{"id":"c0180b6c-7f64-4602-b52e-8895740d1afb","arxiv_id":"2501.11793","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A self-calibrated epipolar reconstruction algorithm reconstructed carotid artery aneurysms from simulated biplane angiograms with Dice scores of 0.654 to 0.759.","lead":"This paper presents software that builds a 3D reconstruction of an artery and aneurysm from two standard x-ray views, using the motion of injected contrast dye to calibrate the two cameras without a calibration object. It could eventually reduce the need for extra CT scans during aneurysm treatment, but so far it has only been tested on computer-simulated angiograms.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The method's self-calibration assumes equal axial contrast transit in both views; the authors concede that routine cranial/caudal gantry angulation violates this, and no correction is implemented, so the reported in-silico Dice scores may not transfer to clinical biplane data.","rationale":"The reader identified the same load-bearing assumption: equal axial contrast transit in both biplane views, which is the foundation of the self-calibration step. My stress-test pass finds no internal contradiction in the in-silico demonstration; the method is coherent, the simulations are described in adequate detail, and the authors honestly list real-data issues. However, the central claim of applicability to 'routinely acquired biplane angiographic data' is not supported because the one mechanism that replaces a calibration object is precisely the condition that routine gantry angulation breaks. The authors' own Discussion concedes this and defers the fix to future work. This does not change the reader's CONDITIONAL verdict: the paper is a plausible, well-documented simulation study, but the promise of clinical utility requires testing with non-ideal gantry geometry or real data. I set verdict_should_be to UNCHANGED to indicate that the stress-test corroborates the reader's assessment rather than moving it.","tokens_in":10175,"tokens_out":3591,"duration_ms":41819,"concrete_test":"Re-run the pipeline on the same three CFD models, but simulate a 10° cranial angulation of the lateral gantry relative to the frontal gantry in ASTRA, keeping all other settings identical. If the self-calibration step (Section 2.2) cannot equalize magnification and FOV without an affine correction, the Dice-Sorensen coefficients should drop measurably (e.g., by more than 0.1) or the reconstruction should show obvious axial misalignment. If Dice values remain essentially unchanged, the concern would be refuted in this regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—an automatic, calibration-object-free reconstruction from routinely acquired biplane angiograms—depends on the Section 2.2 assumption that 'biplane imaging systems share one common axis between the two imagers (typically in the axial direction).' On this basis, the ratio of contrast bolus transit distances l and l' is used to equalize magnification, and residual discrepancies are attributed to FOV misalignment. If the two imagers have different cranial or caudal angulation, the projected axial advancement in each view is foreshortened by different cosines of the gantry angles, so the computed scaling factor k is biased. The axial alignment step then misaligns corresponding slices, corrupting feature matching, essential matrix estimation, and the final reconstruction. The authors explicitly acknowledge in the Discussion that 'cranial or caudal angulation in either gantry may introduce bias in the magnification or FOV equalization steps' and suggest affine transformations as future work, but such corrections are not implemented or tested. Since the in-silico experiments use idealized common-axis geometry, the reported Dice coefficients of 0.745, 0.759, and 0.654 do not demonstrate robustness to this routine clinical condition. This is an external-validity limitation of the central claim, not an internal inconsistency of the algorithm as tested.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a fully automatic, calibration-object-free epipolar reconstruction pipeline for three-dimensional visualization of internal carotid artery (ICA) aneurysms from biplane angiograms. The method equalizes magnification and field of view between the two views by assuming that the axial advancement of a contrast bolus over a fixed time interval is identical in both projections, then performs feature matching, essential-matrix estimation, back-projection, and ellipse fitting to produce a 3D vessel mask. The pipeline is evaluated on three CFD-derived virtual angiograms generated from patient-specific aneurysm geometries, with reported Dice-Sorensen coefficients of 0.745, 0.759, and 0.654 and an average reconstruction time of about 10 seconds. The authors conclude that the method enhances 3D visualization from routine biplane data and could reduce the need for additional CTA or CBCT acquisitions.","tokens_in":10477,"tokens_out":5722,"duration_ms":62188,"significance":"If the central claim is correct, the proposed method could provide a practical, intraprocedural 3D reconstruction capability for neurointerventional suites without requiring calibration objects or additional imaging. The in-silico validation against exact CFD ground truth geometries is a legitimate proof-of-concept design, and the described pipeline is explainable and non-iterative, which is appealing for clinical translation. However, the study is preliminary: it tests only three models, reports no variance or error bars, lacks a comparison with existing reconstruction methods, and its central self-calibration assumption is violated by routine gantry angulations. The reported Dice values should therefore be interpreted as a feasibility demonstration under idealized geometry rather than evidence of robustness to clinical conditions.","major_comments":[{"comment":"The self-calibration step assumes that the axial advancement of contrast is identical in both views, which requires the two imagers to share a common axis without relative cranial/caudal angulation. The authors explicitly acknowledge in the Discussion that 'cranial or caudal angulation in either gantry may introduce bias in the magnification or FOV equalization steps' and suggest affine transformations as future work, but no correction is implemented or tested. Since such angulation is routine in clinical biplane neuroangiography, the reported Dice coefficients from common-axis simulations do not support the abstract's claim that the method works on 'standard, routinely acquired biplane angiographic data.' This is a load-bearing external-validity limitation; the authors must either implement and validate a correction for gantry angulation or substantially narrow the claims to common-axis acquisitions and justify that such acquisitions are representative of the intended clinical setting.","section":"Section 2.2; Discussion, fourth paragraph"},{"comment":"The Dice-Sorensen evaluation protocol is not described. After self-calibration, the reconstruction is expressed in the rescaled image pixel grid, but the method does not determine an absolute physical scale because magnification is only equalized between views, not determined in absolute units. The manuscript does not explain how the reconstructed mask is aligned, scaled, and resampled to the CFD ground-truth volume in order to compute the reported Dice values (0.745, 0.759, 0.654). Without specifying the voxel grid, coordinate registration procedure, and binarization threshold, these numeric results cannot be reproduced or interpreted. This is a load-bearing methodological gap because the Dice coefficients are the primary evidence supporting the central claim.","section":"Section 2.4, final paragraph; Section 3.1"},{"comment":"The claim that the method 'generalized well across the three tested aneurysm models' is weakly supported. Only three models are tested, with no variance measures, no repeated runs, no sensitivity analysis, and no morphological description of the aneurysm geometries (e.g., size, neck width, tortuosity, degree of vessel overlap). Additionally, no comparison is made to existing epipolar reconstruction methods or to an alternative 3D technique such as 3D DSA or CBCT. At minimum, the limitations of a three-model proof-of-concept should be explicitly acknowledged, and the generalizability claim should be tempered accordingly.","section":"Section 3.1"}],"minor_comments":[{"comment":"The definition of the scaling factor k is ambiguous: the text states the primed system is rescaled by the ratio of transit distances l and l', but it is not clear whether k = l/l' or k = l'/l. Please state the direction explicitly and ensure the rescaling operation in the text matches the figures.","section":"Section 2.2, Figures 2 and 3"},{"comment":"The expressions for the semi-major and semi-minor axes a and b appear garbled by typesetting and are difficult to verify. Please rewrite Eq. (7) with unambiguous notation and check the algebra, as this equation is central to the ellipse-fitting refinement.","section":"Section 2.4, Eq. (7)"},{"comment":"The abstract states that the method uses 'standard, routinely acquired biplane angiographic data,' while the simulations are performed at an effective 1000 fps frame rate (1 ms time steps). The Discussion notes that the method may be difficult at clinical frame rates such as 3 fps. Please specify the simulated frame rate in the Methods and qualify the abstract's applicability claim accordingly.","section":"Abstract and Section 2.1"},{"comment":"The use of the essential matrix solver to estimate rotation and translation is described only briefly. The essential matrix yields the relative pose up to an unknown scale; the manuscript does not explain how this scale is determined for back-projection into a common volume. Please clarify whether absolute scale is used or whether the reconstruction is generated in an unscaled coordinate frame.","section":"Section 2.3"},{"comment":"The captions for Figures 8 and 9 do not describe which model or axial slices are shown, nor the color coding for true positives, false positives, and false negatives in a consistent manner. Please make the captions self-contained and define all color labels explicitly.","section":"Section 3.1, Figures 8 and 9"},{"comment":"The equation for x-ray attenuation is numbered as Eq. (12), but the previous numbered equations in Section 2.4 stop at Eq. (8). Please renumber the equations sequentially throughout the manuscript.","section":"Discussion, Eq. (12)"}],"recommendation":"major_revision","confidential_remarks":"The paper presents a promising proof-of-concept for self-calibrated epipolar reconstruction, but the gap between the idealized simulation assumptions and the clinical claims is substantial. The most serious issues are the unaddressed sensitivity to gantry angulation—which the authors themselves acknowledge—and the missing description of the Dice evaluation methodology. Both are fixable in revision. Also, the limited three-model validation without variance or comparisons to existing methods makes the 'generalized well' claim premature. I would suggest the authors either implement a correction for oblique gantry angles, perhaps along the lines of the affine-transformation idea they mention, or clearly frame the contribution as a feasibility study for common-axis systems only."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a legitimate incremental contribution. The new bit is using contrast bolus transit distance to self-calibrate magnification between the two views, which is clever and reasonably well motivated. Combined with structural projection filtering and ellipse fitting, the pipeline runs in about ten seconds and needs no calibration object. The description is detailed enough to reproduce, and the validation against the exact CFD ground truth is appropriate. The authors are also unusually candid about the limitations, which makes the paper more useful than many similar method papers.\n\nThe main soft spot is well known: the self-calibration assumes the axial advancement of contrast is the same in both views, which only holds when the two gantries share the same axial axis and have no cranial/caudal angulation. The authors explicitly acknowledge this in the Discussion and suggest affine transformations as future work, but they don't test or correct it. So the reported Dice scores are for an idealized geometry, and the clinical utility claim is not yet supported. There are also just three simulated models, no error bars or sensitivity analysis, and no comparison with existing epipolar or 3D reconstruction methods, so it's hard to know where this stands relative to prior work. The quantum mottle and x-ray attenuation issues for real data are noted but not addressed.\n\nThat said, the internal reasoning is sound. The circularity concern does not hold: the reconstruction uses only the projection data, and the self-calibration factor is estimated from an independent observable. The paper does not overclaim; it frames the work as a first step toward clinical use.\n\nWho is this for? Researchers working on 3D reconstruction from biplane angiography, especially those interested in calibration-free pipelines. It deserves a serious referee because the method is clear, the validation is honest if thin, and the limitations are made explicit. A referee should push for real-data validation or at least for experiments with varied gantry angles and a baseline comparison. I would not desk-reject this.","headline":"Solid, honest in-silico demonstration of a fast self-calibrated epipolar reconstruction pipeline for ICA aneurysms, but the main calibration assumption is untested on real gantry angulation and validation is lean (3 models, no variance).","tokens_in":10958,"tokens_out":1786,"would_cite":true,"duration_ms":20220,"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":"This paper establishes a fully automatic, calibration-object-free epipolar reconstruction pipeline that recovers 3D internal-carotid-artery and aneurysm geometry from simulated biplane angiograms in about 10 seconds, with Dice-Sorensen…","keywords":["biplane angiography","epipolar reconstruction","intracranial aneurysm","internal carotid artery","self-calibration","Dice-Sorensen coefficient","computational fluid dynamics","3D vessel reconstruction"],"falsifier":"Take a biplane sequence of a phantom with known vessel geometry and tilt one gantry cranially; if the reconstructed volume's Dice coefficient against the known geometry falls markedly compared with the non-tilted case, the shared-axis bolus assumption is the limiting step, not the triangulation. A direct measurement of the ratio of bolus advance distances in the two views across the sequence would show whether the ratio stays near one as assumed.","tokens_in":9923,"feed_emoji":"🩻","tokens_out":6096,"duration_ms":58958,"temperature":0.7,"pith_summary":"This paper argues that a fully automatic, calibration-object-free epipolar reconstruction can turn standard biplane angiograms into a usable 3D view of the internal carotid artery and its aneurysm. It establishes the method on three simulated patient-derived geometries, generating virtual angiograms with CFD and cone-beam projection, and reports Dice-Sorensen coefficients of 0.745, 0.759, and 0.654 against the ground-truth models. The point of the work is that 3D vessel morphology, useful for aneurysm assessment and device sizing, does not require CBCT or an external calibration object; it can be derived from the routine two-view images themselves, with average reconstruction time around 10 seconds. A sympathetic reader would take the paper as showing that the main obstacle to intraprocedural 3D guidance is not geometry recovery but bolus and timing assumptions plus vessel-overlap disambiguation.","feed_headline":"Biplane angiograms yield 3D aneurysm views in 10 seconds","feed_subtitle":"A calibration-object-free pipeline reconstructs carotid-aneurysm geometry from standard two-view angiography, validated on in-silico cases.","key_machinery":"The load-bearing mechanism is the contrast-bolus self-calibration step: because biplane systems share one axial axis, the axial advance of contrast between two time frames should be identical in both views after pixel pitch equalization, so the ratio of measured advances yields the magnification scaling factor k, and the residual distance from the image bottom to the bolus top gives field-of-view alignment. This removes the calibration object and makes all downstream steps operate on geometrically consistent images: essential-matrix estimation via the OpenCV solver, back-projection intersection into bounding boxes, structural projection filtering, and ellipse fitting driven by single-integral line-profile areas.","core_discovery":"The central claim is that the two views of a biplane angiogram contain enough self-referential information to calibrate each other: by matching the axial distance a contrast bolus travels between corresponding time points, the algorithm equalizes pixel size and field of view across imagers without knowing magnification a priori. With the views scaled and aligned, feature matching supplies the essential matrix, back-projection of vessel masks yields per-slice bounding boxes, structural projection filtering removes false candidates using temporal maximum-intensity-projection thickness estimates, and a parametric ellipse fit, constrained by line-profile area integration, refines each cross-section. The result is a non-iterative, roughly 10-second 3D reconstruction that the paper validates on three in-silico aneurysm models with Dice coefficients 0.745, 0.759, and 0.654, with error concentrated where vessels overlap in projection.","pith_inferences":["Inference: if the self-calibration transfers to clinical frame rates near 3 fps, iterative magnification estimation would likely be needed because bolus advancement between frames becomes coarser; this is a testable extension the authors note as future work.","Inference: the bolus-transit equality could be adapted to single-plane rotational angiography by using cardiac phase-gated frames as pseudo-second views, though the shared-axis assumption would need re-derivation.","Inference: the reported Dice values on simulated data set an upper bound for clinical performance; real-image quantum mottle, patient motion, and incomplete contrast filling will likely degrade overlap disambiguation before they degrade the geometry solver.","Inference: the ellipse-fitting with line-profile area could be extended to non-elliptical cross-sections or to estimate lumen area for flow quantification, replacing the closed-form ellipse solve with the same bounding-box-and-area principle."],"forward_implications":["Standard biplane angiography, without CBCT or a calibration object, can produce a 3D vessel estimate accurate enough to show the aneurysm dome and parent vessel, with Dice coefficients of 0.745, 0.759, and 0.654 in simulation.","Because the reconstruction is non-iterative and rule-based, it runs in about 10 seconds, making intraprocedural use during neurointervention plausible.","The main failure mode is partial vessel overlap in projection, so clinical application would target vasculature clearly visible in both views or require an overlap-resolving refinement.","The method could reduce the need for CTA acquisitions and patient transfers between imaging suites by deriving 3D morphology from routine biplane images.","The 3D reconstruction could serve as input for quantitative angiography and device-sizing analyses that require depth information."],"supporting_citations":[{"why":"Supplies the essential-matrix algorithm for reconstructing a scene from two projections, the theoretical basis for the epipolar step.","marker":"[9]"},{"why":"Provides the OpenCV essential-matrix solver and rotation/translation estimation used in the pipeline.","marker":"[15]"},{"why":"Introduced bolus arrival time and centerline velocity constraints for vascular epipolar reconstruction, which this method adapts for magnification equalization.","marker":"[6]"},{"why":"Provides the CFD-based 4D angiogram simulation parameters that generate the synthetic test data used for validation.","marker":"[12]"},{"why":"Offers iterative biplane geometry estimation, the fallback the authors suggest for low-frame-rate clinical sequences.","marker":"[24]"},{"why":"Provides the exponential-attenuation correction model needed to relate real x-ray intensities to vessel thickness in clinical data.","marker":"[26]"}],"fun_headline_variants":["Self-calibrated biplane angiograms reconstruct aneurysms in 10s","Two-view angiograms self-calibrate to yield 3D aneurysm geometry in seconds","Fast self-calibrating 3D aneurysm reconstruction from standard biplane views","Aneurysm 3D from biplane angiograms without calibration objects in 10 seconds","Biplane angiograms self-calibrate for 3D aneurysm reconstruction in 10 seconds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes the contrast dye moves the same distance along the shared head-to-foot direction in both X-ray views over the same time interval, which breaks if either imaging arm is angled toward the head or feet.","fun_headline_variants_meta":{"raw":{"variants":["Self-calibrated biplane angiograms reconstruct aneurysms in 10s","Two-view angiograms self-calibrate to yield 3D aneurysm geometry in seconds","Fast self-calibrating 3D aneurysm reconstruction from standard biplane views","Aneurysm 3D from biplane angiograms without calibration objects in 10 seconds","Biplane angiograms self-calibrate for 3D aneurysm reconstruction in 10 seconds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000653,"raw_usage":{"total_tokens":3052,"prompt_tokens":1061,"completion_tokens":1991,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":677,"completion_tokens_details":{"reasoning_tokens":1881}},"tokens_in":677,"tokens_out":1991,"duration_ms":14239,"temperature":1.0,"reasoning_tokens":1881,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T17:50:58.914764+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a biplane sequence of a phantom with known vessel geometry and tilt one gantry cranially; if the reconstructed volume's Dice coefficient against the known geometry falls markedly compared with the non-tilted case, the shared-axis bolus assumption is the limiting step, not the triangulation. A direct measurement of the ratio of bolus advance distances in the two views across the sequence would show whether the ratio stays near one as assumed.","supporting_citations":[{"cited_title":"A computer algorithm for reconstructing a scene from two projections","cited_arxiv_id":null,"evidence_quote":"Supplies the essential-matrix algorithm for reconstructing a scene from two projections, the theoretical basis for the epipolar step."},{"cited_title":"The OpenCV Library","cited_arxiv_id":null,"evidence_quote":"Provides the OpenCV essential-matrix solver and rotation/translation estimation used in the pipeline."},{"cited_title":"Angiographic velocimetry analysis using contrast dilution gradient method with a 1000 frames per second photon-counting detector","cited_arxiv_id":null,"evidence_quote":"Provides the CFD-based 4D angiogram simulation parameters that generate the synthetic test data used for validation."},{"cited_title":"Improved determination of biplane imaging geometry from two projection images and its application to three‐dimensional reconstruction of coronary arterial trees","cited_arxiv_id":null,"evidence_quote":"Offers iterative biplane geometry estimation, the fallback the authors suggest for low-frame-rate clinical sequences."},{"cited_title":"A Model-Based Consecutive Scanline Tracking Method for Extracting Vascular Networks From 2-D Digital Subtraction Angiograms","cited_arxiv_id":null,"evidence_quote":"Provides the exponential-attenuation correction model needed to relate real x-ray intensities to vessel thickness in clinical data."}],"review_version":1}