Pith. sign in

REVIEW 4 major objections 6 minor 28 references

Standardisation of Convex Ultrasound Data Through Geometric Analysis and Augmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The fan-shaped convex ultrasound plane can be automatically recovered as an annulus sector—origin, opening angle, inner and outer radii—and those parameters make linearisation, scan-line extraction, and invertible augmentation possible.

desk verdict A useful, well-scoped contribution to ultrasound standardisation, but the evaluation is too narrow and the symmetry assumption needs sensitivity analysis. read the letter →

arxiv 2502.09482 v1 pith:ZAT2JUS5 submitted 2025-02-13 cs.CV

Alistair Weld , Giovanni Faoro , Luke Dixon , Sophie Camp , Arianna Menciassi , Stamatia Giannarou This is my paper · ORCID
classification cs.CV
keywords ultrasoundannulussectorconvexprobelinearisationscanlineextractiondataaugmentationRANSACmedicalimagestandardisation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Ultrasound images from convex probes store the signal in a fan-shaped region whose curvature, position, and depth scale change with the machine, settings, and any cropping applied before publication. This paper argues that the fan can be represented by an annulus sector—a centre, an opening angle, an inner radius, and an outer radius—and that these four parameters can be recovered automatically from a still image using background-intensity detection, connected-component masking, a sliding-window centre estimate, and robust line fitting. If the recovery is accurate, the same parameters enable three practical operations without manual labelling: extracting the scan lines, straightening the fan into a rectangular linearised image, and applying invertible geometric augmentation. The authors validate the parameter estimates against manual annotations on private intraoperative brain ultrasound and on three public ultrasound sets, and show that linearisation keeps a circular object circular, round-trips back to the original image with negligible error, and agrees with a radio-frequency-derived ground truth. The broader point is that a purely geometric route can standardise disorganised and sparsely labelled convex ultrasound data before any learning-based processing begins.

What carries the argument

The central object is the annulus sector, a fan parameterised by origin $O$, opening angle $\theta$, inner radius $r_{inner}$, and outer radius $r_{outer}$; it reduces "where is the ultrasound plane?" to four numbers. The extraction pipeline is: binarise by detecting the background intensity spike in the histogram, isolate the largest connected foreground component as the plane, locate the inner arc and a central axis with a sliding-window row-difference accumulator and centre-of-mass estimate, fit the two radial edges with RANSAC, and reflect the better fit across the symmetry axis to repair the noisier side. The origin follows from the intersection of the two fitted lines via Cramer's rule, and the radii come from the distances from that origin to the inner and outer arc points on the central axis. Once the sector is known, scan lines are rays from the origin sampled by spline interpolation between $r_{inner}$ and $r_{outer}$, and the ratio of sector height to width sets the sampling density so the linearised image keeps a consistent resolution.

What would settle it

Take a set of convex ultrasound images whose planes are deliberately asymmetric—one edge obscured by acoustic shadowing or one side cropped—run the pipeline, and compare the estimated origin, opening angle, and radii against manual annotation; if the symmetry assumption is critical, parameter error should grow with the degree of asymmetry, and a synthetic annulus sector with a known partial occlusion could quantify the failure curve.

Watch

Extended reading notes

Core claim

The core discovery, stated on the paper's own terms, is that the visible convex ultrasound plane is a well-defined geometric object: all scan lines share a common origin and sweep through an opening angle between an inner and an outer arc. The paper shows that this annulus sector can be extracted automatically from ordinary B-mode images, even with superimposed GUI text, cropping, and acoustic artefacts, by masking the foreground, finding a central axis from the centre of mass, fitting the two radial legs with robust regression, and reflecting the better fit across the axis to compensate for the noisier side. From the estimated origin, angle, and radii, scan lines are generated by interpolating along radial rays and the fan is linearised into a rectangle whose columns follow the scan-line direction. The validation evidence is that key-point errors on private data are at the level of a few pixels with a mean angular error of 0.30 degrees, that noisy public data still yields angle errors near or below 3 degrees, that circularity of a ping-pong ball is 0.97 after linearisation, that the convex-to-linear-to-convex round trip has a mean squared error of 0.0064 pixels, and that multi-scale structural similarity to an RF-derived linear ground truth is 0.69. The paper's conclusion is that the method estimates the annulus sector parameters accurately enough to remove the need for manual annotation of the US sector in downstream standardisation.

Load-bearing premise

Everything rests on the premise that the centre of mass of the masked ultrasound plane lies on a symmetry axis, so the better-fitting of the two radial edge lines can be mirrored to stand in for the other; if the visible plane is asymmetric because of shadowing, off-centre probe contact, or cropping, the estimated annulus sector inherits that error.

Editorial extensions

If this is right

  • Convex ultrasound images from different machines can be normalised to a common fan geometry without manual annotation of the sector.
  • Scan lines can be extracted automatically, enabling explicit processing along the ultrasound beam direction in methods that previously required manual plane labelling.
  • The estimated annulus sector yields a linearised image whose column direction follows the scan lines, making rectangular-image operations like convolution or matrix multiplication applicable to convex data.
  • The linearisation is invertible: re-projecting the linear image back to convex coordinates reproduces the original image with a reported mean squared error of 0.0064 pixels, so the augmentation does not discard information.
  • The method is anatomy-agnostic and applies to non-rotated or slightly rotated convex and microconvex B-mode images, the typical clinical presentation; the authors note that reconstructed 3D ultrasound slices can fall outside this assumption.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension the paper does not pursue is using the estimated sector parameters as dataset metadata, so that every image in a benchmark carries its own origin, angle, and radii; this would make cross-dataset normalisation a lookup rather than a per-image fitting problem.
  • Because the linearisation is invertible with small round-trip error, the same geometry could support augmentations that stay on the ultrasound manifold, for instance adding realistic shadowing or depth-dependent attenuation before re-projecting to convex coordinates.
  • The method's sensitivity to the symmetry assumption could be tested on synthetic annulus sectors with controlled occlusions; such a study would show how much of the public-data error is explained by asymmetry rather than by resolution or annotation variance.
  • The representation is generic enough that it might transfer to other fan-shaped image geometries, such as phased-array or endoscopic ultrasound; the paper does not assert this transfer.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper presents a non-learning pipeline for standardising convex ultrasound images. The pipeline detects the ultrasound plane by histogram spike analysis and largest connected component, estimates annulus sector parameters (origin, opening angle, inner and outer radii) using a centre-of-mass line, RANSAC fits of the radial boundaries, and a symmetry-based re-projection of the better-fit line, and then extracts scan lines to linearise the plane via spline interpolation. Validation is performed on 30 intraoperative images with small reported key-point MSE and 0.30° mean absolute angular error, on 50 selected images from three public datasets, and through deformation, inversion, and radio-frequency experiments.

Significance. The idea of automatically recovering the convex US sector geometry without anatomy-specific priors or deep learning is practically appealing and, to my knowledge, not covered by prior work. If the robustness claims held, the method could reduce manual annotation effort and enable standardised augmentation and scan-line processing. The method is transparent, and the deformation, inversion, and RF experiments are useful first steps. However, the evaluation design leaves the central robustness claim insufficiently supported: the public dataset is selected to exclude challenging and rotated cases, and the most safety-critical assumption (radial-boundary symmetry) is never stress-tested. The contribution is promising but requires further validation before the paper's claims can be accepted.

major comments (4)
  1. [§2.3, Eqs. (18)–(19)] The symmetry-based re-projection of the better RANSAC radial line is load-bearing for all downstream parameters: O (Eq. 21), θ, r_inner, and r_outer (Eq. 22) are computed from the two radial boundaries. The paper states the symmetry assumption but provides no sensitivity analysis for its violation. Acoustic shadowing, off-centre probe contact, or lateral cropping break the symmetry; the special crop handling in §3.2.2 only addresses top cropping, and the public evaluation explicitly excludes rotated and hard-to-annotate images. I request a quantitative failure-mode study, e.g. simulated shadowing/cropping on the controlled 30-image set with ground-truth parameters, reporting parameter errors as a function of asymmetry magnitude. Without this, the robustness claim in the abstract and Section 4 is not supported.
  2. [§3.2.2] The public-data evaluation uses a selected subset of 50 images with two exclusion criteria: images too challenging for human annotation and rotated planes. These criteria directly remove the cases most likely to break the largest-connected-component and symmetry assumptions. The claim that the method is robust to nonstandardised images is therefore evaluated only on favourable cases. Please report the total number of images considered, the number excluded for each reason, and either results on a random or inclusive sample or an explicit statement that the method is intended only for non-rotated, human-annotatable images, with the claims revised accordingly.
  3. [§3.3.2] The inversion MSE of 0.0064 is a self-consistency check: the forward and inverse warps use the same estimated annulus parameters and the same interpolation scheme, so a low MSE only shows that the numerical transform is nearly invertible, not that the extracted geometry is accurate or that the augmentation preserves information. To support the conclusion that 'the image augmentation process does not affect the underlying US information', compare against an independent reconstruction, e.g. warp the linearised image back using ground-truth parameters, or report the MSE contribution attributable to parameter error separately. The circularity of 0.97 with visibly horizontal stretch in §3.3.1 should also be explicitly reconciled with the claim of deformation-free linearisation in the conclusion.
  4. [§2.1–§2.2] The pipeline contains several manually chosen thresholds and assumptions: the z-score spike threshold factor, the half-height search range, the moving-average kernel, and the largest-connected-component rule. These choices precede the RANSAC and symmetry steps, so their impact on the final parameter estimates should be quantified. I ask for a sensitivity analysis on the controlled 30-image set, varying at least the spike threshold factor and the moving-average kernel over reasonable ranges and reporting the resulting key-point MSE. Without such an analysis, the reproducibility and robustness of the method are difficult to assess.
minor comments (6)
  1. [Fig. 5 caption] The caption contains a typo: 'privaye' should be 'private'.
  2. [Eq. (9)] The trapezoidal-centre-of-mass formula is missing parentheses in the numerator; it should be written as Σ(iS[i]+(i+1)S[i+1])/2 over Σ(S[i]+S[i+1])/2.
  3. [Eqs. (8) and (12)] The symbol S is used both for the accumulated row-difference vector and for the set of vertical intersections; renaming one of them would remove ambiguity.
  4. [Eq. (17)] The expressions min(edgesl,i) and max(edgesl,i) are undefined; state that the minimum/maximum is taken over the image column coordinate of the edge set.
  5. [References] References [5] and [6] are duplicates of the same paper; one should be removed.
  6. [Table 1] Several reported standard deviations exceed the mean (e.g. POCUS O: 23.73 ± 43.38); the authors should discuss the distribution and indicate whether a few outliers drive the result.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-referential round-trip and reverberation checks, but the central annulus-sector estimation is validated against independent manual annotations and RF data; no load-bearing circularity.

  1. self definitional [Section 3.3.2 (Inversion)]
    "The aim of this section is to show that the linearisation process is not lossy and that the image augmentation process does not affect the underlying US information. The 30 intraoperative US images were linearised and then re-projected back to convex, using the extracted annulus sector geometry and ˆθ."

    The forward linearisation and the inverse re-projection both use the same estimated annulus sector parameters (O, θ, rinner, router). The reported MSE of 0.0064 therefore measures round-trip consistency of the interpolation under a fixed fitted geometry, not whether the estimated parameters match the true US plane. The conclusion that augmentation 'does not affect the underlying US information' is, by construction, a property of applying the same fitted transform and its inverse; it cannot independently validate the estimated geometry. This is a self-referential consistency check, not a prediction, and it is not the paper's central claim.

full rationale

The central derivation chain—masking, centre-of-mass estimation, RANSAC radial-boundary fitting, symmetric correction, and annulus-sector parameter calculation—is not circular: each stage processes image intensities and geometry rather than fitting to the target parameters, and the final estimates are compared against independent manually annotated ground truth in Sections 3.2.1 and 3.2.2. The public-data evaluation also relies on independent manual annotations, and the radio-frequency experiment provides an external benchmark by constructing a ground-truth linear image from RF data. The symmetry assumption in Section 2.3 is a genuine robustness limitation for cropped or shadowed planes, but it is a geometric assumption about the data, not a reduction of the output to the input. The inversion test in Section 3.3.2 and the reverberation visualisation in Section 3.3.1 are self-referential by construction: the round-trip uses the same fitted parameters, and any scan-line-aligned feature becomes vertical under the scan-line-to-column mapping. However, these checks are not presented as external predictions and do not carry the paper's main claim, which is the automatic extraction of the convex US plane. Overall, the core contribution retains independent content, so the circularity score is low.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim depends on several hand-chosen thresholds and geometric assumptions that are not sensitivity-tested. No new physical entities are introduced. The most consequential assumptions are the annulus-sector model, the largest-component selection, and the symmetry-based re-projection of radial boundaries.

free parameters (4)
  • spike threshold factor = 0.5 * max(z-score) or 2
    In Eq 3, the background spike threshold is set to max(max(Z)/2, 2); this value is hand-chosen and not sensitivity-tested.
  • vertical search range = h/2
    In Section 2.2, the window slides over half the image height with 'the range of half the height has been arbitrarily chosen.'
  • moving average kernel = [0.2,0.2,0.2,0.2,0.2]
    In Eq 8, S is smoothed with a fixed 5-tap moving average; the kernel is not justified or varied.
  • connected component selection = largest component
    In Eq 6, the US plane is assumed to be the largest connected component; if a GUI element is larger, this fails.
assumptions (5)
  • domain assumption The convex US plane can be represented as an annulus sector
    This is the geometric model underlying the whole method (Fig 2); it holds for convex probes by design, but assumes the visible plane is a full sector with straight radial edges.
  • domain assumption The background intensity distribution contains a distinct peak
    The masking step (Section 2.1) relies on a histogram spike to identify background; heavy noise, text overlays, or non-zero backgrounds can break this.
  • ad hoc to paper The US plane is the largest connected component
    Eq 6 selects the largest component; this is an assumption about the image content that may fail on cluttered images.
  • domain assumption The US plane is not rotated or only minimally rotated
    The method is 'fundamentally designed for non-rotated or minimally rotated US-planes' (Section 3.2.2); rotated images are excluded from evaluation.
  • ad hoc to paper The centre of mass splits the plane symmetrically
    In Section 2.3, the better RANSAC line is re-projected across the axis of symmetry to define the opposite radial boundary.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Standardisation of Convex Ultrasound Data Through Geometric Analysis and Augmentation." pith.science (2026). https://pith.science/paper/ZAT2JUS5

@misc{pith2026250209482,
  author       = {Pith},
  title        = {Pith review of: Standardisation of Convex Ultrasound Data Through Geometric Analysis and Augmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZAT2JUS5}},
  note         = {Machine review of arXiv:2502.09482}
}
read the original abstract

The application of ultrasound in healthcare has seen increased diversity and importance. Unlike other medical imaging modalities, ultrasound research and development has historically lagged, particularly in the case of applications with data-driven algorithms. A significant issue with ultrasound is the extreme variability of the images, due to the number of different machines available and the possible combination of parameter settings. One outcome of this is the lack of standardised and benchmarking ultrasound datasets. The method proposed in this article is an approach to alleviating this issue of disorganisation. For this purpose, the issue of ultrasound data sparsity is examined and a novel perspective, approach, and solution is proposed; involving the extraction of the underlying ultrasound plane within the image and representing it using annulus sector geometry. An application of this methodology is proposed, which is the extraction of scan lines and the linearisation of convex planes. Validation of the robustness of the proposed method is performed on both private and public data. The impact of deformation and the invertibility of augmentation using the estimated annulus sector parameters is also studied. Keywords: Ultrasound, Annulus Sector, Augmentation, Linearisation.

Figures

Figures reproduced from arXiv: 2502.09482 by the authors.

Figure 1
Figure 1. Medical imaging modality deep learning re [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Annulus sector geometric properties, created [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The proposed method can be broken down into [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (7 more)
Figure 3
Figure 3. Figure 3: Extracting the US plane. Left shows the orig [PITH_FULL_IMAGE:figures/full_fig_p003_3.png]
Figure 4
Figure 4. Figure 4: (Right) An example of the calculation of [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Examples from the controlled experiment with the privaye data. The left images show the processing of [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Example images from the public data experiment, one example from each public dataset used. The left [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the circularity of the object in the original convex US plane compared to the linearised [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Sample from POCUS dataset [6] showing re [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Radio-frequency data validation. Left is the [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 27 canonical work pages

  1. [1]

    Cai, Arunnit Boon- rod, Atefeh Zeinoddini, Alexander D

    Zeynettin Akkus, Jason C. Cai, Arunnit Boon- rod, Atefeh Zeinoddini, Alexander D. Weston, Ken- neth A. Philbrick, and Bradley James Erickson. A survey of deep-learning applications in ultra- sound: Artificial intelligence-powered ultrasound for improving clinical workflow.Journal of the American College of Radiology : JACR , 16 9 Pt B:1318–1328, 2019

    work page 2019

  2. [2]

    Al-Dhabyani, Mohammed Mohammed Mo- hammed Gomaa, H Khaled, and Aly A

    Walid S. Al-Dhabyani, Mohammed Mohammed Mo- hammed Gomaa, H Khaled, and Aly A. Fahmy. Dataset of breast ultrasound images. Data in Brief , 28, 2019

    work page 2019

  3. [3]

    Ballard and Christopher M

    Dana H. Ballard and Christopher M. Brown. Com- puter vision. 1982

    work page 1982

  4. [4]

    Baumgartner, Konstantinos Kamnit- sas, Jacqueline Matthew, Tara P

    Christian F. Baumgartner, Konstantinos Kamnit- sas, Jacqueline Matthew, Tara P. Fletcher, Sandra Smith, Lisa M. Koch, Bernhard Kainz, and Daniel Rueckert. Sononet: Real-time detection and locali- sation of fetal standard scan planes in freehand ul- trasound. Ieee Transactions on Medical Imaging , 36:2204 – 2215, 2016

    work page 2016

  5. [6]

    Accelerating detec- tion of lung pathologies with explainable ultrasound image analysis

    Jannis Born, Nina Wiedemann, Manuel Cossio, Charlotte Buhre, Gabriel Br¨ andle, Konstantin Lei- dermann, Avinash Aujayeb, Michael Moor, Bastian Rieck, and Karsten Borgwardt. Accelerating detec- tion of lung pathologies with explainable ultrasound image analysis. Applied Sciences , 11(2):672, Jan 2021

    work page 2021

  6. [7]

    Zhixuan Zeng, Ashkan Ebadi, Adrian Florea, and Alexander Wong

    Sung-Liang Chen, Quan Zhou, E. Zhixuan Zeng, Ashkan Ebadi, Adrian Florea, and Alexander Wong. Covid-net l2c-ultra: An explainable linear-convex ul- trasound augmentation learning framework to im- prove covid-19 assessment and monitoring. Sensors (Basel, Switzerland) , 24, 2024

    work page 2024

  7. [8]

    Luke Dixon, Adrian Lim, Matthew Grech-Sollars, Dipankar Nandi, and Sophie J. Camp. Intraoperative ultrasound in brain tumor surgery: A review and im- plementation guide. Neurosurgical Review, 45:2503 – 2515, 2022

    work page 2022

  8. [9]

    Fischler and Robert C

    Martin A. Fischler and Robert C. Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated car- tography. Commun. ACM, 24:381–395, 1981

    work page 1981

Show all 28 references
  1. [10]

    Mauro Giuffr´ e and Dennis L. Shung. Harnessing the power of synthetic data in healthcare: innovation, application, and privacy. NPJ Digital Medicine , 6, 2023

    2023

  2. [11]

    John C. Gower. Generalized procrustes analysis. Psychometrika, 40:33–51, 1975

    1975

  3. [12]

    The connected-component labeling problem: A review of state-of-the-art algo- rithms

    Lifeng He, Xiwei Ren, Qihang Gao, Xiao Zhao, Bin Yao, and Yuyan Chao. The connected-component labeling problem: A review of state-of-the-art algo- rithms. Pattern Recognit., 70:25–43, 2017

    2017

  4. [13]

    Trahey, Sina Farsiu, and Mark L

    Ouwen Huang, Will Long, Nick Bottenus, Gregg E. Trahey, Sina Farsiu, and Mark L. Palmeri. Mimick- net, matching clinical post-processing under realistic black-box constraints. 2019 IEEE International Ul- trasonics Symposium (IUS) , pages 1145–1151, 2019

    2019

  5. [14]

    Breast cancer classification from ultrasound images using probability-based optimal deep learn- ing feature fusion

    Kiran Jabeen, Muhammad Attique Khan, Majed Alhaisoni, Usman Tariq, Yudong Zhang, Ameer Hamza, Arturas Mickus, and Robertas Damasevi- cius. Breast cancer classification from ultrasound images using probability-based optimal deep learn- ing feature fusion. Sensors (Basel, Switze...

    2022

  6. [15]

    Salcudeanb, and Nas- sir Navab

    Zhongliang Jiang, Septimiu E. Salcudeanb, and Nas- sir Navab. Robotic ultrasound imaging: State-of-the- art and future perspectives. Medical image analysis, 89:102878, 2023

    2023

  7. [16]

    Ultrasound confidence maps using random walks

    Athanasios Karamalis, Wolfgang Wein, Tassilo Klein, and Nassir Navab. Ultrasound confidence maps using random walks. Medical image analysis , 16 6:1101–12, 2012

    2012

  8. [17]

    Nicole M. A. Krekel, Max H. Haloua, Alexander M F Lopes Cardozo, Roos H de Wit, Anne Marie Bosch, Louise M de Widt-Levert, Sandra Muller, Henk van der Veen, Elisabeth Bergers, Elly S. M. de Lange – de Klerk, Sybren L. Meijer, and M. Petrousjka van den Tol. Intraoperative ultra...

    2013

  9. [18]

    Lubner, Lori Mankowski Gettle, David H

    Meghan G. Lubner, Lori Mankowski Gettle, David H. Kim, Timothy J. Ziemlewicz, Nirvikar Dahiya, and Perry J. Pickhardt. Diagnostic and pro- cedural intraoperative ultrasound: technique, tips and tricks for optimizing results. The British journal of radiology, page 20201406, 2021

    2021

  10. [19]

    L¨ ubke, Sven Rothl¨ ubbers, Oudom Somphone, St´ ephane Allaire, Muyinatu A

    Valeria De Luca, Tobias Benz, Satoshi Kondo, Lars K¨ onig, D. L¨ ubke, Sven Rothl¨ ubbers, Oudom Somphone, St´ ephane Allaire, Muyinatu A. Lediju Bell, Daniel Chung, Daniel Chung, Amalia Cifor, Cristian Grozea, Matthias G¨ unther, J¨ urgen W. Jenne, Till Kipshagen, Markus Kowa...

    2014

  11. [20]

    Kirby, Yuliya Burren, Nicole Porz, Jo- hannes Slotboom, Roland Wiest, Levente Lanczi, Elizabeth R

    Bjoern H Menze, Andr´ as Jakab, Stefan Bauer, Jayashree Kalpathy-Cramer, Keyvan Farahani, Justin S. Kirby, Yuliya Burren, Nicole Porz, Jo- hannes Slotboom, Roland Wiest, Levente Lanczi, Elizabeth R. Gerstner, Marc-Andr´ e Weber, Tal Arbel, Brian B. Avants, Nicholas Ayache, Pat...

    1993

  12. [21]

    Da Costa, Virginia Fer- nandez, Parashkev Nachev, S´ ebastien Ourselin, and Manuel Jorge Cardoso

    Walter Hugo Lopez Pinaya, Petru-Daniel Tudosiu, Jessica Dafflon, Pedro F. Da Costa, Virginia Fer- nandez, Parashkev Nachev, S´ ebastien Ourselin, and Manuel Jorge Cardoso. Brain imaging generation with latent diffusion models. ArXiv, abs/2209.07162, 2022

    2022 arXiv

  13. [22]

    Au- tomatic fetal ultrasound standard plane recognition based on deep learning and iiot

    Bin Pu, Kenli Li, Shengli Li, and Ningbo Zhu. Au- tomatic fetal ultrasound standard plane recognition based on deep learning and iiot. IEEE Transactions on Industrial Informatics , 17:7771–7780, 2021

    2021

  14. [23]

    Subhankar Roy, Willi Menapace, Sebastiaan Oei, Ben Luijten, Enrico Fini, Cristiano Saltori, Iris A. M. Huijben, Nishith Chennakeshava, Federico Mento, Alessandro Sentelli, Emanuele Peschiera, Riccardo Trevisan, Giovanni Maschietto, Elena Torri, Ric- cardo Inchingolo, Andrea Sm...

    2020

  15. [24]

    Is synthetic data all we need? benchmarking the robustness of models trained with synthetic im- ages

    Krishnakant Singh, Thanush Navaratnam, Jannik Holmer, Simone Schaub-Meyer, and Stefan Roth. Is synthetic data all we need? benchmarking the robustness of models trained with synthetic im- ages. 2024 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition Workshops (CV...

    2024

  16. [25]

    Cabrera, I

    Daniel Susser, Daniel S Schiff, Sara Gerke, Laura Y. Cabrera, I. Glenn Cohen, Megan Doerr, Jordan Harrod, Kristin Kostick-Quenet, Jasmine McNealy, Michelle N Meyer, W. Nicholson Price, and Jen- nifer K Wagner. Synthetic health data: Real ethical promise and peril. The Hastings...

    2024

  17. [26]

    Deglint, Brian Li, and Robert T

    Bennett VanBerlo, Delaney Smith, Jared Tschirhart, Blake Vanberlo, Derek Wu, Alex Ford, Joseph Mc- Cauley, Benjamin Wu, Rushil Chaudhary, Chintan Dave, Jordan Ho, Jason L. Deglint, Brian Li, and Robert T. Arntfield. Enhancing annotation efficiency with machine learning: Automa...

    2022

  18. [27]

    Simoncelli, and Alan Conrad Bovik

    Zhou Wang, Eero P. Simoncelli, and Alan Conrad Bovik. Multiscale structural similarity for image quality assessment. The Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003 , 2:1398–1402 Vol.2, 2003

    2003

  19. [28]

    Sound the alarm: The sonographer shortage is echo- ing across healthcare

    Daniel Won, James Walker, Russ Horowitz, Sandeep Bharadwaj, Edward Carlton, and Helena Gabriel. Sound the alarm: The sonographer shortage is echo- ing across healthcare. Journal of ultrasound in medicine : official journal of the American Institute of Ultrasound in Medicine , 2024

    2024

  20. [29]

    Fuiqa: Fetal ultra- sound image quality assessment with deep convolu- tional networks

    Lingyun Wu, Jie-Zhi Cheng, Shengli Li, Baiying Lei, Tianfu Wang, and Dong Ni. Fuiqa: Fetal ultra- sound image quality assessment with deep convolu- tional networks. IEEE Transactions on Cybernetics, 47:1336–1349, 2017. 11

    2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.