Acoustic Holography in the Megahertz Frequency Range with Optimal Lens Topologies and Nonlinear Acoustic Feedback

Costas D. Arvanitis; Pradosh Pritam Dash

REVIEW 4 major objections 3 minor 12 references

Acoustic Holography in the Megahertz Frequency Range with Optimal Lens Topologies and Nonlinear Acoustic Feedback

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Megahertz acoustic holograms can be designed by fast differentiable optimization of lens thickness, with skull aberration correction and nonlinear-mixing feedback for alignment and CSF monitoring.

desk verdict The abstract and full text are two different papers; the acoustic-holography claims have no supporting body, while the attached MIL preprint is a separate, honestly argued paper. read the letter →

arxiv 2508.07103 v2 pith:ZDEPQTH2 submitted 2025-08-09 physics.app-ph physics.med-ph

Pradosh Pritam Dash , Costas D. Arvanitis This is my paper · ORCID

classification physics.app-phphysics.med-ph

keywords acousticholographymegahertzultrasoundangularspectrummethodlenstopologyoptimizationtranscranialskullaberrationcorrectionnonlinearfeedbackcerebrospinalfluidmonitoring

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

This paper tries to establish that a heterogeneous angular spectrum method—a wave-propagation scheme that handles in-plane speed-of-sound variations—combined with rapid gradient-based optimization of lens thickness can design high-fidelity acoustic hologram lenses at megahertz frequencies. The same framework is claimed to correct wavefront aberrations caused by tissue such as the skull, enabling high-precision transcranial neuro-interventions. The paper further claims that low-frequency acoustic signals generated by nonlinear mixing of the transmitted high-frequency waves provide a feedback channel for accurate, skull-compensating lens alignment and for monitoring cerebrospinal-fluid build-up and removal in hydrocephalus. If these claims hold, simple, economical, and high-performance MHz ultrasound systems would become feasible. The provided document contains only the abstract; the full-text body is an unrelated manuscript, so none of these claims is accompanied by derivation, simulation, or measurement in the reviewed record.

What carries the argument

The heterogeneous angular spectrum approach: a propagation model that represents the field as a superposition of plane waves while letting the local speed of sound vary across the plane, thereby capturing refraction through the lens and the intervening tissue. It is coupled to differentiable optimization of the lens thickness profile—gradient-based search over the thickness map that minimizes the difference between simulated and target pressure fields. A third mechanism is nonlinear acoustic mixing, in which high-frequency waves interact in the propagation medium to generate a low-frequency component whose properties serve as a feedback signal for alignment and fluid monitoring.

What would settle it

A water-tank test: fabricate a lens produced by the differentiable optimization, drive it in the MHz range, and scan the resulting pressure field with a hydrophone; if the measured field deviates from the target hologram beyond tolerance, the modelling premise fails. For the feedback claim, use a phantom with a variable-thickness fluid layer mimicking CSF: the low-frequency signal from nonlinear mixing should vary monotonically with layer thickness; if it does not, the hydrocephalus-monitoring claim fails.

Watch

Extended reading notes

Core claim

The claimed discovery is that the heterogeneous angular spectrum approach, which naturally incorporates in-plane varying speed-of-sound maps, supports rapid differentiable optimization of lens thickness profiles and thereby generates lens topologies for high-fidelity acoustic holography in the megahertz range. Crucially, the approach is claimed to account for wavefront aberrations in the propagation medium itself, opening the way to skull-compensating, reconfigurable transcranial ultrasound for neuro-interventions. In addition, the paper reports that nonlinear mixing of high-frequency waves produces a low-frequency acoustic feedback signal usable for accurate lens alignment and for monitorin

Load-bearing premise

The load-bearing premise is that the heterogeneous angular spectrum method correctly models MHz propagation through the lens and intervening tissue, so gradient-based thickness optimization truly yields the target field, and that nonlinear mixing produces a low-frequency feedback signal strong enough for alignment and CSF monitoring; in the reviewed document neither premise is supported by derivation, simulation, or measurement.

Editorial extensions

If this is right

MHz-range acoustic holography lens design becomes a tractable optimization problem rather than a case-by-case analytic construction.
Skull-induced wavefront aberrations can be folded into the same design step, enabling high-precision transcranial ultrasound without elaborate external hardware.
Nonlinear-mixing feedback provides an in situ alignment signal that compensates for skull distortions during treatment.
The same feedback channel could track cerebrospinal-fluid build-up and removal, supporting hydrocephalus management.
Together, these capabilities point toward simple, economical, high-performance ultrasound systems for manufacturing, non-destructive testing, and neuro-intervention.

Reading between the lines

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

If the angular-spectrum optimization works for arbitrary in-plane sound-speed maps, the design procedure should transfer to other aberrating or layered media beyond the skull, such as breast or abdominal tissue; a testable extension is to repeat the optimization on a tissue-mimicking phantom with a measured speed map.
The nonlinear-mixing feedback channel, if it tracks fluid-layer thickness, may serve as a general fluid-volume sensor; a direct experiment would sweep a saline-layer thickness in a phantom and check monotonic correspondence with the low-frequency signal.
Because the reviewed record contains no methods or results text for these claims, the correct reading is that the paper asserts a mechanism and its expected consequences; confirming either claim requires the missing technical content or new measurements.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Plain English first: the abstract and the attached full text are two different papers. The abstract advertises a heterogeneous angular spectrum method for MHz acoustic holography, skull-compensated lens alignment, and CSF monitoring. The body is arXiv:2508.07114v2, a Multiple Instance Learning paper on SMEFT parameter estimation by different authors. No equation, simulation, measurement, or section of the body touches acoustics. As submitted, the central claims are unsupported, full stop.

That said, the attached MIL preprint is not junk. It makes a clear argument that bag-level aggregation improves SNR in low-signal regimes, derives an effective Fisher information with an explicit error-variance term, and backs it with ROC and coverage experiments. It also has a genuinely interesting observation: learned LLRs can violate the second Bartlett identity, requiring a calibration constant. The paper is unusually open about its own limitations—Appendix C.6 flags the i.i.d. assumption, the simplified LHC data, and the fact that an unresolved linear error component would prevent reaching the true Fisher information. That is honest, reproducible work (code and data are linked). If it were submitted under its own title, it would deserve peer review.

The soft spot is structural rather than technical. The acoustic abstract's load-bearing claims have no supporting text. The MIL paper's limitations, however honestly stated, do not bear on acoustic holography. You cannot review a paper whose body is a different preprint; you can only desk reject the package and ask the authors to resubmit the correct manuscript. At the same time, the MIL preprint should not be lost in the rejection—it should go back to its authors to be filed as its own submission.

Who gets value from this? Readers working on low-signal inference in high-energy physics could get something from the MIL half. Readers interested in acoustic holography get nothing here. The citation patterns in the MIL half are standard (Nachman-Thaler, CWoLa) and relevant; the acoustic half offers nothing to check.

Recommendation: as an acoustic-holography paper, desk reject. If the MIL preprint is resubmitted separately, send it to a serious referee.

Referee Report

4 major / 3 minor

Summary. The submission, as provided, consists of the abstract for arXiv:2508.07103 on megahertz acoustic holography and a full text that is in fact arXiv:2508.07114v2, "Increasing Information Extraction in Low-Signal Regimes via Multiple Instance Learning" (MIL) for SMEFT parameter estimation at the LHC. The abstract claims that a heterogeneous angular spectrum approach with differentiable optimization of lens thickness profiles yields high-fidelity MHz acoustic holograms, can compensate for wavefront aberrations such as skull-induced distortions, and that nonlinear mixing of high-frequency waves generates low-frequency acoustic feedback suitable for skull-compensating lens alignment and CSF monitoring. The body contains no acoustic content whatsoever: no propagation model, no optimization formulation, no simulations, no experimental measurements, and no discussion of nonlinear acoustics, lens topologies, skull compensation, or CSF. The only limitations stated in the body (Appendix C.6) concern the MIL methodology and are unrelated. The abstract's central claims are therefore unsupported by the submitted document.

Significance. If realized, the acoustic system described in the abstract would be of considerable applied significance: simple, low-cost MHz ultrasound with high-fidelity holography and skull compensation could impact manufacturing, non-destructive testing, and transcranial therapy. However, because no evidence accompanies the claims, the scientific significance cannot be evaluated. No credit can be given for derivations, code, or data on the acoustic topic, since none are present. The unrelated MIL paper may have its own merits, but it is not the paper described by the abstract.

major comments (4)

[Abstract vs. provided full text (whole submission)] The title and abstract promise an acoustic holography framework, but the body is an unrelated MIL/SMEFT paper. I searched the body and found no occurrence of acoustic holography, angular spectrum, lens thickness, aberration, skull, CSF, nonlinear mixing, or any related term. Consequently, there is no equation, figure, table, simulation, or measurement supporting the abstract's central claim. This is a load-bearing absence: the paper's stated contribution has no derivation.
[Abstract, second paragraph (nonlinear acoustic feedback)] The assertion that low-frequency acoustic feedback from nonlinear mixing of high-frequency waves enables skull-compensating alignment and CSF monitoring is presented as an 'investigation' result, but the body contains no model of nonlinear propagation, no estimate of the generated low-frequency pressure, no signal-to-noise analysis, and no phantom or tissue experiment. This claim is not merely underexplained; the evidence for it is absent.
[Abstract, second paragraph (aberration correction)] The claim that the framework 'can also account for wavefront aberrations in the propagation media' and offers 'high precision neuro-interventions' requires, at minimum, a forward model of heterogeneous media and a validation against aberrating layers. The body provides neither. The MIL paper's Appendix C.6 limitations about i.i.d. assumptions and calibration do not address or limit these acoustic claims; they belong to a different paper.
[Entire manuscript (missing technical content)] This is not a case where a flawed derivation can be repaired locally. The central technical object—the heterogeneous angular spectrum method with differentiable optimization—never appears. A major revision would require writing a new paper, not amending this one.

minor comments (3)

[Title and metadata] The submission metadata conflates arXiv:2508.07103 (the abstract) with arXiv:2508.07114v2 (the full text). The identifiers and topics should be reconciled before any further processing.
[Abstract, line 2] Typographical error: 'weave propagation effects' should read 'wave propagation effects.'
[References] The reference list is entirely from the MIL paper and contains no acoustics literature. If the intended paper is the acoustic one, the bibliography must be replaced.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the MIL body's derivations are self-contained; the acoustic abstract is unsupported rather than circular, and the only fitted parameters are transparently labeled as empirical ansatze.

full rationale

The provided full text (arXiv:2508.07114v2) is an unrelated MIL/SMEFT paper, so the acoustic-holography claims in the abstract have no accompanying derivation to examine; absence of support is not circularity. Within the MIL body, the main derivation chain is self-contained: the SNR growth in Eq. (7) follows directly from the i.i.d. assumption and the central limit theorem, and the effective Fisher information formula in Eq. (15) is derived algebraically from the test statistic and the second Bartlett identity with stated approximations. The paper's fitted elements are explicitly identified as such: Section 4.2 and Appendix C.6 describe σ²_ε(NB) = C√NB as an 'ansatz' with a free parameter C, and the calibration constant c_cicc is an empirically determined post-hoc correction, not a prediction. The main empirical claim (MIL mitigates low-signal suboptimality) is supported by ROC curves and comparisons against event-level baselines, independent of these fitted constants. Appendix C.6 candidly lists the limitations: first-order expansions, unproven sublinear scaling of the error term, simplified LHC data, and i.i.d. assumptions; these are correctness risks rather than circular steps. There is no load-bearing self-citation chain and no quantity is shown to be equal to its own input by construction.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The abstract's acoustic claim has no extractable free parameters because no method details are provided. The attached body (MIL preprint) does contain fitted quantities: the error-variance ansatz C*sqrt(N_B) and the calibration constant c_cicc, both fit to the same data used for evaluation. The acoustic claim rests on three domain assumptions stated only in the abstract: the angular spectrum propagation model, the differentiability of the lens optimization, and the existence of usable nonlinear-mixing feedback. No new entities (particles, forces, dimensions) are postulated in either text.

free parameters (2)

C (error variance prefactor) = not reported numerically, fit per bag-size scaling
Body Section 4.2 fits sigma^2_epsilon(N_B) = C*sqrt(N_B) to measured effective Fisher Information; C is a free parameter of the fit. Belongs to the attached MIL preprint, not the acoustic abstract.
c_cicc (confidence interval calibration constant) = ranges 1.283 to 11.982 across bag sizes (Tables 2 and 3)
Post-hoc rescaling of LLR curvature to enforce the second Bartlett identity, fitted from the same pseudo-experiments used for evaluation. Belongs to the attached MIL preprint.

assumptions (3)

domain assumption Heterogeneous angular spectrum method accurately models MHz wave propagation through lenses with in-plane varying speed-of-sound maps
Abstract states this as the enabling property; no derivation or validation is provided in the submitted text.
domain assumption Lens thickness profiles can be optimized by differentiable optimization to achieve target holograms
Abstract claims rapid differentiable optimization as a capability; no algorithm details, convergence guarantees, or examples appear in the text.
domain assumption Nonlinear mixing of high-frequency waves produces low-frequency acoustic feedback usable for skull-compensating lens alignment and CSF monitoring
Abstract asserts this feedback mechanism; no theory, simulation, or experiment supports it in the provided document.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Acoustic Holography in the Megahertz Frequency Range with Optimal Lens Topologies and Nonlinear Acoustic Feedback." pith.science (2026). https://pith.science/paper/ZDEPQTH2

@misc{pith2026250807103,
  author       = {Pith},
  title        = {Pith review of: Acoustic Holography in the Megahertz Frequency Range with Optimal Lens Topologies and Nonlinear Acoustic Feedback},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZDEPQTH2}},
  note         = {Machine review of arXiv:2508.07103}
}

read the original abstract

Acoustic holography in the megahertz frequency range can impact numerous applications, including manufacturing, non-destructive testing, and transcranial ultrasound. However, designing lens topologies for complex acoustic holograms in the megahertz range poses a significant challenge, as weave propagation effects through the lens cannot be ignored. Here, we show that the inherent ability of heterogeneous angular spectrum approach to incorporate in plane varying speed-of-sound maps and support rapid differentiable optimization of lens thickness profiles can generate lens topologies for high fidelity acoustic holography. Crucially, we show that this framework can also account for wavefront aberrations in the propagation media, providing the opportunity to reconfigure this disruptive technology for high precision neuro-interventions. Our investigations also revealed that low frequency acoustic feedback generated by nonlinear mixing of high frequency waves allows attaining accurate skull-compensating lens alignment and creates the possibility to monitor CSF fluid build-up and removal in hydrocephalus. Together, our findings support the design of simple, economical, and high-performance ultrasound systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 7 canonical work pages

[1]

A normalization layer, adapted to the training data

work page
[2]

Three fully-connected hidden layers with 64 neurons each. Each layer uses the ELU activation function [Clevert et al., 2016], Batch Normalization [Ioffe and Szegedy, 2015], and is regularized with Dropout (rate=0.1) [Srivastava et al., 2014] and an L2 kernel regularizer(10 �3 )

work page 2016
[3]

events-in-bag

A global average pooling layer operates across the "events-in-bag" axis of the output embeddings from the final hidden layer. This produces a single, fixed-size summary vector for the entire bag, which we call the��

work page
[4]

probability

A final output is a single neuron with a sigmoid (for binary) or softmax (for multi-class) activation function. 14 Table 1: Features included in the training dataset. The features are categorized into low-level kinematic variables and high-level engineered features. For pairs of particles, the indices0 and 1 (e.g., ℓ0, ℓ1) refer to the leading and sublead...

work page 2019
[11]

The LLR curvature does not match the MLE variance (Icurv ̸=I MLE), violating the second Bartlett identity

work page
[12]

inductive bias

The models are not unbiased estimators, i.e.�[ ˆθ]̸=θ true. This section details the procedures used to correct for these effects and justifies why our primary measurement of the effective Fisher Information remains sound. C.2.1 Procedure 1: LLR Curvature Calibration As it was explained in Section 4.2, and can be seen in Tables 2 and 3, the nominal predic...

work page 2021
[70]

∂ ∂θ lnp(B |θ) # .(19) Since the reference point is fixed, and ∂ ∂θ NX i=1 lnp(� � |θ 0) = 0, we can add the zero term to variance of the score,Var θ

URLhttps://doi.org/10.1007/JHEP12(2017)070. Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dandelion ...

arXiv 2017
[1997]

URL https://royalsocietypublishing.org/doi/10

doi: 10.1098/rsta.1933.0009. URL https://royalsocietypublishing.org/doi/10. 1098/rsta.1933.0009. Publisher: Royal Society. Johann Brehmer, Felix Kling, Irina Espejo, and Kyle Cranmer. MadMiner: Machine Learning- Based Inference for Particle Physics.�� , 4(1):3, January

arXiv 1933

Show all 12 references

[2017]

doi: 10.1007/JHEP10(2017)174

ISSN 1029-8479. doi: 10.1007/JHEP10(2017)174. URL https://doi.org/10.1007/ JHEP10(2017)174. M. S. Bartlett. Approximate Confidence Intervals.��, 40(1-2):12–19, June 1953. ISSN 0006-

2017 doi
[2020]

doi: 10.1007/s41781-020-0035-2

ISSN 2510-2044. doi: 10.1007/s41781-020-0035-2. URLhttps://doi.org/10.1007/ s41781-020-0035-2. R. Frederix, S. Frixione, V . Hirschi, D. Pagani, H.-S. Shao, and M. Zaro. The automation of next- to-leading order electroweak calculations, October 2021. URL http://arxiv.org/abs/1...

arXiv
[2021]

doi: 10.1109/cvpr46437.2021.00529

IEEE. doi: 10.1109/cvpr46437.2021.00529. URL https://ieeexplore.ieee.org/ document/9577904/. Dawid Rymarczyk, Adriana Borowa, Jacek Tabor, and Bartosz Zielinski. Kernel Self-Attention for Weakly-supervised Image Classification using Deep Multiple Instance Learning. In�� ...

2021
[3444]

URL https://doi.org/10.1093/biomet/40.1-2.12

doi: 10.1093/biomet/40.1-2.12. URL https://doi.org/10.1093/biomet/40.1-2.12. Pierre Baldi, Kyle Cranmer, Taylor Faucett, Peter Sadowski, and Daniel Whiteson. Parameterized neural networks for high-energy physics.�� , 76(5):235, April 2016. ISSN 1434...

2016

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

A normalization layer, adapted to the training data

work page

[2] [2]

Three fully-connected hidden layers with 64 neurons each. Each layer uses the ELU activation function [Clevert et al., 2016], Batch Normalization [Ioffe and Szegedy, 2015], and is regularized with Dropout (rate=0.1) [Srivastava et al., 2014] and an L2 kernel regularizer(10 �3 )

work page 2016

[3] [3]

events-in-bag

A global average pooling layer operates across the "events-in-bag" axis of the output embeddings from the final hidden layer. This produces a single, fixed-size summary vector for the entire bag, which we call the��

work page

[4] [4]

probability

A final output is a single neuron with a sigmoid (for binary) or softmax (for multi-class) activation function. 14 Table 1: Features included in the training dataset. The features are categorized into low-level kinematic variables and high-level engineered features. For pairs of particles, the indices0 and 1 (e.g., ℓ0, ℓ1) refer to the leading and sublead...

work page 2019

[5] [11]

The LLR curvature does not match the MLE variance (Icurv ̸=I MLE), violating the second Bartlett identity

work page

[6] [12]

inductive bias

The models are not unbiased estimators, i.e.�[ ˆθ]̸=θ true. This section details the procedures used to correct for these effects and justifies why our primary measurement of the effective Fisher Information remains sound. C.2.1 Procedure 1: LLR Curvature Calibration As it was explained in Section 4.2, and can be seen in Tables 2 and 3, the nominal predic...

work page 2021

[7] [70]

∂ ∂θ lnp(B |θ) # .(19) Since the reference point is fixed, and ∂ ∂θ NX i=1 lnp(� � |θ 0) = 0, we can add the zero term to variance of the score,Var θ

URLhttps://doi.org/10.1007/JHEP12(2017)070. Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dandelion ...

arXiv 2017

[8] [1997]

URL https://royalsocietypublishing.org/doi/10

doi: 10.1098/rsta.1933.0009. URL https://royalsocietypublishing.org/doi/10. 1098/rsta.1933.0009. Publisher: Royal Society. Johann Brehmer, Felix Kling, Irina Espejo, and Kyle Cranmer. MadMiner: Machine Learning- Based Inference for Particle Physics.�� , 4(1):3, January

arXiv 1933