Sequential Diffusion-Guided Deep Image Prior For Medical Image Reconstruction

TL;DR

This paper introduces uDiG-DIP, a novel method combining Deep Image Prior and diffusion models for improved MRI and CT image reconstruction, achieving superior results over existing methods.

Contribution

The paper proposes a sequential approach that integrates DIP and diffusion models, enhancing image reconstruction quality in medical imaging tasks.

Findings

uDiG-DIP outperforms baseline methods in MRI and CT reconstruction.

The method effectively combines DIP's architecture with diffusion model refinement.

Experimental results show improved image quality and data consistency.

Abstract

Deep learning (DL) methods have been extensively applied to various image recovery problems, including magnetic resonance imaging (MRI) and computed tomography (CT) reconstruction. Beyond supervised models, other approaches have been recently explored including two key recent schemes: Deep Image Prior (DIP) that is an unsupervised scan-adaptive method that leverages the network architecture as implicit regularization but can suffer from noise overfitting, and diffusion models (DMs), where the sampling procedure of a pre-trained generative model is modified to allow sampling from the measurement-conditioned distribution through approximations. In this paper, we propose combining DIP and DMs for MRI and CT reconstruction, motivated by (i) the impact of the DIP network input and (ii) the use of DMs as diffusion purifiers (DPs). Specifically, we propose a sequential procedure that iteratively optimizes the DIP network with a DM-refined adaptive input using a loss with data consistency and autoencoding terms. We term the approach Sequential Diffusion-Guided DIP (uDiG-DIP). Our experimental results demonstrate that uDiG-DIP achieves superior reconstruction results compared to leading DM-based baselines and the original DIP for MRI and CT tasks.