Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Learning Contrast Kinetics with Multi-Condition Latent Diffusion Models

Richard Osuala, Daniel M. Lang, Preeti Verma, Smriti Joshi, Apostolia Tsirikoglou, Grzegorz Skorupko, Kaisar Kushibar, Lidia Garrucho, Walter H. L. Pinaya, Oliver Diaz, Julia A. Schnabel, Karim Lekadir

TL;DR

A multi-conditional latent diffusion model capable of acquisition time-conditioned image synthesis of DCE-MRI temporal sequences is proposed and the Fr\'echet radiomics distance is validated as an image quality measure based on biomarker variability between synthetic and real imaging data.

Abstract

Contrast agents in dynamic contrast enhanced magnetic resonance imaging allow to localize tumors and observe their contrast kinetics, which is essential for cancer characterization and respective treatment decision-making. However, contrast agent administration is not only associated with adverse health risks, but also restricted for patients during pregnancy, and for those with kidney malfunction, or other adverse reactions. With contrast uptake as key biomarker for lesion malignancy, cancer recurrence risk, and treatment response, it becomes pivotal to reduce the dependency on intravenous contrast agent administration. To this end, we propose a multi-conditional latent diffusion model capable of acquisition time-conditioned image synthesis of DCE-MRI temporal sequences. To evaluate medical image synthesis, we additionally propose and validate the Fréchet radiomics distance as an image quality measure based on biomarker variability between synthetic and real imaging data. Our results demonstrate our method's ability to generate realistic multi-sequence fat-saturated breast DCE-MRI and uncover the emerging potential of deep learning based contrast kinetics simulation. We publicly share our accessible codebase at https://github.com/RichardObi/ccnet and provide a user-friendly library for Fréchet radiomics distance calculation at https://pypi.org/project/frd-score.

Towards Learning Contrast Kinetics with Multi-Condition Latent Diffusion Models

TL;DR

A multi-conditional latent diffusion model capable of acquisition time-conditioned image synthesis of DCE-MRI temporal sequences is proposed and the Fr\'echet radiomics distance is validated as an image quality measure based on biomarker variability between synthetic and real imaging data.

Abstract

Contrast agents in dynamic contrast enhanced magnetic resonance imaging allow to localize tumors and observe their contrast kinetics, which is essential for cancer characterization and respective treatment decision-making. However, contrast agent administration is not only associated with adverse health risks, but also restricted for patients during pregnancy, and for those with kidney malfunction, or other adverse reactions. With contrast uptake as key biomarker for lesion malignancy, cancer recurrence risk, and treatment response, it becomes pivotal to reduce the dependency on intravenous contrast agent administration. To this end, we propose a multi-conditional latent diffusion model capable of acquisition time-conditioned image synthesis of DCE-MRI temporal sequences. To evaluate medical image synthesis, we additionally propose and validate the Fréchet radiomics distance as an image quality measure based on biomarker variability between synthetic and real imaging data. Our results demonstrate our method's ability to generate realistic multi-sequence fat-saturated breast DCE-MRI and uncover the emerging potential of deep learning based contrast kinetics simulation. We publicly share our accessible codebase at https://github.com/RichardObi/ccnet and provide a user-friendly library for Fréchet radiomics distance calculation at https://pypi.org/project/frd-score.
Paper Structure (9 sections, 3 equations, 3 figures, 1 table)

This paper contains 9 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Multi-Condition Latent Diffusion Model
  4. Biomarker Variability as Image Quality Measure
  5. Experiments and Results
  6. Dataset and Implementation
  7. Fréchet Radiomics Distance as Image Perturbation Correlate
  8. Generation of DCE Sequences from Pre-Contrast Images
  9. Discussion and Conclusion

Figures (3)

  • Figure 1: Overview of our proposed methods, including ContrastControlNet (CCNet) and the Fréchet radiomics distance (FRD). CCNet trains the denoising U-Net and the ControlNet in consecutive stages under contrast enhancement-specific conditioning (pre-contrast image, text, acquisition time). During inference, E is discarded (in violet) and, based on a random latent $z_T$ and $w$-weighted ControlNet guidance, the U-Net generates the post-contrast image latent $z_{T0}$. $z_{T0}$ is divided by factor $S$ and decoded via D into image space. Finally, FRD compares extracted real and synthetic imaging biomarker distributions.
  • Figure 2: (a) Image perturbation scales in breast MRI, (b) resulting Frechét radiomics distance (FRD) values (y-axis) per percentage scale (x-axis) per applied perturbation for 2D axial slices and for 3D volumes based on DCE-MRI post-contrast phase 1 data from 254 patient cases.
  • Figure 3: Qualitative (a) and quantitative (b) test set contrast enhancement. In (b), marker size represents the standard deviation of the mean intensity within the tumor region averaged across all test cases. When normalized, tumor region mean intensity is divided by mean intensity of the remaining tumor-free pixels.