Temporally Consistent Graph Extraction and Matching for Longitudinal Angiographic Images
Linus Kreitner, Laurin Lux, Carmen Baumann, Daniel Rueckert, Daniel Rueckert, Martin J. Menten
cs.CV
Quick verdict
Sound match-before-refine idea for longitudinal vessel graphs, carefully engineered; evidence stays qualitative on private same-day healthy-eye OCTA data.
Summary
The paper addresses instability in vessel-graph pipelines for longitudinal angiography: minuscule differences in segmentation masks propagate into substantially different graphs, breaking temporal matching. The authors invert the standard extract-refine-match order for pairs of time-adjacent scans: after basic skeleton-based graph extraction, graphs are matched early, and refinement steps (bulge removal, H-junction merging) are then applied jointly, deleting a structure only if its matched counterpart in the other graph also falls below threshold. Matching uses a relaxed path mapping (concatenated edge paths instead of strict 1-to-1 edges), six cost terms, and a binary integer linear program solved with CBC; path matches are decomposed to edge level by inserting degree-2 nodes. Experiments use an in-house OCTA set of 160 images (two same-day scans of 80 healthy eyes) with 608x608 segmentations from a public tool. Unrefined graphs average about 1,300 mostly spurious edges, and the pipeline matches graphs of roughly 2,000 edges. Joint refinement yields higher matched area than separate refinement with stable biomarker agreement (Q = 1 - sMAPE), while the Moriconi et al. baseline matches more vessels at clearly lower biomarker correspondence. Main limitations: private data, no code-release statement, qualitative result reporting, hyperparameters tuned on a single annotated graph pair, and only idealized same-day healthy-eye scans.
Review complete
Assessment at a glance
5
risk flags identified
● high 1
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Small variations in the underlying segmentation map cause substantial divergence between refined vessel graphs, inhibiting longitudinal matching (Sections 1, 2; Figure 4).
- Matching raw graphs early and then applying bulge removal and junction merging jointly yields higher matched area without fragmentation than refining each graph separately (Section 2.4; Figures 4-5).
- Joint refinement maintains stable biomarker agreement (Q = 1 - sMAPE), whereas the Moriconi et al. matching baseline produces more matches at substantially lower biomarker correspondence (Section 3).
- The custom extraction pipeline yields graphs comparable to established tools Voreen and VesselVio in edge counts and average vessel size (Section 3).
Strengths
Well-motivated inversion of pipeline order
Section 2.4 directly targets the diagnosed failure mode: refinement decisions are coupled across time points via a precise rule (bulge/H-bridge removed only if the matched counterpart also satisfies the criterion), promoting convergence instead of divergence.
Engineering-complete matching formulation
Section 2.3 relaxes bijective edge matching to path mapping, tolerating edge splintering; the binary ILP (PuLP/CBC) yields a global optimum rather than greedy errors, and decomposition to 1-to-1 edge matches via inserted degree-2 nodes is a clean mechanism.
Controlled experimental design
Guard metric against trivial coverage gains
Honest scoping
Weaknesses
Qualitative-only result reporting
Tuning and validation on a single annotated pair
Private data, no code statement
Idealized scope versus longitudinal claims
Internal inconsistency and thin ablation
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This MICCAI-style method paper targets a narrow but consequential failure mode in longitudinal angiography analysis: pipelines that extract, refine, and match vessel graphs sequentially are so sensitive to the input segmentation that minuscule mask differences yield substantially different graphs, breaking cross-time correspondence (Section 1). The authors claim three contributions: (1) demonstrating this segmentation-to-graph sensitivity, (2) a reordered pipeline that matches raw graphs early and then refines them jointly (bulge removal plus H-junction merging under a cross-graph coupling rule), and (3) evidence on an in-house OCTA dataset that joint refinement increases matched area without graph fragmentation. The claims are internally coherent and modestly scoped; Section 4 confines generalization to 3D, more than two graphs, and learned extractors to explicit postulates without supporting experiments.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What problem does this paper solve?
In longitudinal angiography (e.g., repeated retinal OCTA scans), tiny differences in vessel segmentation maps produce noticeably different vessel graphs, which breaks the matching of the same vessels across time points. The paper proposes matching the graphs first and then refining them jointly so both time points converge to similar structures.
How does the proposed joint refinement work?
After basic skeleton-based graph extraction, the two graphs of a subject are matched early. Refinement steps — removing spurious bulges and merging H-shaped junction pairs — are then applied using information from both graphs: a structure is removed only if its matched counterpart in the other graph also falls below the removal threshold.
What data were used and are they public?
An in-house dataset of 160 retinal OCTA images: two repeated scans taken on the same day for 80 healthy eyes, with 608x608 segmentation masks produced by a public OCTA segmentation tool. The paper does not state that the dataset will be made public, and no code release is mentioned.
What were the main results?
Unrefined graphs averaged about 1,300 mostly spurious edges. Compared with refining each graph separately, joint refinement increased the total matched vessel area while keeping biomarker agreement (Q = 1 - sMAPE) stable; the Moriconi et al. baseline matched more vessels but with substantially lower biomarker correspondence.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1-5 (pipeline diagrams, quantitative comparisons, qualitative examples) cannot be analyzed from the extracted text; the headline quantitative comparisons exist only in figure form.
- The prose reports very few numeric results, so effect sizes cannot be verified from the text.
- Review is based on the embedded arXiv v1 text; any supplementary material was not included.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.