Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MRIQT: Physics-Aware Diffusion Model for Image Quality Transfer in Neonatal Ultra-Low-Field MRI

Malek Al Abed, Sebiha Demir, Anne Groteklaes, Elodie Germani, Shahrooz Faghihroohi, Hemmen Sabir, Shadi Albarqouni

TL;DR

This work addresses the fidelity gap between portable neonatal ultra-low-field MRI and diagnostic high-field MRI by introducing MRIQT, a fully 3D conditional diffusion model for image quality transfer from uLF to HF. It innovates with a physics-aware K-space transfer, classifier-free guidance, v-prediction, and a 3D perceptual loss to preserve anatomy while enhancing resolution. On a neonatal dataset with diverse pathologies, MRIQT outperforms state-of-the-art GAN/CNN baselines in PSNR and perceptual quality, with reader studies suggesting improved interpretability. The approach enables unpaired training via a learned K-space simulator and demonstrates practical potential for bedside neonatal neuroimaging in resource-limited settings.

Abstract

Portable ultra-low-field MRI (uLF-MRI, 0.064 T) offers accessible neuroimaging for neonatal care but suffers from low signal-to-noise ratio and poor diagnostic quality compared to high-field (HF) MRI. We propose MRIQT, a 3D conditional diffusion framework for image quality transfer (IQT) from uLF to HF MRI. MRIQT combines realistic K-space degradation for physics-consistent uLF simulation, v-prediction with classifier-free guidance for stable image-to-image generation, and an SNR-weighted 3D perceptual loss for anatomical fidelity. The model denoises from a noised uLF input conditioned on the same scan, leveraging volumetric attention-UNet architecture for structure-preserving translation. Trained on a neonatal cohort with diverse pathologies, MRIQT surpasses recent GAN and CNN baselines in PSNR 15.3% with 1.78% over the state of the art, while physicians rated 85% of its outputs as good quality with clear pathology present. MRIQT enables high-fidelity, diffusion-based enhancement of portable ultra-low-field (uLF) MRI for deliable neonatal brain assessment.

MRIQT: Physics-Aware Diffusion Model for Image Quality Transfer in Neonatal Ultra-Low-Field MRI

TL;DR

This work addresses the fidelity gap between portable neonatal ultra-low-field MRI and diagnostic high-field MRI by introducing MRIQT, a fully 3D conditional diffusion model for image quality transfer from uLF to HF. It innovates with a physics-aware K-space transfer, classifier-free guidance, v-prediction, and a 3D perceptual loss to preserve anatomy while enhancing resolution. On a neonatal dataset with diverse pathologies, MRIQT outperforms state-of-the-art GAN/CNN baselines in PSNR and perceptual quality, with reader studies suggesting improved interpretability. The approach enables unpaired training via a learned K-space simulator and demonstrates practical potential for bedside neonatal neuroimaging in resource-limited settings.

Abstract

Portable ultra-low-field MRI (uLF-MRI, 0.064 T) offers accessible neuroimaging for neonatal care but suffers from low signal-to-noise ratio and poor diagnostic quality compared to high-field (HF) MRI. We propose MRIQT, a 3D conditional diffusion framework for image quality transfer (IQT) from uLF to HF MRI. MRIQT combines realistic K-space degradation for physics-consistent uLF simulation, v-prediction with classifier-free guidance for stable image-to-image generation, and an SNR-weighted 3D perceptual loss for anatomical fidelity. The model denoises from a noised uLF input conditioned on the same scan, leveraging volumetric attention-UNet architecture for structure-preserving translation. Trained on a neonatal cohort with diverse pathologies, MRIQT surpasses recent GAN and CNN baselines in PSNR 15.3% with 1.78% over the state of the art, while physicians rated 85% of its outputs as good quality with clear pathology present. MRIQT enables high-fidelity, diffusion-based enhancement of portable ultra-low-field (uLF) MRI for deliable neonatal brain assessment.

Paper Structure

This paper contains 19 sections, 2 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Setup
  4. K-space Simulation
  5. 3D Conditional Diffusion
  6. Classifier-Free Guidance
  7. $v$-Parametrization
  8. 3D Perceptual Feature Extractor
  9. Training Objective
  10. Experiments and Results
  11. Experimental Setup
  12. Dataset
  13. Implementation Details
  14. Baselines and Evaluation Metrics
  15. Quantitative Comparison
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Our MRIQT restores fine anatomical structures and contrast in portable uLF scans, producing HF-like images.
  • Figure 2: Overview of the MRIQT framework showing the training (left) and inference (right) stages. Diffusion process from $x_0$ to $x_T$, where $x_0$ is HF reference and $x_t$ is noise ($\xleftarrow{}$). At each timestep $t \in T$, the UNet, conditioned on $c=\text{uLF}_{\text{sim}}$ is trained to predict the added noise/v. Inference/sampling starts from $\hat{x}_{t=K_{start}}=c+\epsilon$, where $c=\text{uLF}_{\text{real}}$, progressively denoising the input until $\hat{x_0}$.
  • Figure 3: Qualitative comparison of 3 samples on the axial view on IQT. Left to right: ULF, LoHiResGAN Islam2023improving, LF-SynthSR$\ddagger$iglesias2022quantitative, SFNet$\dagger$Tap_SuperField_MICCAI2024, GAMBAS$\dagger$baljer2025gambas, Our base model, Ours, reference HF scan. [($\dagger$) trained on T2w-scans, ($\ddagger$) requires both T1w and T2w scans for testing.]