pith. machine review for the scientific record. sign in

arxiv: 2604.09643 · v1 · submitted 2026-03-23 · 💻 cs.CV

Recognition: no theorem link

PA-SFM: Tracker-free differentiable acoustic radiation for freehand 3D photoacoustic imaging

Shuang Li , Jian Gao , Chulhong Kim , Seongwook Choi , Qian Chen , Yibing Wang , Shuang Wu , Yu Zhang
show 5 more authors
Tingting Huang Yucheng Zhou Boxin Yao Yao Yao Changhui Li
Authors on Pith no claims yet

Pith reviewed 2026-05-15 00:50 UTC · model grok-4.3

classification 💻 cs.CV
keywords photoacoustic tomographyfreehand imagingdifferentiable optimizationsensor pose estimationacoustic wave equation3D reconstructiontracker-freevascular imaging
0
0 comments X

The pith

PA-SFM recovers sensor poses and high-resolution 3D images from photoacoustic signals alone by embedding the wave equation in a differentiable optimizer.

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

The paper introduces PA-SFM, a method that treats freehand photoacoustic tomography as a joint optimization problem. It uses gradients computed through an acoustic radiation model to refine both the unknown 3D source distribution and the sensor array positions at the same time. No external tracking hardware is required because the optimization relies only on consistency between the recorded signals and the wave equation. A coarse-to-fine schedule plus rigid-body constraints discards motion outliers so that convergence remains stable even with irregular handheld trajectories. If the approach holds, portable 3D vascular imaging becomes feasible without bulky sensor rigs.

Core claim

By casting acoustic forward modeling as a differentiable operation, the framework simultaneously recovers the three-dimensional photoacoustic source map and the six-degree-of-freedom sensor poses from raw single-modality measurements. Gradient descent on the mismatch between simulated and observed signals drives the joint estimation, while geometric consistency checks and rigid-body constraints prune inconsistent measurements during a coarse-to-fine schedule. Numerical simulations and in-vivo rat carotid imaging confirm that the recovered poses reach sub-millimeter accuracy and that the reconstructed volumes match ground-truth quality.

What carries the argument

Differentiable acoustic radiation kernel that propagates gradients of the wave-equation residual back to both the source distribution and the sensor pose parameters.

If this is right

  • Tracker-free handheld 3D photoacoustic tomography becomes practical for clinical settings.
  • Sub-millimeter positioning accuracy is obtained directly from the acoustic data.
  • High-resolution vascular volumes comparable to sensor-tracked reconstructions are restored.
  • The same differentiable pipeline can be applied to other acoustic modalities that obey the wave equation.
  • Software-only correction removes the hardware cost and bulk of external positioning systems.

Where Pith is reading between the lines

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

  • The method may extend naturally to ultrasound or thermoacoustic imaging where similar wave physics apply.
  • Real-time variants could emerge if the GPU kernel is further accelerated or replaced by learned surrogates.
  • Integration with deep priors on vascular geometry might further stabilize optimization on noisy clinical data.

Load-bearing premise

The wave equation together with rigid-body constraints and geometric consistency checks are enough to make the joint pose-and-source problem well-posed for typical freehand trajectories.

What would settle it

A set of freehand scans collected with deliberately erratic motion that produces many rejected measurements yet still yields pose errors larger than one millimeter when compared to an external optical tracker.

read the original abstract

Three-dimensional (3D) handheld photoacoustic tomography typically relies on bulky and expensive external positioning sensors to correct motion artifacts, which severely limits its clinical flexibility and accessibility. To address this challenge, we present PA-SFM, a tracker-free framework that leverages exclusively single-modality photoacoustic data for both sensor pose recovery and high-fidelity 3D reconstruction via differentiable acoustic radiation modeling. Unlike traditional structure-from-motion (SFM) methods based on visual features, PA-SFM integrates the acoustic wave equation into a differentiable programming pipeline. By leveraging a high-performance, GPU-accelerated acoustic radiation kernel, the framework simultaneously optimizes the 3D photoacoustic source distribution and the sensor array pose via gradient descent. To ensure robust convergence in freehand scenarios, we introduce a coarse-to-fine optimization strategy that incorporates geometric consistency checks and rigid-body constraints to eliminate motion outliers. We validated the proposed method through both numerical simulations and in-vivo rat experiments. The results demonstrate that PA-SFM achieves sub-millimeter positioning accuracy and restores high-resolution 3D vascular structures comparable to ground-truth benchmarks, offering a low-cost, software-defined solution for clinical freehand photoacoustic imaging. The source code is publicly available at \href{https://github.com/JaegerCQ/PA-SFM}{https://github.com/JaegerCQ/PA-SFM}.

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 presents PA-SFM, a tracker-free framework for freehand 3D photoacoustic imaging that integrates the acoustic wave equation into a differentiable GPU-accelerated pipeline to jointly optimize the 3D source distribution and sensor array poses via gradient descent. It introduces a coarse-to-fine optimization strategy with geometric consistency checks and rigid-body constraints to handle motion outliers, validated on numerical simulations and in-vivo rat experiments claiming sub-millimeter positioning accuracy and high-resolution vascular reconstructions comparable to ground-truth benchmarks. Source code is released publicly.

Significance. If the central claims hold under broader conditions, the work would be significant for clinical photoacoustic tomography by removing reliance on external trackers, thereby improving accessibility and flexibility of handheld 3D imaging. The differentiable acoustic radiation approach offers a software-defined alternative to traditional structure-from-motion methods, with potential for extension to other acoustic modalities; public code availability aids reproducibility.

major comments (3)
  1. [Experiments] Experiments section: The reported simulations and in-vivo rat validation omit quantitative error bars, ablation studies on the geometric consistency checks, and explicit details on post-hoc outlier rejection, which are load-bearing for the claim of robust sub-millimeter accuracy across freehand trajectories.
  2. [Method] Method section: The sufficiency of the coarse-to-fine strategy plus rigid-body constraints for guaranteeing convergence in the non-convex joint optimization of sources and poses is not demonstrated beyond the tested trajectories; no quantification of basin of convergence or performance under irregular motion, variable speed-of-sound, or higher noise is provided.
  3. [Forward model] Forward model description: The acoustic wave equation implementation assumes homogeneity and smoothness without sensitivity analysis for tissue heterogeneity or noise levels that could trap the gradient-based optimization, undermining the robustness claim for arbitrary freehand scenarios.
minor comments (2)
  1. [Abstract] Abstract: The statement that reconstructions are 'comparable to ground-truth benchmarks' lacks specification of the quantitative metrics (e.g., RMSE, SSIM) used for the comparison.
  2. [Method] Notation: The definition of the acoustic radiation kernel and its differentiability could be clarified with an explicit equation reference to aid readers unfamiliar with the GPU implementation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We have revised the manuscript to strengthen the experimental validation, add supporting analyses, and clarify limitations where appropriate. Point-by-point responses to the major comments follow.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: The reported simulations and in-vivo rat validation omit quantitative error bars, ablation studies on the geometric consistency checks, and explicit details on post-hoc outlier rejection, which are load-bearing for the claim of robust sub-millimeter accuracy across freehand trajectories.

    Authors: We agree that these elements strengthen the claims. In the revised manuscript we have added mean ± standard deviation error bars to all positioning accuracy results for both simulations and in-vivo rat data. We have also included a new ablation study quantifying the contribution of the geometric consistency checks. Explicit implementation details of the post-hoc outlier rejection (thresholds, rigid-body constraint enforcement, and rejection criteria) have been expanded in the Methods section. revision: yes

  2. Referee: [Method] Method section: The sufficiency of the coarse-to-fine strategy plus rigid-body constraints for guaranteeing convergence in the non-convex joint optimization of sources and poses is not demonstrated beyond the tested trajectories; no quantification of basin of convergence or performance under irregular motion, variable speed-of-sound, or higher noise is provided.

    Authors: We acknowledge that the joint optimization is non-convex and no theoretical convergence guarantee is provided. In revision we added quantitative basin-of-convergence experiments using perturbed initial poses. Results for irregular motion trajectories and elevated noise levels have been included in the supplementary material. The method assumes constant speed-of-sound (standard in photoacoustic imaging); we have added a brief discussion of this assumption and its implications in the Discussion section. revision: partial

  3. Referee: [Forward model] Forward model description: The acoustic wave equation implementation assumes homogeneity and smoothness without sensitivity analysis for tissue heterogeneity or noise levels that could trap the gradient-based optimization, undermining the robustness claim for arbitrary freehand scenarios.

    Authors: The forward model follows the standard homogeneous acoustic wave equation used throughout the photoacoustic literature. To address the concern we have added a sensitivity analysis in the revised manuscript that evaluates convergence under moderate speed-of-sound heterogeneity (up to 5 % variation) and across a range of noise levels. The analysis shows stable behavior within the tested regimes; we have updated the Forward model section to report these results and have noted the limitation for strongly heterogeneous tissues in the Discussion. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation rests on standard wave equation and independent optimization

full rationale

The paper's core pipeline integrates the standard acoustic wave equation into a differentiable forward model, then performs joint gradient descent on source distribution and sensor poses. This is a direct application of physics-based simulation and numerical optimization; no equation or reported accuracy metric is defined in terms of itself or reduced to a fitted parameter from the same dataset. Coarse-to-fine checks and rigid-body constraints are algorithmic safeguards, not self-referential definitions. Validation relies on separate numerical simulations and in-vivo rat experiments whose ground-truth benchmarks are external to the optimization loop. No self-citation chain carries load-bearing uniqueness claims, and no result is presented as a 'prediction' that is statistically forced by its own inputs. The method is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard acoustic wave equation as a domain assumption and on the effectiveness of the introduced coarse-to-fine geometric consistency checks. No new physical entities are postulated. A small number of optimization hyperparameters (step sizes, consistency thresholds) are implicitly present but not quantified in the abstract.

free parameters (1)
  • coarse-to-fine optimization hyperparameters
    Step sizes, consistency thresholds, and rigid-body constraint weights are required for the strategy but not numerically specified in the provided abstract.
axioms (1)
  • domain assumption The acoustic wave equation accurately models propagation in the imaged medium.
    Invoked as the foundation of the differentiable radiation kernel.

pith-pipeline@v0.9.0 · 5585 in / 1235 out tokens · 46574 ms · 2026-05-15T00:50:43.291244+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    Nature Reviews Bioengineering 3(3), 193–212 (2025)

    Park, J., Choi, S., Knieling, F., Clingman, B., Bohndiek, S., Wang, L.V., Kim, C.: Clinical translation of photoacoustic imaging. Nature Reviews Bioengineering 3(3), 193–212 (2025)

    work page 2025

  2. [2]

    science335(6075), 1458–1462 (2012)

    Wang, L.V., Hu, S.: Photoacoustic tomography: in vivo imaging from organelles to organs. science335(6075), 1458–1462 (2012)

    work page 2012

  3. [3]

    Photoacoustics32, 100539 (2023)

    Assi, H., Cao, R., Castelino, M., Cox, B., Gilbert, F.J., Gr¨ ohl, J., Gurusamy, K., Hacker, L., Ivory, A.M., Joseph, J.,et al.: A review of a strategic roadmapping exercise to advance clinical translation of photoacoustic imaging: from current barriers to future adoption. Photoacoustics32, 100539 (2023)

    work page 2023

  4. [4]

    Chemical Society Reviews46(8), 2158–2198 (2017)

    De´ an-Ben, X., Gottschalk, S., Mc Larney, B., Shoham, S., Razansky, D.: Advanced optoacoustic methods for multiscale imaging of in vivo dynamics. Chemical Society Reviews46(8), 2158–2198 (2017)

    work page 2017

  5. [5]

    Nature Reviews Clinical Oncology19(6), 365–384 (2022)

    Lin, L., Wang, L.V.: The emerging role of photoacoustic imaging in clinical oncology. Nature Reviews Clinical Oncology19(6), 365–384 (2022)

    work page 2022

  6. [6]

    Nature Reviews Bioengineering3(3), 182–184 (2025)

    Ntziachristos, V.: Addressing unmet clinical need with optoacoustic imaging. Nature Reviews Bioengineering3(3), 182–184 (2025)

    work page 2025

  7. [7]

    Journal of biomedical optics24(2), 026003–026003 (2019)

    Wray, P., Lin, L., Hu, P., Wang, L.V.: Photoacoustic computed tomography of human extremities. Journal of biomedical optics24(2), 026003–026003 (2019)

    work page 2019

  8. [8]

    Journal of Biomedical Optics29(S1), 11519–11519 (2024)

    Li, S., Zhang, G., Wang, Y., Li, W., Sun, Y., Li, C.: Photoacoustic imaging of peripheral vessels in extremities by large-scale synthetic matrix array. Journal of Biomedical Optics29(S1), 11519–11519 (2024)

    work page 2024

  9. [9]

    Biomedical Optics Express12(3), 1236–1246 (2021)

    Wang, M., Zhao, L., Wei, Y., Li, J., Qi, Z., Su, N., Zhao, C., Zhang, R., Tang, T., Liu, S.,et al.: Functional photoacoustic/ultrasound imaging for the assess- ment of breast intraductal lesions: preliminary clinical findings. Biomedical Optics Express12(3), 1236–1246 (2021)

    work page 2021

  10. [10]

    Photoacoustics21, 100222 (2021)

    Han, T., Yang, M., Yang, F., Zhao, L., Jiang, Y., Li, C.: A three-dimensional modeling method for quantitative photoacoustic breast imaging with handheld probe. Photoacoustics21, 100222 (2021)

    work page 2021

  11. [11]

    Photoacoustics36, 100593 (2024)

    Sun, Y., Wang, Y., Li, W., Li, C.: Real-time dual-modal photoacoustic and fluorescence small animal imaging. Photoacoustics36, 100593 (2024)

    work page 2024

  12. [12]

    Scientific reports 8(1), 786 (2018) 17

    Matsumoto, Y., Asao, Y., Yoshikawa, A., Sekiguchi, H., Takada, M., Furu, M., Saito, S., Kataoka, M., Abe, H., Yagi, T.,et al.: Label-free photoacoustic imaging of human palmar vessels: a structural morphological analysis. Scientific reports 8(1), 786 (2018) 17

    work page 2018

  13. [13]

    Scientific reports8(1), 14930 (2018)

    Matsumoto, Y., Asao, Y., Sekiguchi, H., Yoshikawa, A., Ishii, T., Nagae, K.-i., Kobayashi, S., Tsuge, I., Saito, S., Takada, M.,et al.: Visualising peripheral arte- rioles and venules through high-resolution and large-area photoacoustic imaging. Scientific reports8(1), 14930 (2018)

    work page 2018

  14. [14]

    Radiology291(1), 45–50 (2019)

    Ivankovic, I., Merˇ cep, E., Schmedt, C.-G., De´ an-Ben, X.L., Razansky, D.: Real- time volumetric assessment of the human carotid artery: handheld multispectral optoacoustic tomography. Radiology291(1), 45–50 (2019)

    work page 2019

  15. [15]

    Optics express21(23), 28062–28071 (2013)

    De´ an-Ben, X.L., Razansky, D.: Portable spherical array probe for volumetric real-time optoacoustic imaging at centimeter-scale depths. Optics express21(23), 28062–28071 (2013)

    work page 2013

  16. [16]

    F1000Research 7(2018)

    Nagae, K., Asao, Y., Sudo, Y., Murayama, N., Tanaka, Y., Ohira, K., Ishida, Y., Otsuka, A., Matsumoto, Y., Saito, S., et al.: Real-time 3d photoacoustic visual- ization system with a wide field of view for imaging human limbs. F1000Research 7(2018)

    work page 2018

  17. [17]

    Optics Letters48(2), 343–346 (2023)

    Kim, W., Choi, W., Ahn, J., Lee, C., Kim, C.: Wide-field three-dimensional pho- toacoustic/ultrasound scanner using a two-dimensional matrix transducer array. Optics Letters48(2), 343–346 (2023)

    work page 2023

  18. [18]

    IEEE Journal of Selected Topics in Quantum Electronics16(4), 730–739 (2009)

    Piras, D., Xia, W., Steenbergen, W., Van Leeuwen, T.G., Manohar, S.: Pho- toacoustic imaging of the breast using the twente photoacoustic mammoscope: present status and future perspectives. IEEE Journal of Selected Topics in Quantum Electronics16(4), 730–739 (2009)

    work page 2009

  19. [19]

    Heijblom, M., Piras, D., Xia, W., Van Hespen, J., Klaase, J., Engh, F., Van Leeuwen, T., Steenbergen, W., Manohar, S.: Visualizing breast cancer using the twente photoacoustic mammoscope: what do we learn from twelve new patient measurements? Optics express20(11), 11582–11597 (2012)

    work page 2012

  20. [20]

    Journal of Biomedical Optics29(6), 066006–066006 (2024)

    Wang, Y., Li, C.: Comprehensive framework of gpu-accelerated image reconstruc- tion for photoacoustic computed tomography. Journal of Biomedical Optics29(6), 066006–066006 (2024)

    work page 2024

  21. [21]

    Nature Communications16(1), 10077 (2025)

    Wang, X., Meng, Y., Sun, M., Gao, X., Wang, Y., Wang, S., Wang, K., Wang, R., Ren, D., Yin, Y.,et al.: Cross-regional real-time visualization of systemic physiology and dynamics with 3d panoramic photoacoustic computed tomography (3d-panopact). Nature Communications16(1), 10077 (2025)

    work page 2025

  22. [22]

    Nature Biomedical Engineering, 1–12 (2026)

    Zhang, Y., Na, S., Russin, J.J., Sastry, K., Lin, L., Zheng, J., Luo, Y., Tong, X., An, Y., Hu, P., et al.: Rotational ultrasound and photoacoustic tomography of the human body. Nature Biomedical Engineering, 1–12 (2026)

    work page 2026

  23. [23]

    IEEE Transactions on Medical Imaging (2025)

    Lee, S., Kim, S., Seo, M., Park, S., Imrus, S., Ashok, K., Lee, D., Park, 18 C., Lee, S., Kim, J., et al.: Enhancing free-hand 3d photoacoustic and ultra- sound reconstruction using deep learning. IEEE Transactions on Medical Imaging (2025)

    work page 2025

  24. [24]

    IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control (2024)

    Dou, Y., Mu, F., Li, Y., Varghese, T.: Sensorless end-to-end freehand three- dimensional ultrasound reconstruction with physics guided deep learning. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control (2024)

    work page 2024

  25. [25]

    Sensors23(19), 8149 (2023)

    Lee, C., Kim, C., Park, B.: Review of three-dimensional handheld photoacoustic and ultrasound imaging systems and their applications. Sensors23(19), 8149 (2023)

    work page 2023

  26. [26]

    IEEE transactions on biomedical engineering61(10), 2527–2537 (2014)

    Lasso, A., Heffter, T., Rankin, A., Pinter, C., Ungi, T., Fichtinger, G.: Plus: open- source toolkit for ultrasound-guided intervention systems. IEEE transactions on biomedical engineering61(10), 2527–2537 (2014)

    work page 2014

  27. [27]

    IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 72(11), 1462–1475 (2025)

    Frey, B.N., Hyun, D., Simson, W., Zhuang, L., Hashemi, H.S., Schneider, M., Dahl, J.J.: Ultraflex: Iterative model-based ultrasonic flexible-array shape cali- bration. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 72(11), 1462–1475 (2025)

    work page 2025

  28. [28]

    IEEE Transactions on Medical Imaging39(4), 1160–1169 (2019)

    Knauer, N., Dean-Ben, X.L., Razansky, D.: Spatial compounding of volumetric data enables freehand optoacoustic angiography of large-scale vascular networks. IEEE Transactions on Medical Imaging39(4), 1160–1169 (2019)

    work page 2019

  29. [29]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Schonberger, J.L., Frahm, J.-M.: Structure-from-motion revisited. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4104– 4113 (2016)

    work page 2016

  30. [30]

    In: Proceedings of the 3rd ACM SIGPLAN International Workshop on Machine Learning and Programming Languages, pp

    Tillet, P., Kung, H.-T., Cox, D.: Triton: an intermediate language and compiler for tiled neural network computations. In: Proceedings of the 3rd ACM SIGPLAN International Workshop on Machine Learning and Programming Languages, pp. 10–19 (2019)

    work page 2019

  31. [31]

    Advanced Science10(1), 2202089 (2023)

    Choi, S., Yang, J., Lee, S.Y., Kim, J., Lee, J., Kim, W.J., Lee, S., Kim, C.: Deep learning enhances multiparametric dynamic volumetric photoacoustic computed tomography in vivo (dl-pact). Advanced Science10(1), 2202089 (2023)

    work page 2023

  32. [32]

    Chemical Reviews123(11), 7379–7419 (2023)

    Choi, W., Park, B., Choi, S., Oh, D., Kim, J., Kim, C.: Recent advances in contrast-enhanced photoacoustic imaging: overcoming the physical and practical challenges. Chemical Reviews123(11), 7379–7419 (2023)

    work page 2023

  33. [33]

    ACS nano (2024)

    Kim, J., Lee, J., Choi, S., Lee, H., Yang, J., Jeon, H., Sung, M., Kim, W.J., Kim, C.: 3d multiparametric photoacoustic computed tomography of primary and metastatic tumors in living mice. ACS nano (2024)

    work page 2024

  34. [34]

    Light: science & applications11(1), 131 (2022)

    Kim, J., Kim, G., Li, L., Zhang, P., Kim, J.Y., Kim, Y., Kim, H.H., Wang, 19 L.V., Lee, S., Kim, C.: Deep learning acceleration of multiscale superresolu- tion localization photoacoustic imaging. Light: science & applications11(1), 131 (2022)

    work page 2022

  35. [35]

    Physical Review E—Statistical, Nonlinear, and Soft Matter Physics71(1), 016706 (2005)

    Xu, M., Wang, L.V.: Universal back-projection algorithm for photoacoustic com- puted tomography. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics71(1), 016706 (2005)

    work page 2005

  36. [36]

    Review of scientific instruments77(4) (2006) 20

    Xu, M., Wang, L.V.: Photoacoustic imaging in biomedicine. Review of scientific instruments77(4) (2006) 20

    work page 2006