A physics-driven framework with cross-domain iteration for self-supervised LDCT denoising and shows consistent gains over the evaluated self-supervised baselines across dose levels, with performance comparable to the evaluated supervised baseline.
Abstract
Low-dose computed tomography (LDCT) measurements contain mixed Poisson-Gaussian noise. However, most self-supervised methods rely on generic image statistics and do not explicitly model this noise, which may limit their ability to effectively suppress realistic LDCT noise. To address this issue, we propose a physics-driven framework with cross-domain iteration for self-supervised LDCT denoising. The proposed framework proceeds in three main steps. First, a learned sinogram prior and the LDCT noise model guide posterior inference of photon counts, enabling separation of the Poisson and Gaussian components. Second, the separated Poisson and Gaussian components are respectively processed by binomial thinning and Gaussian data thinning to construct two branches, and residual scaling matches each branch's noise level to that of the observation, yielding a training pair with approximately independent noise realizations from one low-dose measurement. Finally, the pair is used to train an image-domain network whose forward-projected outputs update the prior. Through cross-domain iteration, the prior and the training pair are progressively refined while maintaining consistency with CT acquisition physics. Experiments on simulated data from AAPM, LIDC-IDRI, and LoDoPaB-CT and on real LDCT data show consistent gains over the evaluated self-supervised baselines across dose levels, with performance comparable to the evaluated supervised baseline.
Single-pixel imaging (SPI) is strongly affected by a hybrid Poisson–Gaussian noise. Clean labels are difficult to obtain in many SPI experiments, while conventional label-free denoisers generally operate after reconstruction and do not explicitly use the bucket-measurement physics. To address this issue, we propose a s...
Xiao-Fei Zhang, Hao Tang, Shu-Run Wang et al.· IEEE Sensors Journal· 0 citations
JIRSSD-Net eliminates the error propagation of conventional sequential pipelines and enhances robustness and accuracy for preclinical PAT applications and outperforming image-domain, signal-domain, and joint baselines across all metrics with statistical significance confirmed by the Friedman test.
Zheng Sun, Bing Ai, Mei-Chen Sun et al.· Physica Scripta· 0 citations
This paper proposes an ADMM-inspired unrolled plug-and-play solver for Poisson inverse problems that decouples a closed-form data-consistency update from a parameter-efficient prior, implemented as a lightweight decoder operating on frozen CLIP RN50 dense multi-scale features.
Laura C. Diaz-Delgado, Emmanuel Martinez, Henry Arguello· 1 citation
SinoDiff is a novel self-supervised, physics-consistent diffusion framework for the recovery of PET sinograms across multiple predefined dose levels, and achieves competitive performance against supervised and self-supervised baselines across multiple dose levels.
Ghulam Nabi Ahmad Hassan Yar, Himashi Peiris, S. Jamadar et al.· 0 citations
While deep learning-based denoising has become widely adopted in low-dose CT, conventional models use generic architectures designed for natural images, failing to account for non-stationary and spatially correlated CT noise characteristics. To address this, we propose an Efficient Noise COntext-aware REpresentation (E...
Minwoo Yu, N. R. Bennett, Jongduk Baek et al.· 0 citations
An unsupervised physics-informed deconvolution network framework (UPIDN) that bridges the strengths of physical modeling and deep learning is proposed, providing a reliable approach for high-quality ultrasound reconstruction without paired training data and offers a potential solution for ultrasound imaging scenarios w...
Xiang-Yu Li, Xin Zhang, Li-Zheng Zu et al.· Physics in Medicine and Biol...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.