Meta-Guided Diffusion Models for Zero-Shot Medical Imaging Inverse Problems

Hossein Askari, Fred Roosta, Hongfu Sun

Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Zero-shot Imaging, Inverse Problems, Posterior Sampling, Proximal Optimization

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: The paper introduces a robust method to enhance medical image reconstruction from partial/noisy data in MRI and CT using diffusion models, ensuring data consistency through a proximal optimisation problem, outperforming existing algorithms.

Abstract: In the realm of medical imaging, inverse problems aim to infer high-quality images from incomplete, noisy measurements, with the objective of minimizing expenses and risks to patients in clinical settings. The Diffusion Models have recently emerged as a promising approach to such practical challenges, proving particularly useful for the zero-shot inference of images from partially acquired measurements in Magnetic Resonance Imaging (MRI) and Computed Tomography (CT). A central challenge in this approach, however, is how to guide an unconditional prediction to conform to the measurement information. In this paper, we propose a Meta-Guided Diffusion Model (MGDM) that tackles this challenge through a \emph{bi-level} guidance strategy, where the \emph{outer level} solves a proximal optimization problem to impose measurement consistency and the \emph{inner level} approximates the measurement-conditioned posterior mean as the initial prediction. Furthermore, we introduce a refinement phase, termed the "discrepancy gradient'', designed to reduce the distance between the outputs of the aforementioned levels, thereby acting as an effective regularizer to further enhance data consistency in the recovered samples. Empirical results on publicly available medical datasets in MRI and CT highlight the superior performance of our proposed algorithm, faithfully reproducing high-fidelity medical images consistent with measurements, and notably mitigating the generation of hallucinatory images observed in state-of-the-art methods under similar conditions.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7428