Diffusion Model Regularized Implicit Neural Representation for CT Metal Artifact Reduction

2025-12-08

AI Summary

CT metal artifacts severely degrade image quality. Supervised metal artifact reduction (MAR) methods are hampered by the scarcity of paired metal-contaminated/clean CT data, resulting in poor generalizability; unsupervised approaches often neglect CT’s geometric and physical constraints, while conventional regularization struggles to effectively incorporate structural priors. This paper proposes an unsupervised implicit neural representation (INR) framework that jointly integrates forward geometric modeling with diffusion model priors. It is the first to embed a pre-trained diffusion model as a strong structural prior into the INR reconstruction pipeline; explicitly encodes X-ray scanning geometry to ensure data consistency; and operates entirely without paired training data. Evaluated on both synthetic and real clinical datasets, our method outperforms state-of-the-art supervised and unsupervised MAR techniques, achieving superior artifact suppression, higher anatomical fidelity, and enhanced generalizability—demonstrating strong potential for clinical translation.

Technology Category

Computer Vision: Diffusion Models for VisionMachine Learning: Calibration & Uncertainty QuantificationKnowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingWeb Mining and Content Analysis: Large pretrained models with web dataUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Computed tomography (CT) images are often severely corrupted by artifacts in the presence of metals. Existing supervised metal artifact reduction (MAR) approaches suffer from performance instability on known data due to their reliance on limited paired metal-clean data, which limits their clinical applicability. Moreover, existing unsupervised methods face two main challenges: 1) the CT physical geometry is not effectively incorporated into the MAR process to ensure data fidelity; 2) traditional heuristics regularization terms cannot fully capture the abundant prior knowledge available. To overcome these shortcomings, we propose diffusion model regularized implicit neural representation framework for MAR. The implicit neural representation integrates physical constraints and imposes data fidelity, while the pre-trained diffusion model provides prior knowledge to regularize the solution. Experimental results on both simulated and clinical data demonstrate the effectiveness and generalization ability of our method, highlighting its potential to be applied to clinical settings.

Problem

• Reduces metal artifacts in CT images
• Incorporates physical constraints for data fidelity
• Uses diffusion models as prior knowledge regularization

Innovation

• Implicit neural representation integrates physical constraints
• Pre-trained diffusion model provides prior knowledge regularization
• Framework combines data fidelity with learned priors