{"id":"161a2aa7-463e-410a-896b-6d08a9ce5e43","arxiv_id":"2607.16498","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A new formula and inversion scheme recover point-by-point blood velocity along an artery centerline from the spatial signal decay in standard TOF-MRA images.","lead":"This paper derives a formula for blood speed from the decay of MRI signal along an artery in standard TOF-MRA scans, and tests it on simulated vessels. It could let doctors get a functional blood-flow measurement from images already taken, without extra scan time.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Validation is an oracle experiment: regularization parameters are selected using ground-truth velocities, so the claimed accuracy is not achieved in a blind setting and the central claim is not supported.","rationale":"The Reader's verdict is CONDITIONAL, and I agree that the paper should not be fully accepted as establishing the central claim. My concern is slightly different from the Reader's weakest_assumption: the Reader emphasizes that the forward model is unvalidated, while I emphasize that even within the simulation, the regularization parameters are tuned to the known ground truth, making the reported error metrics an oracle result. Both issues point to the same conclusion: the proof-of-concept is not convincing enough to support the abstract's broad claim. However, the paper is honest about its limitations, explicitly stating that L-curve methods and in vivo validation are needed. Given that the mathematical derivation appears internally consistent and the simulation does demonstrate numerical stability of the inversion, a CONDITIONAL verdict remains appropriate. No verdict change is needed; the condition should explicitly require a blind, pre-registered validation procedure. I chose 'partial' agreement because my attack identifies a distinct additional flaw (oracle parameter selection) while sharing the Reader's concern about the unvalidated forward model. The concrete test I propose would settle both issues in one experiment by comparing blind recovered velocities against a realistic ground truth.","tokens_in":8727,"tokens_out":9796,"duration_ms":113848,"concrete_test":"Run a prospective blind validation on synthetic data with known ground truth: pre-register a single automated selection rule for (λ1, λ2) (e.g., L-curve corner or the discrepancy principle), generate noisy TOF-MRA signals from a forward model that includes a parabolic velocity distribution, partial-volume averaging, and pulsatile flow, then recover v(z) without access to the true velocities. If the blind RMSE is within 20% of the oracle RMSE and the stenotic peak amplitude is preserved to within 30%, the central claim survives. Alternatively, acquire TOF-MRA and PC-MRI in a steady-flow phantom with a focal stenosis and compare the recovered centerline velocities to the PC-MRI measurements.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is that the validation is not a blind test. In Sections 2.5–3.2, synthetic TOF-MRA signals are generated from the same Bloch–McConnell equation (Eq. 10) that the inversion is built upon, so the simulation only demonstrates that a known integral equation can be stably inverted; it provides no evidence that actual TOF-MRA signals obey that equation. More critically, the regularization weights λ1 and λ2 are chosen by sweeping the parameter space and selecting the combination that minimizes RMSE against the known ground-truth velocities—e.g., the claim that 'λ1=5000, λ2=100 established an optimal balance' is determined by inspecting the true velocity profile. In any real application the ground truth is unknown. The paper itself concedes that 'automated, data-driven parameter selection (e.g., L-curve methods) will be necessary for large-scale clinical application' but does not implement or test such a method. Therefore the reported RMSEs (1.94 cm/s for Scenario A; 7.71 cm/s for Scenario B) are oracle bounds, not achievable without knowing the answer. The central claim that quantitative velocity can be extracted from standard TOF-MRA is thus not established. A secondary but related issue: the model assumes a single plug-flow velocity, whereas the measured voxel signal is an average over the parabolic velocity distribution and partial-volume effects; Section 2.4's 'parabolic laminar flow constraint' does not feed this multi-compartment mixing into the forward signal equation, so the inversion's mapping from signal decay to v is calibrated only to the idealized plug-flow model.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a physics-informed framework to estimate point-wise blood velocity from the 1D centerline signal of standard TOF-MRA. Starting from the conventional Bloch-McConnell expression for flow-related enhancement, the author derives a variable-velocity formulation in which the spatial decay of the signal depends on the transit-time integral (Eq. 10). A dual-Tikhonov regularized inversion is set up to recover the inverse-velocity array from the log-transformed signal (Eqs. 15–18). The method is tested on two synthetic tube geometries—a gradually narrowing tube and a focal stenosis—with ground-truth velocities derived from mass conservation. The paper reports RMSEs of 1.94 cm/s (Scenario A) and 7.71 cm/s (Scenario B), claims these recover the variable velocity profiles, and concludes that quantitative hemodynamics can be extracted from structural TOF-MRA without additional scans. An in vivo example (Figure 3) shows only signal intensity decay along traced centerlines, not recovered velocities.","tokens_in":9127,"tokens_out":3909,"duration_ms":45732,"significance":"If the framework were validated against independent measurements, it could offer a low-cost way to augment routine TOF-MRA with hemodynamic information, avoiding phase-contrast or multi-TE acquisitions. The analytical derivation in Sections 2.1–2.2 is transparent and appears algebraically correct, and the regularized inversion is a reasonable, convex approach to an ill-posed integral equation. The author also openly acknowledges limitations, including the need for data-driven regularization selection and the amplitude attenuation of stenotic peaks. However, the current validation is entirely self-referential: synthetic signals are generated from the same model being inverted, and the regularization weights are chosen using the ground-truth velocity. The reported error figures are thus oracle bounds rather than achievable performance. The central claim—that quantitative velocity can be recovered from real TOF-MRA—is not yet supported.","major_comments":[{"comment":"The validation is an oracle experiment. The regularization weights λ1 and λ2 are selected by sweeping the parameter space and choosing the combination with the lowest RMSE against the ground-truth v(z) used to synthesize the signal. The reported RMSEs (1.94 cm/s and 7.71 cm/s) therefore represent an upper bound on achievable accuracy only when the answer is already known. The paper states that 'automated, data-driven parameter selection (e.g., L-curve methods) will be necessary for large-scale clinical application,' which concedes that the current workflow is not blind. This undermines the quantitative claims in the abstract and conclusions.","section":"Section 3.2 / Table 1 / Discussion (p. 8/13)"},{"comment":"The signal model in Eq. 10 assumes plug flow: a coherent packet of blood moves with a single centerline velocity v(z) and experiences the same RF pulse history. In real TOF-MRA, the voxel signal is an average over a parabolic velocity distribution, partial-volume effects, slice profile, inflow angle, and T2* decay. The 'parabolic laminar flow constraint' mentioned in Section 2.4 is applied only to the velocity field used to generate ground truth; it does not enter the signal generation model. Thus the synthetic validation provides no evidence that actual TOF-MRA centerline intensities obey Eq. 10. A multi-compartment simulation or an in-vivo comparison with phase-contrast velocities would be needed.","section":"Section 2.4 and Eq. 10"},{"comment":"The inversion requires S0 (fresh-blood signal) and Sss (steady-state signal) to be known. These quantities are not directly available in standard clinical TOF-MRA, and they vary spatially due to B1 inhomogeneity, coil sensitivity, and tissue properties. The paper does not propose an estimation strategy for S0 and Sss. Without such a strategy, the practical applicability of Eq. 14 and Eq. 16 is not established; the synthetic experiments assume these values are known exactly.","section":"Eqs. 14 and 16"},{"comment":"The in vivo illustration demonstrates only that signal intensity decreases along extracted centerlines; no velocity profile is recovered or compared with an independent reference. The abstract's claim that 'quantitative, localized functional hemodynamic metrics can be extracted from standard structural MRA imaging' is therefore not supported for real data. The paper is appropriately framed elsewhere as a 'proof-of-concept,' but the in vivo figure does not provide any quantitative validation.","section":"Figure 3 and Abstract"}],"minor_comments":[{"comment":"The superscript T on D1 and D2 is not defined; it should be stated that it denotes matrix transpose.","section":"Eq. 17"},{"comment":"For Scenario B, the lowest RMSE (7.71 cm/s) occurs at λ1=1000, λ2=10, but the text selects λ1=5000, λ2=100 as 'optimal' based on qualitative absence of ringing. The distinction between RMSE-optimal and qualitatively-optimal should be stated explicitly.","section":"Table 1 / Section 3.2"},{"comment":"The manuscript retains MDPI template placeholders such as 'Academic Editor: Firstname Last-name' and the DOI '10.3390/xxxxx'. These must be removed or completed before submission.","section":"General formatting"},{"comment":"The ground-truth velocities are derived from a 1D mass-conservation formula, not from a full computational fluid dynamics simulation. The term 'simulated hemodynamics' should be qualified to avoid overstating the physical realism of the test scenarios.","section":"Section 3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper's central derivation is sound, but the validation is entirely circular and the clinical claim is not supported. A revision would need to implement a blind regularization-selection strategy (or at least a hold-out validation) and ideally include a comparison against phase-contrast MRI on a small in-vivo dataset. Without that, the manuscript is best viewed as a theoretical proof-of-concept rather than a demonstration of the stated clinical capability."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the derivation is fine, the validation is not. The paper derives a point-wise velocity formula from 1D TOF-MRA signal decay (Eq. 14) and sets up a dual-Tikhonov inversion that solves the noisy integral equation. That is a real, if modest, theoretical contribution—the constant-velocity exponential is textbook, but the variable-velocity integral and the inversion framing go beyond the cited qTOF and PC-MRA work. The algebra checks out, and the authors are upfront about the model's idealizations.\n\nThe soft spot is the validation. The synthetic signals are generated by the same Eq. 10 the inversion inverts, so the simulation only shows the inverse of a model can recover the model's own outputs. More seriously, the regularization weights λ1 and λ2 are selected by sweeping against the ground-truth velocity profile and picking the lowest RMSE. That makes the reported 1.94 and 7.71 cm/s errors oracle results, not achievable in a blind setting. The paper concedes that automated parameter selection is needed but doesn't implement it. So the abstract's claim that functional hemodynamics can be extracted from standard structural MRA is not supported by the evidence presented.\n\nThere are also modeling gaps that are mentioned but not dealt with: plug-flow assumption, no partial-volume or slice-profile effects, S0/Sss treated as known. The in vivo example only shows intensity decay along centerlines, not recovered velocity against any reference. None of this kills the idea, but it means the paper is a proof-of-concept at best.\n\nWho's it for? Someone working on qTOF or retrospective hemodynamic analysis might find the derivation useful as a starting point. But it needs a real validation study before the claim is taken seriously. I'd send it to peer review because the derivation is worth checking and the idea is of interest to the MR angiography community, but I would expect the reviewers to demand at least a phantom or in vivo velocity comparison and a data-driven parameter selection scheme.","headline":"The derivation is sound and the inversion is well-posed, but the validation is an oracle experiment: regularization parameters are tuned on ground-truth velocities, so the central claim about extracting functional velocity from clinical TOF-MRA is not yet supported.","tokens_in":9628,"tokens_out":1252,"would_cite":false,"duration_ms":13557,"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":"Quantitative blood velocity can be extracted from the spatial decay of standard TOF-MRA centerline signals.","keywords":["TOF-MRA","blood velocity","Bloch-McConnell equations","flow-related enhancement","effective relaxation rate","Dual-Tikhonov regularization","cerebrovascular imaging"],"falsifier":"Measure velocity in a straight uniform vessel with phase-contrast MRI, extract the centerline TOF-MRA signal from the same subject and sequence, and check whether v(z) computed from Eq. 14 matches the independent measurement; if the recovered velocities deviate by much more than the simulated RMSE (≈2 cm/s tapered, ≈8 cm/s stenotic) at the same noise level, the idealized decay model is falsified for real data.","tokens_in":8603,"feed_emoji":"🩸","tokens_out":8140,"duration_ms":75358,"temperature":0.7,"pith_summary":"Standard Time-of-Flight MRA is normally read as a purely structural image, but this paper argues that the spatial decay of its signal along a vessel contains quantitative blood velocity information. By deriving the Bloch-McConnell flow equations into a one-dimensional decay law, the author shows that a vessel's local speed can be inverted from the intensity slope at each centerline point, stabilized by a dual-penalty regularized optimization. Simulations in a tapering vessel and a focal stenosis under synthetic scanner noise recover ground-truth point-wise velocities, with errors of roughly 2 cm/s for gradual narrowing and 8 cm/s for the sharp stenotic jet. If the framework holds in vivo, velocity measurement would be available from a standard clinical MRA without phase-contrast, multi-echo, or additional scan time.","feed_headline":"Derive blood velocity from standard TOF-MRA alone","feed_subtitle":"A physics-based inversion reads local velocity from the spatial decay of the MRA signal along a vessel, skipping extra phase-contrast scans.","key_machinery":"The key machinery is the effective relaxation rate R_eff = 1/T1 - ln(cosα)/TR, which folds T1 relaxation and RF saturation into a single spatial decay constant, together with the inversion identity v(z) = -R_eff (S(z)-S_ss)/S'(z). The forward model maps velocity history into signal through the transit-time integral, and the dual-Tikhonov global matrix inversion (Eq. 18) replaces unstable numerical differentiation with a stabilized linear solve, allowing pointwise instead of constant velocity.","core_discovery":"The paper's central discovery is a mathematical identity linking the spatial decay of TOF-MRA signal intensity to local blood velocity. The author models a blood packet entering the imaging volume fully relaxed and then being repeatedly excited by RF pulses, which yields an effective relaxation rate R_eff = 1/T1 - ln(cosα)/TR that combines true T1 relaxation with artificial RF-driven saturation. The resulting centerline signal obeys S(z) = S_ss + (S0 - S_ss) exp(-R_eff ∫ dz'/v(z')), and differentiating the log-transformed signal gives the pointwise velocity v(z) = -R_eff (S(z)-S_ss)/S'(z). Because this derivative amplifies noise, the inversion is recast as a globally convex least-squares pro","pith_inferences":["A natural next experiment is to compare the solver's output against phase-contrast MRI or transcranial Doppler in the same vessel segment; any systematic mismatch would identify which neglected physics must be added to the model.","The plug-flow assumption ignores the parabolic or pulsatile profile of real arterial flow; extending the transit-time integral to a distributed velocity field would test the method's sensitivity to flow shape.","The inversion is not limited to TOF-MRA: any inflow-weighted contrast that produces a spatial decay profile, such as arterial spin labeling, could be treated with the same log-derivative identity."],"forward_implications":["Pointwise blood velocity becomes measurable from a standard ~4-minute TOF-MRA acquisition, eliminating the need for phase-contrast or 4D-flow scans that add 5-20 minutes.","Previously acquired TOF-MRA images could be re-analyzed retrospectively, since the inversion needs only the centerline signal profile and the known sequence parameters (TR, flip angle, T1).","The method distinguishes smooth tapering from focal stenosis by reconstructing local velocity, where global constant-velocity fits would only return a mean value.","The global inverse problem is convex with a closed-form solution, so no iterative non-linear fitting is required and the computation can be made fast enough for clinical workflows.","The in vivo centerline decay shown in the paper matches the tapering-tube model, suggesting the framework can be applied to real cerebrovascular networks once the inverse solver is run on those topologies."],"fun_headline_variants":["Blood velocity from TOF-MRA signal decay","Speed from MRA: a mathematical identity","No contrast: velocity from signal attenuation","Read blood flow from MRA intensity curves","Physics inversion yields velocity from TOF-MRA"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the measured TOF-MRA centerline signal is produced exactly by the 1D plug-flow Bloch-McConnell model—fully relaxed inflow at z=0, identical RF pulses, no partial-volume averaging, slice-profile, inflow-angle, or T2* effects, and known S0 and S_ss.","fun_headline_variants_meta":{"raw":{"variants":["Blood velocity from TOF-MRA signal decay","Speed from MRA: a mathematical identity","No contrast: velocity from signal attenuation","Read blood flow from MRA intensity curves","Physics inversion yields velocity from TOF-MRA"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000216,"raw_usage":{"total_tokens":1287,"prompt_tokens":778,"completion_tokens":509,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":522,"completion_tokens_details":{"reasoning_tokens":442}},"tokens_in":522,"tokens_out":509,"duration_ms":5674,"temperature":1.0,"reasoning_tokens":442,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T20:47:34.636094+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure velocity in a straight uniform vessel with phase-contrast MRI, extract the centerline TOF-MRA signal from the same subject and sequence, and check whether v(z) computed from Eq. 14 matches the independent measurement; if the recovered velocities deviate by much more than the simulated RMSE (≈2 cm/s tapered, ≈8 cm/s stenotic) at the same noise level, the idealized decay model is falsified for real data.","supporting_citations":[],"review_version":1}