pith. sign in

arxiv: 2605.20073 · v1 · pith:ATKSPVMOnew · submitted 2026-05-19 · 💻 cs.CV

X-Ray cardiac angiographic vessel segmentation based on pixel classification using machine learning and region growing

E O Rodrigues , L O Rodrigues , J J Lima , D Casanova , F Favarim , E R Dosciatti , V Pegorini , L S N Oliveira
show 1 more author
F F C Morais
This is my paper

Pith reviewed 2026-05-20 05:45 UTC · model grok-4.3

classification 💻 cs.CV
keywords vessel segmentationangiogrampixel classificationrandom forestsregion growingcardiac imagingx-ray angiographymachine learning
0
0 comments X p. Extension
pith:ATKSPVMO Add to your LaTeX paper What is a Pith Number? 
\usepackage{pith}
\pithnumber{ATKSPVMO}

Prints a linked pith:ATKSPVMO badge after your title and writes the identifier into PDF metadata. Compiles on arXiv with no extra files. Learn more

The pith

Pixel classification with textural features and region-growing control segments vessels in X-ray angiograms at 95.48 percent accuracy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes classifying each pixel in X-ray cardiac angiograms as vessel or background using features drawn from its local neighborhood. Features include anisotropic diffusion outputs, Hessian matrix properties, mathematical morphology operations, and basic statistics. These feed a Random Forests classifier whose decisions are steered by the ELEMENT region-growing process so that early classifications influence the classification of nearby pixels. The resulting hybrid method reaches 95.48 percent accuracy, higher than previously reported unsupervised techniques on the same task. A reader would care because reliable vessel maps from routine angiograms could support faster and more consistent cardiac diagnosis.

Core claim

The central claim is that extracting textural features such as anisotropic diffusion, Hessian-based, morphological and statistical descriptors from each pixel neighborhood, then feeding them to a Random Forests classifier whose output is controlled by the ELEMENT region-growing mechanism, produces vessel segmentations in X-ray angiograms with 95.48 percent accuracy that exceeds unsupervised state-of-the-art results.

What carries the argument

The ELEMENT methodology, a region-growing loop in which each pixel classification result directly influences the classification of adjacent pixels.

If this is right

  • Vessel structures can be extracted from angiograms more accurately than with prior unsupervised methods.
  • Random Forests predictions become reliable for this task once guided by the region-growing feedback loop.
  • The combination of anisotropic diffusion, Hessian, morphology and statistics features supplies enough discrimination for high-accuracy pixel decisions.
  • The overall pipeline offers a concrete supervised route to cardiac vessel mapping that improves on existing automatic approaches.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Clinical software could incorporate this pipeline to generate vessel overlays that reduce the need for manual tracing.
  • The same feature-plus-controlled-classification pattern might transfer to segmentation in other 2-D medical X-ray or fluoroscopy modalities.
  • Testing the method on images from different acquisition equipment would show how robust the 95.48 percent figure remains under real-world variation.

Load-bearing premise

The chosen neighborhood textural features remain sufficiently different between vessel and background pixels across typical angiographic images when the region-growing control is added.

What would settle it

Running the same trained classifier and region-growing procedure on an independent collection of X-ray angiograms and checking whether accuracy stays at or above 95.48 percent while still beating the unsupervised baselines.

Figures

Figures reproduced from arXiv: 2605.20073 by D Casanova, E O Rodrigues, E R Dosciatti, F Favarim, F F C Morais, J J Lima, L O Rodrigues, L S N Oliveira, V Pegorini.

Figure 2
Figure 2. Figure 2: Figure2. A [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Figure3. G [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: X-ray anglogram image with ground truth data and associated segmentation results using the Random Forests classifier. () Input image. (b) Ground truth. (c) Result: training with one ge. (d) Result: t ining with 6 images, Figure5. X-ray anglogram image with ground truth data and associated segmentation results using the Random Forests classifier. well. These features represent the brush motion. Clas- - 4. E… view at source ↗
Figure 5
Figure 5. Figure 5: Figure5. X [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Figure6. X [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Figure7. X [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

This work proposes a pixel-classification approach for vessel segmentation in x-ray angiograms. The proposal uses textural features such as anisotropic diffusion, features based on the Hessian matrix, mathematical morphology and statistics. These features are extracted from the neighborhood of each pixel. The approach also uses the ELEMENT methodology, which consists of creating a pixel-classification controlled by region-growing where the result of the classification affects further classifications of pixels. The Random Forests classifier is used to predict whether the pixel belongs to the vessel structure. The approach achieved the best accuracy in the literature (95.48%) outperforming unsupervised state-of-the-art approaches.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a supervised pixel-classification method for vessel segmentation in X-ray cardiac angiograms. Textural features (anisotropic diffusion, Hessian-matrix, mathematical morphology, and statistical) are extracted from local pixel neighborhoods and fed to a Random Forest classifier; the ELEMENT region-growing procedure then iteratively uses classification outputs to control subsequent pixel decisions. The central empirical result is an accuracy of 95.48%, presented as the highest reported in the literature and superior to unsupervised state-of-the-art methods.

Significance. If the accuracy figure proves reproducible under standard validation protocols, the combination of neighborhood textural features with an ELEMENT-controlled Random Forest could offer a practical supervised alternative to purely unsupervised angiogram segmentation pipelines. The work explicitly credits the ELEMENT feedback loop as a novel control mechanism, but the absence of dataset size, cross-validation details, and ablation results currently prevents a clear assessment of whether the reported performance generalizes beyond the (unspecified) training images.

major comments (3)
  1. [Abstract and Results] Abstract and Results section: the headline claim of 95.48% accuracy 'outperforming unsupervised state-of-the-art approaches' is presented without any description of the dataset (number of angiograms, patients, resolution, or acquisition variability), the train/test split, or the cross-validation protocol. This information is load-bearing for the central numerical claim and must be supplied before the result can be evaluated.
  2. [Methodology] Methodology section on ELEMENT: the region-growing control loop that feeds classification results back into subsequent pixel decisions is described but never subjected to an ablation study or stability analysis. If the base Random Forest exhibits even moderate error on low-contrast or boundary pixels, the iterative feedback can produce correlated over- or under-segmentation; no evidence is given that the 95.48% figure survives leave-one-patient-out testing or that the loop was validated independently of the final metric.
  3. [Experiments] Experiments/Comparison subsection: the statement that the method outperforms 'unsupervised state-of-the-art approaches' does not identify which specific algorithms were reimplemented, what parameter settings were used, or whether they were evaluated on identical images and metrics. Without these details the comparative claim cannot be verified.
minor comments (2)
  1. [Methodology] The paper would benefit from a dedicated table listing the exact feature set (dimensions, extraction parameters for anisotropic diffusion and Hessian eigenvalues) to improve reproducibility.
  2. [Figures] Figure captions should explicitly state whether displayed segmentations are on training or held-out images and whether ground-truth overlays are included.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, indicating where revisions will be made to improve clarity and reproducibility.

read point-by-point responses

  1. Referee: [Abstract and Results] Abstract and Results section: the headline claim of 95.48% accuracy 'outperforming unsupervised state-of-the-art approaches' is presented without any description of the dataset (number of angiograms, patients, resolution, or acquisition variability), the train/test split, or the cross-validation protocol. This information is load-bearing for the central numerical claim and must be supplied before the result can be evaluated.

    Authors: We agree that these details are essential for evaluating the reported accuracy. In the revised manuscript we will add a dedicated subsection describing the dataset (number of angiograms, patients, resolutions, and acquisition variability), the train/test split used, and the cross-validation protocol. This will directly support the central numerical claim. revision: yes

  2. Referee: [Methodology] Methodology section on ELEMENT: the region-growing control loop that feeds classification results back into subsequent pixel decisions is described but never subjected to an ablation study or stability analysis. If the base Random Forest exhibits even moderate error on low-contrast or boundary pixels, the iterative feedback can produce correlated over- or under-segmentation; no evidence is given that the 95.48% figure survives leave-one-patient-out testing or that the loop was validated independently of the final metric.

    Authors: We recognize the importance of demonstrating the contribution and stability of the ELEMENT feedback loop. The original work presents the integrated performance but does not contain an ablation study or leave-one-patient-out validation. We will expand the methodological description to discuss potential error propagation and design choices intended to mitigate it, but a full ablation and independent validation of the loop were not performed in the submitted experiments. revision: partial

  3. Referee: [Experiments] Experiments/Comparison subsection: the statement that the method outperforms 'unsupervised state-of-the-art approaches' does not identify which specific algorithms were reimplemented, what parameter settings were used, or whether they were evaluated on identical images and metrics. Without these details the comparative claim cannot be verified.

    Authors: We agree that explicit identification of the compared methods is required for verifiability. The revised manuscript will list the specific unsupervised algorithms that were reimplemented, the parameter settings employed for each, and confirmation that all methods were evaluated on the same images and metrics as our approach. revision: yes

standing simulated objections not resolved
  • Full ablation study, stability analysis, and leave-one-patient-out testing of the ELEMENT control loop, as these experiments were not conducted in the original work.

Circularity Check

0 steps flagged

No circularity: empirical ML pipeline with no derivations or self-referential reductions

full rationale

The manuscript describes a standard supervised pixel-classification pipeline that extracts textural features (anisotropic diffusion, Hessian, morphology, statistics) from neighborhoods, trains a Random Forests classifier, and applies an iterative ELEMENT region-growing control. No equations, first-principles derivations, or parameter-fitting steps are present that could reduce any claimed prediction to its own inputs by construction. The 95.48% accuracy figure is reported as an empirical outcome on the evaluated angiograms; it does not arise from renaming a fitted quantity or from a self-citation chain that supplies the central result. The method is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the untested premise that the listed texture features plus region-growing feedback are adequate for the classification task; no independent evidence for this premise is supplied in the abstract.

free parameters (1)
  • Random Forest hyperparameters
    Typical ML tuning parameters that affect the reported accuracy but are not enumerated in the abstract.
axioms (1)
  • domain assumption The selected textural features from pixel neighborhoods are discriminative for vessel pixels in X-ray angiograms
    Invoked by the choice of anisotropic diffusion, Hessian, morphology, and statistical features as input to the classifier.
invented entities (1)
  • ELEMENT methodology no independent evidence
    purpose: To couple pixel classification with region-growing so that early classifications influence later ones
    Presented as the novel control mechanism in the abstract; no external validation cited.

pith-pipeline@v0.9.0 · 5661 in / 1393 out tokens · 39054 ms · 2026-05-20T05:45:14.257285+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

  1. [1]

    Detrano R etal 1996 Prognostic value of coronary calcification and angiographicstenoses in patients undergoing coronary angiography J. Am. College Cardiol. 27 285-90

    work page 1996

  2. [2]

    LimLS, LingLH, OngPG, Foulds W, Tai ESand WongT Y 2017 Dynamic responses in retinal vessel caliber with flicker light stimulation and risk of diabetic retinopathy and its progression Investigative Ophthalmol. Vis. Sci. 58 244955 3] RodriguesE O, Morais FF G, Morais NA OS, Conci LS, Neto L V and Conci A 2016a A novelapproach for the automated segmentation ...

    work page 2017

  3. [3]

    Halland lan H

    Eibe Frank, Mark A. Halland lan H. Witten 2016 Evaluation of hessian-based filters to enhancethe axis of coronary arteries in ctimages The WEKA Workbench. Online Appendixfor ‘Data Mining: Practical Machine Learning Tools and Techniques”, Morgan Kaufinann, Fourth Edition, 2016

    work page 2016

  4. [4]

    Salem NM and NandiA K 2008 Unsupervised segmentation of retinalblood vessels usinga single parameter vesselness measure Computer Vision, Graphicsand Image Processing

    work page 2008

  5. [5]

    Rodrigues E O, Conci A and Liatsis P 2020 Element: Multi- modal retinal vessel segmentation based ona coupled region growingand machine learning approach IEEE Journal of ‘Biomedical and Health Informatics 24350719

    work page 2020

  6. [6]

    Vision 8 167-70

    Dehkordi M, Hoseini AM, Sadri S and Soltanianzadeh H 2014 Local feature fitting active contour for segmenting vessels in angiograms IET Comput. Vision 8 167-70

    work page 2014

  7. [7]

    SanchezF C, AcevesI C, Aguirre A H, Cervantes ] G, Meza S, Rodriguez M O and Gisneros M T 2016 Segmentationof coronary angiograms using gabor filters and boltzmann univariate marginal distribution algorithm Computational Intelligence and Neuroscience 2016

    work page 2016

  8. [8]

    Pattern Anal

    Stanfield S A 1986 Angy: A rule-based expert system for automatic segmentation of coronary vessels from digital subtracted angiograms IEEE Trans. Pattern Anal. Mach. Intell. 8 18899

    work page 1986

  9. [9]

    SPIE 7962, Medical Tmaging 2011: Image Processing

    Zheng, Loziczonek M, Georgescu B, Zhou SK, Vega-Higuera Fand Comaniciu D 2011 Machinelearning based vesselness measurementfor coronary artery segmentation in cardiac ctvolumes Proc. SPIE 7962, Medical Tmaging 2011: Image Processing

    work page 2011

  10. [10]

    GaoZ, Chung], Abdelrazek M, Leung$, Hau W K, Xian Z, ZhangHand Li $ 2020a Privileged modality distillation for vessel border detection in intracoronary imaging IEEE Trans Med Imaging39 1524-34 (18] GaoZ, LiuX, QiS, WuW, Hau WK and Zhang H 2017 Automatic segmentation of coronary tree in ct angiography images Int. J. Adapt Control Signal Process

    work page 2017

  11. [11]

    GaoZ, Wang X, SunS, Wu D, Bai, Yin Y, Liu X, Zhang Hand Albuquerque VH €2020b Learning physical properties in complex visual scenes: Anintelligent machine for perceiving blood flow dynamics from static ct angiography imaging NeuralNetw. 1238293

    work page

  12. [12]

    84 82-96

    Rodrigues E O, Conci A and Liatsis P 2018 Morphological classifiers Pattern Recognit. 84 82-96

    work page 2018

  13. [13]

    66 392403

    Rodrigues E O, Torok L, Liatsis P, Viterbo Jand Conci A 2017b k-ms: A novel clusteringalgorithm based on morphological reconstruction Pattern Recognit. 66 392403

    work page

  14. [14]

    117 66-71 (23] Lupascu CA, Tegolo Dand Trucco E2010 Fabc: Retinal vessel segmentation usingadaboost IEEE Trans

    Rodrigues E O 2021 Anefficient and locality-oriented hausdorff distance algorithm: Proposal and analysis of paradigms and implementations Pattern Recognit. 117 66-71 (23] Lupascu CA, Tegolo Dand Trucco E2010 Fabc: Retinal vessel segmentation usingadaboost IEEE Trans. Inf. Technol. Biomed. 14126774

    work page 2021

  15. [15]

    Dicom image library (https: / /www.osirix- viewer.com /resources /dicom-image-library/)

    OSIRIX, 2020. Dicom image library (https: / /www.osirix- viewer.com /resources /dicom-image-library/)

    work page 2020

  16. [16]

    Rodrigues E O 2020 X-ray angiograms ground truth (https:// github.com /Oyatsumi/X-RayAngiogramDataset)

    work page 2020

  17. [17]

    Lorenz C, Carlsen C, Buzug TM, Fassnacht Cand Weese | 2005 A multi-scale line filter withautomatic scaleselection based on the hessian matrix for medical image segmentation Scale-Space Theory in Computer Vision 1252 152-63

    work page 2005

  18. [18]

    TIKHG, Jane O and TIk O 2011 The effect of laplacian filter in adaptive unsharp masking for infrared image enhancement Infrared Phys. Technol. 54 427-38 28] Vincent L 1994 Morphological area openings and closings for grey-scale images Shape in Picture 126 197-208

    work page 2011

  19. [19]

    Pattern Anal

    Perona P and Malik J 1990 Scale-space and edge detection usinganisotropicdiffusion IEEE Trans. Pattern Anal. Mach. Intell. 1262939 Bartyzel K 2016 Adaptive kuwahara filter Signal, Image and Video Processing 10 66370

    work page 1990

  20. [20]

    Rodrigues E O, Porcino TM, Conci Aand Silva A C2016b A simple approach for biometrics: Finger-knuckle prints recognition based on asobel filter and similarity measures International Conference on Systems, Signals and Image Processing (IWSSIP) Rodrigues E O, Pinheiro V H A, Liatsis P and Conci A 2017a Machine learningin the prediction of cardiac epicardial...

    work page 2018

  21. [21]

    Frank E, Hall M A and Witten 1H 2016 The weka workbench, online appendix for data mining: practical machine learning tools and techniques Morgan Kaufmann

    work page 2016

  22. [22]

    Breiman L 2001 Random forests Mach. Learn. 45532

    work page 2001

  23. [23]

    EihoSand Qian Y 1997 Detection of coronary artery tree using morphological operator Computers in Cardiology

    work page 1997

  24. [24]

    Wang$,LiBand ZhouS 2012 A segmentation method of coronary angiogramsbased on multi-scale filteringand region-growing Interrational Conference on Biomedical EngineeringandBiotechnology KangW, KangW, Li Y and WangQ 2013 The segmentation method of degree-based fusion algorithm for coronary angiogramsinternational Conference on Measurement, Information andControl

    work page 2012

  25. [25]

    Chanwimaluang T, Fan G and FransenS R 2006 Hybrid retinal image registration IEEE Trans. Inf. Technol. Biomed. 10 12942 ChaudhuriS, Chatterjee S, Katz N, NelsonM and Goldbaum M 1989 Detection of blood vessels in retinal images using two-dimensional matched filters IEEE Trans. Med. Imaging8263-9 [30] B2] B3 B8 [4o]

    work page 2006