{"id":"836448b0-6c3f-4b52-aefa-f85d090261f4","arxiv_id":"2502.01011","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"A recycling algorithm based on iterative least-squares selection lets adaptive proton FLASH plans reuse most original ridge-filter pins while restoring dosimetric coverage.","lead":"This paper tests a way to adapt proton FLASH radiotherapy plans when a patient's anatomy changes, by reusing pin-shaped filters from the original plan instead of printing new ones. The approach restored tumor coverage in three liver cancer cases while keeping most of the original hardware and the modeled FLASH effect.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Recycling claim lacks a no-recycling baseline: Eq. (5) measures spot-MU similarity, not dosimetric impact, so the reported ADP-pRF quality may reflect reoptimization alone, leaving the cost of freezing pins unmeasured.","rationale":"The reader's weakest assumption is precisely that PBDs passing the ΔE≤5 MeV and least-squares similarity criteria can be transferred to re-CT without meaningful dosimetric degradation, and I agree. The paper's reported V100 values for ADP-pRF plans are promising, but the design of the experiment does not isolate the effect of recycling. A fully reoptimized adaptive plan on re-CT is the natural control: if the recycling method degrades V100 by more than 2 pp relative to full reoptimization, then the claimed 'comparable to initial pRF' is achieved at a real, unquantified cost. The algorithm specification is also incomplete (how X and Y are aligned when spot vectors differ in length; how β is used after selection; the stopping criterion), but the missing baseline is the more load-bearing gap because it directly tests the central premise. The paper does provide independent support in prior work for pRF modularization and the FLASH effectiveness model is cited; these are not the issue. Internal inconsistencies (recycling percentages reversed in abstract vs. results; V100 below the stated ≥95% goal in cases 1 and 2) further reduce confidence but are secondary. Conditional acceptance is appropriate, conditioned on the baseline comparison and algorithm clarification.","tokens_in":11909,"tokens_out":7605,"duration_ms":74771,"concrete_test":"Reproduce the three liver SBRT cases with the same IMPT-DS optimization settings on re-CT but without any PBD recycling (all PBDs reoptimized, then translated to a fresh pRF plan). Compare V100, HI, and OAR constraints against the reported ADP-pRF plan. If the recycled-pin plan's V100 is within 1 pp of the full-reoptimization plan in all cases, the freezing assumption is supported; if the gap is larger than ~2–3 pp, the paper's claim that recycling preserves plan quality is overstated and the stopping criterion in Section 2.2(c) must be revisited.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that PBDs selected in Section 2.2(c) can be recycled onto re-CT without meaningful dosimetric degradation. The selection criterion, Eq. (5), minimizes ||Yβ−X||, where X and Y are normalized spot-MU vectors from the initial (pCT) and adaptive (re-CT) DS plans. This is a purely spot-weight similarity measure. It does not evaluate the dose distribution delivered by the recycled pin on re-CT: anatomical changes alter water-equivalent path lengths, so a PBD with ΔE≤5 MeV and small least-squares error can still deposit a shifted SOBP that the fixed pin geometry cannot correct. The remaining PBDs are reoptimized, but the paper never quantifies how much the frozen subset degrades the plan relative to a fully reoptimized baseline. Without such a baseline (all PBDs reoptimized on re-CT, no recycling), the reported V100 values (91.4%, 91.7%, 98.8%) and recycling rates (64.7–91.2%) do not establish that recycling is harmless; the quality may be attributable to reoptimization alone, and the recycling cost is unmeasured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an adaptive proton FLASH therapy workflow that recycles modularized pin ridge filter (pRF) pins from an initial plan when replanning on a new CT. Pencil beam directions (PBDs) are classified by energy change, and an iterative least-squares regression selects a subset of PBDs whose normalized spot-MU distributions are most similar to the initial plan; these PBDs are frozen in the background while the remaining PBDs are reoptimized. The method is tested on three liver SBRT cases, reporting that V100 values on re-CT improve from 89.2%, 60.2%, and 69.9% (initial pRF applied to re-CT) to 91.4%, 91.7%, and 98.8% (ADP-pRF), with 91.2%, 64.7%, and 71% of pins recycled, while preserving a modeled FLASH effect.","tokens_in":12206,"tokens_out":4115,"duration_ms":38285,"significance":"If the recycling claim holds, the method would substantially reduce the cost and turnaround time of adaptive FLASH replanning by reusing most hardware. The study is a planning study with direct dosimetric comparisons, and the authors correctly identify that initial pRFs degrade on re-CT, motivating adaptation. The FLASH-effectiveness evaluation is quantitative and the dosimetric metrics are standard. However, the central claim that recycling pins preserves plan quality is currently not isolated from the effect of reoptimization, and there are inconsistencies between stated clinical goals and the reported numbers. These issues need to be resolved before the conclusions can be considered supported.","major_comments":[{"comment":"The stated clinical goal is V100 ≥ 95% for the CTV, but Table 1 shows initial pRF plans on pCT with V100 = 93.7% and 93.5% for cases 1 and 2, i.e., below the goal. Section 3.1.1 also states that the initial pRF plan ensured '98% of the CTV received the full prescription dose,' which is inconsistent with the Table 1 values. The abstract's claim of 'maintaining all clinical goals' is therefore not supported: the ADP-pRF V100 values for cases 1 and 2 (91.4% and 91.7%) also fall below the 95% threshold. Please clarify which dose specification is actually used and reconcile the text with the table.","section":"Section 2.4, Section 3.1.1, Table 1"},{"comment":"The paper never compares the proposed ADP-pRF plan (with recycled pins) against a fully reoptimized adaptive plan on re-CT with no recycling. The selection criterion in Eq. (5) is a normalized spot-MU similarity measure, not a dosimetric measure; it does not quantify the dosimetric impact of freezing a given PBD's pin on the re-CT anatomy, where water-equivalent path lengths may have changed. Without a no-recycling baseline, the reported V100 values and recycling rates do not establish that recycling is harmless—the quality recovery could be entirely due to reoptimization of the remaining PBDs. Please provide a baseline plan with 0% recycling and, ideally, a curve of plan quality versus fraction of recycled pins to quantify the true cost of recycling.","section":"Section 2.2(c), Eq. (5), Section 3.1.2"},{"comment":"The FLASH-effect evaluation compares the ADP-pRF FLASH-1Gy plan on re-CT to 'the initial IMPT plan designed on pCT' rather than to a conventional IMPT plan on re-CT. Because the anatomical changes between pCT and re-CT are substantial (e.g., case 2 V100 drops to 60.2% when the initial pRF is evaluated on re-CT), the reported reductions in OAR doses may be confounded by anatomical differences rather than reflecting the FLASH effect of the adaptive plan. To support the claim that the FLASH effect is preserved, the FLASH-1Gy reductions should be recomputed against a conventional IMPT plan optimized on re-CT.","section":"Section 3.2, Figure 4"}],"minor_comments":[{"comment":"Typo: 'FAST-01 trail' should be 'FAST-01 trial'.","section":"Introduction"},{"comment":"The organ name is misspelled as 'Deudenum' in both the text and Table 1; it should be 'Duodenum'.","section":"Section 3.1.2, Table 1"},{"comment":"The reference 'Figure 1(d)' is incorrect; the ADP-pRF dose distribution is shown in Figure 3(d), not Figure 1(d).","section":"Section 3.1.2"},{"comment":"In Eq. (5), the vector X is used but not explicitly defined; the text defines y_ij and Y, but X should be defined as the normalized spot-MU vector of the initial pCT plan for the corresponding PBD.","section":"Section 2.2(c), Eq. (5)"},{"comment":"The energy-change threshold uses ΔE_max or ΔE_min ≥ 5 MeV, but it is not clear whether these differences are absolute values and whether the threshold applies to the maximum or minimum spot energy within each PBD or to some other per-PBD energy metric; please clarify the definition.","section":"Section 2.2(a)"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a relevant practical problem and the core idea is plausible, but the missing no-recycling baseline is a serious gap for the recycling claim, and the clinical-goal inconsistency needs correction. The FLASH-effect comparison baseline also needs to be re-evaluated. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe headline: this paper has a genuinely practical idea—recycling modular pin ridge filter pins in adaptive proton FLASH replanning—and its three liver cases do show that the adapted plan restores target coverage that the initial plan loses on the new CT. But the manuscript is not ready as written. The numbers have several internal inconsistencies, and the paper never measures the cost of freezing pins relative to a fully reoptimized plan.\n\nWhat is new: the iterative least-squares selection of pencil beam directions (PBDs) to retain from the original plan, while reoptimizing the rest, is a sensible extension of the authors' prior modular pRF work. The recycling rates (65–91%) would be a real logistical win for adaptive FLASH, where reprinting ridge filters is currently slow and expensive. The motivation is clearly laid out, and Table 1 supports the qualitative claim that reusing most pins yields a plan comparable to the original on pCT.\n\nSoft spots, in rough order of importance. First, there is no baseline of a fully reoptimized adaptive pRF plan (no recycling) on the re-CT. The reported V100 values could reflect the reoptimization of the remaining PBDs alone; without that baseline, the dosimetric benefit of the recycling step itself is unmeasured. The selection criterion in Eq. (5) is spot-MU similarity, not dose impact—anatomical changes shift water-equivalent path lengths, so a pin with similar spot weights can still deliver a shifted SOBP. A proper comparison would show whether the frozen pins actually hold up or merely get compensated for. Second, the manuscript contains multiple numeric errors: the text claims initial plans 'ensuring 98% of the CTV received the full prescription dose' but Table 1 lists V100 of 93.7% and 93.5%; the Discussion swaps cases 2 and 3 in the largest V100 drop (case 2, not case 3, drops 33.3 pp); and case 2's CTV volume is given as 21.35 cc in the Discussion but 25.2 cc in the methods. These are likely typos, but there are enough of them that a careful reader loses trust in the exact numbers. Third, the algorithm's stopping criterion is vague ('until sufficient selected PBDs are recycled'), and the iterative reoptimization is not spelled out precisely enough to reproduce. Fourth, the claim of 'maintaining all clinical goals' is overstated—case 3's duodenum V0.5cc reportedly exceeds the BR001 tolerance.\n\nThe stress-test note is fair on the missing baseline; it doesn't kill the practical claim, but it does mean you can't yet say recycling is harmless. A reviewer should ask for the full-reoptimization comparison and for spot-level dose validation of frozen pins.\n\nWho is this for? Anyone working on proton FLASH planning, particularly adaptive workflows. It deserves a serious referee, but with major revision. I'd recommend accept-with-revisions only after the baseline and numeric fixes are in.\n\nBest.","headline":"Genuinely practical idea for adaptive proton FLASH pin reuse, with three supportive cases, but the manuscript's internal inconsistencies and missing no-recycling baseline keep this at major-revision status.","tokens_in":12776,"tokens_out":6202,"would_cite":true,"duration_ms":58309,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Adaptive proton FLASH replans can recycle 65–91% of ridge-filter pins and restore coverage to within 2.3 percentage points of the original plan.","keywords":["adaptive proton therapy","FLASH radiotherapy","pin ridge filter","pencil beam direction recycling","least-squares regression","liver SBRT","treatment plan adaptation","FLASH effectiveness model"],"falsifier":"Reoptimize each adaptive pRF plan from scratch on the re-CT with no frozen PBDs, and compare CTV V100, homogeneity index, and OAR DVHs against the recycled-pin plans; if the recycled plans are worse than the fully reoptimized ones by more than the roughly 2 percentage points already seen, or if an independent phantom delivery with film and ion chambers shows the frozen-weight recycled pins do not reproduce the planned dose, the recycling criterion is not sufficient.","tokens_in":11700,"feed_emoji":"⚡","tokens_out":8098,"duration_ms":70686,"temperature":0.7,"pith_summary":"This paper tries to establish that adaptive replanning of proton FLASH therapy can be done by reusing most of the physical ridge-filter pins from the original treatment plan instead of manufacturing new filters. The method identifies pencil beam directions whose spot-energy pattern changed little when the patient's anatomy changed, then iteratively freezes a subset of those directions with their beam weights intact while reoptimizing the rest. Across three liver stereotactic body radiotherapy cases, the adaptive plans restored target coverage that applying the unmodified original plan to the new CT lost, reaching V100 within 2.3 percentage points of the original plan (and slightly better in one case) while recycling 64.7–91.2% of ridge pins. The modeled FLASH effect, scored with a 40 Gy/s dose-rate threshold and 1 Gy dose threshold, was preserved in the recycled plans. If correct, this makes adaptive FLASH replanning faster and cheaper by turning a full re-manufacturing problem into a reuse problem.","feed_headline":"Adaptive FLASH replanning reuses up to 91% of ridge-filter pins","feed_subtitle":"New plans match original target coverage within 2.3 percentage points while preserving the FLASH effect.","key_machinery":"The central object is the modularized pin ridge filter (pRF), a stack of pyramid-shaped steps made from 1–6 mm cuboid modules with 5 mm water-equivalent thickness each, which spreads a 250 MeV single-energy beam into a spread-out Bragg peak. The argument is carried by the PBD recycling loop: pencil beam directions whose maximum or minimum spot energy shifts by $\\Delta E \\ge 5$ MeV are tagged new/changed, and the rest become recycling candidates; an iterative least-squares regression $\\min_\\beta \\| \\mathbf{Y}\\beta - \\mathbf{X}\\|_2$ on normalized spot ion MUs selects the two PBDs per iteration whose spot-weight distribution best matches the adaptive beamset, freezes their weights, and places them in a background plan while the remaining PBDs are reoptimized. The transfer criterion is that a small normalized spot-MU error means the physical pin geometry can be reused without reoptimizing that PBD's weight.","core_discovery":"The central claim is that adaptive FLASH replanning can be reduced to a pin-recycling problem: a plan adapted to new anatomy can be built by freezing most of the original single-energy 250 MeV pencil beam directions and reoptimizing only the remainder, provided the frozen PBDs are chosen by a least-squares criterion on normalized spot MU distributions. This recovers dosimetric quality lost when the original pin ridge filter plan is simply recalculated on the re-CT: V100 went from 89.2% to 91.4%, 60.2% to 91.7%, and 69.9% to 98.8% across the three cases, against benchmarks of 93.7%, 93.5%, and 97.3% for the original plan on the planning CT. The recycled fraction was 91.2%, 64.7%, and 71% of PBDs respectively, and the modeled FLASH effect remained comparable to the original plan.","pith_inferences":["If the least-squares transfer criterion generalizes, the same recycling loop could be tested in other hypofractionated sites where anatomy changes between fractions, such as lung or pancreas; the paper only demonstrates liver SBRT.","A sensitivity study that varies the 5 MeV energy-change threshold and the two-PBD-per-iteration selection rate could reveal the maximum recyclable fraction before dosimetric degradation, and whether the roughly 2 percentage point V100 gap is a floor or an artifact of these choices.","The clinical benefit still rests on the FLASH effectiveness model's parameters; direct measurement of normal-tissue sparing with recycled pRFs would be needed before the modeled 33% dose reduction is relied on.","The least-squares error used for pin selection could double as an online plan-quality early warning, triggering adaptation automatically when anatomical change passes a threshold."],"forward_implications":["Adaptive pRF plans can be produced without full hardware re-manufacture; 65–91% of the existing ridge pins carry over to the new plan.","Target coverage on the re-CT is restored to within 2.3 percentage points of the original pCT plan, and in one case slightly exceeds it.","Organs at risk remain within protocol tolerance in the recycled plans, apart from one duodenum V0.5cc value that rises 2.03 Gy above the initial plan in case 3.","The modeled FLASH effect is preserved: for example, PTV-GTV mean dose is reduced by 35.5% in the adaptive plan versus 35% in the initial plan for case 1.","The recycling rate tracks anatomical change, with the smallest CTV change recycling over 90% of pins and larger or shifted CTVs recycling roughly 65–71%."],"supporting_citations":[{"why":"Supplies the range shifter and range compensator concept that lets single-energy Bragg peak beams be shaped, the delivery basis for the IMPT-DS plans.","marker":"[25]"},{"why":"Establishes the streamlined modularized pin-ridge-filter design and the spot-reduction translation from IMPT-DS to pRF plans.","marker":"[26]"},{"why":"Makes the case for adaptive proton therapy, the clinical problem this recycling method addresses.","marker":"[28]"},{"why":"Shows single-energy PBS Bragg peak FLASH planning for liver hypofractionation, providing the planning setup adopted here.","marker":"[29]"},{"why":"Provides the FLASH effectiveness model used to score dose and dose-rate thresholds in the plans.","marker":"[31]"},{"why":"Supplies the modularized pRF FLASH planning parameters (40 Gy/s, 200 ms, 0.67) used in the FLASH evaluation.","marker":"[32]"},{"why":"Defines the liver SABR protocol constraints and 50 Gy in 5 fraction prescription that the plans must satisfy.","marker":"[33]"},{"why":"Feasibility study of modularized pRF implementation in liver SBRT, the benchmark for the FLASH effect comparison.","marker":"[34]"}],"fun_headline_variants":["Most beam directions reused in adaptive FLASH plan","Adaptive FLASH recovers coverage lost by replanning","Pin recycling cuts adaptive FLASH replanning effort","91% of beam directions recycled in adaptive FLASH","Least-squares picks reusable beams for adaptive FLASH"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a pencil beam direction whose spot-energy limits change by less than 5 MeV and whose normalized spot-MU distribution has small least-squares error can be transferred to the new anatomy with its beam weight frozen, without clinically meaningful dosimetric degradation.","fun_headline_variants_meta":{"raw":{"variants":["Most beam directions reused in adaptive FLASH plan","Adaptive FLASH recovers coverage lost by replanning","Pin recycling cuts adaptive FLASH replanning effort","91% of beam directions recycled in adaptive FLASH","Least-squares picks reusable beams for adaptive FLASH"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000409,"raw_usage":{"total_tokens":2223,"prompt_tokens":1151,"completion_tokens":1072,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":767,"completion_tokens_details":{"reasoning_tokens":995}},"tokens_in":767,"tokens_out":1072,"duration_ms":7805,"temperature":1.0,"reasoning_tokens":995,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T16:53:20.000246+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reoptimize each adaptive pRF plan from scratch on the re-CT with no frozen PBDs, and compare CTV V100, homogeneity index, and OAR DVHs against the recycled-pin plans; if the recycled plans are worse than the fully reoptimized ones by more than the roughly 2 percentage points already seen, or if an independent phantom delivery with film and ion chambers shows the frozen-weight recycled pins do not reproduce the planned dose, the recycling criterion is not sufficient.","supporting_citations":[],"review_version":1}