Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Counterfactual MRI Data Augmentation using Conditional Denoising Diffusion Generative Models

Pedro Morão, Joao Santinha, Yasna Forghani, Nuno Loução, Pedro Gouveia, Mario A. T. Figueiredo

TL;DR

This work introduces a novel method using conditional denoising diffusion generative models (cDDGMs) to generate counterfactual magnetic resonance (MR) images that simulate different IAP without altering patient anatomy, and shows promise in addressing domain and covariate shifts in medical imaging.

Abstract

Deep learning (DL) models in medical imaging face challenges in generalizability and robustness due to variations in image acquisition parameters (IAP). In this work, we introduce a novel method using conditional denoising diffusion generative models (cDDGMs) to generate counterfactual magnetic resonance (MR) images that simulate different IAP without altering patient anatomy. We demonstrate that using these counterfactual images for data augmentation can improve segmentation accuracy, particularly in out-of-distribution settings, enhancing the overall generalizability and robustness of DL models across diverse imaging conditions. Our approach shows promise in addressing domain and covariate shifts in medical imaging. The code is publicly available at https: //github.com/pedromorao/Counterfactual-MRI-Data-Augmentation

Counterfactual MRI Data Augmentation using Conditional Denoising Diffusion Generative Models

TL;DR

This work introduces a novel method using conditional denoising diffusion generative models (cDDGMs) to generate counterfactual magnetic resonance (MR) images that simulate different IAP without altering patient anatomy, and shows promise in addressing domain and covariate shifts in medical imaging.

Abstract

Deep learning (DL) models in medical imaging face challenges in generalizability and robustness due to variations in image acquisition parameters (IAP). In this work, we introduce a novel method using conditional denoising diffusion generative models (cDDGMs) to generate counterfactual magnetic resonance (MR) images that simulate different IAP without altering patient anatomy. We demonstrate that using these counterfactual images for data augmentation can improve segmentation accuracy, particularly in out-of-distribution settings, enhancing the overall generalizability and robustness of DL models across diverse imaging conditions. Our approach shows promise in addressing domain and covariate shifts in medical imaging. The code is publicly available at https: //github.com/pedromorao/Counterfactual-MRI-Data-Augmentation

Paper Structure

This paper contains 17 sections, 2 equations, 2 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Image Acquisition Parameters Prediction Model
  5. Conditional Denoising Diffusion Generative Model
  6. Breast Tissue Segmentation Model
  7. IAP, cDDGM, and Segmentation Models Training
  8. Evaluation metrics
  9. Results and Discussion
  10. Conclusions
  11. Appendix
  12. Data Normalization and Preprocessing
  13. Train, Validation and Test sets
  14. cDDGM Arquitecture
  15. Training Setup
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Comparison between the ground truth (True) and DL breast segmentation models trained without data augmentation (Pred.) and with data augmentation using cDDGM (Pred.Aug.), in out-of-distribution settings. (A) results of models trained in GE when applied to Siemens MRIs. (B) results of models trained in Siemens when applied to GE MRIs. Blue - Fat mask; Orange - FGT mask.
  • Figure 2: Comparison between the ground truth (True) and DL breast segmentation models trained without data augmentation (Pred.) and with data augmentation using cDDGM (Pred.Aug.), in in-distribution settings. (A) results of models trained in GE when applied to GE MRIs. (B) results of models trained in GE when applied to GE MRIs. Blue - Fat mask; Orange - FGT mask.