REVIEW 3 major objections 3 minor 11 references
Do We Need Pre-Processing for Deep Learning Based Ultrasound Shear Wave Elastography?
T0 review · 3 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read Deep learning can estimate tissue elasticity from raw, unprocessed ultrasound data.
desk verdict The abstract and full text are two different papers; the elastography claim is unreviewable from the posted manuscript. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the spatio-temporal radiofrequency data cube fed to a 3D convolutional neural network. The network learns to map the time-resolved shear wave propagation pattern directly to shear wave velocity, bypassing beamforming, filtering, and envelope detection. The comparison baseline is a conventional time-of-flight estimator that tracks wave arrivals to compute velocity; the CNN's capacity to work from raw radiofrequency data is what carries the argument.
What would settle it
A concrete test is to train the same 3D CNN on raw RF data from one set of phantoms and evaluate it on new phantoms with a different geometry, transducer, or scanning depth; if group separation collapses or predicted velocities no longer track independently measured stiffness, the raw-RF result was phantom- or system-specific rather than general shear-wave physics.
Extended reading notes
Core claim
The central claim is that a 3D convolutional neural network can differentiate soft-tissue elasticity groups from spatio-temporal ultrasound data without any image preprocessing, using raw radiofrequency signals directly. The evidence is that, across four gelatin phantoms with different elasticity levels, the predicted shear wave velocities differ significantly between all groups for every tested degree of preprocessing, including none. Preprocessing does improve performance metrics slightly, but it is not required for the network to learn the physical information needed for elasticity discrimination.
Load-bearing premise
The load-bearing premise is that the raw radiofrequency data from the four gelatin phantoms contains the full shear-wave propagation information the network uses, so the network is learning physical tissue stiffness rather than artifacts specific to those phantoms.
Editorial extensions
If this is right
- Ultrasound shear-wave elastography systems could place the deep network earlier in the acquisition chain, eliminating beamforming, filtering, and envelope-detection steps.
- Because raw radiofrequency data are less vendor-processed than final images, models trained on them may generalize more easily across different ultrasound machines.
- The conventional time-of-flight method remains the baseline; the CNN matches its ability to separate stiffness groups without relying on the same hand-crafted feature assumptions.
- Preprocessing still buys small performance improvements, so clinical systems that already have a processing pipeline can keep it while simpler or faster setups can omit it.
Reading between the lines
- The full text attached to this record describes a different study, so the claims above are grounded in the abstract alone.
- A natural next test is whether the raw-RF result transfers across ultrasound machines, probe geometries, and tissue-mimicking materials; the four gelatin phantoms share one acquisition setup.
- If the network learns shear-wave physics rather than phantom-specific artifacts, adding simulated noise or changing phantom shape should preserve the between-group separations.
- Clinical deployment could skip vendor preprocessing and use raw acquisitions directly, but the statistical separation shown here would need to become clinically acceptable accuracy on real tissues first.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The abstract of arXiv:2508.03744 claims a study of deep learning-based ultrasound shear wave elastography, in which a 3D convolutional neural network predicts shear wave velocities from spatio-temporal ultrasound images with varying degrees of preprocessing, from fully beamformed/filtered images down to raw radiofrequency data, and compares against a conventional time-of-flight method across four gelatin phantoms of different elasticity. The full text provided, however, is an entirely different paper on bone fracture detection using a modified VGG-19 network applied to X-ray images, with no mention of ultrasound, shear waves, radiofrequency data, gelatin phantoms, or a 3D CNN. Consequently, the submitted manuscript does not contain the methods, data, analyses, or results that the abstract reports, and the central claim cannot be verified or even evaluated from the posted work.
Significance. If substantiated, the abstract's claim would be significant: showing that a deep learning model can perform elastography directly from raw, unprocessed radiofrequency data would reduce the need for standardized preprocessing and potentially lower bias in clinical elasticity assessment. However, the manuscript provides no evidence for this claim. There is no reproducible code, no data release, no machine-checked derivation, and no falsifiable prediction grounded in the submitted text; the only evidence is an abstract that is contradicted by the accompanying full text. The actual full-text study on bone fracture detection, even if sound, is irrelevant to the stated contribution.
major comments (3)
- [Entire manuscript (abstract vs. full text)] The abstract describes a study on ultrasound shear wave elastography using a 3D CNN on spatio-temporal ultrasound images, raw radiofrequency data, four gelatin phantoms, and a comparison with a time-of-flight method, but the full text is an entirely different paper on bone fracture detection using a modified VGG-19 on X-ray images, with no mention of ultrasound, shear waves, RF data, phantoms, or elastography. This mismatch means the central claim of the abstract is completely unsupported by the submitted manuscript, so the reported results cannot be verified or evaluated.
- [IV (Results and Discussion), general] The abstract claims 'statistically significant differences in the predicted shear wave velocity among all elasticity groups,' but the full text reports no statistical tests, p-values, confidence intervals, or group comparisons; the only quantitative results are classification accuracies and AUC for bone fracture detection. The statistical significance claim is therefore unsupported by any evidence in the manuscript.
- [III.A (Data Collection)] The full text's data collection section describes a publicly available X-ray dataset of 9,463 samples for bone fracture classification, not the four gelatin phantoms with different elasticity levels stated in the abstract. Without any description of phantom construction, ultrasound acquisition, raw RF data, or preprocessing ablation, the claim that raw unprocessed RF data suffices for deep learning-based elastography is not testable from this submission.
minor comments (3)
- [Author affiliations] The affiliations of the authors are formatted inconsistently, with some institutions concatenated in a single line (e.g., 'Islamic University of Technology Bangladesh Agricultural University Brac University East West University'), making the author list difficult to parse.
- [Full-text abstract] The abstract of the full text appears to be rendered as garbled characters in the provided PDF, so even the full text's own abstract is unreadable; the authors should ensure proper encoding.
- [IV.C (VGG-19 Decision Making Process)] The sentence 'The model fails to correctly classify a non-fractured forearm' is ambiguous, and the subsequent description of diffuse attention suggests the opposite of a failure; the text should be clarified.
Circularity Check
No circular reasoning found; the paper's abstract and full text are mismatched, which is a verifiability problem, not a circularity problem.
full rationale
The submitted manuscript contains no derivation chain connecting its inputs to its claims. The abstract reports an empirical comparison of a 3D CNN for ultrasound shear wave elastography against a time-of-flight method on gelatin phantoms, but the full text is an unrelated bone-fracture detection study using a modified VGG-19 on X-ray images. There is no equation, fitted parameter, or cited prior result that is defined in terms of the target outcome, so none of the seven circularity patterns applies. In particular, the central claim is not equivalent to its inputs by construction, and no self-citation is load-bearing. The mismatch between the abstract and the manuscript body is a serious completeness and integrity concern, but it is not circular reasoning, so the circularity score remains 0.
Assumptions & free parameters
assumptions (3)
- domain assumption Gelatin phantoms accurately represent soft tissue elasticity for evaluating shear wave elastography.
- domain assumption Raw radiofrequency data contains sufficient information for a 3D CNN to estimate shear wave velocities.
- domain assumption The CNN does not exploit phantom-specific or dataset-specific shortcuts, so its discriminative performance reflects generalizable elasticity estimation.
Cite this review
Pith. "Pith review of Do We Need Pre-Processing for Deep Learning Based Ultrasound Shear Wave Elastography?." pith.science (2026). https://pith.science/paper/R3HJ276S
@misc{pith2026250803744,
author = {Pith},
title = {Pith review of: Do We Need Pre-Processing for Deep Learning Based Ultrasound Shear Wave Elastography?},
year = {2026},
howpublished = {\url{https://pith.science/paper/R3HJ276S}},
note = {Machine review of arXiv:2508.03744}
}
read the original abstract
Estimating the elasticity of soft tissue can provide useful information for various diagnostic applications. Ultrasound shear wave elastography offers a non-invasive approach. However, its generalizability and standardization across different systems and processing pipelines remain limited. Considering the influence of image processing on ultrasound based diagnostics, recent literature has discussed the impact of different image processing steps on reliable and reproducible elasticity analysis. In this work, we investigate the need of ultrasound pre-processing steps for deep learning-based ultrasound shear wave elastography. We evaluate the performance of a 3D convolutional neural network in predicting shear wave velocities from spatio-temporal ultrasound images, studying different degrees of pre-processing on the input images, ranging from fully beamformed and filtered ultrasound images to raw radiofrequency data. We compare the predictions from our deep learning approach to a conventional time-of-flight method across four gelatin phantoms with different elasticity levels. Our results demonstrate statistically significant differences in the predicted shear wave velocity among all elasticity groups, regardless of the degree of pre-processing. Although pre-processing slightly improves performance metrics, our results show that the deep learning approach can reliably differentiate between elasticity groups using raw, unprocessed radiofrequency data. These results show that deep learning-based approaches could reduce the need for and the bias of traditional ultrasound pre-processing steps in ultrasound shear wave elastography, enabling faster and more reliable clinical elasticity assessments.
Reference graph
Works this paper leans on
-
[1]
[1]� Cleveland Clinic, “Bone fractures,” https://my.clevelandclinic. org/health/diseases/15241-bone-fractures, n.d., accessed: 2025-05-
work page 2025
-
[4]
[Online]. Available: https://my.clevelandclinic.org/health/diseases/ 15241-bone-fractures [2]� Better Health Channel, “Bone fractures,” https://www.betterhealth.vic. gov.au/health/conditionsandtreatments/bone-fractures, n.d., accessed: 2025-05-04. [Online]. Available: https://www.betterhealth.vic.gov.au/ health/conditionsandtreatments/bone-fractures [3]� ...
work page 2025
-
[5]
[7]� M. E. Haque, S. M. J. Islam, J. Maliha, M. S. H. Sumon, R. Sharmin, and S. Rokoni, “Improving chronic kidney disease detection efficiency: Fine tuned catboost and nature-inspired algorithms with explainable ai,” in 2025 IEEE 14th International Conference on Communication Systems and Network Technologies (CSNT), 2025, pp. 811–818. [8]� F. N. Jahan, S....
work page 2025
-
[6]
Fracnet: An end - to-end deep learning framework for bone fracture detection,
[Online]. Available: https://doi.org/10.3389/fmed.2024.1506686 [11]� H. A. Alwzwazy, L. Alzubaidi, Z. Zhao, and Y. Gu, “Fracnet: An end - to-end deep learning framework for bone fracture detection,” Pattern Recognition Letters , vol. 190, pp. 1 –7,
-
[7]
Enhancing diagnosis: ensemble deep-learning model for fracture detection using x-ray images,
[Online]. Available: https://doi.org/10.1016/j.patrec.2025.01.034 [12]� A. Tahir, A. Saadia, K. Khan, A. Gul, A. Qahmash, and R. Akram, “Enhancing diagnosis: ensemble deep-learning model for fracture detection using x-ray images,” Clinical Radiology, vol. 79, no. 11, pp. e1394 –e1402,
-
[8]
Analysis and classification of bone fractures using machine learning techniques,
[Online]. Available: https://doi.org/10.1016/j.crad.2024.08.006 [13]� A. Noureen, M. A. Zia, A. Adnan, and M. Hashim, “Analysis and classification of bone fractures using machine learning techniques,” in E3S Web of Conferences , vol
-
[409]
Detection of bone fracture based on machine learning techniques,
EDP Sciences, 2023, p. 02015. [Online]. Available: https://doi.org/10.1051/e3sconf/202340902015 [14]� K. D. Ahmed and R. Hawezi, “Detection of bone fracture based on machine learning techniques,” Measurement: Sensors, vol. 27, p. 100723,
-
[1990]
A threshold selection method from gray -level histograms,
Proceedings of the First Conference on Visualization in Biomedical Computing, 1990, pp. 337–345. [16]� N. Otsu, “A threshold selection method from gray -level histograms,” IEEE Transactions on Systems, Man, and Cybernetics , vol. 9, no. 1, pp. 62–66,
work page 1990
Show all 11 references
-
[2023]
Contrast-limited adaptive histogram equalization: speed and effective - ness,
[Online]. Available: https://doi.org/10.1016/j.measen. 2023.100723 [15]� S. Pizer, R. Johnston, J. Ericksen, B. Yankaskas, and K. Muller, “Contrast-limited adaptive histogram equalization: speed and effective - ness,” in
2023
-
[2024]
Leveraging gene expression data and explainable machine learning for enhanced early detection of type 2 diabetes,
[Online]. Available: https://www.mdpi.com/2673-6470/4/3/27 [6]� A. L. Roy, M. K. Siam, N. N. I. Prova, S. Jahan, and A. A. Maruf, “Leveraging gene expression data and explainable machine learning for enhanced early detection of type 2 diabetes,” arXiv preprint arXiv:2411.14471,
-
[2025]
Diagnosis and detection of bone fracture in radiographic images using deep learning approaches,
[Online]. Available: https://doi.org/10.1186/s12880-024-01546-4 [10]� T. Aldhyani, Z. A. T. Ahmed, B. M. Alsharbi, S. Ahmad, M. H. Al-Adhaileh, A. H. Kamal, M. Almaiah, and J. Nazeer, “Diagnosis and detection of bone fracture in radiographic images using deep learning approach...
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.