{"id":"b19f01ce-3e1d-4006-8437-760f386914ae","arxiv_id":"2411.08185","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Combining path-length correction with SVD-based deconvolution stabilizes quantitative angiographic parameters across projection views and injection conditions in in-silico and in-vitro aneurysm models.","lead":"Digital subtraction angiograms of aneurysms are distorted by foreshortening and hand-injection variability. This paper combines 3D-geometry-based path-length correction with SVD deconvolution and reports more stable quantitative angiographic parameters across views and injection rates in simulation and phantom studies.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"In-vitro PLC accuracy rests on unquantified 3D-2D co-registration; no registration error metric is reported, so the observed view-bias corrections could reflect misregistration artifacts rather than true path-length normalization.","rationale":"The paper's strongest claim is that PLC combined with SVD makes quantitative angiographic parameters independent of projection view and injection duration. The reader identified the unquantified 3D-2D co-registration accuracy in the in-vitro pipeline as the weakest assumption, and I agree that this is the most load-bearing concern. The entire PLC correction rests on the geometric correctness of the path-length weight matrix; if the registration is off, the correction is wrong and could introduce artifacts. The in-silico experiments cannot validate the registration because the geometry is perfectly aligned by construction, leaving only the in-vitro results as evidence, and those results lack any registration error metric or sensitivity analysis. The other limitations noted by the reader (small sample, no statistical testing, empirical SVD threshold, no code/data) are real but secondary; they affect generalizability and reproducibility, whereas the registration issue threatens the internal validity of the central correction mechanism. A perturbation test on the in-vitro registration would directly assess whether the observed benefits survive realistic misalignment. If they do, the core claim is strengthened; if they do not, the paper would need to demonstrate that clinically achievable registration accuracy is sufficient. Thus the verdict remains CONDITIONAL, as the method is plausible but not yet fully supported.","tokens_in":7945,"tokens_out":2361,"duration_ms":26589,"concrete_test":"Using the in-vitro phantom dataset, apply controlled rigid perturbations (translations of 1-3 mm and rotations of 1-3 degrees) to the estimated registration transformation before computing the PLC weight matrix, then recompute AIF RMSE (Table 1) and MTT slopes (Figure 6) for one injection rate. If MTT slopes and RMSE reductions remain within a small tolerance (e.g., MTT slope change < 0.02 s per second of injection duration), the PLC is robust to realistic misregistration. If they degrade substantially, report a fiducial-based target registration error measurement and state the required registration accuracy for the method to remain valid.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that PLC plus SVD yields view- and injection-independent QA parameters depends on the path-length normalization being geometrically correct. For in-silico data, the geometry is known by construction, so PLC is exact by design. For in-vitro data (Section 2.2), the 3D reconstruction is aligned to the averaged DSA frame using an affine plus B-spline deformable registration, but the paper reports no registration accuracy metric (e.g., Dice overlap, target registration error, or fiducial-based error) and no sensitivity analysis. Since PLC divides each pixel by a ray-traced vessel-intersection length, a small misalignment can nonuniformly scale the TDCs, altering AIF RMSE and MTT slopes in either direction. The observed post-PLC reductions in RMSE and flattened MTT slopes could therefore be partly produced by the registration procedure itself rather than by true path-length correction. The in-silico experiments cannot rule this out because they assume perfect alignment. Consequently, the evidence does not yet establish that the method will be reliable when 3D-2D registration is imperfect, which is exactly the clinical scenario the conclusion targets.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript proposes a pipeline that combines path-length correction (PLC) with singular value decomposition (SVD) based deconvolution to reduce projection-induced foreshortening and injection-related variability in two-dimensional quantitative angiography. The PLC step co-registers a 3D vascular mask or reconstruction with the 2D DSA projection, ray-traces vessel-intersection lengths, and divides the logarithmic angiogram by this path-length map. Three SVD variants (sSVD, bSVD, oSVD) are then used to deconvolve the aneurysm time-density curve with an arterial input function, yielding impulse response parameters PH_IRF, AUC_IRF, and MTT. The method is evaluated in three in-silico intracranial aneurysm models and one in-vitro patient-specific phantom under different views, flow velocities, and injection durations. The authors report reduced RMSE between AIFs from different views/locations after PLC and flattening of MTT versus injection-duration slopes after PLC+SVD, and conclude that the combination significantly enhances the reliability of quantitative angiographic measurements.","tokens_in":8099,"tokens_out":4468,"duration_ms":43331,"significance":"If the proposed method truly removes projection and injection biases, it would fill a practical clinical need for stable quantitative angiographic measurements during neurovascular procedures. The paper's strengths are its combined in-silico and in-vitro experimental design, the systematic comparison of three standard SVD deconvolution variants, and a workflow that relies on routinely acquired 3D rotational angiography. However, the current validation is almost entirely based on self-consistency metrics—agreement between AIFs and constancy of MTT across injection durations—rather than on comparisons with independent ground truth. In addition, the in-vitro branch of the pipeline depends on an unquantified 3D-to-2D registration step. The contribution is therefore promising and potentially useful, but the central claim of enhanced reliability requires stronger validation before it can be considered established.","major_comments":[{"comment":"The accuracy of the PLC pipeline in the in-vitro experiments depends on the co-registration between the forward-projected 3D reconstruction and the averaged DSA frame, but the paper reports no quantitative registration error (e.g., target registration error, Dice overlap, or manual landmark distances) and no sensitivity analysis to misalignment. Because PLC divides each pixel by a ray-traced path length, even small registration errors could nonuniformly scale the time-density curves and potentially produce or mask the RMSE reductions and slope flattening reported in Section 3. The in-silico experiments assume perfect alignment by construction and therefore cannot rule out such artifacts. Please add a registration accuracy metric and a perturbation study (e.g., applying realistic translations or rotations to the projected mask) to show that the correction is robust.","section":"Section 2.2"},{"comment":"The validation is based entirely on self-consistency metrics: RMSE between AIFs from different views/locations and the flattening of MTT versus injection-duration slopes. There is no comparison to a gold-standard measurement, even though the in-silico CFD simulations provide an independent ground-truth transit time that could be computed directly from particle trajectories, and the in-vitro phantom could potentially be assessed with an independent measurement technique. Without such a comparison, the claim that PLC+SVD 'significantly enhances the reliability of quantitative angiographic measurements' is not fully supported; the method could be imposing an internally consistent but biologically or physically incorrect result. Please include direct comparisons with ground-truth MTT or other independent quantities where available.","section":"Section 3 (Table 1 and Figures 5-6)"},{"comment":"The SVD truncation threshold is described as 'determined empirically, retaining singular values above a certain percentage of the maximum value,' but the manuscript does not specify the exact selection rule, the range of thresholds tested, or whether the same threshold was applied across all cases. If the threshold was chosen by inspecting the same datasets used to compute the MTT slopes, the reported flattening could partly reflect tuning rather than genuine removal of injection bias. Please provide a principled threshold selection criterion (e.g., based on noise level or the L-curve) and report the sensitivity of the MTT results to the threshold choice.","section":"Section 2.3"},{"comment":"The in-vitro conclusions are derived from a single patient-specific phantom, two views, and no repeated acquisitions, and the reported MTT slopes are means across views without measures of variability or statistical tests. The statement that the method 'nearly eliminated' injection effects is therefore based on a very small sample. Please report per-view slopes, standard errors, and formal tests of whether the slopes are consistent with zero, or explicitly state the exploratory nature of these in-vitro results.","section":"Section 3 (in-vitro results)"}],"minor_comments":[{"comment":"The phrase 'path-length correction (PLC) correction' is redundant; the second 'correction' should be deleted.","section":"Abstract and Section 1"},{"comment":"The term 'corregistration' is used in several places; it should be 'co-registration.'","section":"Section 2.2 and Figure 3 captions"},{"comment":"The caption states that RMSEs are averaged across three in-silico models but does not indicate the spread of the values; please include standard deviations or error bars so the reader can assess consistency across models.","section":"Table 1"},{"comment":"The hypothesis tests are described only qualitatively (e.g., 'significant reduction in RMSE') without statistical tests; please add confidence intervals or p-values, or soften the wording to describe observed reductions rather than significant effects.","section":"Section 2.4"},{"comment":"The captions do not define the line colors and styles in the figures (e.g., which lines correspond to 'without PLC and deconvolution,' 'SVD alone,' and 'PLC+SVD'); please ensure every plot element is identified in the caption or legend.","section":"Figures 5 and 6"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal and addresses a practical problem in quantitative angiography. My central concerns are the absence of ground-truth validation and the unquantified 3D-to-2D registration error in the in-vitro branch. These are not fatal flaws, but they are load-bearing for the paper's main claim, and I would like to see them addressed before publication. The paper would also benefit from a clear statement that the in-vitro results are preliminary given the single-phantom design."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about arXiv:2411.08185. First, the paper is a straightforward integration of the same group's earlier path-length correction (ref 8) and SVD-based injection variability removal (ref 6), now applied together on new in-silico and in-vitro cases. Second, the validation is self-consistency, not accuracy: reduced AIF RMSE across views and flattened MTT versus injection-duration slopes. That is a reasonable demonstration that the pipeline removes the intended biases, but it does not show that the corrected parameters are closer to true hemodynamics.\n\nWhat the paper does well: the in-silico experiments are clean geometry is known by construction, so PLC is exact there and the reported stability of AIFs and MTT across views and injection durations is genuine evidence that the concept works in simulation. The in-vitro setup uses a real biplane system and a patient-specific phantom, which is a step beyond pure simulation. The writing is clear and the limitations section is honest about the need for clinical validation. The authors do not overclaim in the abstract; they claim improved reliability, which is supported by their metrics.\n\nThe soft spots are real but not fatal. Most importantly, the stress-test concern about 3D-2D co-registration for the in-vitro data is on target: the paper reports no registration error (Dice, TRE, or similar) and no sensitivity analysis for misalignment. Because PLC divides by ray-traced path lengths, a small misregistration could nonuniformly scale the TDCs and produce exactly the kind of RMSE reduction and slope flattening they report. The in-silico experiments cannot rule this out because they assume perfect alignment. This does not invalidate the method, but it does mean the clinical reliability claim is not yet established.\n\nSecond, the MTT slope flattening after SVD is partly tautological: if the deconvolution successfully removes injection effects by construction, the slope should go to zero. A ground-truth comparison, e.g., MTT from the CFD simulation or from an independent flow measurement, would have made the result much stronger. Third, the SVD truncation threshold is described as empirically tuned and never disclosed, which makes the experiments hard to reproduce. No code or data are provided.\n\nOverall, this is a solid incremental paper from a group that knows this problem well. The integration is legitimate and the in-silico evidence is decent. But the current evidence supports \"the method behaves consistently,\" not \"the measurements are reliable in the clinic.\" It deserves peer review because the underlying question matters and the in-vitro gap is addressable: report registration error, add a sensitivity analysis, and compare against a gold-standard MTT. The paper would be stronger after major revision.\n\nIf you work on quantitative angiography or intraprocedural hemodynamics, this is worth reading. If not, you can skip it. It is not a breakthrough, but it is neither sloppy nor dishonest.","headline":"A legitimate integration of the group's own PLC and SVD methods with clear self-consistency evidence, but the clinical claims outrun the validation, especially on 3D-2D registration and ground truth.","tokens_in":8703,"tokens_out":1347,"would_cite":false,"duration_ms":15912,"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":"Two-dimensional quantitative angiography can be made view- and injection-independent by dividing the angiogram by a 3D-derived path-length map, then deconvolving the dome curve with the arterial input function.","keywords":["quantitative angiography","digital subtraction angiography","path-length correction","foreshortening bias","injection variability","singular value decomposition","deconvolution","intracranial aneurysm hemodynamics"],"falsifier":"Acquire DSA in two views at two injection rates for a single phantom, then deliberately mis-register the 3D mask by increasing translational offsets (for example, 1, 2, and 5 mm) and recompute the mean transit time; if the mean-transit-time-versus-duration slope grows with mis-registration rather than staying flat, the claimed injection independence is bounded by registration accuracy and fails outside that tolerance.","tokens_in":7713,"feed_emoji":"🩻","tokens_out":13489,"duration_ms":112074,"temperature":0.7,"pith_summary":"This paper argues that two known biases in 2D quantitative angiography — foreshortening, where the X-ray path through a vessel changes with viewing angle, and injection variability, where manual contrast delivery changes the time-density curves — can be removed by a two-step correction pipeline. The first step divides the logarithmic angiogram by a 3D-derived path-length map, converting projection intensity into something closer to contrast concentration. The second step deconvolves the aneurysm's time-density curve with the arterial input function using SVD-based methods, stripping the injection waveform from the hemodynamic signal. In in-silico aneurysm models and a patient-specific in-vitro phantom, the corrected parameters — peak height, area under the curve, and mean transit time — become nearly independent of projection view and injection duration, with root-mean-square errors between arterial input functions falling sharply and mean-transit-time slopes versus injection duration approaching zero. If true, the same lesion can be compared across different C-arm angles and injection runs during a single procedure, which would make quantitative angiography a more trustworthy intraoperative tool.","feed_headline":"Aneurysm blood-flow readings become view- and injection-independent","feed_subtitle":"A geometry-aware correction plus deconvolution makes aneurysm flow metrics stable across views and injection rates.","key_machinery":"The load-bearing object is the path-length weight matrix: for each detector ray, the length of its intersection with the co-registered 3D vessel mask, obtained by forward-projecting the reconstruction onto the DSA geometry and ray tracing. Dividing the log-angiogram by this weight matrix converts the signal from path-length-weighted absorption into a concentration-like signal, which is what removes view dependence. The second mechanism is SVD deconvolution of the aneurysm dome time-density curve by the arterial input function, which removes the injection waveform; the paper compares sSVD, bSVD, and oSVD variants. The corrected impulse response function then yields the three parameters that the paper treats as the stable hemodynamic readouts.","core_discovery":"The authors' central claim is that the projection and injection biases that make quantitative angiography unstable are not intrinsic to the hemodynamics, and that they can be corrected with a co-registered 3D geometry plus deconvolution. After ray tracing through the 3D vessel mask to build a path-length weight matrix, each logarithmically subtracted angiographic frame is divided by that matrix to remove foreshortening. Then the dome time-density curve is deconvolved by the arterial input function with one of three SVD variants (standard Tikhonov-regularized, block-circulant, and oscillation-index), producing an impulse response function whose peak height, area under the curve, and mean transit time are reported. The paper presents RMSE reductions between arterial input curves from frontal and lateral views after path-length correction, and $\\mathrm{MTT}$-versus-injection-duration slopes near zero after the combined correction: $0.015 \\pm 0.017$, $0.014 \\pm 0.032$, and $0.013 \\pm 0.025$ for sSVD, bSVD, and oSVD in silico, and $0.031 \\pm 0.015$, $0.047 \\pm 0.031$, and $0.013 \\pm 0.004$ in vitro.","pith_inferences":["A testable extension would use the residual mean-transit-time-versus-duration slope as a per-case quality check: a large slope after correction would flag a failed 3D-to-2D co-registration or an incomplete vessel mask.","The same path-length weight matrix could be applied to other DSA-derived biomarkers, such as parametric color maps or wash-in/wash-out indices, not just the three impulse-response parameters studied here.","Registration accuracy is the unquantified tolerance: perturbing the 3D-to-2D alignment by millimeters and measuring the resulting mean-transit-time drift would define the clinical geometry requirement.","The method depends on having a pre-existing 3D volume, so workflows that acquire only biplane DSA would need a surrogate depth model before this correction could be applied."],"forward_implications":["Post-correction arterial input functions from frontal and lateral views agree, so angiographic comparisons no longer need to be restricted to a single projection angle.","Mean transit time stays nearly flat over injection durations from 0.25 to 2 seconds, so the speed of a 5 ml bolus does not change the hemodynamic readout.","The pipeline was demonstrated on both CFD-generated virtual angiograms and a physical patient-specific phantom, suggesting it can be transferred to clinical C-arm DSA data.","Stable parameters allow pre- and post-treatment comparisons to be made even when the C-arm angle or injection settings differ between runs, which is the common intraoperative situation.","All three SVD variants flattened the mean-transit-time dependence, so the choice among them can be made on noise handling, computational cost, or other practical grounds."],"supporting_citations":[{"why":"Defines the patient-specific phantom circulation setup (pump, contrast injector, DSA acquisition) that the in-vitro validation is built on.","marker":"[3]"},{"why":"Supplies the path-length-corrected 2D angiographic parametric imaging method and the ray-weighted projection concept that the PLC step extends.","marker":"[8]"},{"why":"Provides the three SVD deconvolution variants and the definitions of the impulse-response parameters used as outcomes.","marker":"[6]"},{"why":"Gives the Tikhonov regularization background for the standard SVD variant.","marker":"[12]"},{"why":"Provides the singular-value properties of block-circulant matrices that underpin bSVD and oSVD.","marker":"[13]"},{"why":"Supplies the oscillation-index block-circulant deconvolution technique for tracer-arrival-insensitive flow estimation.","marker":"[17]"}],"fun_headline_variants":["Path-length correction ends view and injection bias in aneurysm flow readings","Geometry-aware deconvolution stabilizes aneurysm flow metrics","Aneurysm flow metrics made view- and injection-invariant","New correction removes biases from quantitative angiography","Combined corrections make aneurysm flow readings stable across angles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire correction rests on the 3D vessel geometry being accurately co-registered with the 2D DSA frame; if that alignment is off, the ray-traced path-length map is wrong and the correction could inject new artifacts instead of removing bias.","fun_headline_variants_meta":{"raw":{"variants":["Path-length correction ends view and injection bias in aneurysm flow readings","Geometry-aware deconvolution stabilizes aneurysm flow metrics","Aneurysm flow metrics made view- and injection-invariant","New correction removes biases from quantitative angiography","Combined corrections make aneurysm flow readings stable across angles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001008,"raw_usage":{"total_tokens":4328,"prompt_tokens":1077,"completion_tokens":3251,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":693,"completion_tokens_details":{"reasoning_tokens":3174}},"tokens_in":693,"tokens_out":3251,"duration_ms":24549,"temperature":1.0,"reasoning_tokens":3174,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:52:09.773224+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Acquire DSA in two views at two injection rates for a single phantom, then deliberately mis-register the 3D mask by increasing translational offsets (for example, 1, 2, and 5 mm) and recompute the mean transit time; if the mean-transit-time-versus-duration slope grows with mis-registration rather than staying flat, the claimed injection independence is bounded by registration accuracy and fails outside that tolerance.","supporting_citations":[{"cited_title":"Effect of injection technique on temporal parametric imaging derived from digital subtraction angiography in patient specific phantoms","cited_arxiv_id":null,"evidence_quote":"Defines the patient-specific phantom circulation setup (pump, contrast injector, DSA acquisition) that the in-vitro validation is built on."},{"cited_title":"Enhancing cerebral vasculature analysis with pathlength-corrected 2D angiographic parametric imaging: A feasibility study","cited_arxiv_id":null,"evidence_quote":"Supplies the path-length-corrected 2D angiographic parametric imaging method and the ray-weighted projection concept that the PLC step extends."},{"cited_title":"Effect of singular value decomposition on removing injection variability in 2D quantitative angiography: An in silico and in vitro phantoms study","cited_arxiv_id":null,"evidence_quote":"Provides the three SVD deconvolution variants and the definitions of the impulse-response parameters used as outcomes."},{"cited_title":"On Tikhonov regularization, bias and variance in nonlinear system identification","cited_arxiv_id":null,"evidence_quote":"Gives the Tikhonov regularization background for the standard SVD variant."},{"cited_title":"On singular values of block circulant matrices","cited_arxiv_id":null,"evidence_quote":"Provides the singular-value properties of block-circulant matrices that underpin bSVD and oSVD."},{"cited_title":"Tracer arrival timing- insensitive technique for estimating flow in MR perfusion-weighted imaging using singular value decomposition with a block-circulant deconvolution matrix","cited_arxiv_id":null,"evidence_quote":"Supplies the oscillation-index block-circulant deconvolution technique for tracer-arrival-insensitive flow estimation."}],"review_version":1}