pith. machine review for the scientific record. sign in

arxiv: 2605.12567 · v1 · submitted 2026-05-12 · 💻 cs.CV · cs.AI

Recognition: unknown

Pyramid Self-contrastive Learning Framework for Test-time Ultrasound Image Denoising

Jiajing Zhang , Bingze Dai , Xi Zhang , Yue Xu , Wei-Ning Lee
Authors on Pith no claims yet

Pith reviewed 2026-05-14 20:57 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords ultrasound denoisingtest-time trainingself-contrastive learningsynthetic aperture ultrasoundpyramid latent spacesone-shot learningspeckle noiseimage reconstruction
0
0 comments X

The pith

Self-contrastive learning on sub-aperture signals produces denoised ultrasound images from a single test sample.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a test-time training framework for ultrasound image denoising that operates on one noisy sample from synthetic aperture ultrasound without any pretraining or labeled data. It shuffles sub-aperture transmissions and uses self-contrastive learning in pyramid latent spaces to isolate shared anatomical content from independent noise realizations. The clean image is reconstructed solely from the anatomy component after discarding the noise component. This sidesteps domain shifts that limit conventional pre-trained denoisers and works under composite electronic and speckle noise. Reported gains reach 69.3 percent SNR and 34.4 percent CNR in simulations plus 84.8 percent SNR and 25.7 percent CNR in vivo across heart, liver, and kidney views.

Core claim

The Aperture-to-Aperture (A2A) framework disentangles anatomical similarity and noise randomness from shuffled sub-apertures through self-contrastive learning in pyramid latent spaces. The clean image is then decoded from the anatomy space, while discarding the noise space. A2A is trained at test time on one noisy sample of SAU signals, so it fundamentally eliminates the domain shift and pretraining costs.

What carries the argument

Aperture-to-Aperture (A2A) self-contrastive learning in pyramid latent spaces that contrasts shuffled sub-aperture signals to separate shared anatomy from differing noise.

If this is right

  • Denoising requires no large labeled datasets or explicit noise models.
  • The approach avoids domain shift because training occurs on the exact test sample.
  • Only two aperture signals suffice for the reported in vivo quality gains.
  • Clearer images support improved anatomical visualization and functional assessment.
  • The method handles both electronic noise from 0 to 30 dB and varying inclusion geometries.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same sub-aperture contrastive principle could apply to other multi-transmission ultrasound modes.
  • One-shot adaptation may enable patient-specific denoising directly on clinical scanners.
  • Pyramid levels likely help preserve fine structural details that single-scale methods blur.

Load-bearing premise

Sub-aperture signals share enough common anatomical content while their noise differs independently enough for contrastive learning to separate the two reliably.

What would settle it

Running the method on sub-aperture data where noise is strongly correlated across apertures or where anatomy varies markedly between them produces no measurable SNR or CNR improvement.

Figures

Figures reproduced from arXiv: 2605.12567 by Bingze Dai, Jiajing Zhang, Wei-Ning Lee, Xi Zhang, Yue Xu.

Figure 1
Figure 1. Figure 1: (a) illustrates major noise sources in ultrasound images, including electronic noise, speckle noise, and side lobes. Electronic noise primarily arises from electromag￾netic interference and electronics and appears as random fluctuations following a Gaussian distribution. As grainy patterns (green boxes in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Examples of in vivo ultrasound images sampled from (b) various imaging domains. (c) Illustration of domain shift under different SNRs and inclusion geometries. The dashed and solid boxes show noisy input and clean GT, respectively, for training (blue) and testing (red). Despite extensive efforts in ultrasound denoising (Sec￾tion 2), domain shift is still one of the most critical chal￾lenges. Convention… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between (a) conventional test-time training and (b) our A2A strategy; comparison between (c) conventional contrastive learning and (d) our pyramid self-contrastive learning; Dashed lines indicate loss functions. These methods are derived from explicit noise assump￾tions and are sensitive to parameters such as window size, threshold, and noise coefficients (intensity, variance, scale). 2.2. Coher… view at source ↗
Figure 4
Figure 4. Figure 4: Pipeline of (a) one step of A2A iteration, (b) A2A inference (denoising) process. achieved color-Doppler denoising via recursive dealiasing after segmentation. Both supervised and self-supervised methods require large datasets for pretraining, which implies a potential domain shift out of the training distribution 𝑃 . In fact, DIP (Deep image prior) [54] proved that a CNN could serve as an implicit denoise… view at source ↗
Figure 5
Figure 5. Figure 5: Performance comparison of denoising in simulations under SNR of (a) 0 dB and (b) 30 dB, and imaging configurations of four and eight sub-apertures. For each method, the denoised image is shown on the left side; the upper-right shows its intensity probability density functions (PDFs) of the signal (red curve) and noise (yellow curve) regions; the lower-right color image shows the difference map compared wit… view at source ↗
Figure 6
Figure 6. Figure 6: The radar plots of CNR, gCNR, SNR, PSNR, and SSIM on simulation data under four noise levels. For each metric, results of four or eight apertures and star or circle inclusion are shown at four corners, respectively. decoder asymmetrical to 𝑓𝑎 . A2A uses summation for skip connections, which connects 𝑘 1∕2 + 𝜂 𝑘 1∕2 to implement Eq. 4, and connects only 𝑘 1∕2 to implement Eq. 5. A2A is built upon pure com… view at source ↗
Figure 7
Figure 7. Figure 7: Denoising comparison on in vivo echocardiograms of (a) A4C, (b) PSAX, and (c) PLAX views. The imaging configuration includes two and eight apertures for each view. The visualization layout is identical to that in [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The box plots of (a-f) SNR, (g-l) CNR, and (m-r) gCNR calculated from the original image and the images obtained by A2A, and seven comparison methods on in vivo echocardiograms in six different views (A4C, A2C, PSAX-MV, PSAX-Pap, PSAX-Apex, and PLAX). Original A2A (Ours) CF PCF SRAD BM3D DIP N2N N2S 2 Apertures 8 Apertures (a) Liver 2 Apertures 8 Apertures (b) Kidney 0 5 10 15 20 25 30 35 40 45 50 -50 -45 … view at source ↗
Figure 9
Figure 9. Figure 9: Denoising comparison on in vivo abdominal ultrasound of (a) liver and (b) kidney. The imaging configuration includes two and eight sub-apertures for each organ. The visualizing layout is identical to [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: The 𝑐𝑜𝑛 and 𝑠𝑤𝑎𝑝 monitored with A2A test-time training process using (a) 2, (b) 4, and (c) 8 apertures. Anatomy Encoder Noise Encoder Sub-aperture Image (a) Layer #1 (b) Layer #2 Layer #3 Original Denoised Difference [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: The (a) SNR, (b) CNR, and (c) gCNR among A2A and seven comparison methods on in vivo ultrasound images of different organs, including human heart, liver, and kidney. methods still struggle to remove noise effectively in low SNR environment [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: (a) The feature maps extracted by 𝑓𝑎 and 𝑓𝑛 . (b) The original and denoised image from a single sub-aperture. clean signal is then reconstructed from the anatomy space using this 𝑓𝑑 . Therefore, A2A relaxes the i.i.d condition and can address any noise type that shows distinct characteristics between sub-apertures. 6.2. Self-contrastive Learning (Self-CL) In theory, the proposed Self-CL maximizes a lower … view at source ↗
Figure 13
Figure 13. Figure 13: (a) The feature distributions visualized by t-SNE before and after PSCL in the 8 sub-aperture configuration. (b) The feature distributions learned from 2, 4, and 6 sub-apertures in the 3-rd layer. 𝑘 1 , 𝑘 2 , 𝜂𝑘 1 , 𝜂𝑘 2 are coded in blue, pink, cyan, and red, respectively. 2 3 4 5 6 7 4 5 6 7 8 9 10 11 Sub-aperture Number 8 (a) CNR (dB) 2 3 4 5 6 7 6 8 10 12 14 16 18 20 Sub-aperture Number 8 (b) SNR (d… view at source ↗
Figure 14
Figure 14. Figure 14: The (a) CNR and (b) SNR of in vivo ultrasound images under different sub-aperture settings. in the legend, the attraction between (𝑘 1 , 𝑘 2 ) is optimized by Eq. 10, while the repulsion between (𝜂 𝑘 1 , 𝜂𝑘 2 ) and between (𝑘 1∕2, 𝜂𝑘 1∕2) are optimized by Eq. 11 and Eq. 12, respec￾tively. 6.3. Pyramid Learning Space Unlike conventional CL that merely manipulates the deepest features after a projection … view at source ↗
read the original abstract

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 require massive labeled data and model parameters. These pre-defined and pre-trained manners entail an inevitable domain shift in complex in vivo environments, so they are limited to a specific noise type and often blur structural details. In this study, we propose a pure test-time training framework for one-shot ultrasound image denoising and apply it to synthetic aperture ultrasound (SAU), which synthesizes transmit focus from sub-aperture transmissions. Our Aperture-to-Aperture (A2A) framework disentangles anatomical similarity and noise randomness from shuffled sub-apertures through self-contrastive learning in pyramid latent spaces. The clean image is then decoded from the anatomy space, while discarding the noise space. A2A is trained at test time on one noisy sample of SAU signals, so it fundamentally eliminates the domain shift and pretraining costs. Simulation experiments, including electronic noise levels of 0 to 30 dB and different inclusion geometries, demonstrated an improvement of 69.3% SNR and 34.4% CNR by A2A. 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. A2A delivers clear images/signals across diverse imaging targets and configurations, paving the way for more reliable anatomical visualization and functional assessment by ultrasound.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes the Aperture-to-Aperture (A2A) framework, a pure test-time training method for one-shot denoising of synthetic aperture ultrasound (SAU) images. It applies self-contrastive learning in pyramid latent spaces to shuffled sub-aperture transmissions in order to disentangle shared anatomical content from noise randomness; the clean image is decoded from the anatomy latent space while the noise space is discarded. Simulation experiments report 69.3% SNR and 34.4% CNR gains across electronic noise levels and inclusion geometries; in-vivo results on heart, liver, and kidney data using only two apertures report 84.8% SNR and 25.7% CNR gains.

Significance. If the disentanglement is reliable, the approach would be significant for clinical ultrasound by enabling domain-shift-free denoising without pre-training or large labeled datasets, directly addressing composite electronic-plus-speckle noise while preserving structural detail.

major comments (2)
  1. [A2A framework description] The core claim in the A2A framework description rests on the assumption that sub-aperture signals share identical anatomical content and differ solely by independent noise realizations. Speckle, however, is multiplicative and arises from coherent interference within the resolution cell, making it statistically dependent on local tissue structure rather than additive and independent across apertures. With only two apertures in the in-vivo experiments, the contrastive objective has limited negative samples and no explicit speckle model, raising the risk that structural speckle leaks into the anatomy space and undermines the one-shot decoding step that discards the noise space.
  2. [Experimental results] Table or figure reporting quantitative results (simulation and in-vivo SNR/CNR gains) provides no error bars, statistical tests, or detailed baseline comparisons, and the abstract supplies no equations, loss formulations, or architecture diagrams. This weakens support for the reported 69.3% SNR and 84.8% SNR improvements and makes it difficult to evaluate robustness against overfitting in one-shot training on complex noise.
minor comments (1)
  1. [Method] Pyramid depth and contrastive hyperparameters are free parameters but are not specified or ablated in the method or experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and insightful comments on our manuscript. We address each major comment point by point below and indicate the planned revisions.

read point-by-point responses

  1. Referee: [A2A framework description] The core claim in the A2A framework description rests on the assumption that sub-aperture signals share identical anatomical content and differ solely by independent noise realizations. Speckle, however, is multiplicative and arises from coherent interference within the resolution cell, making it statistically dependent on local tissue structure rather than additive and independent across apertures. With only two apertures in the in-vivo experiments, the contrastive objective has limited negative samples and no explicit speckle model, raising the risk that structural speckle leaks into the anatomy space and undermines the one-shot decoding step that discards the noise space.

    Authors: We appreciate the referee highlighting the structure-dependent nature of speckle. In SAU sub-aperture transmissions, different apertures provide angular diversity that decorrelates speckle realizations for the same underlying anatomy, which our pyramid self-contrastive objective exploits by contrasting multi-scale features across shuffled apertures to isolate shared content. We acknowledge that this is an approximation without an explicit speckle model and that two apertures limit negative sample diversity, potentially allowing some leakage. In revision we will expand the method section with a dedicated discussion of these assumptions, their validity in SAU, and risks of leakage, supported by additional qualitative visualizations of the disentangled spaces. The core framework remains unchanged. revision: partial

  2. Referee: [Experimental results] Table or figure reporting quantitative results (simulation and in-vivo SNR/CNR gains) provides no error bars, statistical tests, or detailed baseline comparisons, and the abstract supplies no equations, loss formulations, or architecture diagrams. This weakens support for the reported 69.3% SNR and 84.8% SNR improvements and makes it difficult to evaluate robustness against overfitting in one-shot training on complex noise.

    Authors: We agree that stronger statistical reporting and presentation are needed. In the revised manuscript we will add error bars (standard deviation across repeated noise realizations and training runs) to all SNR/CNR tables and figures, include paired statistical tests (e.g., Wilcoxon signed-rank) for significance, and expand baseline comparisons with additional methods plus ablation studies on pyramid depth and contrastive terms to assess one-shot robustness. Key loss equations and a compact architecture diagram will be added to the main text (or supplementary), and the abstract will be updated to reference the loss formulation. These changes directly address the concerns about support and overfitting evaluation. revision: yes

Circularity Check

0 steps flagged

No circularity in A2A framework derivation

full rationale

The paper introduces a test-time self-contrastive learning method on shuffled sub-aperture signals to separate anatomy and noise in pyramid latent spaces, with the clean image decoded from the anatomy space. This relies on the stated premise that sub-apertures share anatomical content while differing in noise realizations, applied via standard contrastive objectives without any fitted parameters being renamed as predictions, self-definitional loops, or load-bearing self-citations. Reported SNR/CNR gains are empirical outcomes from simulation and in vivo tests on single samples, not quantities forced by construction from the inputs. The derivation chain is self-contained and independent of the target results.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that sub-apertures contain identical anatomy with independent noise; no free parameters or invented entities are explicitly introduced in the abstract, but the method implicitly depends on contrastive loss hyperparameters and pyramid depth choices.

free parameters (1)
  • pyramid depth and contrastive hyperparameters
    Number of pyramid levels and loss weighting parameters are not specified and must be chosen or tuned for the disentanglement to succeed.
axioms (1)
  • domain assumption Sub-aperture signals share identical anatomical content while noise realizations are statistically independent.
    Invoked to justify the contrastive separation of anatomy and noise spaces.

pith-pipeline@v0.9.0 · 5584 in / 1512 out tokens · 44861 ms · 2026-05-14T20:57:03.150351+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

68 extracted references · 5 canonical work pages · 2 internal anchors

  1. [1]

    J. A. Jensen, S. I. Nikolov, K. L. Gammelmark, M. H. Pedersen, Syntheticapertureultrasoundimaging,Ultrasonics44(2006)e5–e15

    2006

  2. [2]

    Papadacci, M

    C. Papadacci, M. Pernot, M. Couade, M. Fink, M. Tanter, High- contrastultrafastimagingoftheheart,IEEEtransactionsonultrason- ics, ferroelectrics, and frequency control 61 (2) (2014) 288–301

    2014

  3. [3]

    Krissian, R

    K. Krissian, R. Kikinis, C.-F. Westin, K. Vosburgh, Speckle- constrained filtering of ultrasound images, in: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), Vol. 2, IEEE, 2005, pp. 547–552

    2005

  4. [4]

    H. Xie, L. E. Pierce, F. T. Ulaby, Statistical properties of logarith- mically transformed speckle, IEEE transactions on geoscience and remote sensing 40 (3) (2002) 721–727

    2002

  5. [5]

    Loupas, W

    T. Loupas, W. McDicken, P. Allan, Noise reduction in ultrasonic images by digital filtering, The British journal of radiology 60 (712) (1987) 389–392

    1987

  6. [6]

    M.Gupta,H.Taneja,L.Chand,Performanceenhancementandanaly- sisoffiltersinultrasoundimagedenoising,Procediacomputerscience 132 (2018) 643–652

    2018

  7. [7]

    Dabov, A

    K. Dabov, A. Foi, V. Katkovnik, K. Egiazarian, Image denoising by sparse 3-d transform-domain collaborative filtering, IEEE Transac- tions on image processing 16 (8) (2007) 2080–2095

    2007

  8. [8]

    Y. Yu, S. T. Acton, Speckle reducing anisotropic diffusion, IEEE Transactions on image processing 11 (11) (2002) 1260–1270

    2002

  9. [9]

    Buades, B

    A. Buades, B. Coll, J.-M. Morel, Non-local means denoising, Image processing on line 1 (2011) 208–212

    2011

  10. [10]

    Coupé, P

    P. Coupé, P. Hellier, C. Kervrann, C. Barillot, Nonlocal means-based speckle filtering for ultrasound images, IEEE transactions on image processing 18 (10) (2009) 2221–2229

    2009

  11. [11]

    J. Yang, J. Fan, D. Ai, X. Wang, Y. Zheng, S. Tang, Y. Wang, Local statistics and non-local mean filter for speckle noise reduction in medical ultrasound image, Neurocomputing 195 (2016) 88–95

    2016

  12. [12]

    X.Li,T.Wang,Researchonspinalultrasoundimagedenoisingbased on an improved bm3d algorithm, in: Fifth International Conference onImageProcessingandIntelligentControl(IPIC2025),Vol.13782, SPIE, 2025, pp. 152–156

    2025

  13. [13]

    Y. Gan, E. Angelini, A. Laine, C. Hendon, Bm3d-based ultrasound image denoising via brushlet thresholding, in: 2015 IEEE 12th Inter- nationalSymposiumonBiomedicalImaging(ISBI),IEEE,2015,pp. 667–670

    2015

  14. [14]

    Hollman, K

    K. Hollman, K. Rigby, M. O’donnell, Coherence factor of speckle fromamulti-rowprobe,in:1999IEEEUltrasonicsSymposium.Pro- ceedings. International Symposium (Cat. No. 99CH37027), Vol. 2, IEEE, 1999, pp. 1257–1260

    1999

  15. [15]

    P.-C.Li,M.-L.Li,Adaptiveimagingusingthegeneralizedcoherence factor,IEEEtransactionsonultrasonics,ferroelectrics,andfrequency control 50 (2) (2003) 128–141

    2003

  16. [16]

    Camacho, M

    J. Camacho, M. Parrilla, C. Fritsch, Phase coherence imaging, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 56 (5) (2009) 958–974

    2009

  17. [17]

    Wang, C.-H

    S.-L. Wang, C.-H. Chang, H.-C. Yang, Y.-H. Chou, P.-C. Li, Perfor- mance evaluation of coherence-based adaptive imaging using clini- cal breast data, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 54 (8) (2007) 1669–1679

    2007

  18. [18]

    B. M. Asl, A. Mahloojifar, Minimum variance beamforming com- bined with adaptive coherence weighting applied to medical ultra- sound imaging, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 56 (9) (2009) 1923–1931

    2009

  19. [19]

    O’Donnell, Y

    M. O’Donnell, Y. Wang, Coded excitation for synthetic aperture ultrasound imaging, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 52 (2) (2005) 171–176

    2005

  20. [20]

    X. Li, N. Navab, Z. Jiang, Speckle2self: Self-supervised ultrasound specklereductionwithoutcleandata,MedicalImageAnalysis(2025) 103755

    2025

  21. [21]

    K. J. Parker, Ultrasonic attenuation and absorption in liver tissue, Ultrasound in medicine & biology 9 (4) (1983) 363–369

    1983

  22. [22]

    P.C.Li,M.ODonnell,Evaluationalspatialcompounding,Ultrasonic imaging 16 (3) (1994) 176–189

    1994

  23. [23]

    M. A. L. Bell, R. Goswami, J. A. Kisslo, J. J. Dahl, G. E. Trahey, Short-lagspatialcoherenceimagingofcardiacultrasounddata:Initial clinical results, Ultrasound in medicine & biology 39 (10) (2013) 1861–1874

    2013

  24. [24]

    Y. Wang, C. Zheng, H. Peng, X. Chen, Short-lag spatial coherence combined with eigenspace-based minimum variance beamformer for synthetic aperture ultrasound imaging, Computers in Biology and Medicine 91 (2017) 267–276

    2017

  25. [25]

    R. B. Kuc, Application of kalman filtering techniques to diagnostic ultrasound, Ultrasonic Imaging 1 (2) (1979) 105–120

    1979

  26. [26]

    3917–3920

    F.Y.Rizi,H.A.Noubari,S.K.Setarehdan,Wavelet-basedultrasound imagedenoising:Performanceanalysisandcomparison,in:2011An- nual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE, 2011, pp. 3917–3920

    2011

  27. [27]

    Rabbani, M

    H. Rabbani, M. Vafadust, P. Abolmaesumi, S. Gazor, Speckle noise reduction of medical ultrasound images in complex wavelet domain using mixture priors, IEEE transactions on biomedical engineering 55 (9) (2008) 2152–2160

    2008

  28. [28]

    Khare, P

    S. Khare, P. Kaushik, Speckle filtering of ultrasonic images using weighted nuclear norm minimization in wavelet domain, Biomedical Signal Processing and Control 70 (2021) 102997

    2021

  29. [29]

    Zhu, C.-W

    L. Zhu, C.-W. Fu, M. S. Brown, P.-A. Heng, A non-local low-rank framework for ultrasound speckle reduction, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 5650–5658

    2017

  30. [30]

    M. A. Lediju, G. E. Trahey, B. C. Byram, J. J. Dahl, Short-lag spatial coherence of backscattered echoes: Imaging characteristics, IEEEtransactionsonultrasonics,ferroelectrics,andfrequencycontrol 58 (7) (2011) 1377–1388

    2011

  31. [31]

    Kokil, S

    P. Kokil, S. Sudharson, Despeckling of clinical ultrasound images using deep residual learning, Computer Methods and Programs in Biomedicine 194 (2020) 105477. J. Zhang et al.:Preprint submitted to ElsevierPage 13 of 14 Pyramid Self-contrastive Learning Framework for Test-Time Ultrasound Image Denoising

    2020

  32. [32]

    K. M. Mohamed, M. H. Ali, Ultrasound images enhancement using unet-deep learning according to resolution and speckle noise, Inter- national Journal of Mechanical Engineering 7 (5)

  33. [33]

    Zhang, J

    L. Zhang, J. Zhang, Ultrasound image denoising using generative adversarial networks with residual dense connectivity and weighted joint loss, PeerJ Computer Science 8 (2022) e873

    2022

  34. [34]

    Karaoğlu, H

    O. Karaoğlu, H. Ş. Bilge, I. Uluer, Removal of speckle noises from ultrasound images using five different deep learning networks, Engi- neering Science and Technology, an International Journal 29 (2022) 101030

    2022

  35. [35]

    Singh, A

    H. Singh, A. S. Ahmed, F. Melandsø, A. Habib, Ultrasonic image denoising using machine learning in point contact excitation and detection method, Ultrasonics 127 (2023) 106834

    2023

  36. [36]

    Asgariandehkordi, S

    H. Asgariandehkordi, S. Goudarzi, M. Sharifzadeh, A. Basarab, H. Rivaz, Denoising plane wave ultrasound images using diffu- sion probabilistic models, IEEE Transactions on Ultrasonics, Ferro- electrics, and Frequency Control

  37. [37]

    Zhang, C

    Y. Zhang, C. Huneau, J. Idier, D. Mateus, Ultrasound image re- construction with denoising diffusion restoration models, in: Inter- national Conference on Medical Image Computing and Computer- Assisted Intervention, Springer, 2023, pp. 193–203

    2023

  38. [38]

    Yancheng, X

    L. Yancheng, X. Zeng, Q. Dong, X. Wang, Red-mam: A residual encoder-decoder network based on multi-attention fusion for ultra- sound image denoising, Biomedical Signal Processing and Control 79 (2023) 104062

    2023

  39. [39]

    D. Hyun, L. L. Brickson, K. T. Looby, J. J. Dahl, Beamforming and speckle reduction using neural networks, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 66 (5) (2019) 898– 910

    2019

  40. [40]

    X. Yu, S. Luan, S. Lei, J. Huang, Z. Liu, X. Xue, T. Ma, Y. Ding, B. Zhu, Deep learning for fast denoising filtering in ultrasound local- ization microscopy, Physics in Medicine & Biology 68 (20) (2023) 205002

    2023

  41. [41]

    Cammarasana, P

    S. Cammarasana, P. Nicolardi, G. Patanè, Real-time denoising of ultrasound images based on deep learning, Medical & Biological Engineering & Computing 60 (8) (2022) 2229–2244

    2022

  42. [42]

    T.-T.Zhang,H.Shu,K.-Y.Lam,C.-Y.Chow,A.Li,Featuredecompo- sition and enhancement for unsupervised medical ultrasound image denoising and instance segmentation, Applied Intelligence 53 (8) (2023) 9548–9561

    2023

  43. [43]

    H. Wu, T. T. Huynh, R. Souvenir, Echocardiogram enhancement using supervised manifold denoising, Medical image analysis 24 (1) (2015) 41–51

    2015

  44. [44]

    Jarosik, M

    P. Jarosik, M. Lewandowski, Z. Klimonda, M. Byra, Pixel-wise deep reinforcement learning approach for ultrasound image denoising, in: 2021 IEEE International Ultrasonics Symposium (IUS), IEEE, 2021, pp. 1–4

    2021

  45. [45]

    Jiang, C

    M. Jiang, C. You, M. Wang, H. Zhang, Z. Gao, D. Wu, T. Tan, Controllable deep learning denoising model for ultrasound images using synthetic noisy image, in: Computer Graphics International Conference, Springer, 2023, pp. 297–308

    2023

  46. [46]

    Noise2Noise: Learning Image Restoration without Clean Data

    J. Lehtinen, J. Munkberg, J. Hasselgren, S. Laine, T. Karras, M. Ait- tala, T. Aila, Noise2noise: Learning image restoration without clean data, arXiv preprint arXiv:1803.04189

    work page internal anchor Pith review Pith/arXiv arXiv

  47. [47]

    Huang, J

    L. Huang, J. Yin, J. Zhang, U.-W. Lok, R. M. DeRuiter, J. Jin, K. M. Knoll, K. E. Petersen, J. D. Krier, X.-y. Zhu, et al., Self-supervised deep learning for denoising in ultrasound microvascular imaging, arXiv preprint arXiv:2507.05451

    work page arXiv

  48. [48]

    H.Asgariandehkordi,M.Sharifzadeh,H.Rivaz,Lightweightphysics- informed zero-shot ultrasound plane wave denoising, arXiv preprint arXiv:2506.21499

    work page arXiv

  49. [49]

    D. Jung, M. Kang, S. H. Park, N. Guezzi, J. Yu, Unsupervised speckle noise reduction technique for clinical ultrasound imaging, Ultrasonography 43 (5) (2024) 327–344

    2024

  50. [50]

    C.Yu,F.Ren,S.Bao,Y.Yang,X.Xu,Self-supervisedultrasoundim- agedenoisingbasedonweightedjointloss,DigitalSignalProcessing 162 (2025) 105151

    2025

  51. [51]

    Zhang, N

    Y. Zhang, N. Jiang, Z. Xie, J. Cao, Y. Teng, Ultrasonic image’s annotation removal: A self-supervised noise2noise approach, arXiv preprint arXiv:2307.04133

    work page arXiv

  52. [52]

    J. Huh, S. Khan, S. Choi, D. Shin, J. E. Lee, E. S. Lee, J. C. Ye, Tunable image quality control of 3-d ultrasound using switchable cyclegan, Medical Image Analysis 83 (2023) 102651

    2023

  53. [53]

    S. Muth, S. Dort, I. A. Sebag, M.-J. Blais, D. Garcia, Unsupervised dealiasing and denoising of color-doppler data, Medical image anal- ysis 15 (4) (2011) 577–588

    2011

  54. [54]

    Ulyanov, A

    D. Ulyanov, A. Vedaldi, V. Lempitsky, Deep image prior, in: Pro- ceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 9446–9454

    2018

  55. [55]

    Batson, L

    J. Batson, L. Royer, Noise2self: Blind denoising by self-supervision, in: International conference on machine learning, PMLR, 2019, pp. 524–533

    2019

  56. [56]

    A. M. Christensen, I. M. Rosado-Mendez, T. J. Hall, A systematized review of quantitative ultrasound based on first-order speckle statis- tics,IEEETransactionsonUltrasonics,Ferroelectrics,andFrequency Control 71 (7) (2024) 872–886

    2024

  57. [57]

    Zhang, Y

    Y. Zhang, Y. Guo, W.-N. Lee, Ultrafast ultrasound imaging with cascadeddual-polaritywaves,IEEETransactionsonMedicalImaging 37 (4) (2017) 906–917

    2017

  58. [58]

    T. D. Mast, Empirical relationships between acoustic parameters in human soft tissues, Acoustics Research Letters Online 1 (2) (2000) 37–42

    2000

  59. [59]

    Y. Sun, X. Wang, Z. Liu, J. Miller, A. Efros, M. Hardt, Test-time training with self-supervision for generalization under distribution shifts,in:Internationalconferenceonmachinelearning,PMLR,2020, pp. 9229–9248

    2020

  60. [60]

    Jaiswal, A

    A. Jaiswal, A. R. Babu, M. Z. Zadeh, D. Banerjee, F. Makedon, A survey on contrastive self-supervised learning, Technologies 9 (1) (2020) 2

    2020

  61. [61]

    T. Chen, S. Kornblith, M. Norouzi, G. Hinton, A simple framework for contrastive learning of visual representations, in: International conference on machine learning, PmLR, 2020, pp. 1597–1607

    2020

  62. [62]

    M.Gutmann,A.Hyvärinen,Noise-contrastiveestimation:Anewesti- mationprincipleforunnormalizedstatisticalmodels,in:Proceedings ofthethirteenthinternationalconferenceonartificialintelligenceand statistics, JMLR Workshop and Conference Proceedings, 2010, pp. 297–304

    2010

  63. [63]

    A. v. d. Oord, Y. Li, O. Vinyals, Representation learning with con- trastive predictive coding, arXiv preprint arXiv:1807.03748

    work page internal anchor Pith review Pith/arXiv arXiv

  64. [64]

    X. Wang, R. Zhang, C. Shen, T. Kong, L. Li, Dense contrastive learningforself-supervisedvisualpre-training,in:Proceedingsofthe IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 3024–3033

    2021

  65. [65]

    Zhang, B

    J. Zhang, B. Dai, H. Guan, W.-N. Lee, Msaf: A multi-level sensing adversarial framework for signal recovery in synthetic aperture ultra- sound, IEEE Journal of Biomedical and Health Informatics

  66. [66]

    B. E. Treeby, J. Jaros, A. P. Rendell, B. T. Cox, Modeling nonlinear ultrasound propagation in heterogeneous media with power law ab- sorption using a k-space pseudospectral method, The Journal of the Acoustical Society of America 131 (6) (2012) 4324–4336

    2012

  67. [67]

    Mitchell, P

    C. Mitchell, P. S. Rahko, L. A. Blauwet, B. Canaday, J. A. Finstuen, M. C. Foster, K. Horton, K. O. Ogunyankin, R. A. Palma, E. J. Velazquez, Guidelines for performing a comprehensive transthoracic echocardiographic examination in adults: recommendations from the americansocietyofechocardiography,JournaloftheAmericanSoci- ety of Echocardiography 32 (1) (2...

    2019

  68. [68]

    F. Wang, H. Liu, Understanding the behaviour of contrastive loss, in:ProceedingsoftheIEEE/CVFconferenceoncomputervisionand pattern recognition, 2021, pp. 2495–2504. J. Zhang et al.:Preprint submitted to ElsevierPage 14 of 14

    2021