{"id":"e30749ce-5383-46f2-b4a8-4fd877378cd4","arxiv_id":"2501.00666","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A coded-mask gamma camera measured prompt-gamma distal falloff positions in a PMMA phantom with 1.7 mm precision at 10^8 protons, matching Monte Carlo predictions and surviving clinical beam rates.","lead":"Researchers tested a coded-mask gamma camera that watches gamma rays from a proton beam stopping in plastic, measuring where the beam stops to within 1.7 mm in a clinical-style experiment. The result is a candidate technology for checking proton-therapy dose placement in real time.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 1.7 mm precision at 10^8 protons is never traced to the bootstrap analysis; the only 1.7 mm computed in the text is the 10^10-proton RMSE across energies, so the headline precision may be a mislabeled accuracy metric.","rationale":"The paper is a genuine first experiment and the feasibility conclusion is plausible; the concern is not about fraud or hardware, but about the exact support for the headline number. I read the central claim as the quantitative precision value, not only the qualitative demonstration that a coded-mask camera can image prompt gammas. The reader's weakest assumption targets the Monte Carlo system matrix, which is a legitimate systematic risk for absolute range accuracy, but it does not directly attack the claimed statistical precision: a biased system matrix can alter DFP positions without changing the spread of repeated estimates. The more load-bearing weakness is that the text never connects 1.7 mm at 10^8 protons to the bootstrap analysis. Section 3.4 computes RMSE = 1.7 mm for 10^10-proton data across seven energies; Section 3.5 gives only IQR statements and plot ranges, with no numerical standard deviation for S4 at 10^8. The abstract, conclusions, and Table 3 repeat the 1.7 mm value for 10^8 protons at S4, which happens to equal the RMSE. Footnote 5 strengthens the concern because the authors themselves previously withdrew a claim based on the same distinction between aggregated shift precision and single-spot statistical precision. If the bootstrap value is different, the comparison in Table 3 against MPS and KES systems changes, and the quantitative basis of the central claim is weakened. The proposed check is inexpensive: it only requires exposing the bootstrap summary statistics and rerunning a few reconstructions with non-optimized parameters. Therefore the reader's CONDITIONAL verdict is appropriate and should remain; the condition should explicitly include reporting the bootstrap distribution and tracing the 1.7 mm value to it.","tokens_in":18512,"tokens_out":9326,"duration_ms":94787,"concrete_test":"Ask the authors to provide the 100 bootstrap DFP values for spot S4 at 10^8 protons that underpin Fig. 13, and compute the sample standard deviation and 68% interval; also draw 10 independent thinned 10^8-proton subsets from the raw S4 run and reconstruct them with reconstruction parameters fixed a priori, rather than the parameters optimized on the full 10^10 dataset. If the S4/10^8 standard deviation is not 1.7 ± 0.3 mm, or if the only traceable 1.7 mm in the paper is the Table 2 RMSE, the abstract and Table 3 have substituted an aggregated, in-sample accuracy figure for the claimed statistical precision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim is that the camera reaches 1.7 mm statistical precision for 10^8 protons at 90.86 MeV. The only 1.7 mm value actually derived in the text is the RMSE in Section 3.4 and Table 2, computed from full 10^10-proton profiles for all seven beam energies by comparing reconstructed DFPs with PSTAR ranges. That RMSE is an across-energy accuracy metric, not a per-spot statistical precision. The bootstrap analysis described in Section 2.5.5 and shown in Figs. 10 and 13 is never summarized numerically: Section 3.5 reports only IQR thresholds, and no standard deviation for spot S4 at 10^8 protons is given. The abstract, conclusions, and Table 3 nonetheless assert 1.7 mm for 10^8 protons at S4 without pointing to the bootstrap distribution. This matters because the same number appears in two different roles: RMSE for 10^10-proton multi-energy data and statistical precision for 10^8-proton single-spot data. Footnote 5 in Section 4.2 explicitly retracts a prior simulation claim for precisely this conflation, between aggregated range-shift RMSE and statistically driven single-spot precision, which makes the risk concrete rather than hypothetical. If the S4/10^8 bootstrap standard deviation is not 1.7 mm, the headline comparison with MPS and KES systems in Table 3, and the quantitative feasibility claim, would need revision.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports the first experimental test of a coded-mask (CM) gamma camera for proton therapy monitoring. The system uses a scintillating-fiber detector with a tungsten MURA mask, tested at the Heidelberg Ion Therapy Center with a PMMA phantom and proton beams of seven energies (70.51–108.15 MeV). Prompt-gamma depth profiles are reconstructed with MLEM, and the distal falloff position (DFP) is compared with PSTAR proton ranges. The authors report a statistical precision of DFP determination of 1.7 mm for 10^8 protons at 90.86 MeV, rate capability at clinical beam intensities, and benchmarking against Geant4 simulations. They conclude that the CM camera is a competitive alternative to slit-based systems.","tokens_in":18824,"tokens_out":6690,"duration_ms":59546,"significance":"If the quantitative claim is correct, this would be the first experimental demonstration that a coded-mask camera can track proton range shifts with precision comparable to knife-edge slit and multi-parallel slit systems in a clinically realistic setting. The paper has notable strengths: a seven-energy experimental dataset, rate tests up to 3.2×10^9 protons/s, transparent handling of dead pixels and background, and a detailed simulation chain with optical-photon tracking. However, the main precision claim is not consistently derived in the text: the only 1.7 mm value computed is an RMSE accuracy metric for full-statistics data, not a per-spot statistical precision from the bootstrap analysis. In addition, Table 3 includes a precision value for prior simulation work that the authors themselves retract in footnote 5. The qualitative feasibility result is credible, but the quantitative headline needs substantial revision or re-derivation.","major_comments":[{"comment":"The paper claims a statistical precision of 1.7 mm for 10^8 protons at the reference energy of 90.86 MeV, but the only 1.7 mm value derived in the text is the RMSE computed in Section 3.4 (Eq. 2) from full 10^10-proton profiles across all seven beam energies. The bootstrap analysis in Section 2.5.5 and Figures 10 and 13 is never summarized as a standard deviation for spot S4 at 10^8 protons; Section 3.5 reports only interquartile ranges. The headline precision therefore appears to be a mislabeled accuracy metric rather than a statistically derived per-spot precision. Because this number is the basis of the abstract, the conclusions, and the comparison in Table 3, the authors must either provide the actual bootstrap standard deviation for S4 at 10^8 protons or reword the claims to refer to RMSE accuracy; footnote 5 shows that a similar conflation previously led to an incorrect published value.","section":"Abstract; Section 3.4; Section 3.5; Section 5"},{"comment":"The reconstruction parameters (MLEM iteration count, energy threshold, Gaussian smoothing kernel, excluded detector columns, and event classes) are optimized using the same experimental data set on which the RMSE, correlation, and slope are then reported. No held-out validation is described. Because the optimization explicitly targets low RMSE against PSTAR ranges, the quoted RMSE of 1.7 mm is an in-sample fit statistic and may be optimistic. I recommend a cross-validation scheme, such as leave-one-spot-out, or at least an explicit description of how the optimization was prevented from overfitting, to demonstrate that the reported performance is not an artifact of parameter tuning.","section":"Section 2.5.4; Section 3.4"},{"comment":"Table 3 lists a precision of 0.72 mm for CM simulation [25], but footnote 5 in Section 4.2 explicitly retracts the 0.7–1.3 mm range from that prior work as incorrect, stating that it corresponded to aggregated range-shift precision rather than single-spot statistical precision. Using the retracted value in the comparison table is misleading and undermines the claim that the current experimental result is consistent with prior simulation predictions. The entry should be removed, corrected, or annotated with the retraction.","section":"Table 3; Section 4.2, footnote 5"},{"comment":"The system matrix A used in the MLEM update (Eq. 1) is generated entirely from Monte Carlo simulations of point-like gamma sources, using the S4 energy spectrum, and the same simulation chain is then validated by its agreement with the experimental DFP data (Table 2). Consequently, the agreement between reconstructed and PSTAR ranges could reflect internal consistency of the simulation rather than the camera's physical response. I ask the authors to provide a sensitivity analysis with respect to the assumed gamma energy spectrum and the detector efficiency model, or to cross-validate using a system matrix computed from an independent simulation or analytical model, to establish that the DFP reconstruction is robust to simulation uncertainties.","section":"Section 2.3.3; Section 3.6"}],"minor_comments":[{"comment":"The abstract contains typographical and formatting issues, such as \"to108.15 MeV\" (missing space) and \"10 8\" instead of \"10^8\" in several places; please correct these throughout.","section":"Abstract"},{"comment":"The sentence \"For all but the deepest beam spot (S1-S6), the IQR is below 3 mm\" is confusing because S1-S6 is not a single spot; presumably it means all spots except S7, which is the deepest. Please reword for clarity.","section":"Section 3.5"},{"comment":"The text states that the statistical-precision studies are \"translated to the usual metric of standard deviation (1σ)\" and summarized in Fig. 13, but Figure 13 shows no numeric standard deviations. A table of the 1σ values, especially for spot S4 at 10^8 protons, would be necessary to support the claimed 1.7 mm precision.","section":"Section 4.2, first paragraph; Figure 13"},{"comment":"The y-position resolution of 74 ± 10 mm FWHM is very poor and the authors note this precludes 2D imaging. The abstract and conclusions should perhaps qualify that the current prototype provides 1D imaging only, to avoid overstating the system's current capabilities.","section":"Section 3.1; Section 4.3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript repeats a precision/accuracy conflation that the authors have already retracted in a prior publication (footnote 5), and Table 3 still cites the retracted value. I urge the editor to require a thorough audit of all numeric claims related to precision, including the bootstrap analysis and the comparison table, before the paper can be considered for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a genuine first: a full-scale coded-mask gamma camera used with a clinical proton beam at HIT, realistic PMMA phantom, seven energies, and all the supporting calibration work. The feasibility conclusion holds up. The camera produced PG depth profiles whose distal falloff positions track PSTAR ranges with an RMSE of 1.7 mm at 10^10 protons per spot, and the rate capability tests show no significant dead time at clinical intensities. That is solid, useful experimental evidence for a technology that previously had only simulations and small-scale prototypes behind it. The paper is also honest about the hardware problems it hit, dead SiPMs and all, and it benchmarks the simulation chain carefully.\n\nThe soft spot is the headline precision claim. The abstract and conclusions say 1.7 mm statistical precision for 10^8 protons at 90.86 MeV, but the only 1.7 mm actually computed in the text is the RMSE from the full 10^10-proton reconstructions across all energies, reported in Section 3.4 and Table 2. That is an accuracy metric, not a per-spot statistical precision. The bootstrap analysis in Section 3.5 is shown only as box plots and IQRs; no standard deviation for spot S4 at 10^8 is given numerically. Figure 13 apparently displays those standard deviations, but the text never quotes the value. Given that footnote 5 explicitly retracts an earlier claim that confused range-shift RMSE with statistically-driven single-spot precision, the authors should have made this distinction unmistakable. They may well have read 1.7 mm off the Figure 13 point for S4 at 10^8, but the reader cannot verify it from the text.\n\nTwo more moderate issues. First, the reconstruction parameters were optimized on the same 10^10-proton data whose RMSE is then quoted. This is not fatal, but it should be acknowledged, ideally with a validation split or at least a statement that the parameters were chosen to maximize correlation and slope, not to minimize the headline RMSE. Second, the system matrix comes from the same Geant4 simulation chain that the paper validates against experiment. The agreement with PSTAR is therefore partly a check of the simulation and the camera as a combined system, rather than a pure camera measurement. A point-source response measurement or an independent reconstruction model would strengthen the argument.\n\nOn balance, this deserves a serious referee. It is the first experimental demonstration that a coded-mask camera can operate at clinical intensities and track proton range with a competitive field of view. The main revision is to report the bootstrap distribution's numerical values for S4 at 10^8, clarify the relationship between the 1.7 mm RMSE and the precision claim, and address the in-sample optimization. With those changes it would be a solid contribution to prompt-gamma imaging.","headline":"The first full-scale coded-mask gamma camera test at a clinical proton facility is a real step forward, but the headline 1.7 mm precision at 10^8 protons is not actually derived in the text.","tokens_in":868,"tokens_out":850,"would_cite":true,"duration_ms":42708,"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":"A coded-mask gamma camera, tested for the first time under clinical proton-therapy conditions, can locate the distal falloff of the prompt-gamma depth profile — and hence the proton range — with a statistical precision of about 1.7 mm for…","keywords":["coded mask","prompt-gamma imaging","proton therapy","range verification","distal falloff position","MLEM","scintillating fibers"],"falsifier":"Take the same camera into the same geometry but irradiate the PMMA phantom at an energy not used in building the system matrix (say 75 MeV), reconstruct the distal falloff position, and compare with the PSTAR range; if the residual grows beyond the claimed 1.7 mm or the slope departs from the tested energies' fit, the simulated system matrix is not transferable.","tokens_in":18337,"feed_emoji":"🎯","tokens_out":5466,"duration_ms":51589,"temperature":0.7,"pith_summary":"The paper reports the first experimental test of a coded-mask gamma camera for real-time proton therapy monitoring. The central claim is that the camera can determine the prompt-gamma distal falloff position, a proxy for the proton range, with a statistical precision of 1.7 mm for $10^{8}$ protons at a reference beam energy of 90.86 MeV, in a clinical beam environment. This precision matches the paper's Monte Carlo predictions, holds despite non-functional detector pixels, and is comparable to the precision of established knife-edge slit and multi-slit cameras. The authors also show the detector can sustain clinical beam intensities without measurable dead time, making real-time range verification a practical prospect.","feed_headline":"Coded-mask gamma camera tracks proton range to 1.7 mm","feed_subtitle":"First clinical-condition test shows prompt-gamma imaging can catch beam falloff shifts at therapy dose rates.","key_machinery":"The central object is the coded-mask camera: a structured tungsten collimator with a modified uniformly redundant array (MURA) pattern that casts a position-dependent shadow of the prompt-gamma source onto a pixelated scintillating-fiber detector. The reconstruction machinery is maximum-likelihood expectation maximization (MLEM), iteratively inverting the recorded hit map against a simulated system matrix to recover the one-dimensional prompt-gamma depth profile, from which the distal falloff position is read at half maximum of the falling edge. The system matrix is the load-bearing link between detector hits and source depth: it encodes, for each depth bin, the probability that a prompt gamma is registered in each detector pixel, corrected per-pixel for measured detection efficiency.","core_discovery":"In this experiment, a prototype coded-mask camera — a 7-layer stack of LYSO:Ce,Ca scintillating fibers read out by silicon photomultipliers, shadowed by a 476-rank MURA tungsten mask — was placed beside a PMMA phantom irradiated by proton beams of seven energies from 70.51 to 108.15 MeV at the Heidelberg Ion Therapy Center. Prompt-gamma depth profiles were reconstructed with an MLEM algorithm whose system matrix came from Geant4 simulations of point-source responses. The distal falloff position extracted from each profile tracked the PSTAR-calculated proton range with a Pearson correlation of 0.996 and an RMSE of 1.7 mm; for the reference spot S4 at 90.86 MeV and $10^{8}$ protons, the statistical precision of the distal falloff determination was 1.7 mm (1σ). The agreement with simulations, which reproduce the experimental profiles including artifacts, supports the conclusion that the camera works as modeled. The paper interprets this as validating coded-mask imaging as a competitive alternative for online range verification.","pith_inferences":["The 1.7 mm precision is established in a homogeneous PMMA phantom; clinical tissue has density and composition variations that will add systematic errors, so the clinically achievable precision may be worse.","The system matrix is built from a single beam energy's gamma spectrum and assumed valid across energies; measuring the camera's response at untested energies or with a radioactive line source would directly test this transferability.","If the simulated factor-of-four improvement is realized after fixing dead pixels, coded-mask cameras could reach sub-millimetre range precision per spot, which would make margin reduction a quantitative trade-off rather than a safety risk.","The paper's comparison suggests coded masks offer a larger field of view and smaller material budget than slit cameras at similar precision; a head-to-head clinical study would be needed to see whether this translates into better patient outcomes."],"forward_implications":["A coded-mask camera can determine proton range shifts in a phantom with about 2 mm precision per 10^8-proton spot, putting it on par with existing slit-based prompt-gamma cameras.","The detector's rate capability (up to 3.2×10^9 protons/s without dead time, with headroom for higher rates) is sufficient for synchrotron and cyclotron clinical beam intensities, so the approach is not limited by count-rate saturation.","Simulations of a fully operational detector with no dead pixels predict a fourfold improvement in the RMSE of distal falloff determination (1.7 mm to 0.4 mm), identifying hardware repair as the direct path to sub-millimetre precision.","The camera is insensitive to lateral beam position changes within ±1 cm at fixed depth, so small patient setup shifts do not corrupt range readings.","Once y-position resolution is restored, the same detector geometry can be extended to 2D prompt-gamma imaging, as shown in prior simulations."],"supporting_citations":[{"why":"Simulation and design basis for the near-field coded-mask technique and the full-scale detector, including the precision estimate this experiment tests.","marker":"[25]"},{"why":"Establishes the correlation between prompt-gamma emission and Bragg-peak position that motivates distal falloff tracking.","marker":"[3]"},{"why":"Provides the MLEM reconstruction algorithm used to invert the detector hit maps into depth profiles.","marker":"[47]"},{"why":"Supplements the EM reconstruction formalism used in the MLEM implementation.","marker":"[48]"},{"why":"NIST PSTAR proton ranges in PMMA are the reference values the reconstructed falloff positions are compared against.","marker":"[39]"},{"why":"Geant4 is the simulation toolkit used to produce the system matrix and the benchmarked detector response.","marker":"[40]"},{"why":"Selects and validates the QGSP_BIC_HP_EMZ physics list for prompt-gamma production, which the simulated system matrix inherits.","marker":"[42]"},{"why":"Clinical multi-slit camera results serving as the main experimental precision benchmark for comparison.","marker":"[15]"},{"why":"Clinical knife-edge slit camera results providing the established precision standard that the coded-mask result is compared with.","marker":"[9]"}],"fun_headline_variants":["Coded-mask gamma camera verifies proton range in therapy","Proton therapy range pinpointed to 1.7 mm by coded mask","First clinical test: coded-mask imager tracks proton range","Coded-mask imager hits 1.7 mm precision in proton therapy","Real-time proton range check: coded-mask hits 1.7 mm"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reconstructed depth profile is only as trustworthy as the simulated system matrix: Monte Carlo point-source responses with an energy spectrum from one beam energy are assumed to hold for all seven beam energies, all lateral positions, and the real detector's efficiency pattern, so any depth- or energy-dependent simulation bias becomes a range-measurement bias.","fun_headline_variants_meta":{"raw":{"variants":["Coded-mask gamma camera verifies proton range in therapy","Proton therapy range pinpointed to 1.7 mm by coded mask","First clinical test: coded-mask imager tracks proton range","Coded-mask imager hits 1.7 mm precision in proton therapy","Real-time proton range check: coded-mask hits 1.7 mm"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000967,"raw_usage":{"total_tokens":4140,"prompt_tokens":1000,"completion_tokens":3140,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":616,"completion_tokens_details":{"reasoning_tokens":3044}},"tokens_in":616,"tokens_out":3140,"duration_ms":23449,"temperature":1.0,"reasoning_tokens":3044,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:46:10.612883+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same camera into the same geometry but irradiate the PMMA phantom at an energy not used in building the system matrix (say 75 MeV), reconstruct the distal falloff position, and compare with the PSTAR range; if the residual grows beyond the claimed 1.7 mm or the slope departs from the tested energies' fit, the simulated system matrix is not transferable.","supporting_citations":[{"cited_title":"Near-field coded-mask technique and its potential for proton therapy monitoring","cited_arxiv_id":null,"evidence_quote":"Simulation and design basis for the near-field coded-mask technique and the full-scale detector, including the precision estimate this experiment tests."},{"cited_title":"Prompt gamma measurements for locating the dose falloff region in the proton therapy","cited_arxiv_id":null,"evidence_quote":"Establishes the correlation between prompt-gamma emission and Bragg-peak position that motivates distal falloff tracking."},{"cited_title":"Em reconstruction algorithms for emission and transmission tomography","cited_arxiv_id":null,"evidence_quote":"Supplements the EM reconstruction formalism used in the MLEM implementation."},{"cited_title":"Stopping-power and range tables for protons","cited_arxiv_id":null,"evidence_quote":"NIST PSTAR proton ranges in PMMA are the reference values the reconstructed falloff positions are compared against."},{"cited_title":"Prompt-gamma emission in geant4 revisited and confronted with experiment","cited_arxiv_id":null,"evidence_quote":"Selects and validates the QGSP_BIC_HP_EMZ physics list for prompt-gamma production, which the simulated system matrix inherits."},{"cited_title":"Tackling range uncertainty in proton therapy: De- velopment and evaluation of a new multi-slit prompt-gamma camera (mspgc) system","cited_arxiv_id":null,"evidence_quote":"Clinical multi-slit camera results serving as the main experimental precision benchmark for comparison."},{"cited_title":"First-in-human validation of CT-based proton range prediction using prompt gamma imaging in prostate cancer treatments","cited_arxiv_id":null,"evidence_quote":"Clinical knife-edge slit camera results providing the established precision standard that the coded-mask result is compared with."}],"review_version":1}