Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion Models for conditional MRI generation

Miguel Herencia García del Castillo, Ricardo Moya Garcia, Manuel Jesús Cerezo Mazón, Ekaitz Arriola Garcia, Pablo Menéndez Fernández-Miranda

TL;DR

This work tackles data scarcity and privacy constraints in medical imaging by proposing a conditioned MRI generator based on a Latent Diffusion Model. The model operates in a compressed latent space using an encoder–U-Net–decoder architecture and DDIM-based inference to generate MRI samples conditioned on pathology and modality, including unseen combinations. Evaluation with Fréchet Inception Distance (FID) and MS-SSIM shows high fidelity to real data and substantial structural diversity, evidencing extrapolation capabilities beyond the training distribution. The approach enables dataset augmentation, mitigates class imbalance, and provides a privacy-preserving tool for AI model development and evaluation in radiology.

Abstract

In this article, we present a Latent Diffusion Model (LDM) for the generation of brain Magnetic Resonance Imaging (MRI), conditioning its generation based on pathology (Healthy, Glioblastoma, Sclerosis, Dementia) and acquisition modality (T1w, T1ce, T2w, Flair, PD). To evaluate the quality of the generated images, the Fréchet Inception Distance (FID) and Multi-Scale Structural Similarity Index (MS-SSIM) metrics were employed. The results indicate that the model generates images with a distribution similar to real ones, maintaining a balance between visual fidelity and diversity. Additionally, the model demonstrates extrapolation capability, enabling the generation of configurations that were not present in the training data. The results validate the potential of the model to increase in the number of samples in clinical datasets, balancing underrepresented classes, and evaluating AI models in medicine, contributing to the development of diagnostic tools in radiology without compromising patient privacy.

Diffusion Models for conditional MRI generation

TL;DR

This work tackles data scarcity and privacy constraints in medical imaging by proposing a conditioned MRI generator based on a Latent Diffusion Model. The model operates in a compressed latent space using an encoder–U-Net–decoder architecture and DDIM-based inference to generate MRI samples conditioned on pathology and modality, including unseen combinations. Evaluation with Fréchet Inception Distance (FID) and MS-SSIM shows high fidelity to real data and substantial structural diversity, evidencing extrapolation capabilities beyond the training distribution. The approach enables dataset augmentation, mitigates class imbalance, and provides a privacy-preserving tool for AI model development and evaluation in radiology.

Abstract

In this article, we present a Latent Diffusion Model (LDM) for the generation of brain Magnetic Resonance Imaging (MRI), conditioning its generation based on pathology (Healthy, Glioblastoma, Sclerosis, Dementia) and acquisition modality (T1w, T1ce, T2w, Flair, PD). To evaluate the quality of the generated images, the Fréchet Inception Distance (FID) and Multi-Scale Structural Similarity Index (MS-SSIM) metrics were employed. The results indicate that the model generates images with a distribution similar to real ones, maintaining a balance between visual fidelity and diversity. Additionally, the model demonstrates extrapolation capability, enabling the generation of configurations that were not present in the training data. The results validate the potential of the model to increase in the number of samples in clinical datasets, balancing underrepresented classes, and evaluating AI models in medicine, contributing to the development of diagnostic tools in radiology without compromising patient privacy.

Paper Structure

This paper contains 8 sections, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Solution
  3. Generative AI Models in Imaging
  4. Proposed Solution
  5. Experiments and Results
  6. Datasets
  7. Results
  8. Conclusions

Figures (3)

  • Figure 1: Training process scheme of the generative model.
  • Figure 2: Inference process scheme of the generative model.
  • Figure 3: Examples of images generated by the model for each combination of pathology and MRI modality. Orange borders represent data whose modality and disease were not in the training set.