pith. sign in

arxiv: 2510.27487 · v2 · submitted 2025-10-31 · 📡 eess.IV

Towards robust quantitative photoacoustic tomography via learned iterative methods

Anssi Manninen , Janek Gr\"ohl , Felix Lucka , Andreas Hauptmann This is my paper

Pith reviewed 2026-05-18 03:15 UTC · model grok-4.3

classification 📡 eess.IV
keywords quantitative photoacoustic tomographylearned iterative methodsmodel-based reconstructiondeep learningscarce training datageneralizabilityinverse problemsdigital twin
0
0 comments X

The pith

By iteratively providing model information to updating networks, learned methods achieve robust quantitative photoacoustic tomography even with scarce training data.

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

The paper aims to show that a model-based learned iterative approach improves generalizability for quantitative photoacoustic tomography when training data is scarce. It does this by supplying information from the physical forward model to the networks at each iteration step, combining the speed of learned methods with the constraints of the imaging physics. A reader would care because large training datasets are hard to obtain in medical imaging applications, while purely model-based reconstructions are computationally slow. The authors compare updates drawn from gradient descent, Gauss-Newton and Quasi-Newton schemes, trained either greedily or end-to-end, and evaluate them on both ideal simulations and a digital twin dataset built to mimic limited data and modeling inaccuracies.

Core claim

Adopting the model-based learned iterative approach for QPAT provides better generalizability with scarce training data by iteratively providing additional information from the model to the updating networks. The work compares learned updates based on gradient descent, Gauss-Newton, and Quasi-Newton methods, formulated as either greedy iterate-wise or end-to-end training tasks, and tests the resulting reconstructions on ideal simulated data as well as a digital twin dataset that emulates scarce training data and high modeling error.

What carries the argument

The model-based learned iterative update, which injects forward-model information into each network step to refine the estimated optical absorption coefficients.

If this is right

  • The iterative supply of model information yields reconstructions that generalize better than purely data-driven networks when training examples are few.
  • Learned versions of gradient-descent, Gauss-Newton and Quasi-Newton updates are all viable for the task.
  • Both greedy per-iterate training and joint end-to-end training produce usable networks.
  • Performance gains persist when the test data include the kinds of modeling errors captured by the digital twin.

Where Pith is reading between the lines

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

  • The same hybrid update structure could be tried on other quantitative tomography problems that have an accurate forward model but limited training data.
  • If the robustness carries over, the method might reduce the data-collection burden for clinical deployment of photoacoustic imaging.
  • Further tests on dynamic or time-resolved acquisitions would show whether the added model steps still permit real-time operation.

Load-bearing premise

The digital twin dataset sufficiently emulates scarce training data and high modeling error in a way that transfers to real measurements.

What would settle it

A comparison of the learned iterative reconstructions against baselines on experimental data collected from a real photoacoustic system, using only a small number of training examples and the actual modeling discrepancies present in the hardware.

Figures

Figures reproduced from arXiv: 2510.27487 by Andreas Hauptmann, Anssi Manninen, Felix Lucka, Janek Gr\"ohl.

Figure 1
Figure 1. Figure 1: Left: Setup of the photoacoustic hardware. Right: Absorption ( [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Ideal problem: Example of drawn absorption (left) and scattering (mid) coefficients and respective simulated absorbed energy density (right), when the top side was illuminated. To simulate the absorbed energy densities of each sample, the ValoMC [13] open Monte Carlo software package for Matlab was used. ValoMC utilizes the photon package method to simulate the fluence in the given geometry and light sourc… view at source ↗
Figure 4
Figure 4. Figure 4: Digital twin problem: Used finite element mesh for the digital twin problem (left). The blue regions represent the water nodes, red the estimated nodes, and yellow the light source (700 nm) locations. The right figure shows nodewise relative difference between predicted absorbed energy density from diffusion approximation and simulated measurements for a single twin sample. We emphasize that since the data… view at source ↗
Figure 5
Figure 5. Figure 5: shows the average relative error of absorption and reduced scattering over the test set of 125 samples for GD, SR1, and GN based learned iterative methods up to 9 updating networks Λθk . After 9 iterations, the difference in average relative errors was observed to be marginal. The relative error of the fully learned (single-step) reconstructions is shown as the vertical dashed line in [PITH_FULL_IMAGE:fig… view at source ↗
Figure 6
Figure 6. Figure 6: Ideal problem: absorption (µa) and reduced scattering (µ ′ s ) recon￾structions of a test sample (left) using classical GN solver (14 iterations) and residual U-Net [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Ideal problem: Absorption (µa) and reduced scattering (µ ′ s ) recon￾structions of a single test sample using a) greedily and b) end-to-end trained learned iterative SR1, GD, and GN with 9 updating networks Λθk . reconstructions with an average relative error of 7% for absorption and 9% for reduced scattering. As can be seen from [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Digital twin problem: relative errors of absorption (µa) and reduced scattering (µ ′ s ) reconstruction from U-Net (left), learned iterative Gauss-Newton (mid), and Gradient descent (right) for circular, twin, and outlier samples. The dashed black line shows the average relative error of training samples. The relative error of each plotted sample is averaged over the six solved wavelengths. The learned gra… view at source ↗
Figure 9
Figure 9. Figure 9: Digital twin problem: absorption (µa) and reduced scattering (µ ′ s ) reconstructions of a) outlier, b) twin, and c) circular test samples (leftmost column). The methods used for the reconstructions starting from the second leftmost column are classical Gauss-Newton with total variation regularizer after around 30 iterations, fully learned U-Net, learned gradient descent with K = 4, and learned Gauss-Newto… view at source ↗
read the original abstract

Photoacoustic tomography (PAT) is a medical imaging modality that can provide high-resolution tissue images based on the optical absorption. Classical reconstruction methods for quantifying the absorption coefficients rely on sufficient prior information to overcome noisy and imperfect measurements. As these methods utilize computationally expensive forward models, the computation becomes slow, limiting their potential for time-critical applications. As an alternative approach, deep learning-based reconstruction methods have been established for faster and more accurate reconstructions. However, most of these methods rely on having a large amount of training data, which is not the case in practice. In this work, we adopt the model-based learned iterative approach for the use in Quantitative PAT (QPAT), in which additional information from the model is iteratively provided to the updating networks, allowing better generalizability with scarce training data. We compare the performance of different learned updates based on gradient descent, Gauss-Newton, and Quasi-Newton methods. The learning tasks are formulated as greedy, requiring iterate-wise optimality, as well as end-to-end, where all networks are trained jointly. The implemented methods are tested with ideal simulated data as well as against a digital twin dataset that emulates scarce training data and high modeling error.

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

2 major / 1 minor

Summary. The manuscript proposes adopting model-based learned iterative methods for quantitative photoacoustic tomography (QPAT), where updates derived from gradient descent, Gauss-Newton, and Quasi-Newton schemes iteratively inject information from the physical forward model into the networks. This is intended to improve generalizability under scarce training data compared to purely data-driven approaches. The methods are formulated in both greedy (iterate-wise) and end-to-end training modes and are evaluated on ideal simulated data as well as a digital twin dataset constructed to emulate data scarcity and high modeling error.

Significance. If the central claim holds, the work offers a practical route to robust QPAT reconstruction by hybridizing model-based iteration with learned updates, addressing the common limitation of insufficient training data in medical imaging applications. The explicit comparison of optimizer variants and training strategies provides a useful benchmark for the field.

major comments (2)
  1. [methods / digital twin] Digital twin construction (methods section): The robustness claim rests on the digital twin emulating scarce training data and high modeling error, yet the manuscript provides insufficient detail on the specific injected mismatches (e.g., acoustic heterogeneity, wavelength-dependent fluence, or transducer directivity). Without quantitative characterization of these statistics and a sensitivity analysis, it is unclear whether the observed gains transfer to physical measurements.
  2. [results] Results on generalizability (results section): The abstract and results lack explicit quantitative metrics (e.g., mean absolute error, structural similarity, or error bars across multiple realizations) comparing the learned iterative variants against baselines under the scarce-data regime. This makes it difficult to assess the magnitude and statistical significance of the reported improvement.
minor comments (1)
  1. [methods] Notation for the learned update operators should be unified across the gradient-descent, Gauss-Newton, and Quasi-Newton variants to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their positive evaluation of the significance of our work and for the constructive major comments. We address each point below and describe the revisions we will implement to improve the manuscript.

read point-by-point responses

  1. Referee: [methods / digital twin] Digital twin construction (methods section): The robustness claim rests on the digital twin emulating scarce training data and high modeling error, yet the manuscript provides insufficient detail on the specific injected mismatches (e.g., acoustic heterogeneity, wavelength-dependent fluence, or transducer directivity). Without quantitative characterization of these statistics and a sensitivity analysis, it is unclear whether the observed gains transfer to physical measurements.

    Authors: We agree that a more detailed and quantitative description of the digital twin is warranted to support the robustness claims. In the revised manuscript we will expand the Methods section with explicit statistics on the injected mismatches, including mean and standard deviation of acoustic speed-of-sound heterogeneity, quantitative bounds on wavelength-dependent fluence errors, and the modeled transducer directivity function. We will also add a sensitivity analysis that varies these parameters within realistic ranges and reports the resulting changes in reconstruction error for the learned iterative methods. While the current study is confined to simulated and digital-twin data, we will include a brief discussion of how the chosen mismatch statistics were derived from published tissue-property measurements, thereby clarifying the intended bridge toward physical experiments. revision: yes

  2. Referee: [results] Results on generalizability (results section): The abstract and results lack explicit quantitative metrics (e.g., mean absolute error, structural similarity, or error bars across multiple realizations) comparing the learned iterative variants against baselines under the scarce-data regime. This makes it difficult to assess the magnitude and statistical significance of the reported improvement.

    Authors: We thank the referee for pointing this out. Although comparative figures are present, we acknowledge that tabulated quantitative metrics with error bars were not provided for the scarce-data regime. In the revision we will insert a new table (and corresponding text) that reports mean absolute error, structural similarity index, peak signal-to-noise ratio, and their standard deviations computed over multiple independent noise realizations and training/validation splits for all learned iterative variants and the baseline methods. This will enable a direct, statistically grounded comparison of generalization performance. revision: yes

Circularity Check

0 steps flagged

No circularity: model-based learned updates remain independent of target results

full rationale

The derivation introduces model-based learned iterative schemes (gradient descent, Gauss-Newton, Quasi-Newton variants) that inject the physical forward operator into each network update step. Training is performed either greedily or end-to-end on simulated and digital-twin data; performance metrics are then measured on those same data distributions. No equation reduces a claimed prediction to a fitted parameter by construction, no uniqueness theorem is imported from prior self-work, and no ansatz is smuggled via citation. The digital-twin emulation is an external testbed rather than a definitional input, so the reported robustness claims do not collapse into tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that iterative injection of model information into networks yields better generalization than standard learned methods when data is scarce; no free parameters, new axioms, or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption The physical forward model for photoacoustic wave propagation provides useful additional information that can be iteratively fed into the network updates.
    Invoked when stating that additional information from the model is iteratively provided to the updating networks.

pith-pipeline@v0.9.0 · 5740 in / 1332 out tokens · 25201 ms · 2026-05-18T03:15:18.467013+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

52 extracted references · 52 canonical work pages

  1. [1]

    Biomedical photoacoustic imaging,

    P . Beard, “Biomedical photoacoustic imaging,” Interface F ocus, vol. 1, 2011

    work page 2011

  2. [2]

    L. V . Wang, Photoacoustic imaging and spectroscopy . CRC press, 2017

    work page 2017

  3. [3]

    Photoacoustic tomography and sensing in biomedicine,

    C. Li and L. V . Wang, “Photoacoustic tomography and sensing in biomedicine,” Physics in Medicine & Biology , vol. 54, no. 19, p. R59, 2009

    work page 2009

  4. [4]

    Quantitative spectroscopic photoacoustic imaging: a review,

    B. Cox, J. G. Laufer, S. R. Arridge, and P . C. Beard, “Quantitative spectroscopic photoacoustic imaging: a review,” Journal of biomedical optics, vol. 17, no. 6, pp. 061 202–061 202, 2012

    work page 2012

  5. [5]

    Recent advances in photoacoustic tomography,

    L. Li and L. V . Wang, “Recent advances in photoacoustic tomography,” BME frontiers, 2021

    work page 2021

  6. [6]

    Small-animal whole-body photoacoustic to- mography: a review,

    J. Xia and L. V . Wang, “Small-animal whole-body photoacoustic to- mography: a review,” IEEE Trans. Biomed. Eng. , vol. 61, no. 5, pp. 1380–1389, 2013

    work page 2013

  7. [7]

    A practical guide to photoacoustic tomography in the life sciences,

    L. V . Wang and J. Y ao, “A practical guide to photoacoustic tomography in the life sciences,” Nature methods, vol. 13, no. 8, pp. 627–638, 2016

    work page 2016

  8. [8]

    Model-based reconstructions for quantitative imaging in photoacoustic tomography,

    A. Hauptmann and T. Tarvainen, “Model-based reconstructions for quantitative imaging in photoacoustic tomography,” in Biomedical Pho- toacoustics: Technology and Applications. Springer, 2024, pp. 133–153

    work page 2024

  9. [9]

    Optical tomography in medical imaging,

    S. R. Arridge, “Optical tomography in medical imaging,” Inverse prob- lems, vol. 15, no. 2, p. R41, 1999

    work page 1999

  10. [10]

    Multiple illumination quantitative photoacoustic tomography using transport and diffusion models,

    B. Cox, T. Tarvainen, and S. Arridge, “Multiple illumination quantitative photoacoustic tomography using transport and diffusion models,” in Tomography and Inverse Transport Theory , 2011, vol. 559, pp. 1–12

    work page 2011

  11. [11]

    Image reconstruction in quantitative photoacoustic tomography using the radiative transfer equation and the diffusion approximation,

    T. Tarvainen, A. Pulkkinen, B. T. Cox, J. P . Kaipio, and S. R. Arridge, “Image reconstruction in quantitative photoacoustic tomography using the radiative transfer equation and the diffusion approximation,” in European Conference on Biomedical Optics . Optica Publishing Group, 2013, p. 880006

    work page 2013

  12. [12]

    An investigation of light transport through scattering bodies with non- scattering regions,

    M. Firbank, S. R. Arridge, M. Schweiger, and D. T. Delpy, “An investigation of light transport through scattering bodies with non- scattering regions,” Physics in Medicine & Biology , vol. 41, no. 4, p. 767, 1996. 13

    work page 1996

  13. [13]

    V alomc: a monte carlo software and matlab toolbox for simulating light transport in biological tissue,

    A. A. Leino, A. Pulkkinen, and T. Tarvainen, “V alomc: a monte carlo software and matlab toolbox for simulating light transport in biological tissue,” Osa Continuum , vol. 2, no. 3, pp. 957–972, 2019

    work page 2019

  14. [14]

    Nocedal and S

    J. Nocedal and S. J. Wright, Numerical optimization . Springer, 1999

    work page 1999

  15. [15]

    A gradient-based method for quantitative photoacoustic tomography using the radiative transfer equation,

    T. Saratoon, T. Tarvainen, B. Cox, and S. Arridge, “A gradient-based method for quantitative photoacoustic tomography using the radiative transfer equation,” Inverse Problems, vol. 29, no. 7, p. 075006, 2013

    work page 2013

  16. [16]

    Quantitative photoacoustic imaging: fitting a model of light transport to the initial pressure distribution,

    B. T. Cox, S. Arridge, K. Kostli, and P . C. Beard, “Quantitative photoacoustic imaging: fitting a model of light transport to the initial pressure distribution,” in Photons Plus Ultrasound: Imaging and Sensing 2005, vol. 5697. SPIE, 2005, pp. 49–55

    work page 2005

  17. [17]

    Estimating chromophore distributions from multiwavelength photoacoustic images,

    B. T. Cox, S. R. Arridge, and P . C. Beard, “Estimating chromophore distributions from multiwavelength photoacoustic images,” JOSA A , vol. 26, no. 2, pp. 443–455, 2009

    work page 2009

  18. [18]

    Quantitative photoacoustic tomography: modeling and inverse problems,

    T. Tarvainen and B. Cox, “Quantitative photoacoustic tomography: modeling and inverse problems,” Journal of Biomedical Optics , vol. 29, no. S1, pp. S11 509–S11 509, 2024

    work page 2024

  19. [19]

    Deep learning in photoacoustic tomogra- phy: current approaches and future directions,

    A. Hauptmann and B. Cox, “Deep learning in photoacoustic tomogra- phy: current approaches and future directions,” Journal of Biomedical Optics, vol. 25, no. 11, pp. 112 903–112 903, 2020

    work page 2020

  20. [20]

    Deep learn- ing for biomedical photoacoustic imaging: A review,

    J. Gr ¨ohl, M. Schellenberg, K. Dreher, and L. Maier-Hein, “Deep learn- ing for biomedical photoacoustic imaging: A review,” Photoacoustics, vol. 22, p. 100241, 2021

    work page 2021

  21. [21]

    A deep learning method based on u-net for quantitative photoacoustic imaging,

    T. Chen, T. Lu, S. Song, S. Miao, F. Gao, and J. Li, “A deep learning method based on u-net for quantitative photoacoustic imaging,” in Photons Plus Ultrasound: Imaging and Sensing 2020 , vol. 11240. SPIE, 2020, pp. 216–223

    work page 2020

  22. [22]

    End-to-end deep neural network for optical inversion in quantitative photoacoustic imaging,

    C. Cai, K. Deng, C. Ma, and J. Luo, “End-to-end deep neural network for optical inversion in quantitative photoacoustic imaging,” Optics letters , vol. 43, no. 12, pp. 2752–2755, 2018

    work page 2018

  23. [23]

    Quantitative photoacoustic blood oxygenation imaging using deep residual and recurrent neural network,

    C. Y ang, H. Lan, H. Zhong, and F. Gao, “Quantitative photoacoustic blood oxygenation imaging using deep residual and recurrent neural network,” in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019) . IEEE, 2019, pp. 741–744

    work page 2019

  24. [24]

    Toward accurate quantitative photoacoustic imaging: learning vascular blood oxygen saturation in three dimensions,

    C. Bench, A. Hauptmann, and B. Cox, “Toward accurate quantitative photoacoustic imaging: learning vascular blood oxygen saturation in three dimensions,” Journal of Biomedical Optics , vol. 25, no. 8, pp. 085 003–085 003, 2020

    work page 2020

  25. [25]

    Distribution-informed and wavelength- flexible data-driven photoacoustic oximetry,

    J. Gr ¨ohl, K. Y eung, K. Gu, T. R. Else, M. Golinska, E. V . Bunce, L. Hacker, and S. E. Bohndiek, “Distribution-informed and wavelength- flexible data-driven photoacoustic oximetry,” Journal of Biomedical Optics, vol. 29, no. S3, pp. S33 303–S33 303, 2024

    work page 2024

  26. [26]

    Moving beyond simulation: data-driven quantitative photoacoustic imaging using tissue-mimicking phantoms,

    J. Gr ¨ohl, T. R. Else, L. Hacker, E. V . Bunce, P . W. Sweeney, and S. E. Bohndiek, “Moving beyond simulation: data-driven quantitative photoacoustic imaging using tissue-mimicking phantoms,” IEEE Trans. Med. Imag. , vol. 43, no. 3, pp. 1214–1224, 2023

    work page 2023

  27. [27]

    Deep learning acceleration of iterative model-based light fluence correction for photoacoustic tomography,

    Z. Liang, S. Zhang, Z. Liang, Z. Mo, X. Zhang, Y . Zhong, W. Chen, and L. Qi, “Deep learning acceleration of iterative model-based light fluence correction for photoacoustic tomography,” Photoacoustics, vol. 37, p. 100601, 2024

    work page 2024

  28. [28]

    Self-supervised light fluence correction network for photoacoustic tomography based on diffusion equation,

    Z. Liang, Z. Mo, S. Zhang, L. Chen, D. Wang, C. Hu, and L. Qi, “Self-supervised light fluence correction network for photoacoustic tomography based on diffusion equation,” Photoacoustics, vol. 42, p. 100684, 2025

    work page 2025

  29. [29]

    Solving inverse problems using data-driven models,

    S. Arridge, P . Maass, O. ¨Oktem, and C.-B. Sch ¨onlieb, “Solving inverse problems using data-driven models,” Acta Numerica, vol. 28, pp. 1–174, 2019

    work page 2019

  30. [30]

    Learned primal-dual reconstruction,

    J. Adler and O. ¨Oktem, “Learned primal-dual reconstruction,” IEEE Trans. Med. Imag. , vol. 37, no. 6, pp. 1322–1332, 2018

    work page 2018

  31. [31]

    Model-based learning for accelerated, limited-view 3-d photoacoustic tomography,

    A. Hauptmann, F. Lucka, M. Betcke, N. Huynh, J. Adler, B. Cox, P . Beard, S. Ourselin, and S. Arridge, “Model-based learning for accelerated, limited-view 3-d photoacoustic tomography,” IEEE Trans. Med. Imag. , vol. 37, no. 6, pp. 1382–1393, 2018

    work page 2018

  32. [32]

    Learning a variational network for reconstruction of accelerated mri data,

    K. Hammernik, T. Klatzer, E. Kobler, M. P . Recht, D. K. Sodickson, T. Pock, and F. Knoll, “Learning a variational network for reconstruction of accelerated mri data,” Magnetic resonance in medicine , vol. 79, no. 6, pp. 3055–3071, 2018

    work page 2018

  33. [33]

    A model-based iterative learning approach for diffuse optical tomography,

    M. Mozumder, A. Hauptmann, I. Nissil ¨a, S. R. Arridge, and T. Tar- vainen, “A model-based iterative learning approach for diffuse optical tomography,” IEEE Trans. Med. Imag. , vol. 41, pp. 1289–1299, 2021

    work page 2021

  34. [34]

    Graph convolutional networks for model-based learning in nonlinear inverse problems,

    W. Herzberg, D. B. Rowe, A. Hauptmann, and S. J. Hamilton, “Graph convolutional networks for model-based learning in nonlinear inverse problems,” IEEE Trans. Comput. Imag. , vol. 7, pp. 1341–1353, 2021

    work page 2021

  35. [35]

    Deep-plug-and- play proximal gauss-newton method with applications to nonlinear, ill- posed inverse problems,

    F. Colibazzi, D. Lazzaro, S. Morigi, and A. Samor ´e, “Deep-plug-and- play proximal gauss-newton method with applications to nonlinear, ill- posed inverse problems,” Inverse Problems and Imaging , vol. 17, no. 6, pp. 1226–1248, 2023

    work page 2023

  36. [36]

    Deep unfolding network for nonlinear multi-frequency electrical impedance tomography,

    G. S. Alberti, D. Lazzaro, S. Morigi, L. Ratti, and M. Santace- saria, “Deep unfolding network for nonlinear multi-frequency electrical impedance tomography,” arXiv preprint arXiv:2507.16678 , 2025

    work page arXiv 2025

  37. [37]

    Digital twins enable full- reference quality assessment of photoacoustic image reconstructions,

    J. Gr ¨ohl, L. Kunyansky, J. Poimala, T. R. Else, F. Di Cecio, S. E. Bohndiek, B. T. Cox, and A. Hauptmann, “Digital twins enable full- reference quality assessment of photoacoustic image reconstructions,” The Journal of the Acoustical Society of America , vol. 158, no. 1, pp. 590–601, 07 2025

    work page 2025

  38. [38]

    Ishimaru et al

    A. Ishimaru et al. , Wave propagation and scattering in random media . Academic press New Y ork, 1978, vol. 2

    work page 1978

  39. [39]

    Diffuse radiation in the galaxy,

    L. G. Henyey and J. L. Greenstein, “Diffuse radiation in the galaxy,” Astrophysical Journal, vol. 93, p. 70-83 (1941). , vol. 93, pp. 70–83, 1941

    work page 1941

  40. [40]

    Reconstruction- classification method for quantitative photoacoustic tomography,

    E. Malone, S. Powell, B. T. Cox, and S. Arridge, “Reconstruction- classification method for quantitative photoacoustic tomography,” Jour- nal of Biomedical Optics , vol. 20, no. 12, pp. 126 004–126 004, 2015

    work page 2015

  41. [41]

    Reconstructing absorption and scattering distributions in quantitative photoacoustic tomography,

    T. Tarvainen, B. T. Cox, J. Kaipio, and S. R. Arridge, “Reconstructing absorption and scattering distributions in quantitative photoacoustic tomography,” Inverse Problems, vol. 28, no. 8, p. 084009, 2012

    work page 2012

  42. [42]

    Optical properties of biological tissues: a review,

    S. L. Jacques, “Optical properties of biological tissues: a review,” Physics in Medicine & Biology , vol. 58, no. 11, p. R37, 2013

    work page 2013

  43. [43]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

    work page 2016

  44. [44]

    Dataset for: Moving beyond simulation: data-driven quantitative photoacoustic imaging using tissue-mimicking phantoms,

    J. Gr ¨ohl, T. Else, L. Hacker, E. Bunce, P . Sweeney, and S. Bohndiek, “Dataset for: Moving beyond simulation: data-driven quantitative photoacoustic imaging using tissue-mimicking phantoms,” Apollo - University of Cambridge Repository , 2023. [Online]. Available: https://www.repository.cam.ac.uk/handle/1810/359968

    work page 2023

  45. [45]

    A copolymer-in-oil tissue-mimicking material with tuneable acoustic and optical characteristics for photoacoustic imaging phantoms,

    L. Hacker, J. Joseph, A. M. Ivory, M. O. Saed, B. Zeqiri, S. Rajagopal, and S. E. Bohndiek, “A copolymer-in-oil tissue-mimicking material with tuneable acoustic and optical characteristics for photoacoustic imaging phantoms,” IEEE Trans. Med. Imag. , vol. 40, no. 12, pp. 3593–3603, 2021

    work page 2021

  46. [46]

    Double-integrating-sphere system for measuring the optical properties of tissue,

    J. W. Pickering, S. A. Prahl, N. V an Wieringen, J. F. Beek, H. J. Sterenborg, and M. J. V an Gemert, “Double-integrating-sphere system for measuring the optical properties of tissue,” Applied optics , vol. 32, no. 4, pp. 399–410, 1993

    work page 1993

  47. [47]

    The medical imaging interaction toolkit: challenges and advances: 10 years of open-source development,

    M. Nolden, S. Zelzer, A. Seitel, D. Wald, M. M ¨uller, A. M. Franz, D. Maleike, M. Fangerau, M. Baumhauer, L. Maier-Hein et al. , “The medical imaging interaction toolkit: challenges and advances: 10 years of open-source development,” International journal of computer assisted radiology and surgery , vol. 8, pp. 607–620, 2013

    work page 2013

  48. [48]

    Monte carlo simulation of photon migration in 3d turbid media accelerated by graphics processing units,

    Q. Fang and D. A. Boas, “Monte carlo simulation of photon migration in 3d turbid media accelerated by graphics processing units,” Optics express, vol. 17, no. 22, pp. 20 178–20 190, 2009

    work page 2009

  49. [49]

    k-wave: Matlab toolbox for the simulation and reconstruction of photoacoustic wave fields,

    B. E. Treeby and B. T. Cox, “k-wave: Matlab toolbox for the simulation and reconstruction of photoacoustic wave fields,” Journal of biomedical optics, vol. 15, no. 2, pp. 021 314–021 314, 2010

    work page 2010

  50. [50]

    Simpa: an open-source toolkit for simulation and image processing for photonics and acoustics,

    J. Gr ¨ohl, K. K. Dreher, M. Schellenberg, T. Rix, N. Holzwarth, P . Vieten, L. Ayala, S. E. Bohndiek, A. Seitel, and L. Maier-Hein, “Simpa: an open-source toolkit for simulation and image processing for photonics and acoustics,” Journal of biomedical optics , vol. 27, no. 8, pp. 083 010– 083 010, 2022

    work page 2022

  51. [51]

    Gauss–Newton method for image reconstruction in diffuse optical tomography,

    M. Schweiger, S. R. Arridge, and I. Nissil ¨a, “Gauss–Newton method for image reconstruction in diffuse optical tomography,” Physics in Medicine & Biology , vol. 50, no. 10, p. 2365, 2005

    work page 2005

  52. [52]

    An efficient primal-dual interior-point method for minimizing a sum of euclidean norms,

    K. D. Andersen, E. Christiansen, A. R. Conn, and M. L. Overton, “An efficient primal-dual interior-point method for minimizing a sum of euclidean norms,” SIAM Journal on Scientific Computing , vol. 22, no. 1, pp. 243–262, 2000. 1 SUPPLEMENTARY MATERIAL A. Radiative transform equation for QPAT For QPA T, the time-independent RTE in r ∈ Ω ⊂ Rn (n = 2 or 3) wi...

    work page 2000