Medical imaging aims to recover underlying tissue properties, using inexact (simplified/linearized) imaging models and often from inaccurate and incomplete measurements. Analytical reconstruction methods rely on hand-crafted regularization, sensitive to noise assumptions and parameter tuning. Among deep learning alternatives, plug-and-play (PnP) approaches learn regularization while incorporating imaging physics during inference, outperforming purely data-driven methods. The performance of all these approaches, however, still strongly depends on measurement quality and imaging model accuracy. In this work, we propose DenOiS, a framework that denoises both input observations and resulting solution in their respective domains. It consists of an observation refinement strategy that corrects degraded measurements while compensating for imaging model simplifications, and a diffusion-based PnP reconstruction approach that remains robust under missing measurements. DenOiS enables generalization to real data from training only in simulations, resulting in high-fidelity image reconstruction with noisy observations and inexact imaging models. We demonstrate this for speed-of-sound imaging as a challenging setting of quantitative ultrasound image reconstruction.
Figures & tables
Figure 1: Deep learning approaches in image reconstruction (a) without using the imaging model, (b,c) using it disjointly, and (d) jointly with data consistency in a plug-and-play (PnP) framework. (e) Our proposed iterative dual-domain denoising framework (DenOiS), alternating between observation and solution domain refinement. (f) Illustration on an ultrasound application for speed-of-sound imaging from incomplete and noisy observations with an inexact model.
Input
L2 -TV
L1 -TV
Unet [ 5 ]
Dolce [ 14 ]
Unet( MU )
RD
b′
11.95
( 36.25 )
10.39
( 32.37 )
10.73
( 42.31 )
6.75
( 15.73 )
43.35
( 37.27 )
6.78
( 18.31 )
MU(b′)
8.45
( 32.23 )
8.13
( 30.43 )
34.82
( 42.27 )
10.15
( 14.46 )
7.01
( 21.22 )
7.21
( 14.35 )
MU(b′,P)
8.36
( 31.39 )
7.70
( 27.35 )
34.23
( 42.27 )
10.42
( 17.56 )
6.70
( 19.38 )
7.21
( 12.28 )
MU(b′,P,A′x′)
8.26
( 30.12 )
8.04
( 28.19 )
34.20
( 42.27 )
11.78
( 27.42 )
7.10
( 24.10 )
7.55
( 11.53 )
MD(b′)
9.44
( 31.04 )
9.11
( 28.87 )
34.95
( 42.27 )
11.86
( 22.65 )
6.97
( 18.04 )
8.29
( 13.16 )
MD(b′,P)
8.58
( 30.32 )
8.27
( 28.23 )
34.33
( 42.27 )
10.97
( 23.62 )
6.34
( 16.58 )
7.43
( 11.67 )
Table 1: In-silico results on (C 2 ) reporting RMSE (ACE). Gray marks baseline configurations. Rows are observation and columns are reconstruction approaches.
Input
L2 -TV
L1 -TV
Dolce [ 14 ]
Unet( MU )
RD
b′
18.04
( 37.56 )
21.94
( 28.01 )
14.13
( 28.80 )
34.36
( 29.96 )
13.71
( 16.34 )
MU(b′)
13.60
( 38.41 )
13.60
( 38.25 )
10.87
( 35.83 )
14.14
( 42.10 )
10.26
( 24.45 )
MU(b′,P)
13.12
( 38.93 )
13.18
( 38.74 )
11.22
( 38.25 )
14.94
( 40.60 )
10.61
( 23.85 )
MU(b′,P,A′x′)
13.93
( 34.32 )
13.94
( 33.21 )
12.11
( 31.47 )
13.49
( 30.80 )
10.62
( 15.09 )
MD(b′)
12.95
( 36.27 )
12.88
( 35.81 )
11.69
( 35.18 )
14.69
( 39.52 )
10.04
( 21.98 )
MD(b′,P)
14.01
( 34.90 )
13.99
( 34.78 )
10.51
( 30.31 )
16.50
( 40.03 )
9.79
( 20.18 )
Table 2: Real phantom data results reporting RMSE (ACE). Values within half a standard deviation of the best results 9.79 ± 5.54 (13.48 ± 4.0) are highlighted. Gray marks baseline configurations.
Figure 2: (a) Sample reconstructions for C 2 and phantom data, with RMSE (ACE) values reported. (b) An in-vivo breast cancer example reporting Δc / gCNR. (c) Corrections of two acquired observations with different MD variants.
Plug-and-Play diffusion prior (PnPDP) frameworks have emerged as a powerful paradigm for solving imaging inverse problems by treating pretrained generative models as modular priors. However, we identify a critical flaw in prevailing PnP solvers (e.g., based on HQS or Proximal Gradient): they function as memoryless operators, updating estimates solely based on instantaneous gradients. This lack of historical tracking inevitably leads to non-vanishing steady-state bias, where the reconstruction fails to strictly satisfy physical measurements under heavy corruption. To resolve this, we propose Dual-Coupled PnP Diffusion (DC-PnPDP), which restores the classical dual variable to provide integral feedback, progressively enforce agreement between the data-consistency and prior. However, this rigorous geometric coupling introduces a secondary challenge: the accumulated dual residuals exhibit spectrally colored, structured artifacts that violate the Additive White Gaussian Noise (AWGN) assumption of diffusion priors, causing severe hallucinations. To bridge this gap, we introduce Spectral Homogenization (SH), a frequency-domain adaptation mechanism that modulates these structured residuals into statistically compliant pseudo-AWGN inputs. This effectively aligns the solver's rigorous optimization trajectory with the denoiser's valid statistical manifold. Extensive experiments on CT and MRI reconstruction demonstrate that our approach resolves the bias-hallucination trade-off, achieving state-of-the-art fidelity with significantly accelerated convergence. The code is available at https://github.com/duchenhe/DC-PnPDP
Chenhe Du, Xuanyu Tian, Qing Wu +4
ShanghaiTech University · Shanghai Jiao Tong University
The inherent electronic and speckle noise complicates clinical interpretation of ultrasound images. Conventional denoising methods rely on explicit noise assumptions whose validity diminishes under composite noise conditions. Learning-based methods are usually pretrained in a limited image domain using a labeled dataset, which implies inevitable domain shift in complex in vivo environments. This study proposes a Pyramid Self-Contrastive Learning (PSCL) framework for test-time ultrasound image denoising without pretraining. Given multiple noisy samples from only one-shot imaging, PSCL disentangles anatomical similarity and noise randomness into separate pyramid latent spaces. The clean image is then decoded from the anatomy space while discarding the noise space. We first apply PSCL to synthetic aperture ultrasound (SAU), where an Aperture-to-Aperture loop serves as a self-supervised proxy task to ensure denoising fidelity. Simulation experiments, including noise levels from 0 to 30 dB and inclusion geometries from simple to complex, demonstrated improvements of 69.3% in SNR and 34.4% in CNR. The in vivo results showed 84.8% SNR and 25.7% CNR gains using only two aperture data of the heart in six echocardiographic views, liver, and kidney. PSCL delivers clear images across diverse imaging targets and configurations, paving the way for more reliable anatomical visualization without domain shift and pretraining costs.
Jiajing Zhang, Bingze Dai, Xi Zhang +2
Department of Electrical and Computer Engineering, The University of Hong Kong, Hong Kong · Department of Biomedical Engineering, Duke University, North Carolina, United States
Ultrasound imaging is widely used for real-time, noninvasive diagnosis, but speckle and related artifacts reduce image quality and can hinder interpretation. We present a diffusion-based ultrasound despeckling method built on the Image Restoration Stochastic Differential Equations framework. To enable supervised training, we curate large paired datasets by simulating ultrasound images from speckle-free magnetic resonance images using the Matlab UltraSound Toolbox. The proposed model reconstructs speckle-suppressed images while preserving anatomically meaningful edges and contrast. On a held-out simulated test set, our approach consistently outperforms classical filters and recent learning-based despeckling baselines. We quantify prediction uncertainty via cross-model variance and show that higher uncertainty correlates with higher reconstruction error, providing a practical indicator of difficult or failure-prone regions. Finally, we evaluate sensitivity to simulation probe settings and observe domain shift, motivating diversified training and adaptation for robust clinical deployment.
Shuoqi Chen, Yujia Wu, Geoffrey P. Luke
Thayer School of Engineering, Dartmouth College, Hanover, NH, USA