{"id":"1950b033-ab59-4827-ba53-0e3c219e01f4","arxiv_id":"2607.29478","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Phase-induced resolution enhancement can occur in ordinary micro-CT without visible fringes when the detector blur matches the required Paganin filtering.","lead":"Conventional micro-CT images can contain hidden phase-contrast sharpening even when no fringes are visible, and in some geometries the detector itself performs phase retrieval during acquisition. The authors provide a framework to detect and exploit this effect, with experimental demonstrations on custom and commercial scanners.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (6)'s Gaussian-variance cancellation is invalid for the measured Lorentzian detector PSFs; if Lorentzian tails don't cancel the propagation Laplacian, the quantitative Regime-B resolution predictions are unsupported, though the dithering experiment may still stand.","rationale":"The paper has independent support: the Nikon dithering experiment directly demonstrates resolution beyond the conventional blur prediction, and the reproduced periodic texture in simulation rules out some detector artifacts. However, the analytical engine for the regime classification and quantitative resolution predictions is the Gaussian variance-subtraction of Eq. (6), which the paper itself concedes is not strictly applicable to the measured mixed Gaussian-Lorentzian detector PSFs. This does not by itself overturn the central qualitative claim, but it justifies the reader's CONDITIONAL verdict: the quantitative predictions and DIPR optimization metrics need to be re-derived or validated against the measured PSF. If the proposed delta=0 test confirms the paper's claim, the verdict need not change; if it fails, the central claim would be seriously weakened. I agree with the reader's identification of the weakest assumption, and the appropriate disposition remains CONDITIONAL pending that check.","tokens_in":19564,"tokens_out":19013,"duration_ms":204788,"concrete_test":"Using the paper's 1D wave-optics simulator, compute the reconstructed PTFE edge LSF for the Nikon dithering geometry (R1=130 mm, M=7, measured mixed Gaussian-Lorentzian PSF) in three cases: (i) full phase model (delta/beta for PTFE); (ii) phase-disabled (delta=0); (iii) a pure Gaussian PSF with FWHM matched to the measured mixture, phase on. If case (ii) yields FWHM <=28.5 um, the claimed phase-induced resolution enhancement is an artifact of the non-Gaussian PSF and the conventional-blur baseline. If case (iii) differs from case (i) by >20%, Eq. (6)'s Gaussian variance subtraction is not a quantitatively reliable predictor for the actual PSFs.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The framework's central quantitative step is Eq. (6), sigma_eff^2 = sigma_s_eff^2 + sigma_det_eff^2 - 2a^2, which treats propagation-induced sharpening as a negative Gaussian variance. This follows from multiplying the forward propagator T_prop(f)=1+a^2(2pi f)^2 by Gaussian MTFs in Eq. (5). The measured detector PSFs are not Gaussian: Methods states the custom detector is 67.9% Lorentzian (HWHM 50 um) + 32.1% Gaussian (sigma=130 um), and the Nikon detector is 40% Lorentzian (HWHM 120 um) + 60% Gaussian (sigma=286 um). A Lorentzian MTF decays as exp(-2pi gamma |f|) and has an infinite second moment, so the quadratic variance subtraction underlying Eq. (6) is undefined; the authors themselves concede in Methods that the Gaussian sigma is 'strictly, no longer applicable.' Yet Eq. (6) is used to predict the Regime A/B transition and the resolution improvement (e.g., 100 um -> 54.4 um in the custom system, and the 'phase-induced reduction in effective system width predicted by Eq. (6)' in Results). If the Laplacian does not cancel Lorentzian tails the way it cancels Gaussian variance, the magnitude of the claimed enhancement and the location of the DIPR optimum could be substantially different. The strongest experimental evidence -- the Nikon dithering result (28.5 um vs 55 um conventional blur) -- does not directly test Eq. (6), because it is sampling-limited; it shows the effective system response is <28.5 um, but a sharp Lorentzian core could contribute to this even without phase effects. The paper's Supp. Fig. 7(b) delta=0 simulation is therefore the key check of whether the observed sharpening truly requires phase contrast.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper argues that propagation-based phase contrast is present and functionally important in many micro-CT systems even when Fresnel fringes are not visible. The authors classify imaging into an absorption-only Regime A and a phase-transfer Regime B, introduce hardware-induced phase retrieval (HIPR) categories including detector-induced phase retrieval (DIPR), and derive Gaussian-based analytical conditions, frequency-domain metrics nφ and Ematch, and an effective-width formula. They support the framework with simulations, custom-system scans over varying geometry, and a commercial Nikon scanner with half-pixel dithering, reporting spatial resolution finer than the conventional source/detector blur prediction.","tokens_in":19996,"tokens_out":8128,"duration_ms":88304,"significance":"If established, the central claim would be important: it would change the standard interpretation of image formation and spatial resolution in laboratory micro-CT, where 'no visible fringes' is often equated with attenuation-only imaging. The experimental work contains genuinely strong elements: the Nikon dithering experiment shows measured resolution (28.5 μm) below the conventional blur prediction (55.0 μm), with δ=0 simulations indicating the effect is phase-related; the PTFE fringe-amplitude measurements across geometries agree with simulations. However, the quantitative theory currently rests on a Gaussian variance-subtraction step that the authors themselves concede is not applicable to their measured mixed Gaussian-Lorentzian detector PSFs. The load-bearing quantitative predictions therefore need to be re-derived, replaced, or carefully bounded before the central claims can be accepted at the stated strength.","major_comments":[{"comment":"Equation (6), σ_eff² = σ_s_eff² + σ_det_eff² − 2a², is obtained by combining the quadratic propagation transfer T_prop(f)=1+a²(2πf)² with Gaussian MTFs. Even for perfectly Gaussian PSFs this is a small-frequency expansion, not an exact identity. More importantly, it is undefined for the measured detector PSFs: Methods reports the custom detector as 67.9% Lorentzian (50 μm HWHM) plus 32.1% Gaussian, and the Nikon detector as 40% Lorentzian (120 μm HWHM) plus 60% Gaussian; a Lorentzian MTF has an infinite second moment, so variance subtraction has no well-defined meaning. The paper explicitly concedes in Methods that Gaussian σ is 'strictly, no longer applicable' in this case. Yet Eq. (6) is used to predict the Regime A/B transition, the intrinsic DIPR resolutions (5.9 μm custom, 14.2 μm Nikon), and the phase-induced reduction in effective system width. Please replace Eq. (6) with a calcul","section":"Methods, Eq. (6); Results (intrinsic resolutions)"},{"comment":"DIPR is defined as the condition H_det ≈ H_Pag, and Ematch is defined as the RMS difference between H_det and H_Pag over the usable frequency band. Finding that Ematch has a minimum, and calling that geometry the DIPR optimum, is therefore partly a restatement of the metric's definition. The independent experimental evidence (especially the dithering scan) is valuable, but the theoretical 'optimality' claim needs a separate argument that the minimum of this particular RMS metric is the optimum operating point in a resolution/SNR tradeoff sense. This is especially important because H_Pag contains the sample-dependent propagation parameter a², and the integration band f1..fmax is defined system-by-system.","section":"Methods, Eq. (14); Results, Fig. 1(c)"},{"comment":"Several of the measured resolutions coincide exactly with the sampling limit, so they are upper bounds rather than direct measurements of the intrinsic phase-enhanced width. Scan 3's 'improvement from 100 μm to 54.4 μm FWHM' equals the two-pixel sample-plane scale at that geometry (27.2 μm pixel), and Scan II's 28.5 μm is exactly the dithered sampling limit at 14.3 μm pixels. The comparison with the conventional blur prediction (99.9 μm and 55.0 μm, respectively) is still informative, but the text should state that the measured LSF is sampling-limited at ≤54.4/≤28.5 μm, not that the intrinsic DIPR resolution was directly measured at those values. Please report fitted LSF FWHM values with uncertainties and show that the edge profiles are not truncated by the sampling grid.","section":"Results, Scan 3 and Scan II; Table I"}],"minor_comments":[{"comment":"Equation (8) appears to contain a typo: the detector contribution is written as a factor (−2π²|f|²(...)) rather than as an exponential. It should presumably be exp(−2π²|f|²(1/M)²σ_det²).","section":"Methods, Eq. (8)"},{"comment":"Typo: 'freqiencies' should be 'frequencies'.","section":"Results, first subsection"},{"comment":"The forward model in Eq. (2) uses the transport-of-intensity transfer function T_prop(f)=1+a²(2π|f|)², while the phase-transfer metric in Eq. (11) uses sin(πλR′f²), a different Fresnel weak-phase form. Please clarify which model is used for which prediction and confirm that the difference does not affect the comparison with the Fresnel-Kirchhoff simulations.","section":"Methods, Eq. (11) vs Eq. (2)"},{"comment":"The word 'universal' in the title/abstract is stronger than what two systems can establish. Consider tempering to 'common' or 'prevalent' unless a broader survey is provided.","section":"Abstract and Discussion"}],"recommendation":"major_revision","confidential_remarks":"The paper is worth pursuing. The qualitative claim is plausible and the dithering experiment provides direct evidence that should be preserved. The main obstacle is the Gaussian variance-subtraction step, which the manuscript itself flags as inapplicable to the measured PSFs; if the authors can re-derive or validate Eq. (6) for their actual detector response, and separate the sampling-limited bounds from intrinsic resolution values, I would be willing to reconsider favorably. I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nWorth reading: this is the first paper I've seen that treats phase transfer as a default operating condition for commercial micro-CT, not a special mode requiring visible fringes. The central claim—that propagation sharpening can beat the conventional source+detector blur even when no Fresnel fringes are visible—is supported by a direct experimental result on a Nikon scanner: half-pixel dithering resolves 28.5 µm against a conventional blur prediction of 55 µm. That result is independent of Eq. (6) and is the paper's strongest evidence.\n\nWhat's genuinely new: the Regime A/B classification, the DIPR concept, and the n_phi and E_match metrics. The physics backbone is Gureyev's deblur-by-defocus, and the authors say so openly. The custom-system scans (Scan 1–4) show a monotonic fringe-amplitude drop and SNR trend consistent with under- to over-retrieval; that part is convincing and well-executed.\n\nThe soft spot is real: Eq. (6) subtracts a Gaussian variance to predict the resolution enhancement, but the measured detector PSFs are mixed Gaussian-Lorentzian with Lorentzian fractions of 68% and 40%. The Methods section explicitly says that in such cases the Gaussian σ is “strictly, no longer applicable.” Yet the paper uses Eq. (6) to predict the 100→54.4 µm improvement and the DIPR optimum. A Lorentzian MTF decays exponentially, not quadratically, so the cancellation of the propagation Laplacian is not guaranteed. The dithering experiment shows the system responds below 28.5 µm, but a sharp Lorentzian core could contribute to that even without phase effects. The δ=0 simulation in Supp. Fig. 7(b) is the key control; I'd want to see that prominently and confirmed.\n\nThe E_match metric is partly circular—it defines DIPR as detector≈Paganin and then finds where they match—but the resolution claim doesn't depend on that circularity.\n\nNo missing references or invented entities. Self-citation is appropriate given their earlier optimization framework. Data availability is only “upon request,” which is weak for a paper making these claims.\n\nBottom line: the experimental demonstration deserves referee time. The theoretical machinery needs a second pass, especially Eq. (6), before this is published as-is. My recommendation: send to peer review with a request for careful treatment of non-Gaussian PSFs and better data availability.\n\nYours,","headline":"Phase-transfer micro-CT reframing with a strong dithering experiment, but Eq. (6)'s Gaussian assumption is load-bearing and the paper itself concedes it doesn't hold.","tokens_in":20539,"tokens_out":2018,"would_cite":true,"duration_ms":20575,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Micro-CT systems with finite source and detector blur can operate in a phase-transfer regime where propagation sharpens images beyond the conventional blur limit, even when no Fresnel fringes are visible, and the detector itself can perform","keywords":["phase contrast","micro-CT","propagation-based imaging","phase retrieval","detector-induced phase retrieval","spatial resolution","hardware-induced phase retrieval","Fresnel fringes"],"falsifier":"Measure a sharp edge with a detector whose PSF is strongly non-Gaussian (Lorentzian-dominated) across a range of propagation distances, and compare the measured edge widths to Eq. (6). If the resolution minimum occurs at a different position, or the improvement is far smaller than predicted, the Gaussian cancellation that defines the DIPR resolution advantage is the point of failure. Conversely, a system with a nearly Gaussian detector PSF should reproduce the predicted resolution gain.","tokens_in":19448,"feed_emoji":"🩻","tokens_out":5826,"duration_ms":57708,"temperature":0.7,"pith_summary":"The paper argues that conventional high-resolution micro-CT is rarely purely absorption-based: whenever propagation distance and phase shifts are large enough relative to the system's source and detector blur, free-space propagation actively sharpens the image, even if no Fresnel fringes are visible. In that phase-transfer regime, the detector's blur can act as a built-in phase-retrieval filter—called detector-induced phase retrieval (DIPR)—which removes fringes and suppresses high-frequency noise while keeping the resolution gain. The authors derive analytic conditions for this regime, introduce frequency-domain metrics to identify it, and validate it numerically and experimentally on a custom and a commercial micro-CT scanner. If correct, this changes how spatial resolution and image optimization in micro-CT are interpreted, since routine scans may already be benefiting from phase contrast without showing any fringes.","feed_headline":"Propagation sharpens micro-CT images even when fringes are invisible","feed_subtitle":"In a commercial micro-CT, measured resolution hit 28.5 µm where conventional blur predicted 55 µm.","key_machinery":"The load-bearing identity is the Gaussian-equivalent effective system width, σ_eff² = σ_s_eff² + σ_det_eff² − 2a², where 2a² is the negative variance contributed by free-space propagation (a² = γR′λ/4π). This expresses propagation as a sharpening that subtracts from the blur variances, predicting finer resolution whenever 2a² is non-negligible relative to σ_sys². The DIPR condition is T_prop H_det ≈ 1: the detector MTF H_det should match the inverse of the propagation transfer function—the standard single-distance phase-retrieval filter—over the usable spatial-frequency band, so the detector performs phase retrieval at acquisition. The paper also introduces two metrics: n_φ, the fraction of","core_discovery":"The central claim is that micro-CT systems with non-negligible propagation-induced phase transfer operate in a phase-transfer regime (Regime B) in which propagation modifies the effective system response and can provide spatial resolution finer than that predicted from conventional source and detector blur alone, irrespective of whether Fresnel fringes remain visible. Within Regime B, hardware-induced phase retrieval (HIPR) occurs in degrees: under-HIPR leaves residual edge enhancement, matched-HIPR compensates propagation exactly, and over-HIPR suppresses high frequencies. Detector-induced phase retrieval (DIPR) is the matched case in which the detector's modulation transfer function approx","pith_inferences":["If the Gaussian variance-subtraction argument is weakened by non-Gaussian detector PSFs, the quantitative resolution gain may be smaller than Eq. (6) predicts, but the qualitative existence of a phase-transfer regime would likely survive; a Lorentzian-tail-aware analysis would put numbers on this.","The framework suggests that many existing micro-CT datasets acquired with small focal spots and moderate propagation distances may already contain phase-induced resolution enhancement, implying that retrospective resolution estimates in the literature could be revisited.","A direct testable extension: scanning a sharp-edged phantom at fixed source-detector distance while sweeping source-to-sample distance should show a dip in measured edge width near the E_match minimum, exactly as the paper's experiments begin to show.","The DIPR concept could extend to other in-line imaging modalities with pixelated detectors, such as visible-light or electron microscopy, where a similar detector-blur-matches-propagation condition might explain resolution beyond conventional PSF predictions."],"forward_implications":["Absence of visible Fresnel fringes cannot be taken as evidence that image formation is purely absorption-based; it may instead indicate matched hardware phase retrieval.","The source-to-sample distance becomes a tunable parameter for jointly optimizing phase transfer, hardware retrieval, and field of view, and the optimum is not where geometric magnification is highest.","System optimization should allocate as much of the required low-pass filtering to the detector as possible (DIPR) rather than to source blur, to preserve the phase-induced resolution gain and gain noise suppression.","Spatial resolution estimates based on source and detector blur alone will underestimate performance in Regime B, where detector sampling (and sub-pixel dithering) often becomes the limiting factor.","Software phase retrieval is only strictly needed in the under-HIPR sub-regime; in matched or over-HIPR, the hardware already provides the retrieval during acquisition."],"fun_headline_variants":["Phase contrast sharpens micro-CT without visible fringes","Micro-CT resolution boosted by hidden phase effects","Sharp micro-CT images: phase beats blur even without fringes","Invisible phase fringes still improve micro-CT resolution","Phase transfer sharpens micro-CT beyond blur predictions"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The central calculation assumes the detector point-spread function can be treated as Gaussian when subtracting the propagation term in quadrature (σ_eff² = σ_s_eff² + σ_det_eff² − 2a²); real detector PSFs have significant Lorentzian tails, and the paper itself notes the Gaussian variance rule is strictly no longer applicable for such PSFs.","fun_headline_variants_meta":{"raw":{"variants":["Phase contrast sharpens micro-CT without visible fringes","Micro-CT resolution boosted by hidden phase effects","Sharp micro-CT images: phase beats blur even without fringes","Invisible phase fringes still improve micro-CT resolution","Phase transfer sharpens micro-CT beyond blur predictions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000503,"raw_usage":{"total_tokens":2297,"prompt_tokens":748,"completion_tokens":1549,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":492,"completion_tokens_details":{"reasoning_tokens":1474}},"tokens_in":492,"tokens_out":1549,"duration_ms":10383,"temperature":1.0,"reasoning_tokens":1474,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T06:17:09.044409+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure a sharp edge with a detector whose PSF is strongly non-Gaussian (Lorentzian-dominated) across a range of propagation distances, and compare the measured edge widths to Eq. (6). If the resolution minimum occurs at a different position, or the improvement is far smaller than predicted, the Gaussian cancellation that defines the DIPR resolution advantage is the point of failure. Conversely, a system with a nearly Gaussian detector PSF should reproduce the predicted resolution gain.","supporting_citations":[],"review_version":1}