Problem
Research questions and friction points this paper is trying to address.
De-noising Diffusion Models
Medical Imaging
Accuracy Improvement
Steerable Conditional Diffusion for Out-of-Distribution Adaptation in Medical Image Reconstruction
🤖 AI Summary
Diffusion models often generate hallucinatory artifacts in out-of-distribution (OOD) medical image reconstruction, degrading fidelity and clinical reliability. To address this, we propose a measurement-driven real-time adaptive sampling framework. Our method introduces a novel guidance-conditioned diffusion mechanism that dynamically calibrates the denoising process using only a single measurement signal—requiring no model fine-tuning or additional training—thereby jointly optimizing reconstruction and implicit model adaptation. By integrating measurement consistency constraints with latent-space directional control, and employing a gradient-guided iterative reweighted sampling strategy, our approach significantly enhances reconstruction accuracy and structural preservation. Extensive experiments across OOD MRI and CT reconstruction tasks demonstrate substantial average improvements in PSNR and SSIM, over 40% reduction in hallucinatory artifacts, and markedly improved generalization and clinical robustness.
📝 Abstract
Denoising diffusion models have emerged as the go-to generative framework for solving inverse problems in imaging. A critical concern regarding these models is their performance on out-of-distribution tasks, which remains an under-explored challenge. Using a diffusion model on an out-of-distribution dataset, realistic reconstructions can be generated, but with hallucinating image features that are uniquely present in the training dataset. To address this discrepancy during train-test time and improve reconstruction accuracy, we introduce a novel sampling framework called Steerable Conditional Diffusion. Specifically, this framework adapts the diffusion model, concurrently with image reconstruction, based solely on the information provided by the available measurement. Utilising our proposed method, we achieve substantial enhancements in out-of-distribution performance across diverse imaging modalities, advancing the robust deployment of denoising diffusion models in real-world applications.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Guided Conditional Diffusion
Parameter Adjustment
Medical Imaging Applications
🔎 Similar Papers
No similar papers found.
