Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ContextMRI: Enhancing Compressed Sensing MRI through Metadata Conditioning

Hyungjin Chung, Dohun Lee, Zihui Wu, Byung-Hoon Kim, Katherine L. Bouman, Jong Chul Ye

TL;DR

ContextMRI is proposed, a text-conditioned diffusion model for MRI that integrates granular metadata into the reconstruction process and demonstrates that increasing the fidelity of metadata, ranging from slice location and contrast to patient age, sex, and pathology, systematically boosts reconstruction performance.

Abstract

Compressed sensing MRI seeks to accelerate MRI acquisition processes by sampling fewer k-space measurements and then reconstructing the missing data algorithmically. The success of these approaches often relies on strong priors or learned statistical models. While recent diffusion model-based priors have shown great potential, previous methods typically ignore clinically available metadata (e.g. patient demographics, imaging parameters, slice-specific information). In practice, metadata contains meaningful cues about the anatomy and acquisition protocol, suggesting it could further constrain the reconstruction problem. In this work, we propose ContextMRI, a text-conditioned diffusion model for MRI that integrates granular metadata into the reconstruction process. We train a pixel-space diffusion model directly on minimally processed, complex-valued MRI images. During inference, metadata is converted into a structured text prompt and fed to the model via CLIP text embeddings. By conditioning the prior on metadata, we unlock more accurate reconstructions and show consistent gains across multiple datasets, acceleration factors, and undersampling patterns. Our experiments demonstrate that increasing the fidelity of metadata, ranging from slice location and contrast to patient age, sex, and pathology, systematically boosts reconstruction performance. This work highlights the untapped potential of leveraging clinical context for inverse problems and opens a new direction for metadata-driven MRI reconstruction.

ContextMRI: Enhancing Compressed Sensing MRI through Metadata Conditioning

TL;DR

ContextMRI is proposed, a text-conditioned diffusion model for MRI that integrates granular metadata into the reconstruction process and demonstrates that increasing the fidelity of metadata, ranging from slice location and contrast to patient age, sex, and pathology, systematically boosts reconstruction performance.

Abstract

Compressed sensing MRI seeks to accelerate MRI acquisition processes by sampling fewer k-space measurements and then reconstructing the missing data algorithmically. The success of these approaches often relies on strong priors or learned statistical models. While recent diffusion model-based priors have shown great potential, previous methods typically ignore clinically available metadata (e.g. patient demographics, imaging parameters, slice-specific information). In practice, metadata contains meaningful cues about the anatomy and acquisition protocol, suggesting it could further constrain the reconstruction problem. In this work, we propose ContextMRI, a text-conditioned diffusion model for MRI that integrates granular metadata into the reconstruction process. We train a pixel-space diffusion model directly on minimally processed, complex-valued MRI images. During inference, metadata is converted into a structured text prompt and fed to the model via CLIP text embeddings. By conditioning the prior on metadata, we unlock more accurate reconstructions and show consistent gains across multiple datasets, acceleration factors, and undersampling patterns. Our experiments demonstrate that increasing the fidelity of metadata, ranging from slice location and contrast to patient age, sex, and pathology, systematically boosts reconstruction performance. This work highlights the untapped potential of leveraging clinical context for inverse problems and opens a new direction for metadata-driven MRI reconstruction.
Paper Structure (14 sections, 4 equations, 10 figures, 2 tables)

This paper contains 14 sections, 4 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Result
  3. ContextMRI captures intricate metadata cues
  4. Conditional inference outperforms unconditional inference
  5. Impact of MR Imaging Parameters
  6. Patient Demographics as a Reconstruction Prior
  7. Pathology Priors and Robustness to Misinformation
  8. Discussion
  9. Method
  10. Data curation
  11. Metadata conditioning
  12. Diffusion model training
  13. Diffusion model architecture
  14. Inverse problem solving with diffusion models

Figures (10)

  • Figure 1: Illustration of the method used in ContextMRI. (a) We convert available metadata into text format, which is encoded as a feature vector as an additional input to the diffusion model. The diffusion model is trained in pixel space with MVUE complex-valued images. (b) ContextMRI can be used for CS-MRI by leveraging off-the-shelf diffusion model-based inverse problem solvers while additionally incorporating available metadata.
  • Figure 2: CFG vs. PSNR on fastMRI knee
  • Figure 3: CFG vs. PSNR on fastMRI brain
  • Figure 4: Qualitative comparison of ContextMRI against unconditional reconstruction
  • Figure 6: PSNR vs. CFG by varying the amount of information contained in the metadata. Experiments are conducted on a subset of fastMRI knee data with uniform 1D $\times 4$ acceleration.
  • ...and 5 more figures