{"id":"51e5617a-6bc7-45fc-9808-ec71ace714f8","arxiv_id":"2507.20793","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Mechanical supersampling with photon-counting detectors plus MLEM reconstruction produces roughly 3x sharper X-ray radiographs at clinical flux rates.","lead":"A medical physics team in Freiburg shows that moving a photon-counting X-ray detector in tiny steps and merging the frames with an iterative algorithm yields radiographs about three times sharper than a single static exposure. The approach works at clinical dose levels and could let mammography systems image fine detail without bulky geometric magnification.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 3x resolution gain may be inflated by MLEM edge overshoot from an unmodeled charge-sharing response; ESF/MTF metrics are not enough to distinguish true resolution from reconstruction artifact.","rationale":"The paper is a solid proof-of-concept: mechanical supersampling with ECC-based shifts and MLEM visibly sharpens phantom images, and long-range sampling demonstrably homogenizes the image. Those claims are supported by the figures. The load-bearing issue is whether the quantitative threefold resolution gain is real spatial resolution or an artifact of an incomplete forward model. The reader identified the same forward-model risk; I agree and add a mechanism: unregularized MLEM can commit an inverse crime when the model omits charge sharing and field distortions. The self-reported ringing in Fig. 3 is direct evidence that the model is not capturing the detector response. Since the only quantitative resolution metrics are ESF 10-90% and MTF10%, both of which are sensitive to edge overshoot, the headline factor ~3 is not yet established. A synthetic end-to-end test with a known object and a charge-sharing-aware simulator would settle this. I therefore keep the CONDITIONAL verdict: acceptance requires either forward-model augmentation or an independent ground-truth resolution check.","tokens_in":9547,"tokens_out":3474,"duration_ms":45155,"concrete_test":"Simulate the full acquisition with a known high-resolution phantom and a Monte Carlo PCD model that includes a position- and energy-dependent charge-sharing kernel calibrated to the SANTIS detector (e.g., from microbeam pixel-response maps, cf. [17]). Reconstruct the simulated frames with the paper's unregularized MLEM using the current distance-driven forward model, and again using the same MLEM with the forward model augmented by the charge-sharing kernel. If the MTF10% or 10-90% ESF width of the reconstructed image differs by more than about 10% between the two forward models, or if the reconstructed MTF exceeds the known system MTF that includes charge sharing, then the reported resolution gain is not established as true object resolution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on the forward model in Section II.C accurately describing the PCD response. The distance-driven ray tracer models geometric pixel sensitivity but does not model charge sharing or field distortions; the authors themselves attribute the ringing in Fig. 3 and the geometric warping to these effects (Section III, Section IV). Because MLEM is run without regularization (Section IV), a wrong forward model will be compensated by placing energy at high spatial frequencies. The resolution evidence is an ESF from a single tungsten edge (Fig. 4) and an MTF from a single region (Fig. 7). Any Gibbs-like overshoot at the edge narrows the 10-90% rise and raises MTF10%, so the reported threefold improvement could be, in part, an artifact of the reconstruction rather than recovered object information. The comparison to linear/cubic interpolation is not a ground truth; it only shows that iterative deconvolution is sharper than interpolation under the assumed model.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a super-resolution radiography method (PC-SSI) that combines mechanical detector motion over sub-pixel steps, ECC-based registration for shift estimation, and MLEM reconstruction with a distance-driven ray tracer. The authors report a near threefold improvement in effective spatial resolution, quantified by a 10-90% edge spread function (31.25 µm versus 93.75 µm for linear interpolation) and by MTF10% analysis, along with qualitative improvements in bar-pattern visibility, suppression of pixel/ASIC defects for long-range trajectories, and a comparison with a clinical mammography system at comparable or lower dose. The central quantitative claim rests on single ESF and MTF measurements without error bars, and the forward model does not include charge sharing or sensor field distortions that the authors themselves identify as sources of artifacts; these points need additional validation.","tokens_in":9724,"tokens_out":6125,"duration_ms":77623,"significance":"If the reported resolution gain is real, the method is practically significant: it operates in frame-based photon-counting mode at clinically relevant flux, avoids the flux limitations of event-based centroiding, and could reduce the need for geometric magnification and large-area detectors in mammography. The qualitative phantom results, the dose-conscious acquisition design, and the explicit comparison with a clinical system are valuable strengths. The method's feasibility is plausible, but the quantitative threefold resolution claim is not yet established because the evidence is a single-edge ESF and a single-region MTF, both produced by an unregularized iterative reconstruction whose forward model omits known detector physics.","major_comments":[{"comment":"The central quantitative claim of a threefold resolution improvement is based on a single ESF from one tungsten edge and a single MTF from one image region, with no repeated acquisitions, error bars, or sensitivity analysis. The reported numbers (10-90% rise in 31.25 µm versus 93.75 µm, and MTF10% values) are load-bearing and need to be shown to be reproducible and stable against ROI selection, edge-angle estimation, and registration uncertainty. Please provide multiple independent measurements with spread, and state how the metrics were computed.","section":"Section III, Figs. 4 and 7"},{"comment":"The distance-driven ray tracer models geometric pixel sensitivity but not charge sharing or sensor field distortions, and the paper attributes the ringing in Fig. 3 and the geometric warping to these unmodeled effects. Because MLEM is run without regularization (Section IV), an incorrect forward model can be compensated by placing energy at high spatial frequencies, which can artificially narrow the 10-90% ESF and raise MTF10%. The comparison with linear/cubic interpolation therefore does not by itself demonstrate recovery of true object information. Please validate against known ground truth (for example, an independently measured edge profile, a phantom with known feature dimensions, or a high-resolution reference method), include a charge-sharing response in the forward model, or report how the ESF/MTF metrics change with iteration count, regularization, and energy threshold settings.","section":"Section II.C, Section III, Section IV"},{"comment":"The acquisition and reconstruction protocol is under-specified for reproducibility: no MLEM iteration count or stopping criterion, no TV regularization weight (even though it was disabled), no ECC registration parameters, no exact trajectory or dwell pattern (only approximate strides in figure captions), and no description of the ASIC-border interpolation parameters. Since these choices directly affect the resolution metrics, the full protocol should be reported or the code/data should be made available.","section":"Section II.C and Section IV"}],"minor_comments":[{"comment":"\"Module Transfer Function\" should be \"Modulation Transfer Function\".","section":"Fig. 7"},{"comment":"There are spelling errors such as \"imhomongeneities\" and \"magni ication\" in the Fig. 8 caption; the manuscript should be proofread.","section":"Section III"},{"comment":"The ballistic imaging paragraph refers to a \"5 kW X-ray source\" while the experimental settings are described elsewhere as 5 W; please clarify which value is intended.","section":"Section IV"},{"comment":"The acronym \"PCD SSI\" is used without definition; elsewhere the method is called \"PC-SSI\".","section":"Section III"},{"comment":"The phantom is referred to as both \"Gammex 156 Mammographic Accreditation Phantom\" and \"Gammex ACR MAP\"; please use consistent terminology.","section":"Section II.B and Fig. 8"},{"comment":"The caption's \"TH40kV\" is ambiguous; please specify the tube setting or threshold condition it denotes.","section":"Fig. 3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope and addresses a practical problem, but the quantitative resolution claim needs stronger evidence. The main risk is that model mismatch and unregularized MLEM contribute to the reported sharpening; a revision that adds ground-truth validation, repeatability statistics, and a complete protocol would substantially strengthen the paper. The DECTRIS employee co-authorship and detector loan are disclosed, and I see no novelty concern."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid, novel proof-of-concept that combines mechanical detector motion with MLEM reconstruction to sharpen planar radiographs in true photon-counting mode at clinical flux. The key new thing is the combination itself—supersampling with sub-pixel registration and a distance-driven forward model on a high-Z GaAs PCD—and the phantom evidence backs up the qualitative claim. The bar phantom, ESF, and MTF all point the same direction, and the paper is unusually candid about the artifacts (ringing, geometric warping, ASIC borders) that appear when you push resolution.\n\nWhere I'd push back: the quantitative 'threefold improvement' is measured on single edges and a single MTF region, with no error bars or repeated acquisitions, so the uncertainty is unknown. The stress-test worry that some of the gain could be MLEM edge overshoot from an unmodeled charge-sharing response is legitimate, and the paper's own attribution of ringing to charge sharing cuts both ways: it shows the forward model isn't complete. Because they intentionally disabled TV regularization to isolate supersampling, the reconstruction has no penalty to stop it from parking energy at high spatial frequencies. So the ESF 10-90% and MTF10% numbers are probably optimistic as estimates of true recoverable detail. That said, I think the stress-test note goes too far if it implies the whole gain is artifact: the QRM bar phantom visibly resolves 20-25 µm structures, which ringing alone wouldn't produce. But the factor of three needs a ground-truth or simulation check to separate real recovery from overshoot.\n\nThe clinical comparison to the Mammomat is illustrative, not a head-to-head: different detector, geometry, and dose, with STC calibration making contrast non-comparable. The abstract's 'clinically viable' overreaches a single phantom study, though the discussion itself is more measured. The scan-time estimate for ballistic acquisition is back-of-envelope, not measured.\n\nCitation pattern is honest—they credit Dreier and Tremsin/Khalil for prior super-resolution ideas, and self-citations [12]–[14] genuinely anchor the reconstruction framework. No data or code shipped, which is typical for a hardware-adjacent med-phys preprint and not a dealbreaker.\n\nBottom line: worth a serious referee. I'd send it to review and ask for uncertainty quantification, a ground-truth or simulation check on the resolution gain, and a toned-down clinical claim. The core idea is sound and the experiments are a real step beyond interpolation-based supersampling.","headline":"A credible proof-of-concept for mechanical supersampling on high-Z PCDs at clinical flux; the resolution gain is likely real, but the factor-of-three quantitative claim needs uncertainty analysis and a ground-truth check.","tokens_in":10320,"tokens_out":3224,"would_cite":true,"duration_ms":34306,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["87.59.-e","87.59.Dj"],"model":"deepseek-v4-flash","headline":"Mechanical supersampling with a high-Z photon-counting detector plus model-based iterative reconstruction raises radiographic spatial resolution by roughly threefold at clinical flux rates and dose.","keywords":["photon-counting detectors","super-resolution","mechanical supersampling","iterative reconstruction","MLEM","digital mammography","edge spread function","high-Z semiconductor detectors"],"falsifier":"Take a physical phantom with line pairs or a sharp edge, and image it twice: once with the supersampling pipeline presented here, and once with a detector whose true pixel pitch equals the claimed effective resolution (about 25 µm). If the supersampled image resolves features that the genuinely finer-pitched detector does not, or if the measured edge-spread width changes when the trajectory model is changed without touching the hardware, then the apparent resolution gain is a reconstruction artifact rather than recovered information.","tokens_in":9358,"feed_emoji":"🩻","tokens_out":6372,"duration_ms":66689,"temperature":0.7,"pith_summary":"This paper claims that planar radiography can be pushed well beyond the pixel-pitch limit of a high-Z photon-counting detector without the flux restrictions of event-based super-resolution, by physically translating the detector in sub-pixel steps during exposure and merging the frames with an iterative, model-based reconstruction. The authors show in phantom studies that this mechanical supersampling plus distance-driven MLEM roughly triples the effective spatial resolution (MTF10%) and narrows the 10–90% edge-spread width from 93.75 µm to 31.25 µm relative to interpolation-based upscaling. They further demonstrate that a prototype GaAs detector so processed yields sharper, more homogeneous images of a mammography phantom than a clinical mammography system, at comparable or lower dose. The clinical significance, if the result holds, is that micron-scale mammographic detail such as microcalcifications could be obtained without geometric magnification, large detector areas, or increased radiation exposure.","feed_headline":"Moving detector triples X-ray resolution at clinical dose","feed_subtitle":"Sub-pixel frame merging with model-based reconstruction reveals ~25-micron detail, matching or beating mammography dose for dose.","key_machinery":"The carrying mechanism is a two-stage pipeline. First, the detector is translated along a trajectory spanning several pixels while frames are accumulated; sub-pixel displacements are estimated afterward with enhanced correlation coefficient (ECC) registration, so no mechanical precision is needed. Second, the low-resolution frames are plugged into an inverse problem treated as pseudo-laminographic, solved with a maximum-likelihood expectation-maximization (MLEM) algorithm whose forward projector is a distance-driven ray tracer that models the PCD pixel response. The long-range trajectory additionally acts as a moving average over pixel-level response variations, which suppresses ASIC borders, sensor defects, and spectral inhomogeneities that would otherwise masquerade as detail.","core_discovery":"The paper establishes that mechanically supersampled acquisition combined with model-based iterative reconstruction produces a near threefold improvement in effective spatial resolution in planar radiography with high-Z photon-counting detectors operating in true photon-counting mode at clinical flux rates. The central evidence is the edge-spread measurement: the 10–90% rise of a tungsten edge falls from 93.75 µm (linear interpolation of registered frames) to 31.25 µm with the proposed iterative deconvolution, and the MTF10% threshold improves by a comparable factor. Long-range supersampling over many pixels is shown to average out sensor defects, ASIC boundary pixels, and inter-pixel spectral variations, yielding more homogeneous absorption and spectral line profiles, at the cost of some edge sharpness relative to short-range sampling. The authors position the result as making micron-scale radiography clinically viable, with digital mammography as the most immediate application.","pith_inferences":["If charge sharing is included in the forward model as a threshold-dependent term, the same pipeline could plausibly push resolution toward the sensor's intrinsic edge-response limit, perhaps beyond the threefold factor demonstrated here.","The ECC-based shift estimation should generalize to patient motion compensation, not just detector motion, which could make the method robust in live breast imaging.","The spectral-averaging benefit is a testable precondition for energy-resolved phase-contrast imaging; a bench experiment comparing spectral homogeneity with and without long-range supersampling would settle it.","A reader trial on clinical mammograms (supersampled versus conventional, dose-matched) would be the decisive test of whether the improved phantom resolution translates into better detection of microcalcifications."],"forward_implications":["Digital mammography could detect smaller microcalcifications and sharper spiculations without geometric magnification or a large-area detector, at comparable or reduced dose.","The resolution gain relaxes the pixel-pitch-versus-spectral-pixel-size trade-off in high-Z PCDs, since larger pixels can be used and supersampling recovers the lost detail.","Long-range supersampling improves spectral fidelity enough to support quantitative contrast-agent imaging from a single projection.","Continuous 'ballistic' detector motion with burst readout could bring a complete supersampled scan into the 100–400 ms range of clinical mammography exposures.","Explicit spectral field mapping, proposed for future work, may cure the geometric distortion and ringing artifacts that currently limit short-range supersampling."],"supporting_citations":[{"why":"Supplies the distance-driven projection and backprojection model used as the forward projector in the MLEM reconstruction.","marker":"[11]"},{"why":"Provides the ECC sub-pixel registration that estimates detector shifts without requiring mechanical precision.","marker":"[10]"},{"why":"Defines the interpolation-based supersampling recovery approach used as the baseline in the edge-spread and MTF comparisons.","marker":"[15]"},{"why":"Offers microbeam pixel-response studies that the authors cite to explain the observed ringing artifacts as charge sharing.","marker":"[17]"},{"why":"Supplies the signal-to-equivalent-thickness (STC) normalization method used to calibrate projection data.","marker":"[8]"},{"why":"Provides the beam-hardening correction used in radiogram linearization within the STC calibration.","marker":"[9]"},{"why":"Describes the Medipix3RX PCD readout with charge-sharing mitigation, giving the architectural context of pixel-pitch-limited resolution.","marker":"[1]"},{"why":"Describes cluster-based Timepix3 subpixel imaging, the prior low-flux approach that the new method is designed to surpass.","marker":"[3]"},{"why":"Documents that higher spatial resolution improves CNN-based lesion detection in mammography, motivating the clinical value.","marker":"[20]"}],"fun_headline_variants":["Detector motion triples X-ray resolution at clinical dose","Sub-pixel sampling yields 3x sharper X-rays without extra dose","Mechanical supersampling sharpens radiography threefold","Photon-counting detector motion boosts X-ray resolution 3x","X-ray resolution tripled with clinical-dose supersampled imaging"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The key load-bearing premise is that the distance-driven ray model accurately describes what the detector actually measures, including charge sharing and sensor field distortions; the paper's own observation of ringing artifacts and geometric warping at supersampled resolution shows this premise is only approximately true.","fun_headline_variants_meta":{"raw":{"variants":["Detector motion triples X-ray resolution at clinical dose","Sub-pixel sampling yields 3x sharper X-rays without extra dose","Mechanical supersampling sharpens radiography threefold","Photon-counting detector motion boosts X-ray resolution 3x","X-ray resolution tripled with clinical-dose supersampled imaging"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000595,"raw_usage":{"total_tokens":2786,"prompt_tokens":949,"completion_tokens":1837,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":565,"completion_tokens_details":{"reasoning_tokens":1765}},"tokens_in":565,"tokens_out":1837,"duration_ms":19126,"temperature":1.0,"reasoning_tokens":1765,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T13:14:23.567735+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a physical phantom with line pairs or a sharp edge, and image it twice: once with the supersampling pipeline presented here, and once with a detector whose true pixel pitch equals the claimed effective resolution (about 25 µm). If the supersampled image resolves features that the genuinely finer-pitched detector does not, or if the measured edge-spread width changes when the trajectory model is changed without touching the hardware, then the apparent resolution gain is a reconstruction artifact rather than recovered information.","supporting_citations":[{"cited_title":"Distance -driven projection and backprojection in three dimensio ns,","cited_arxiv_id":null,"evidence_quote":"Supplies the distance-driven projection and backprojection model used as the forward projector in the MLEM reconstruction."},{"cited_title":"Parametri c image alignment using enhanced correlation coeffici ent maximization,","cited_arxiv_id":null,"evidence_quote":"Provides the ECC sub-pixel registration that estimates detector shifts without requiring mechanical precision."},{"cited_title":"Super - resolution X -ray imaging with hybrid pixel detector s using electromagnetic source stepping,","cited_arxiv_id":null,"evidence_quote":"Defines the interpolation-based supersampling recovery approach used as the baseline in the edge-spread and MTF comparisons."},{"cited_title":"Edge pixel response studies of edgeless silicon sensor technology for pixellated imaging detectors,","cited_arxiv_id":null,"evidence_quote":"Offers microbeam pixel-response studies that the authors cite to explain the observed ringing artifacts as charge sharing."},{"cited_title":"Data processing and image reconstruction methods for pixel detectors,","cited_arxiv_id":null,"evidence_quote":"Supplies the signal-to-equivalent-thickness (STC) normalization method used to calibrate projection data."},{"cited_title":"Radiogram enhancement and linearization using the beam hardening correction method,","cited_arxiv_id":null,"evidence_quote":"Provides the beam-hardening correction used in radiogram linearization within the STC calibration."},{"cited_title":"The medipix3RX: A high resolution, zero dead -time pixel detector readout chip allowing spectroscopic imaging,","cited_arxiv_id":null,"evidence_quote":"Describes the Medipix3RX PCD readout with charge-sharing mitigation, giving the architectural context of pixel-pitch-limited resolution."},{"cited_title":"Subpixel resolution in CdTe Timepix3 pixel detectors,","cited_arxiv_id":null,"evidence_quote":"Describes cluster-based Timepix3 subpixel imaging, the prior low-flux approach that the new method is designed to surpass."},{"cited_title":"Ultra -high resolution, multi -scale, c ontext -aware approach for detection of small cancers on mammography,","cited_arxiv_id":null,"evidence_quote":"Documents that higher spatial resolution improves CNN-based lesion detection in mammography, motivating the clinical value."}],"review_version":1}