condition diffusion on both the initial healthy image and a synthetic pathological target, using a bridge-score matching objective and diffusion-bridge implicit model (DBIM) sampling to map between healthy and pathological domains. Training on healthy–pseudo-pathology pairs generated via UNA, the approach yields state-of-the-art performance in brain tissue segmentation and anomaly detection compared with DDPMs and supervised baselines, across multiple MR datasets. By enabling accurate pseudo-healthy references, DDBMs facilitate the application of healthy-brain analysis tools to diseased scans and improve lesion localization, offering a practical impact for medical image analysis and clinical decision support.

Abstract

Generating healthy counterfactuals from pathological images holds significant promise in medical imaging, e.g., in anomaly detection or for application of analysis tools that are designed for healthy scans. These counterfactuals should represent what a patient's scan would plausibly look like in the absence of pathology, preserving individual anatomical characteristics while modifying only the pathological regions. Denoising diffusion probabilistic models (DDPMs) have become popular methods for generating healthy counterfactuals of pathology data. Typically, this involves training on solely healthy data with the assumption that a partial denoising process will be unable to model disease regions and will instead reconstruct a closely matched healthy counterpart. More recent methods have incorporated synthetic pathological images to better guide the diffusion process. However, it remains challenging to guide the generative process in a way that effectively balances the removal of anomalies with the retention of subject-specific features. To solve this problem, we propose a novel application of denoising diffusion bridge models (DDBMs) - which, unlike DDPMs, condition the diffusion process not only on the initial point (i.e., the healthy image), but also on the final point (i.e., a corresponding synthetically generated pathological image). Treating the pathological image as a structurally informative prior enables us to generate counterfactuals that closely match the patient's anatomy while selectively removing pathology. The results show that our DDBM outperforms previously proposed diffusion models and fully supervised approaches at segmentation and anomaly detection tasks.
Paper Structure (14 sections, 8 equations, 2 figures, 2 tables)

This paper contains 14 sections, 8 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Denoising diffusion probabilistic models
  4. Denoising diffusion bridge models
  5. DDBMs for healthy counterfactual generation
  6. Experimental setup
  7. Datasets and comparison methods
  8. Segmentation results: counterfactuals vs. ground-truth
  9. Anomaly detection in real pathology
  10. Conclusion
  11. Acknowledgements
  12. Technical Appendices and Supplementary Material
  13. DDBM implementation details
  14. Data pre-processing

Figures (2)

  • Figure 1: Forward and reverse process of the proposed DDBM framework to map between synthetic pseudo-pathology and real healthy volumes. The denoising bridge score matching objective (Equation \ref{['eq:DBSM']}) is used to match a parameterized network $\mathbf{s}_\theta$ to the unknown bridge score function $\nabla_{\mathbf{x}_t} \log p(\mathbf{x}_t)$.
  • Figure 2: Qualitative examples of anomaly detection on test subjects from the ATLAS dataset. Each subfigure shows, from left to right: the original pathological scan, the healthy counterfactual generated by the DDBM, the corresponding anomaly maps, and the ground-truth manual segmentation.