Low-Field Magnetic Resonance Image Quality Enhancement using a Conditional Flow Matching Model

TL;DR

This study introduces a conditional flow matching (CFM) framework for image quality transfer to enhance low-field MRI (LF-MRI) images to high-field-like quality. By regressing an optimal velocity field, CFM defines a continuous flow from noise to high-quality data and samples via solving the ODE $\frac{d x_t}{d t} = v_\theta(x_t, t \mid x_{\text{low}})$, avoiding iterative denoising. When applied to LF-MRI, the proposed IQT-CFM method demonstrates state-of-the-art reconstruction performance with far fewer parameters than competing methods, maintaining robustness in both in-distribution and out-of-distribution data. These results suggest that CFM is a scalable and efficient tool for MRI reconstruction in resource-limited clinical settings, with potential for broader adoption in IQT tasks. Future work includes incorporating uncertainty estimation to improve clinical reliability.

Abstract

This paper introduces a novel framework for image quality transfer based on conditional flow matching (CFM). Unlike conventional generative models that rely on iterative sampling or adversarial objectives, CFM learns a continuous flow between a noise distribution and target data distributions through the direct regression of an optimal velocity field. We evaluate this approach in the context of low-field magnetic resonance imaging (LF-MRI), a rapidly emerging modality that offers affordable and portable scanning but suffers from inherently low signal-to-noise ratio and reduced diagnostic quality. Our framework is designed to reconstruct high-field-like MR images from their corresponding low-field inputs, thereby bridging the quality gap without requiring expensive infrastructure. Experiments demonstrate that CFM not only achieves state-of-the-art performance, but also generalizes robustly to both in-distribution and out-of-distribution data. Importantly, it does so while utilizing significantly fewer parameters than competing deep learning methods. These results underline the potential of CFM as a powerful and scalable tool for MRI reconstruction, particularly in resource-limited clinical environments.

Figures (5)

• Figure 1: CFM learns velocity fields between noise and data distributions, then generates samples by following these fields.
• Figure 2: Integration of CFM with U-Net backbone: multi-scale inputs, residual blocks with channel attention, transformer-enhanced bottleneck, and advanced sampling methods.
• Figure 3: ResnetBlock serves as the core feature processing component in both encoder and decoder.
• Figure 4: Example of enhanced brain image using InD data and the corresponding error maps.
• Figure 5: Example of enhanced brain image using OOD data and the corresponding error maps.