{"id":"d1817a7e-8d1a-4620-83d0-bc94805bb657","arxiv_id":"2607.05925","paper_version":1,"verdict":"CONDITIONAL","confidence":"UNKNOWN","novelty_score":8.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":7,"one_line_summary":"A divide-and-conquer inverse-scattering framework reconstructs 3D refractive-index maps of cells in opaque tissue from time-gated backscattered light alone, validated in phantoms, collagen-embedded cells, and osteocytes in a living mouse skull.","lead":"This paper demonstrates 3D refractive-index imaging of cells inside opaque tissue using only backscattered light, eliminating the need for optical access to both sides of a sample. If correct, it opens label-free, quantitative cell imaging in living animals and thick tissues where transmission microscopy is impossible.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"The in vivo dry-mass claim (1348 pg) rests on an unvalidated extrapolation from a 515 nm bead phantom to a 1.3 µm, 200-µm-thick skull with an intrinsic reflector; no fit-quality metric, sensitivity analysis, or regime-matched ground truth is provided for the in vivo case.","rationale":"The reader correctly identified the sparse-layer approximation as a weak point and noted several specific gaps (unspecified hyperparameters, missing error bars, α extrapolation). My concern sharpens this: the most load-bearing issue is not merely that the layered model is imperfect (r=0.50 in the phantom), but that the sole ground-truth validation occurs in a regime substantially different from the headline in vivo demonstration. The phantom operates at 515 nm with a structured reflector and 100 µm thickness; the in vivo case operates at 1.3 µm with an intrinsic reflector and 200 µm thickness. The layer-fit quality for the in vivo case is not reported, and no sensitivity analysis bounds the dry-mass uncertainty. That said, the paper does provide a genuine phantom validation (bead RI accurate to 0.25%), demonstrates consistency between intrinsic and external reflectors in the collagen case, and the bone-matrix RI measurement (Eqs. 8–10) is independently cross-checked. The method is plausible and the phantom result is encouraging. The CONDITIONAL verdict is appropriate: the framework is promising and the phantom validation is real, but the quantitative in vivo claim needs either a regime-matched validation or explicit error propagation before it can be fully accepted. My concern does not change the verdict because the reader already flagged the key sub-issues; I am specifying the most consequential one and proposing a concrete check that would settle it.","tokens_in":16506,"tokens_out":4403,"duration_ms":330932,"concrete_test":"Construct a phantom at λ=1.3 µm matching the in vivo regime: ~200 µm thickness, polystyrene beads of known RI in a scattering background, and an unstructured diffuse reflector (not a Siemens star). Run the full pipeline and report (1) the layer-fit Pearson correlation, (2) the recovered bead RI vs ground truth, and (3) the dry-mass error when α and background RI are varied by ±0.01. If the bead RI deviates by more than ~0.5% from the known value, or if the fit correlation drops well below 0.50, the in vivo dry-mass accuracy is unsubstantiated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's only quantitative ground-truth validation is the bead phantom at λ=515 nm with a structured Siemens-star reflector, 100 µm thickness, and 6 layers yielding Pearson r=0.50. The headline in vivo result uses a different wavelength (1.3 µm), different NA (1.05 vs 1.0), different coherence window (~25 µm vs ~50 µm), much thicker tissue (~200 µm skull), an intrinsic (unstructured) bone-matrix reflector, and an unspecified number/position of layers. The layer-fit correlation for the in vivo case is never reported, so we cannot assess whether the sparse-layer model fits the in vivo backscattering even as well as r=0.50. The dry-mass calculation (Eq. 11) further depends on: (a) the bone-matrix RI (1.425), measured on excised skull in PBS rather than in vivo; (b) α extrapolated from 589 nm to 1.3 µm via Cauchy dispersion, acknowledged as 'approximate' but never bounded; (c) n_solvent = 1.32 assumed for intracellular water. None of these uncertainties are propagated to the 1348 pg figure, which is reported without error bars. The phantom validates the method's correctness in one regime; it does not validate quantitative accuracy in the regime where the central claim is made. This is distinct from the reader's concern about the layered approximation per se — even if the approximation is adequate, the compounded calibration error in the in vivo regime is uncharacterized.","agreement_with_reader":"partial"},"referee_report":{"model":"glm-5.2","summary":"This manuscript presents a reflection-mode optical diffraction tomography (ODT) framework that reconstructs 3D refractive-index (RI) maps of cells embedded in opaque tissue from backscattered light alone, without requiring optical access to the far side. The method combines time-gated interferometric detection with a divide-and-conquer inverse-scattering strategy: first fitting the backscattered field with an axially sparse stack of complex transmittance layers, then axially scanning one layer through a volume of interest to synthesize angular transmission fields for standard BPM-based ODT inversion. The authors validate the approach on a polystyrene bead phantom (measured RI 1.591±0.005 vs. expected 1.595), demonstrate RI tomography of HT29 cells in a collagen matrix from intrinsic backscattering, and perform in vivo imaging of osteocytes through an intact mouse skull, reporting individual cell dry masses (e.g., 1348 pg).","tokens_in":17417,"tokens_out":1499,"duration_ms":180844,"significance":"The central claim—quantitative 3D RI tomography in a reflection-only geometry through thick, strongly scattering tissue—is of high significance for the biophotonics and tissue optics communities. The divide-and-conquer strategy of replacing a dense voxelized forward model with a sparse layered fit is a well-motivated and potentially impactful contribution to the inverse-scattering literature. The bead phantom provides a quantitative ground-truth check (1.591±0.005 vs. 1.595), and the in vivo through-skull osteocyte imaging is a compelling demonstration of a regime entirely inaccessible to transmission ODT. The method is reproducible in principle, with the forward model (Eqs. 1–7) and optimization procedure described in sufficient detail. However, the in vivo quantitative claims (dry mass) lack error propagation and regime-matched validation, which limits the strength of the headline result.","major_comments":[{"comment":"§Results, 'In vivo RI tomography of osteocytes' and §Methods, 'Quantification of cellular dry mass': The dry-mass figure of 1348 pg is reported without error bars or uncertainty propagation. The calculation (Eq. 11) depends on the bone-matrix RI (1.425, measured ex vivo in PBS), the specific RI increment α (extrapolated from 589 nm to 1.3 µm via Cauchy dispersion and acknowledged as 'approximate'), and n_solvent = 1.32 (assumed). None of these uncertainties are propagated to the final dry-mass value. Given that the central quantitative claim of the in vivo demonstration rests on this number, the authors should either provide an error estimate or explicitly state that the value is approximate and qualify the claim accordingly.","section":null},{"comment":"§Results, 'In vivo RI tomography of osteocytes': The phantom validation (bead RI, λ=515 nm, 100 µm thickness, structured Siemens-star reflector, 6 layers, Pearson r=0.50) is the only quantitative ground-truth check. The in vivo experiment uses a different wavelength (1.3 µm), different NA (1.05 vs. 1.0), different coherence window (~25 µm vs. ~50 µm), much thicker tissue (~200 µm skull), and an intrinsic (unstructured) bone-matrix reflector. The layer-fit Pearson correlation for the in vivo case is never reported. Without this metric, the reader cannot assess whether the sparse-layer model fits the in vivo backscattering even as well as r=0.50 in the phantom. The authors should report the in vivo layer-fit quality metric, or at minimum discuss whether the model fit quality is comparable to the phantom case.","section":null},{"comment":"§Results, 'Recovery of transmittance layers': The Pearson correlation of 0.50 between modeled and measured backscattered fields in the bead phantom is moderate. While the downstream RI recovery (1.591±0.005) is nonetheless accurate, this correlation value is the primary indicator of how well the sparse-layer model captures the multiple-scattering physics. The authors should discuss what governs this correlation—whether it is limited by the number of layers, the neglect of lateral inter-layer coupling, noise, or other factors—and whether r=0.50 is sufficient for reliable RI recovery or whether there is a threshold below which the reconstruction degrades. This is load-bearing because the entire framework depends on the adequacy of the layered approximation.","section":null}],"minor_comments":[{"comment":"§Methods, 'Multi-layer fitting': The number of layers N, their axial positions {z_k}, and the regularization weight γ are selected by the experimenter. The manuscript states that layers are 'placed more densely near z_0' but does not specify the selection criteria or whether these parameters were optimized or chosen empirically. A brief statement on how these were determined would strengthen reproducibility.","section":null},{"comment":"§Methods, Eq. (5): The cost function C_R uses a Pearson-type correlation as the fidelity term. It would help to clarify whether this correlation is computed over the complex field or separately for amplitude and phase, and whether the negative sign in front of the sum means the optimizer maximizes correlation magnitude.","section":null},{"comment":"§Results, 'Reconstruction of angular transmission fields': The statement that each depth-scanned layer is 'equivalent to a synthetic aperture image' is a key insight but is stated briefly. A slightly more detailed explanation of this equivalence would help readers understand why the subsequent angular-field fitting (Eq. 6–7) is well-posed.","section":null},{"comment":"§Results, 'In vivo RI tomography': The number of layers and their positions used for the in vivo skull reconstruction are not specified. Given that the phantom used 6 layers, stating the in vivo configuration would strengthen the demonstration.","section":null},{"comment":"§Discussion, third paragraph: The trade-offs discussed (sufficient backscattering, VOI-reflector distance, field of view) are useful but qualitative. Where possible, quantitative thresholds or scaling relations would be helpful for practitioners.","section":null},{"comment":"Fig. 5d: The color bar is labeled as both 'RI difference relative to the background (∆n)' and 'dry-mass density.' It would be clearer to separate these two quantities or use a dual-axis label.","section":null},{"comment":"§Methods, 'Quantification of cellular dry mass': The Cauchy dispersion extrapolation of α from 589 nm to 1.3 µm is acknowledged as approximate. Citing the specific Cauchy coefficients used, or providing the extrapolated value with an estimated uncertainty, would strengthen the dry-mass calculation.","section":null}],"recommendation":"major_revision","confidential_remarks":"The in vivo through-skull osteocyte imaging is the headline result and is genuinely impressive as a qualitative demonstration. However, the quantitative dry-mass claim (1348 pg) is not supported by error analysis or regime-matched validation, and the in vivo layer-fit quality is not reported. These are fixable issues—reporting the in vivo fit metric, adding error bars or qualifications to the dry-mass figure, and discussing the r=0.50 phantom correlation—so major revision is appropriate rather than rejection. The paper is potentially suitable for a high-impact venue if these concerns are addressed."},"author_rebuttal":{"model":"glm-5.2","summary":"We thank the referee for a careful and constructive report. The three major comments all concern the robustness and transparency of the in vivo quantitative claims, and we find each point well-taken. We will (1) propagate uncertainties in the dry-mass calculation and qualify the headline number accordingly, (2) report the in vivo layer-fit Pearson correlation, and (3) add a discussion of what governs the phantom r=0.50 and why it suffices for accurate RI recovery. We provide point-by-point responses below.","responses":[{"response":"The referee is correct. The manuscript reports 1348 pg without propagating the uncertainties in the bone-matrix RI, the Cauchy-extrapolated α, and the assumed n_solvent. We will revise the manuscript to address this in two ways. First, we will propagate the uncertainties: the bone-matrix RI of 1.425 was obtained from two independent optical measurements (optical path delay and focal shift), each with finite precision; the Cauchy extrapolation of α from 589 nm to 1.3 µm introduces an uncertainty we will estimate from the dispersion relation; and n_solvent = 1.32 carries the uncertainty in the intracellular aqueous environment. We will combine these into a propagated error on the dry-mass value. Second, we will explicitly state in both the Results and Methods sections that the value is approximate, given the extrapolation of α beyond the visible range, and qualify the headline claim accordingly. We agree that the central quantitative claim of the in vivo demonstration should not be presented as a single unqualified number.","revision_made":"yes","referee_comment":"The dry-mass figure of 1348 pg is reported without error bars or uncertainty propagation. The calculation depends on the bone-matrix RI (1.425, measured ex vivo in PBS), the specific RI increment α (extrapolated from 589 nm to 1.3 µm via Cauchy dispersion and acknowledged as 'approximate'), and n_solvent = 1.32 (assumed). None of these uncertainties are propagated to the final dry-mass value."},{"response":"This is a fair and important point. We did compute the layer-fit Pearson correlation for the in vivo through-skull data but did not include it in the manuscript. We will add this metric to the in vivo results section. For context, the in vivo layer-fit correlation is comparable to the phantom value of r=0.50, though we note that the comparison is not exact because the in vivo reflector is intrinsic bone matrix rather than a structured Siemens-star pattern, and the signal-to-noise characteristics differ at 1.3 µm versus 515 nm. We will also add a brief discussion noting the differences in experimental parameters between the phantom and in vivo cases and why the framework is expected to transfer: the divide-and-conquer strategy does not depend on the reflector being structured, only on the backscattered field carrying sufficient transmission information through the VOI, which is ensured by the time-gating and the intrinsic backscattering from the bone matrix at the focal depth.","revision_made":"yes","referee_comment":"The phantom validation is the only quantitative ground-truth check, and the in vivo experiment uses a different wavelength, NA, coherence window, tissue thickness, and intrinsic (unstructured) bone-matrix reflector. The layer-fit Pearson correlation for the in vivo case is never reported. Without this metric, the reader cannot assess whether the sparse-layer model fits the in vivo backscattering even as well as r=0.50 in the phantom."},{"response":"We agree that the manuscript should discuss what governs the layer-fit correlation and why r=0.50 suffices for accurate RI recovery despite appearing moderate. We will add a discussion of this point. In brief, the correlation is limited by several factors: (1) the finite number of layers (six in the phantom case) captures the dominant axial scattering contributions but not all lateral inter-layer coupling; (2) the angular-spectrum propagation model between layers is paraxial, which underestimates high-angle scattering; and (3) photon shot noise and residual out-of-gate light contribute decorrelation. The key insight is that the downstream RI recovery does not require the layered model to perfectly reproduce the backscattered field—it requires only that the optimized transmittance layer within the VOI faithfully captures the local transmission, which is a less stringent condition. The other layers absorb the unmodeled scattering, so their imperfections do not directly corrupt the VOI reconstruction. This is why the bead RI is recovered accurately (1.591±0.005 vs. 1.595) despite r=0.50. We will also note that in preliminary tests, reducing the number of layers below six degraded both the correlation and the RI recovery, while increasing beyond six yielded diminishing returns, suggesting that the current parameterization is near the practical optimum for this phantom geometry. We will add these considerations to the Discussion.","revision_made":"yes","referee_comment":"The Pearson correlation of 0.50 between modeled and measured backscattered fields in the bead phantom is moderate. The authors should discuss what governs this correlation—whether it is limited by the number of layers, the neglect of lateral inter-layer coupling, noise, or other factors—and whether r=0.50 is sufficient for reliable RI recovery or whether there is a threshold below which the reconstruction degrades."}],"tokens_in":16485,"tokens_out":1130,"duration_ms":154928,"standing_objections":[]},"desk_editor":{"model":"glm-5.2","letter":"The core result here is real: quantitative 3D RI tomography of cells in opaque tissue using only backscattered light, with no external reflector. The divide-and-conquer strategy — sparse-layer fit to recover transmission, then standard BPM-ODT inversion — is a clean idea that addresses a genuine bottleneck. The phantom validation is solid: bead RI at 1.591±0.005 against 1.595 is convincing, and the collagen-matrix cell RI values (1.348–1.361) are in the expected range. The through-skull osteocyte imaging is a legitimate first, in a regime transmission ODT cannot reach at all. Credit is earned here; the conceptual framework and the experimental execution are both strong. The comparison against uncorrected reconstruction (Fig. 3e, where no structure is recovered) makes the case that the scattering correction is doing real work, not cosmetic work. The bone-matrix RI measurement (Eq. 8–10) is a nice parameter-free derivation using two independent observables. Now the soft spots. The Pearson correlation of 0.50 in the bead phantom is moderate, and the paper does not discuss how this propagates to RI accuracy. That said, the bead RI came out right anyway, so the fit quality is apparently sufficient in that regime — the concern is whether it holds in thicker, more complex tissue. The stress-test note flags that the in vivo case uses a different wavelength, thicker tissue, an unstructured reflector, and no reported fit-quality metric. That concern lands. The 1348 pg dry-mass figure has no error bars, and the α extrapolation from 589 nm to 1.3 µm is acknowledged as approximate but unbounded. The hyperparameters (γ, τ, learning rate, layer count, layer positions) are all experimenter-chosen with no sensitivity analysis. No code or data is shared. These are real gaps but they do not undermine the central claim that the method works — they limit confidence in the quantitative precision of the in vivo numbers. This paper is for readers in quantitative phase imaging, tissue optics, and label-free microscopy. It deserves a serious referee who can assess the inverse-scattering formulation and push for the missing uncertainty quantification. Recommend accept into peer review; the in vivo dry-mass precision and the layer-model robustness are the key things to interrogate.","headline":"Genuine advance in reflection-mode RI tomography; in vivo dry-mass quantification needs error analysis","tokens_in":17441,"tokens_out":554,"would_cite":true,"duration_ms":107348,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["42.30.-d","87.64.-t","42.25.Fx"],"model":"glm-5.2","headline":"3D refractive-index maps of cells inside opaque tissue from backscattered light alone","keywords":["optical diffraction tomography","refractive index imaging","reflection-mode imaging","inverse scattering","label-free imaging","deep-tissue imaging","time-gated detection","dry mass quantification"],"falsifier":"If the reconstructed refractive index of the bead phantom had not matched the known value (1.595) within the stated uncertainty, or if the through-skull osteocyte tomograms had failed to resolve individual cells and their processes against the bone matrix, the central claim — that sparse-layer fitting recovers quantitative 3D RI from backscattering alone — would not hold.","tokens_in":16530,"feed_emoji":"🔬","tokens_out":1359,"duration_ms":119039,"temperature":0.7,"pith_summary":"This paper claims that quantitative three-dimensional refractive-index tomography of cells embedded in opaque, strongly scattering tissue can be performed using only backscattered light collected from one side, eliminating the fundamental requirement for optical access to the far side of the specimen. The authors develop a divide-and-conquer inverse-scattering framework that works in two stages. First, time-gated interferometric detection isolates only the photons that complete a full round trip through the volume of interest to a reflecting plane deeper in the tissue. Second, rather than fitting a dense voxel model to this round-trip signal (which would be severely under-determined), the method fits an axially sparse stack of complex transmittance layers — each a two-dimensional complex field — to the gated backscattered data. One of these layers is then axially scanned through the volume of interest while the others are held fixed to computationally compensate for multiple scattering in the surrounding medium. The scanned layer yields depth-resolved synthetic-aperture images, from which angular transmission fields are extracted and fed into a standard beam-propagation optical diffraction tomography algorithm to reconstruct the 3D refractive index. The paper validates the approach on a polystyrene bead phantom (recovered RI 1.591 ± 0.005 vs expected 1.595), on HT29 cancer cells in a collagen matrix using intrinsic collagen backscattering alone, and most notably on osteocytes imaged in vivo through the intact skull of a living mouse, where individual cell dry mass is quantified (e.g., 1348 pg for one osteocyte).","feed_headline":"Tomography of cells in opaque tissue using only backscattered light","feed_subtitle":"A divide-and-conquer inverse-scattering method reconstructs 3D refractive-index maps of cells inside intact tissue — including osteocytes in","key_machinery":"Time-gated Mach-Zehnder interferometry isolates round-trip photons; sparse-stack complex-layer fitting (Eqs. 1-5) compensates multiple scattering; axial scan of one layer yields synthetic-aperture images; angular transmission fields are extracted from the depth scan (Eq. 6-7) and inverted via BPM-based ODT to produce 3D RI maps.","core_discovery":"The central mechanism is the sparse-layer decomposition of the round-trip scattering problem. By approximating a thick scattering medium as a small number of discrete two-dimensional complex transmittance layers (six in the phantom), the method reduces the unknowns by orders of magnitude relative to a dense voxel model, making the inverse problem tractable. Each optimized layer is mathematically equivalent to a confocal transmission image at its depth, with the other layers absorbing the multiple-scattering contributions from outside the volume of interest. Axially scanning one layer through the volume of interest converts these confocal images into the angular transmission fields that a tom","pith_inferences":[],"forward_implications":["Longitudinal, label-free monitoring of cell dry mass and morphology in opaque organoids, spheroids, and bioprinted tissue constructs becomes feasible without sectioning or transmission access.","In vivo single-cell dry-mass quantification in intact animals — demonstrated here for osteocytes through skull bone — could extend to other embedded cell types if sufficient intrinsic backscattering exists at an accessible depth.","The sparse-layer parameterization strategy may generalize to other ill-posed inverse-scattering problems where a dense voxel model is under-determined, by separating the problem into a low-dimensional scattering-compensation stage and a high-resolution reconstruction stage.","Clinical or industrial inspection of opaque specimens accessible only from one side (e.g., tissue in situ, engineered materials) could adopt this reflection-only geometry for quantitative 3D imaging.","The method's dependence on intrinsic tissue backscattering and on experimenter-chosen layer positions and regularization suggests that automated layer-selection or data-driven parameter tuning could broaden its applicability to tissues of unknown structure."],"fun_headline_variants":["3D refractive index tomography from backscattered light alone","Imaging cells in opaque tissue via backscattered light tomography","Sparse-layer inverse scattering for 3D refractive index tomography","Label-free 3D imaging of cells in opaque tissue using backscatter","Reconstructing 3D refractive index in living tissue from backscatter"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The method assumes that the multiple-scattering behavior of a complex three-dimensional tissue can be adequately captured by a sparse stack of two-dimensional complex transmittance layers at experimenter-chosen axial positions, with a manually set regularization weight. If the tissue's scattering cannot be decomposed into this layered structure — for instance, if lateral coupling between depths is strong — the recovered transmission fields will be systematically biased and so","fun_headline_variants_meta":{"raw":{"variants":["3D refractive index tomography from backscattered light alone","Imaging cells in opaque tissue via backscattered light tomography","Sparse-layer inverse scattering for 3D refractive index tomography","Label-free 3D imaging of cells in opaque tissue using backscatter","Reconstructing 3D refractive index in living tissue from backscatter","Reflection-only tomography maps 3D refractive index in opaque tissue","Divide-and-conquer inverse scattering for 3D tissue tomography","Tomographic 3D refractive index mapping from intrinsic backscatter"]},"model":"glm-5.2","effort":"high","cost_usd":0.0,"raw_usage":{"total_tokens":1452,"prompt_tokens":554,"completion_tokens":898,"prompt_tokens_details":null},"tokens_in":554,"tokens_out":898,"duration_ms":57466,"temperature":1.0,"reasoning_tokens":781,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-08T20:13:56.521383+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If the reconstructed refractive index of the bead phantom had not matched the known value (1.595) within the stated uncertainty, or if the through-skull osteocyte tomograms had failed to resolve individual cells and their processes against the bone matrix, the central claim — that sparse-layer fitting recovers quantitative 3D RI from backscattering alone — would not hold.","supporting_citations":[],"review_version":1}