A Generative Diffusion Model to Solve Inverse Problems for Robust in-NICU Neonatal MRI

TL;DR

This work tackles the challenge of long scan times and motion artifacts in low-field, NICU MRI by introducing the first acquisition-agnostic diffusion probabilistic model trained on real-world neonatal data. The authors develop a diffusion-prior framework that decouples the prior from the measurement model, enabling posterior sampling via a reverse ODE to solve diverse inverse problems without retraining. They implement a specialized training pipeline with mixed-contrast/orientation data, class embeddings, and self-supervised denoising, and validate the approach on accelerated reconstruction, motion correction, and super-resolution tasks. The results indicate improved image quality and robustness across scenarios, suggesting meaningful potential for faster, more robust neonatal MRI pending broader clinical validation.

Abstract

We present the first acquisition-agnostic diffusion generative model for Magnetic Resonance Imaging (MRI) in the neonatal intensive care unit (NICU) to solve a range of inverse problems for shortening scan time and improving motion robustness. In-NICU MRI scanners leverage permanent magnets at lower field-strengths (i.e., below 1.5 Tesla) for non-invasive assessment of potential brain abnormalities during the critical phase of early live development, but suffer from long scan times and motion artifacts. In this setting, training data sizes are small and intrinsically suffer from low signal-to-noise ratio (SNR). This work trains a diffusion probabilistic generative model using such a real-world training dataset of clinical neonatal MRI by applying several novel signal processing and machine learning methods to handle the low SNR and low quantity of data. The model is then used as a statistical image prior to solve various inverse problems at inference time without requiring any retraining. Experiments demonstrate the generative model's utility for three real-world applications of neonatal MRI: accelerated reconstruction, motion correction, and super-resolution.

Figures (4)

• Figure 1: We propose training a generative model on low-SNR, low-quantity in-NICU neonatal data by combining all contrasts and orientations with class embedding and applying self-supervised denoising before training. The generative model applies to varies inverse problems to accelerate in-NICU neonatal MRI.
• Figure 2: Example accelerated ($2\times$ scan time reduction) MRI reconstructions using generative models trained on each contrast and orientation separately versus models trained on all data simultaneously with and without class embeddings. The proposed approach of training on all data with class embeddings achieves the best quantitative results and lower values in the error maps.
• Figure 3: Motion correction experiments on prospectively acquired clinical data in the presence of motion. The standard clinical image suffers from motion artifacts where our method uses the proposed generative model to reconstruct images with fewer artifacts and estimate the associated 2D rigid motion parameters.
• Figure 4: Applying the proposed generative model to the super-resolution inverse problem. Limited extent in k-space sampling, corresponding to a $2.5\times$ reduction in scan time, results in low resolution images with Gibbs Ringing, but solving the super-resolution inverse problem with the proposed generative model improves image sharpness.