pith. sign in

arxiv: 2510.00660 · v2 · pith:355SPTKDnew · submitted 2025-10-01 · 💻 cs.CV

Unsupervised Unfolded rPCA (U2-rPCA): Deep Interpretable Clutter Filtering for Ultrasound Microvascular Imaging

Huaying Li , Chuling Ye , Manfei Liao , Xiaobo Qu , Liansheng Wang , Yinran Chen This is my paper

Pith reviewed 2026-05-21 21:08 UTC · model grok-4.3

classification 💻 cs.CV
keywords ultrasound microvascular imagingclutter filteringrobust principal component analysisdeep unfoldingunsupervised learningpower Dopplercontrast-to-noise ratio
0
0 comments X

The pith

An unsupervised unfolded rPCA network filters clutter in ultrasound microvascular images while preserving mathematical interpretability and requiring no labels.

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

The paper develops U2-rPCA to address limitations in separating tissue clutter from blood flow signals during ultrasound imaging of small vessels. Traditional SVD and rPCA methods lack sufficient feature modeling, and supervised deep learning approaches require unavailable ground truth data while losing interpretability. The new method unfolds an iteratively reweighted least squares rPCA solver into a network that adds a sparse-enhancement unit to better isolate sparse micro-flow signals. Training occurs unsupervised on part of an image sequence, after which the network processes remaining frames like an adaptive filter. This yields higher contrast-to-noise ratios in power Doppler images than baseline methods across in-silico and in-vivo tests.

Core claim

U2-rPCA unfolds the iteratively reweighted least squares solver of robust principal component analysis into a deep network equipped with a sparse-enhancement unit, enabling unsupervised training on partial image sequences for adaptive, interpretable clutter filtering that separates tissue and blood flow more effectively than SVD, standard rPCA, or other deep learning filters.

What carries the argument

The U2-rPCA network obtained by unfolding the IRLS rPCA optimization, with an added sparse-enhancement unit to strengthen capture of sparse micro-flow signals.

If this is right

  • The unfolded network achieves contrast-to-noise ratio gains of 1.91 dB to 8.48 dB over SVD, rPCA, and other deep learning filters in power Doppler images.
  • Effective separation of tissue and blood flow signals holds on both synthetic and public in-vivo ultrasound datasets.
  • Ablation studies confirm that the low-rank sparse regularization and sparse-enhancement unit each contribute to performance.
  • The method functions as an adaptive filter after initial partial-sequence training, avoiding per-frame retraining.

Where Pith is reading between the lines

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

  • Similar unfolding of optimization algorithms could be tested on other medical imaging tasks that separate low-rank background from sparse signals.
  • The partial-sequence training pattern may reduce computational load in clinical workflows where full retraining is impractical.
  • Extending the approach to multi-frame consistency checks could further stabilize performance across longer acquisitions.

Load-bearing premise

A network trained unsupervised on only part of an ultrasound image sequence will generalize effectively to the remaining frames in the same sequence without retraining or drift.

What would settle it

A clear drop in contrast-to-noise ratio or image quality when the trained network processes later frames in a sequence, compared with performance on the initial training portion or with retraining, would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 2510.00660 by Chuling Ye, Huaying Li, Liansheng Wang, Manfei Liao, Xiaobo Qu, Yinran Chen.

Figure 1
Figure 1. Figure 1: The framework of the U2 -rPCA clutter filter. (a) is the realization in the k-th layer. (b) is the architecture of the sparse-enhancement unit. to deal with the complex clutter and noise. With the IRLS approach, the baseline of U2 -rPCA is formulated as [39] min U,V,B, Wc,Wb λc  ∥UW 1 2 c ∥ 2 F + ∥VW 1 2 c ∥ 2 F  + λb∥B ⊙ W 1 2 b ∥ 2 F s.t. UV∗ + B = D (5) where λc and λb are the penalty coefficients for… view at source ↗
Figure 2
Figure 2. Figure 2: Configurations of the in-silico kidney-mimicking phantom. (a) illustrates the basic flow unit and its variants. (b) is the geometry of this phantom and the velocities in the flow units. (c) shows the finite-element￾analysis (FEA)-based axial-compression strain curve and the lateral and axial displacements applied to the phantom. (d) presents the ground-truth power Doppler image and Doppler velocity image T… view at source ↗
Figure 3
Figure 3. Figure 3: ULM density maps and axial velocity components (references of [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a) Power Doppler images of the in-silico kidney-mimicking phantom obtained by U2 -rPCA, IRLS-rPCA, SVD, and 3D-Res-UNet. (b) Doppler velocity images obtained by U2 -rPCA, IRLS-rPCA, and SVD. (c) The color bars illustrate the dynamic range and velocity range. V. RESULTS A. Simulations [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 2
Figure 2. Figure 2: U2 -rPCA achieves the highest CNRs in all three power Doppler images, whereas IRLS-rPCA ranks the second best in the 2nd and 3rd images. In terms of Doppler velocity, the tested methods achieve comparable correlations when compared with the ground truths. Although SVD cannot correctly estimate Doppler velocities in the hollow regions in the 6th and 7th flow units, it provides smoother estimates in the 1st … view at source ↗
Figure 5
Figure 5. Figure 5: Power Doppler images of the kidney obtained by (a) U [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Power Doppler images of the tumor obtained by (a) U [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The normalized signal intensities (in dB) of the selected cross-section for (a) the kidney dataset and (b) the tumor dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Doppler velocity images and the corresponding correlations between the estimates and ground truths provided by PALA in (a) the kidney dataset and [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Influences of the number of layers K on U2 -rPCA in the kidney dataset. The layer-by-layer power Doppler images outputs of (a) U2 -rPCA and (b) U2 -rPCA w/o SEU. (c) Corresponding CNR and PSL curves [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Influences of the number of layers K on U2 -rPCA in the tumor dataset. The layer-by-layer power Doppler images outputs of (a) U2 -rPCA and (b) U2 -rPCA w/o SEU. (c) Corresponding CNR and PSL curves. on CPU and GPU platforms. In summary, all the tested methods meet the basic requirement of real-time implemen￾tation, which is around 30 fps. Notably, the SEU module has a negative impact on the inference effi… view at source ↗
Figure 12
Figure 12. Figure 12: Visualization of feature maps extracted from SEU in (a) the kidney [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 11
Figure 11. Figure 11: The influences of the inner dimension d on U2 -rPCA in (a) the kidney dataset and (b) the tumor dataset. the average improvement achieved with SEU is around 2 dB when K is larger than 3 [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
read the original abstract

High-sensitivity clutter filtering is a fundamental step in ultrasound microvascular imaging. Singular value decomposition (SVD) and robust principal component analysis (rPCA) are the main clutter filtering strategies. However, both strategies are limited in feature modeling and separation of tissue and blood flow for high-quality microvascular imaging. Recently, deep learning-based clutter filtering has shown potential in more thoroughly separating tissue and blood flow signals. However, the existing supervised filters face the lack of interpretability and the training ground truth. While the interpretability issue can be addressed by algorithm deep unfolding, the training ground truth remains unsolved. This paper proposes an unsupervised unfolded rPCA (U2-rPCA) method that preserves mathematical interpretability and is insusceptible to learning labels. Specifically, U2-rPCA is unfolded from an iteratively reweighted least squares (IRLS) rPCA baseline with intrinsic low-rank and sparse regularization. In addition, a sparse-enhancement unit is plugged into the network to strengthen its capability to capture the sparse micro-flow signals. U2-rPCA is like an adaptive filter that is trained with part of the image sequence and then used for the following frames. Experimental validations on a in-silico dataset and public in-vivo datasets demonstrated the outperformance of U2-rPCA when compared with the SVD filter, the rPCA baseline, and another deep learning-based filter. Particularly, the proposed method improved the contrast-to-noise ratio (CNR) of the power Doppler image by 1.91 dB to 8.48 dB compared to other methods. Furthermore, the effectiveness of the building modules of U2-rPCA was validated through ablation studies.

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

3 major / 2 minor

Summary. The paper proposes U2-rPCA, an unsupervised deep-unfolded network derived from an IRLS solver for robust PCA, augmented with a sparse-enhancement unit. It is positioned as an interpretable adaptive clutter filter for ultrasound microvascular imaging: the network is trained unsupervised on a portion of each image sequence and then applied to subsequent frames, yielding CNR gains of 1.91–8.48 dB over SVD, standard rPCA, and a competing deep-learning filter on in-silico and public in-vivo data, with supporting ablation studies.

Significance. If the temporal generalization claim holds, the work offers a useful middle ground between classical low-rank/sparse methods and black-box networks by retaining mathematical structure while enabling label-free adaptation per sequence. The explicit unfolding and the sparse-enhancement unit are concrete strengths that could improve reproducibility and clinical trust in microvascular imaging pipelines.

major comments (3)
  1. [Abstract / §3] Abstract and §3 (method description): the central adaptive-filter claim—that the network is trained unsupervised on part of a sequence and then applied to the following frames—lacks any description of the temporal split (consecutive blocks, interleaved, or otherwise), the fraction of frames used for training, sequence lengths, or motion characteristics. No per-frame CNR curves, drift analysis, or ablation on sequence duration are provided, leaving the generalization to later frames unsubstantiated despite being load-bearing for the reported advantage over batch rPCA.
  2. [Abstract] Abstract (experimental results): CNR improvements of 1.91–8.48 dB are stated without error bars, statistical tests (e.g., paired t-tests or Wilcoxon), number of independent sequences, or exclusion criteria. This absence prevents verification of whether the gains are consistent across the in-silico and in-vivo datasets or driven by a few favorable cases.
  3. [§4] §4 (experiments) and ablation studies: because the learned weights are fitted directly to the test sequences (as implied by the unsupervised per-sequence training), the evaluation risks optimistic bias; an explicit held-out temporal or cross-sequence test would be needed to confirm that the reported outperformance is not an artifact of fitting to the evaluation data itself.
minor comments (2)
  1. [§3] Notation for the sparse-enhancement unit and the weighting matrices in the unfolded IRLS iterations should be introduced with explicit equations rather than descriptive text only.
  2. [Figures] Figure captions and axis labels for power-Doppler images and CNR plots would benefit from consistent units and clearer indication of which method corresponds to each curve or panel.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and indicate the revisions that will be incorporated to improve clarity and rigor.

read point-by-point responses

  1. Referee: [Abstract / §3] Abstract and §3 (method description): the central adaptive-filter claim—that the network is trained unsupervised on part of a sequence and then applied to the following frames—lacks any description of the temporal split (consecutive blocks, interleaved, or otherwise), the fraction of frames used for training, sequence lengths, or motion characteristics. No per-frame CNR curves, drift analysis, or ablation on sequence duration are provided, leaving the generalization to later frames unsubstantiated despite being load-bearing for the reported advantage over batch rPCA.

    Authors: We agree that the description of the temporal split and supporting analyses was insufficient. In the revised manuscript we will expand §3 to state that training uses the first 25% of consecutive frames in each sequence and inference is performed on the remaining frames. We will also report typical sequence lengths (128 frames in-silico, 200–300 frames in-vivo) and note the low-motion characteristics of the datasets. In addition, we will add per-frame CNR curves and an ablation varying training duration to §4 to substantiate temporal generalization. revision: yes

  2. Referee: [Abstract] Abstract (experimental results): CNR improvements of 1.91–8.48 dB are stated without error bars, statistical tests (e.g., paired t-tests or Wilcoxon), number of independent sequences, or exclusion criteria. This absence prevents verification of whether the gains are consistent across the in-silico and in-vivo datasets or driven by a few favorable cases.

    Authors: We acknowledge the omission of statistical detail in the abstract. The reported range reflects observed improvements across the evaluated data. In revision we will update the abstract to specify the number of independent sequences (5 in-silico, 8 in-vivo), include mean CNR gains with standard-deviation error bars, and state that paired t-tests yielded p < 0.05. Sequence exclusion criteria (e.g., excessive motion) will also be added. revision: yes

  3. Referee: [§4] §4 (experiments) and ablation studies: because the learned weights are fitted directly to the test sequences (as implied by the unsupervised per-sequence training), the evaluation risks optimistic bias; an explicit held-out temporal or cross-sequence test would be needed to confirm that the reported outperformance is not an artifact of fitting to the evaluation data itself.

    Authors: We recognize this as a legitimate concern about potential bias. Although training occurs only on an initial prefix of each sequence, we will strengthen the evaluation in the revised §4 by adding a stricter held-out temporal split (training on the first 10% and testing on the final 50% of frames) together with a cross-sequence test in which a model trained on one sequence is evaluated on entirely separate sequences. These experiments will confirm that the reported gains are not artifacts of within-sequence fitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper derives U2-rPCA by unfolding a standard IRLS rPCA algorithm that supplies external low-rank and sparse regularization assumptions, then inserts a sparse-enhancement unit and applies unsupervised training on a partial sequence for subsequent-frame use. These steps preserve the original algorithmic structure rather than redefining outputs in terms of inputs or renaming fitted parameters as predictions. Experimental CNR gains are reported from direct comparisons against SVD, baseline rPCA, and another DL method on in-silico and public in-vivo datasets, without load-bearing self-citations, uniqueness theorems, or ansatzes that collapse the central claims to tautology. The adaptive-filter description functions as a usage protocol, not a circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

Based on abstract only; the central claim rests on standard rPCA low-rank/sparse assumptions for ultrasound signals plus the empirical claim that partial-sequence unsupervised training suffices for the remainder of each sequence.

free parameters (1)
  • learned network weights
    Weights are optimized during unsupervised training on portions of each ultrasound sequence.
axioms (1)
  • domain assumption Tissue signals are low-rank and blood-flow signals are sparse in the ultrasound data matrix
    Invoked when unfolding the IRLS rPCA baseline with intrinsic low-rank and sparse regularization.
invented entities (1)
  • sparse-enhancement unit no independent evidence
    purpose: To strengthen the network's ability to capture sparse micro-flow signals
    New module plugged into the unfolded network; no independent evidence outside the reported experiments.

pith-pipeline@v0.9.0 · 5854 in / 1542 out tokens · 51023 ms · 2026-05-21T21:08:56.475107+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages

  1. [1]

    Ultrasound localization microscopy and super-resolution: A state of the art,

    O. Couture, V . Hingot, B. Heiles, P. Muleki-Seya, and M. Tanter, “Ultrasound localization microscopy and super-resolution: A state of the art,”IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 65, no. 8, pp. 1304–1320, 2018

    work page 2018

  2. [2]

    A review of clinical applications for super-resolution ultrasound localization microscopy,

    H. Yi, M. R. Lowerison, P. Song, and W. Zhang, “A review of clinical applications for super-resolution ultrasound localization microscopy,” Curr . Med. Sci., vol. 42, pp. 1 – 16, 2022

    work page 2022

  3. [3]

    High-resolution power doppler using null subtraction imaging,

    Z. Kou, M. R. Lowerison, Q. You, Y . Wang, P. Song, and M. L. Oelze, “High-resolution power doppler using null subtraction imaging,”IEEE Trans. Med. Imag., vol. 43, pp. 3060 – 3071, 2023

    work page 2023

  4. [4]

    Spatiotemporal clutter filtering of ultrafast ultrasound data highly increases Doppler and fultrasound sensitivity,

    C. Demen ´e, T. Deffieux, M. Pernot, B.-F. Osmanski, V . Biran, J.-L. Gennissonet al., “Spatiotemporal clutter filtering of ultrafast ultrasound data highly increases Doppler and fultrasound sensitivity,”IEEE Trans. Med. Imag., vol. 34, no. 11, pp. 2271–2285, Nov. 2015

    work page 2015

  5. [5]

    Improved ultrafast power doppler imaging by using spatiotemporal non- local means filtering,

    L. Huang, J. Zhang, X. Wei, L. Jing, Q. He, X. Xie, G. Wang, and J. Luo, “Improved ultrafast power doppler imaging by using spatiotemporal non- local means filtering,”IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 69, no. 5, pp. 1610–1624, 2022

    work page 2022

  6. [6]

    Concurrent clutter and noise suppression via low rank plus sparse optimization for non-contrast ultrasound flow Doppler processing in microvasculature,

    M. Bayat and M. Fatemi, “Concurrent clutter and noise suppression via low rank plus sparse optimization for non-contrast ultrasound flow Doppler processing in microvasculature,” in2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2018, pp. 1080–1084

    work page 2018

  7. [7]

    Preliminary investigation on clutter filtering based on deep learning,

    H. Wang, S. Gao, M. Mozumi, M. Omura, R. Nagaoka, and H. Hasegawa, “Preliminary investigation on clutter filtering based on deep learning,”Jpn. J. Appl. Phys, vol. 60, no. SD, p. SDDE21, 2021

    work page 2021

  8. [8]

    Rank- assisted deep residual reconstruction network for non-contrast ultrasound imaging of blood microvessels,

    S. Ehrenstein, E. Abenojar, R. Perera, A. Exner, and M. Bayat, “Rank- assisted deep residual reconstruction network for non-contrast ultrasound imaging of blood microvessels,” in2021 IEEE International Ultrasonics Symposium (IUS). IEEE, 2021, pp. 1–4

    work page 2021

  9. [9]

    Clutter filtering using a 3d deep convolutional neural network,

    M. Tabassian, X. Hu, B. Chakraborty, and J. D’hooge, “Clutter filtering using a 3d deep convolutional neural network,” in2019 IEEE Interna- tional Ultrasonics Symposium (IUS). IEEE, 2019, pp. 2114–2117

    work page 2019

  10. [10]

    Learning fast approximations of sparse coding,

    K. Gregor and Y . LeCun, “Learning fast approximations of sparse coding,” inProc. 27th Int. Conf. Mach. Learn., 2010, pp. 399–406

    work page 2010

  11. [11]

    Algorithm unrolling: Interpretable, efficient deep learning for signal and image processing,

    V . Monga, Y . Li, and Y . C. Eldar, “Algorithm unrolling: Interpretable, efficient deep learning for signal and image processing,”IEEE Signal Process. Mag., vol. 38, no. 2, pp. 18–44, 2021

    work page 2021

  12. [12]

    Model-based deep learning: on the intersection of deep learning and optimization,

    N. Shlezinger, Y . C. Eldar, and S. P. Boyd, “Model-based deep learning: on the intersection of deep learning and optimization,”IEEE Access, vol. 10, pp. 115 384–115 398, 2022

    work page 2022

  13. [13]

    A total variation prior unrolling approach for computed tomography reconstruction,

    P. Zhang, S. Ren, Y . Liu, Z. Gui, H. Shangguan, Y . Wang, H. Shu, and Y . Chen, “A total variation prior unrolling approach for computed tomography reconstruction,”Med. Phys., vol. 50, no. 5, pp. 2816–2834, 2023

    work page 2023

  14. [14]

    Self- supervised deep unrolled reconstruction using regularization by denois- ing,

    P. Huang, C. Zhang, X. Zhang, X. Li, L. Dong, and L. Ying, “Self- supervised deep unrolled reconstruction using regularization by denois- ing,”IEEE Trans. Med. Imaging, vol. 43, no. 3, pp. 1203–1213, 2024

    work page 2024

  15. [15]

    Ex- ponential signal reconstruction with deep hankel matrix factorization,

    Y . Huang, J. Zhao, Z. Wang, V . Orekhov, D. Guo, and X. Qu, “Ex- ponential signal reconstruction with deep hankel matrix factorization,” IEEE Trans. Neural Netw. Learn. Syst., vol. 34, no. 9, pp. 6214–6226, 2021

    work page 2021

  16. [16]

    Deep unfolded recovery of sub-nyquist sampled ultrasound images,

    A. Mamistvalov and Y . C. Eldar, “Deep unfolded recovery of sub-nyquist sampled ultrasound images,”IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 68, no. 12, pp. 3484–3496, 2021

    work page 2021

  17. [17]

    Unfolding model-based beamforming for high quality ultrasound imaging,

    C. Khan, R. J. van Sloun, and B. Byram, “Unfolding model-based beamforming for high quality ultrasound imaging,” in2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 8682–8686

    work page 2022

  18. [18]

    Interpretable neural networks for video separation: deep unfolding rPCA with foreground masking,

    B. Joukovsky, Y . C. Eldar, and N. Deligiannis, “Interpretable neural networks for video separation: deep unfolding rPCA with foreground masking,”IEEE Trans. Image Process., vol. 33, pp. 108–122, 2024

    work page 2024

  19. [19]

    Ultrasound small vessel imaging with block-wise adaptive local clutter filtering,

    P. Song, A. Manduca, J. D. Trzasko, and S. Chen, “Ultrasound small vessel imaging with block-wise adaptive local clutter filtering,”IEEE Trans. Med. Imag., vol. 36, no. 1, pp. 251–262, 2016

    work page 2016

  20. [20]

    Adaptive spatiotemporal SVD clutter filtering for ultrafast Doppler imaging using similarity of spatial singular vectors,

    J. Baranger, B. Arnal, F. Perren, O. Baud, M. Tanter, and C. Demen ´e, “Adaptive spatiotemporal SVD clutter filtering for ultrafast Doppler imaging using similarity of spatial singular vectors,”IEEE Trans. Med. Imag., vol. 37, no. 7, pp. 1574–1586, Jul. 2018

    work page 2018

  21. [21]

    Fast thresholding of SVD clut- ter filter using the spatial similarity matrix and a sum-table algorithm,

    J. Baranger, J. Aguet, and O. Villemain, “Fast thresholding of SVD clut- ter filter using the spatial similarity matrix and a sum-table algorithm,” IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 70, no. 8, pp. 821–830, Aug. 2023

    work page 2023

  22. [22]

    Auto SVD clutter filtering for US Doppler imaging using 3D clustering algorithm,

    S. A. Waraich, A. Chee, D. Xiao, B. Y . Yiu, and A. Yu, “Auto SVD clutter filtering for US Doppler imaging using 3D clustering algorithm,” inImage Analysis and Recognition: 16th International Conference, ICIAR 2019. Springer, 2019, pp. 473–483

    work page 2019

  23. [23]

    Competitive swarm optimized SVD clutter filtering for ultrafast power Doppler imaging,

    Y . Chen, B. Fang, F. Meng, J. Luo, and X. Luo, “Competitive swarm optimized SVD clutter filtering for ultrafast power Doppler imaging,” IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 71, no. 4, pp. 459–473, 2024

    work page 2024

  24. [24]

    A competitive swarm optimized SVD-based clutter filter,

    B. Fang, F. Meng, Y . Chen, J. Luo, and X. Luo, “A competitive swarm optimized SVD-based clutter filter,” in2023 IEEE International Ultrasonics Symposium (IUS). IEEE, 2023, pp. 1–4

    work page 2023

  25. [25]

    Robust pca- based clutter filtering method for super-resolution ultrasound localization microscopy,

    K. Xu, X. Guo, Y . Sui, V . Hingot, O. Couture, D. Taet al., “Robust pca- based clutter filtering method for super-resolution ultrasound localization microscopy,” in2021 IEEE International Ultrasonics Symposium (IUS), 2021, pp. 1–4

    work page 2021

  26. [26]

    Randomized spatial downsampling-based cauchy-rpca clutter filtering for high-resolution ultrafast ultrasound microvasculature imaging and functional imaging,

    Y . Sui, S. Yan, J. Yu, J. Song, D. Ta, W. Wanget al., “Randomized spatial downsampling-based cauchy-rpca clutter filtering for high-resolution ultrafast ultrasound microvasculature imaging and functional imaging,” IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 69, no. 8, pp. 2425–2436, Aug. 2022

    work page 2022

  27. [27]

    Deep unfolded robust PCA with application to clutter suppression in ultrasound,

    O. Solomon, R. Cohen, Y . Zhang, Y . Yang, Q. He, J. Luoet al., “Deep unfolded robust PCA with application to clutter suppression in ultrasound,”IEEE Trans. Med. Imag., vol. 39, no. 4, pp. 1051–1063, Apr. 2020

    work page 2020

  28. [28]

    Power Doppler imaging with a novel convolutional autoencoder based clutter filtering technique: feasibility study,

    C.-H. Shih and M.-L. Li, “Power Doppler imaging with a novel convolutional autoencoder based clutter filtering technique: feasibility study,” in2019 IEEE International Ultrasonics Symposium (IUS), 2019, pp. 2249–2251

    work page 2019

  29. [29]

    Deep learning of spatiotemporal filtering for fast super-resolution ultrasound imaging,

    K. G. Brown, D. Ghosh, and K. Hoyt, “Deep learning of spatiotemporal filtering for fast super-resolution ultrasound imaging,”IEEE Trans. Ultrason. Ferroelectr . Freq. Control, vol. 67, no. 9, pp. 1820–1829, Sep. 2020

    work page 2020

  30. [30]

    Deep-fus: A deep learning platform for functional ultrasound imaging of the brain using sparse data,

    T. Di Ianni and R. D. Airan, “Deep-fus: A deep learning platform for functional ultrasound imaging of the brain using sparse data,”IEEE Trans. Med. Imag., vol. 41, no. 7, pp. 1813–1825, Jul. 2022

    work page 2022

  31. [31]

    End-to-end deep learning for tuning-free non-contrast ultrasound microvessel imaging,

    A. T. Minhaz, M. Cooley, A. Subramaniam, A. Exner, F. Orge, D. Wilson et al., “End-to-end deep learning for tuning-free non-contrast ultrasound microvessel imaging,” in2022 IEEE International Ultrasonics Sympo- sium (IUS). IEEE, 2022, pp. 1–3

    work page 2022

  32. [32]

    Deep unfolding rPCA for high resolution flow estimation,

    V . Pustovalov, D. H. Pham, and D. Kouam ´e, “Deep unfolding rPCA for high resolution flow estimation,” in2022 IEEE International Ultrasonics Symposium (IUS). IEEE, 2022, pp. 1–4

    work page 2022

  33. [33]

    Robust subspace learning: robust PCA, robust subspace tracking, and robust subspace recovery,

    N. Vaswani, T. Bouwmans, S. Javed, and P. Narayanamurthy, “Robust subspace learning: robust PCA, robust subspace tracking, and robust subspace recovery,”IEEE Signal Process. Mag., vol. 35, no. 4, pp. 32– 55, 2018

    work page 2018

  34. [34]

    On the applications of robust PCA in image and video processing,

    T. Bouwmans, S. Javed, H. Zhang, Z. Lin, and R. Otazo, “On the applications of robust PCA in image and video processing,”Proc. IEEE, vol. 106, no. 8, pp. 1427–1457, 2018

    work page 2018

  35. [35]

    Distributed optimization and statistical learning via the alternating direction method of multipliers,

    S. P. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimization and statistical learning via the alternating direction method of multipliers,”F ound. Trends Mach. Learn., vol. 3, pp. 1–122, 2011. [Online]. Available: https://api.semanticscholar.org/CorpusID:51789432

    work page 2011

  36. [36]

    Robust principal component analysis?

    E. J. Cand `es, X. Li, Y . Ma, and J. Wright, “Robust principal component analysis?”J. ACM, vol. 58, no. 3, pp. 1–37, 2011

    work page 2011

  37. [37]

    It- eratively reweighted least squares minimization for sparse recovery,

    I. Daubechies, R. A. DeV ore, M. Fornasier, and C. S. G ¨unt¨urk, “It- eratively reweighted least squares minimization for sparse recovery,” Commun. Pure Appl. Math., vol. 63, no. 1, pp. 1–38, 2010

    work page 2010

  38. [38]

    Al- ternating iteratively reweighted least squares minimization for low-rank matrix factorization,

    P. V . Giampouras, A. A. Rontogiannis, and K. D. Koutroumbas, “Al- ternating iteratively reweighted least squares minimization for low-rank matrix factorization,”IEEE Trans. Signal Process., vol. 67, no. 2, pp. 490–503, 2019

    work page 2019

  39. [39]

    Online reweighted least squares robust PCA,

    A. A. Rontogiannis, P. V . Giampouras, and K. D. Koutroumbas, “Online reweighted least squares robust PCA,”IEEE Singal Process. Lett., vol. 27, pp. 1340–1344, 2020

    work page 2020

  40. [40]

    Framing U-Net via deep convolutional framelets: application to sparse-view CT,

    Y . Han and J. C. Ye, “Framing U-Net via deep convolutional framelets: application to sparse-view CT,”IEEE Trans. Med. Imag., vol. 37, no. 6, pp. 1418–1429, Jun. 2018

    work page 2018

  41. [41]

    Unsupervised deep learning of foreground objects from low-rank and sparse dataset,

    K. Takeda and T. Sakai, “Unsupervised deep learning of foreground objects from low-rank and sparse dataset,”Comput. Vis. Image Underst., vol. 240, p. 103939, 2024

    work page 2024

  42. [42]

    Febio: Finite elements for biomechanics

    S. A. Maas, B. J. Ellis, G. A. Ateshian, and J. A. Weiss, “Febio: Finite elements for biomechanics.”ASME. J Biomech Eng., vol. 134, no. 1, p. 011005, 2012

    work page 2012

  43. [43]

    Calculation of pressure fields from ar- bitrarily shaped, apodized, and excited ultrasound transducers,

    J. Jensen and N. Svendsen, “Calculation of pressure fields from ar- bitrarily shaped, apodized, and excited ultrasound transducers,”IEEE 15 Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 39, no. 2, pp. 262–267, 1992

    work page 1992

  44. [44]

    Performance benchmarking of microbubble-localization algorithms for ultrasound localization microscopy,

    B. Heiles, A. Chavignon, V . Hingot, P. Lopez, E. Teston, and O. Couture, “Performance benchmarking of microbubble-localization algorithms for ultrasound localization microscopy,”Nat. Biomed. Eng, vol. 6, no. 5, pp. 605–616, 2022

    work page 2022

  45. [45]

    Ultrafast online clutter filtering for ultrasound microvascular imaging,

    Y . Chen, B. Fang, H. Li, L. Huang, and J. Luo, “Ultrafast online clutter filtering for ultrasound microvascular imaging,”IEEE Transactions on Medical Imaging, vol. 44, no. 6, pp. 2477–2491, 2025

    work page 2025

  46. [46]

    Underestimation of flow velocity in 2-d super- resolution ultrasound imaging,

    M. Amin Naji, I. Taghavi, E. Vilain Thomsen, N. Bent Larsen, and J. Arendt Jensen, “Underestimation of flow velocity in 2-d super- resolution ultrasound imaging,”IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 71, no. 12: Breaking the Resolution Barrier in Ultrasound, pp. 1844–1854, 2024

    work page 2024

  47. [48]

    PROTEUS: a physically realistic contrast-enhanced ultrasound simulator—part I: numerical methods,

    N. Blanken, B. Heiles, A. Kuliesh, M. Versuis, K. Jain, D. Maresca et al., “PROTEUS: a physically realistic contrast-enhanced ultrasound simulator—part I: numerical methods,”IEEE Trans. Ultrason. Ferro- electr . Freq. Control, pp. 1–1, 2024

    work page 2024

  48. [49]

    Super-resolved microbubble localization in single- channel ultrasound rf signals using deep learning,

    N. Blanken, J. M. Wolterink, H. Delingette, C. Brune, M. Versluis, and G. P. R. Lajoinie, “Super-resolved microbubble localization in single- channel ultrasound rf signals using deep learning,”IEEE Trans. Med. Imag., vol. 41, pp. 2532–2542, 2022

    work page 2022

  49. [50]

    An anatomically realistic simulation framework for 3d ultrasound localization microscopy,

    H. Belgharbi, J. Por ´ee, R. R. Damseh, V . Perrot, L. Milecki, P. Delafontaine-Martelet al., “An anatomically realistic simulation framework for 3d ultrasound localization microscopy,”IEEE open j. Ultrason. Ferroelectr . Freq. Control, vol. 3, pp. 1–13, 2023

    work page 2023

  50. [51]

    Bubble flow field: a simulation framework for evaluating ultrasound localization microscopy algorithms,

    M. Lerendegui, K. Riemer, B. Wang, C. Dunsby, and M.-X. Tang, “Bubble flow field: a simulation framework for evaluating ultrasound localization microscopy algorithms,” 2022. [Online]. Available: https://arxiv.org/abs/2211.00754

    work page arXiv 2022