The multiple scattering tracing (MST) algorithm of Kang et al. (Nature Communications 14, 6871, 2023) demonstrated that a time-gated reflection matrix, recorded in a reflection microscope, suffices to identify a stack of phase plates equivalent to the scattering medium and thereby computationally convert object-bearing multiply scattered light into ballistic signal waves — tracing 17 scattering events, amplifying the ballistic signal by almost 580 times, and recovering microscopic images of neurons and osteocytes through an intact mouse skull. The authors explicitly acknowledge that the method does not trace all object-bearing multiple scattering, that collection geometry and the time-gating window only partially select the exploitable trajectories, that backscattering generated inside the medium is absent from the forward model, that the matrix-multiplication cost grows as O(N06)O(N_0^6), and that the ballistic-fraction parameter used to quantify enhancement is ill-determined when the initial ballistic signal is extremely weak. The 2023–2026 follow-up literature leaves these questions open: the same group's scattering-angle-basis successor (2025) extends the traceable fraction but is declared by its own authors to be complementary to the layer-based model, with a synergistic combination left as a long-term goal; reflection-only refractive-index tomography from internal backscattering (2026) exploits backscattering for a different objective; algorithmic work (2025) partially mitigates but does not resolve the computational scaling; and the optimal collection geometry, the optimal time-gating window, and a robust ballistic-fraction estimator remain untouched. The problem — maximizing the degree to which multiple scattering carrying object information can be inverted and exploited for deep optical imaging, across parameterization, geometry, gating, forward model, and computational cost — remains open.

Background

When light travels through a turbid medium such as biological tissue, it is deflected by refractive-index inhomogeneities. After one or two such events the wavefront is distorted but still spatially correlated with the object ('aberration' and weak forward multiple scattering); after many events the field is scrambled into an apparently random speckle. Conventional deep-tissue microscopy keeps only the ballistic component — the light that traverses the medium without scattering — and treats everything else as noise; since the ballistic intensity decays exponentially with depth (typically a few transport mean free paths in tissue), this strategy fails beyond roughly a millimeter in brain or skull. The field of wavefront shaping and matrix-based imaging asks instead whether the scattered light itself, which still carries object information in scrambled form, can be measured and computationally inverted. The central instrument is the reflection matrix R: an interferometric, laser-scanning confocal microscope records the complex (amplitude and phase) field returned from the sample for each of N0×N0N_0 \times N_0 focused input positions, yielding an N02×N02N_0^2 \times N_0^2 matrix mapping input to output fields. Time gating — here a 25 µm coherence window from a broadband pulsed source — selects the recording depth slice. Adaptive-optics and matrix-imaging methods correct aberrations and weak forward multiple scattering by finding per-voxel focusing laws in R; the optical memory effect (the spatial range over which the medium's distortion is correlated) limits their field of view. Kang et al. (Nature Communications 14, 6871, 2023) went decisively further with the multiple scattering tracing (MST) algorithm: modeling the medium as a stack of thin, mutually independent phase plates (random phase masks), it iteratively identifies the plate sequence whose combined transmission matrix reproduces the correlations measured in the experimentally recorded time-gated reflection matrix, and then applies the inverse of the identified plates to the data. This rectified high-order multiple scattering — up to 17 scattering events — and amplified the ballistic signal by almost 580 times, enabling label-free, better-than-1.05 µm resolution imaging of myelinated axons and osteocytes through the intact mouse skull, a regime in which confocal imaging shows nothing. The authors themselves framed this as a first step toward the 'holy grail' of deep imaging, the general high-order inverse scattering problem, and stated the specific limitations that define the open problem. First, MST traces only a subset of the multiple scattering that carries object information; the remainder still acts as noise. Second, the imaging geometry (illumination/collection aperture and configuration) and the width of the time-gating window determine which multiple-scattering trajectories are recorded at all, and how to choose them to maximize recoverable object information is unknown — conventional practice shortens the gate to reject multiple scattering, which may also discard useful signal. Third, the forward model includes only forward scattering between plates; backscattering generated inside the medium (non-forward contributions and weak internal reflections) is neglected. Fourth, the matrix-multiplication cost grows as O(N06)O(N_0^6) in the number of sampled input positions per side, so large regions of interest are impractical (GPU acceleration and patchwise processing mitigate but do not remove this). Fifth, the image-based enhancement metric divides the confocal enhancement by a parameter α\alpha, the ballistic fraction in the diagonal of R, whose estimation becomes ambiguous when the initial ballistic signal is extremely weak — precisely the deep regime the method targets. Two successor works from the same group set the current baselines and sharpen the open frontier. The angular-path framework of Gupta et al. (arXiv:2509.11491, 2025) replaces the discrete phase-plate layers with a decomposition of the scattered waves into angular-deflection components, each associated with a spatially invariant point spread function, and retrieves the dominant components progressively from reflection measurements; it converted multiply scattered light into usable signal about eight times the ballistic component and recovered subcellular osteocyte networks in vivo through intact mouse skulls. The same work shows experimentally that the layer-based MST fails in volumetric media where scattering is continuously distributed near the object plane (dense phase plates over-parameterize and converge unstably), and its authors state that the layer-based and scattering-angle models are complementary — in the long term, a synergistic combination of both approaches could capture a broader range of multiple-scattering events — with the achievable imaging depth fundamentally dictated by the extent to which multiple scattering can be inverted. Separately, Hoang et al. (arXiv:2607.05925, 2026) show that light backscattered by tissue structures beneath a cell can serve as rear illumination for reflection-only refractive-index tomography — a different objective that nevertheless demonstrates that internal backscattering carries recoverable information. Adjacent lines clarify what would and would not answer the question: the Langevin group's reflection-matrix methods (Najar et al. 2024; Balondrade et al. 2024) compensate forward multiple scattering locally but not the round-trip high-order paths MST addresses; the Katz group's acousto-optically gated and virtual reflection matrices (Sunray et al. 2024; Weinberg et al. 2024; Sunray et al. 2025) extend matrix imaging beyond the memory effect, to incoherent fluorescence, and to dynamic media; and reflection-mode optical diffraction tomography (Wasik et al. 2026) exploits background multiply scattered waves for refractive-index reconstruction. None of these inverts the object-bearing round-trip multiple scattering of an embedded object in the generality MST aims at.

Problem Statement

Extend high-order inverse scattering for reflection-mode deep optical imaging so that the fraction of object-bearing multiply scattered light that is computationally traceable and convertible into usable imaging signal is substantially increased beyond the layer-based MST demonstration of Kang et al. (2023) — which traced 17 scattering events with a stack of phase plates identified from a time-gated reflection matrix — in the general geometry the source poses: an object embedded in a thick scattering medium, with only backscattered round-trip waves accessible. Per the source's own formulation, the degree of multiple-scattering coverage is set by coupled factors: (i) the parameterization of the inverse problem, so that continuously distributed volumetric scattering, including near the object plane, is captured without unstable over-parameterization; (ii) the imaging/collection geometry and the temporal gating window, chosen to maximize the object information recoverable from the recorded multiple scattering rather than merely to reject it; (iii) the forward model, extended to incorporate backscattering generated inside the medium (non-forward-scattering contributions and internal reflections) rather than forward scattering between plates only; and (iv) the computational treatment, so that the matrix-multiply cost (scaling as O(N06)O(N_0^6) in the number of sampled input positions per side) and memory footprint do not restrict the traceable region-of-interest size or layer count. The objective is to maximize how much object information is recovered from multiply scattered light at depths and in media where ballistic-only imaging and the 2023/2025 baseline methods fail.

References

  1. Kang, S.; Kwon, Y.; Lee, H.; Kim, S.; Hong, J. H.; Yoon, S.; Choi, W. 'Tracing multiple scattering trajectories for deep optical imaging in scattering media.' Nature Communications 14, 6871 (2023). DOI: 10.1038/s41467-023-42525-7, https://doi.org/10.1038/s41467-023-42525-7 (preprint: arXiv:2302.09503, https://arxiv.org/abs/2302.09503)
  2. Kang, S.; Yoon, S.; Choi, W. 'Implementation of reflection matrix microscopy: an algorithm perspective.' Journal of Physics: Photonics 7, 023002 (2025). DOI: 10.1088/2515-7647/adb7af, https://doi.org/10.1088/2515-7647/adb7af (preprint: arXiv:2407.02321, https://arxiv.org/abs/2407.02321)
  3. Gupta, K.; Tran, D. H.; Kang, S.; Kwon, Y.; Yoon, S.; Hong, J. H.; Lee, Y.-R.; Choi, W. 'Decoding angular light paths for solving the inverse scattering problem.' arXiv:2509.11491 (2025), https://arxiv.org/abs/2509.11491
  4. Hoang, T. D.; Cho, J.; Nguyen, T. V. A.; Seong, E.; Lim, J.; Hong, J. H.; Kwon, Y.; Kim, J. W.; Yang, J.; Yoon, S.; Kang, S.; Choi, W. 'Refractive-index tomography of opaque tissue from its own backscattered light.' arXiv:2607.05925 (2026), https://arxiv.org/abs/2607.05925
  5. Najar, U.; Barolle, V.; Balondrade, P.; Fink, M.; Boccara, A. C.; Aubry, A. 'Harnessing forward multiple scattering for optical imaging deep inside an opaque medium.' Nature Communications 15, 7349 (2024). DOI: 10.1038/s41467-024-51619-9, https://doi.org/10.1038/s41467-024-51619-9 (preprint: arXiv:2303.06119, https://arxiv.org/abs/2303.06119)
  6. Wasik, T.; Barolle, V.; Aubry, A.; Garnier, J. 'Taking Advantage of Multiple Scattering for Optical Reflection Tomography.' IEEE Transactions on Computational Imaging 12, 533–547 (2026). DOI: 10.1109/TCI.2025.3644225, https://doi.org/10.1109/TCI.2025.3644225 (preprint: arXiv:2506.04232, https://arxiv.org/abs/2506.04232)
  7. Balondrade, P.; Barolle, V.; Guigui, N.; Auriant, E.; Rougier, N.; Boccara, C.; Fink, M.; Aubry, A. 'Multi-spectral reflection matrix for ultrafast 3D label-free microscopy.' Nature Photonics 18, 1097–1104 (2024). DOI: 10.1038/s41566-024-01479-y, https://doi.org/10.1038/s41566-024-01479-y
  8. Giraudat, E.; Barolle, V.; Bureau, F.; Guigui, N.; Balondrade, P.; Ho, C.; Brochard, V.; Dubois, O.; Bonnet-Garnier, A.; Aubry, A. 'Label-free subcellular 3D imaging of oocytes and embryos via reflection matrix microscopy.' arXiv:2410.11126 (2024), https://arxiv.org/abs/2410.11126
  9. Sunray, E.; Weinberg, G.; Rosenfeld, M.; Katz, O. 'Beyond memory-effect matrix-based imaging in scattering media by acousto-optic gating.' APL Photonics 9, 096112 (2024). DOI: 10.1063/5.0219316, https://doi.org/10.1063/5.0219316
  10. Weinberg, G.; Sunray, E.; Katz, O. 'Noninvasive megapixel fluorescence microscopy through scattering layers by a virtual incoherent reflection matrix.' Science Advances 10, eadl5218 (2024). DOI: 10.1126/sciadv.adl5218, https://doi.org/10.1126/sciadv.adl5218
  11. Sunray, E.; Weinberg, G.; Laufer, B.; Katz, O. 'Matrix-based imaging through dynamic scattering.' Nature Communications 16, 9413 (2025). DOI: 10.1038/s41467-025-64422-x, https://doi.org/10.1038/s41467-025-64422-x
  12. Aubry, A. 'A digital solution towards deep optical imaging.' Nature Photonics 19, 3–4 (2025). DOI: 10.1038/s41566-024-01594-w, https://doi.org/10.1038/s41566-024-01594-w