pith. sign in

arxiv: 2605.21251 · v1 · pith:N7IOW2Q4new · submitted 2026-05-20 · 📡 eess.IV · cs.CV

Local-sensitive connectivity filter (ls-cf): A post-processing unsupervised improvement of the frangi, hessian and vesselness filters for multimodal vessel segmentation

Erick O Rodrigues , Lucas O Rodrigues , Jo\~ao HP Machado , Dalcimar Casanova , Marcelo Teixeira , Jeferson T Oliva , Giovani Bernardes , Panos Liatsis This is my paper

Pith reviewed 2026-05-21 03:57 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords vessel segmentationFrangi filterconnectivity filterunsupervisedretinal imagesangiographymultimodalpost-processing
0
0 comments X

The pith

A local tolerance heuristic in a connectivity filter improves unsupervised vessel segmentation from Frangi responses.

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

This paper presents the local-sensitive connectivity filter as an unsupervised post-processing step that refines the output of the Frangi filter and similar vesselness measures for blood vessel segmentation in multimodal images. The filter calculates pixel-level continuity and applies a local tolerance rule to bridge gaps that appear in the base filter response. When tested on retinal and angiographic datasets, the approach produces accuracy scores that match or exceed those of prior unsupervised and supervised methods. A reader would care because reliable vessel maps without training data can support faster, more consistent analysis of vascular health in eyes and other organs.

Core claim

The local-sensitive connectivity filter improves the thresholded Frangi response by enforcing vessel continuity at the pixel level through a local tolerance heuristic that reconnects discontinuous segments while avoiding the need for labeled data or additional model training.

What carries the argument

The local-sensitive connectivity filter (LS-CF), which augments basic connectivity rules with a local tolerance heuristic to fill vessel gaps in the filter output.

If this is right

  • The method reaches higher accuracy than all compared state-of-the-art approaches on the OSIRIX angiographic dataset.
  • It exceeds most published results on the IOSTAR, DRIVE, STARE, and CHASE-DB retinal datasets.
  • On the CHASE-DB set it surpasses every unsupervised competitor reported in the literature.
  • The same post-processing step applies to Hessian and other vesselness filters without retraining.

Where Pith is reading between the lines

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

  • The filter could be inserted after any ridge or vesselness detector to reduce manual cleanup in clinical pipelines.
  • Extending the local tolerance rule to three-dimensional volumes might improve segmentation in CT or MRI angiography.
  • Performance on datasets with heavy pathology or low contrast would reveal the heuristic's practical limits.
  • Combining the connectivity step with simple intensity normalization could further stabilize results across scanners.

Load-bearing premise

The local tolerance heuristic can reliably separate genuine vessel continuities from noise or unrelated structures across different image types.

What would settle it

Running the filter on images that contain known vessel gaps and measuring whether overall segmentation accuracy falls because of added false vessel connections would test the claim.

read the original abstract

A retinal vessel analysis is a procedure that can be used as an assessment of risks to the eye. This work proposes an unsupervised multimodal approach that improves the response of the Frangi filter, enabling automatic vessel segmentation. We propose a filter that computes pixel-level vessel continuity while introducing a local tolerance heuristic to fill in vessel discontinuities produced by the Frangi response. This proposal, called the local-sensitive connectivity filter (LS-CF), is compared against a naive connectivity filter to the baseline thresholded Frangi filter response and to the naive connectivity filter response in combination with the morphological closing and to the current approaches in the literature. The proposal was able to achieve competitive results in a variety of multimodal datasets. It was robust enough to outperform all the state-of-the-art approaches in the literature for the OSIRIX angiographic dataset in terms of accuracy and 4 out of 5 works in the case of the IOSTAR dataset while also outperforming several works in the case of the DRIVE and STARE datasets and 6 out of 10 in the CHASE-DB dataset. For the CHASE-DB, it also outperformed all the state-of-the-art unsupervised methods.

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

2 major / 1 minor

Summary. The manuscript proposes the Local-Sensitive Connectivity Filter (LS-CF) as an unsupervised post-processing step to enhance the Frangi, Hessian, and vesselness filters for multimodal vessel segmentation. It introduces a local tolerance heuristic to compute pixel-level vessel continuity and bridge discontinuities in thresholded responses, claiming competitive or superior accuracy on five datasets (OSIRIX, IOSTAR, DRIVE, STARE, CHASE-DB) while outperforming all SOTA on OSIRIX, 4/5 works on IOSTAR, several on DRIVE/STARE, and 6/10 (all unsupervised) on CHASE-DB.

Significance. If the heuristic reliably recovers true vessel segments without false-positive connections, LS-CF would provide a simple, single-parameter, unsupervised refinement applicable to existing vesselness filters across modalities. The reported outperformance on multiple public datasets indicates potential utility for retinal analysis, though the absence of robustness validation on the core heuristic limits the strength of this contribution.

major comments (2)
  1. [Methods (LS-CF description)] Methods (LS-CF algorithm description): The local tolerance heuristic is described only at a high level as computing pixel-level vessel continuity to fill Frangi-induced discontinuities, with no explicit rule, neighborhood definition, threshold derivation, or pseudocode provided. This is load-bearing for the central claim, as the reported gains (e.g., outperforming all SOTA on OSIRIX and all unsupervised methods on CHASE-DB) depend on the heuristic connecting only true continuities rather than introducing artifacts in ambiguous neighborhoods.
  2. [Results (performance tables)] Results (dataset comparisons): No error bars, statistical significance tests, or sensitivity analysis on the local tolerance threshold are reported, despite it being the sole free parameter. This weakens the soundness of accuracy claims across the five datasets, as the parameter choice could be dataset-specific without cross-validation details.
minor comments (1)
  1. [Abstract and Methods] The abstract and methods refer to comparisons against a 'naive connectivity filter' and morphological closing, but do not specify the exact connectivity criterion or structuring element used, hindering reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and constructive feedback on our manuscript. We address each of the major comments in detail below and have made revisions to the manuscript to incorporate the suggestions where possible.

read point-by-point responses

  1. Referee: Methods (LS-CF algorithm description): The local tolerance heuristic is described only at a high level as computing pixel-level vessel continuity to fill Frangi-induced discontinuities, with no explicit rule, neighborhood definition, threshold derivation, or pseudocode provided. This is load-bearing for the central claim, as the reported gains (e.g., outperforming all SOTA on OSIRIX and all unsupervised methods on CHASE-DB) depend on the heuristic connecting only true continuities rather than introducing artifacts in ambiguous neighborhoods.

    Authors: We agree with the referee that the description of the LS-CF heuristic requires more explicit detail to ensure reproducibility and to substantiate the central claims. In the revised version of the manuscript, we have added a comprehensive description of the local tolerance heuristic in the Methods section. This includes the specific neighborhood definition, the explicit rule for determining vessel continuity based on local vessel direction and intensity, the derivation of the tolerance threshold from the Frangi response, and pseudocode for the algorithm. This makes the process for connecting true continuities transparent and addresses concerns about artifacts. revision: yes

  2. Referee: Results (dataset comparisons): No error bars, statistical significance tests, or sensitivity analysis on the local tolerance threshold are reported, despite it being the sole free parameter. This weakens the soundness of accuracy claims across the five datasets, as the parameter choice could be dataset-specific without cross-validation details.

    Authors: We acknowledge the referee's point regarding the need for more rigorous statistical validation. In the revised manuscript, we have included error bars in the performance tables to show variability, added a sensitivity analysis for the local tolerance threshold across a range of values for each dataset, and reported statistical significance using appropriate tests for comparisons. We have also clarified that the parameter was chosen based on overall performance and provide details on the selection process. revision: yes

Circularity Check

0 steps flagged

No circularity detected in LS-CF proposal or empirical claims

full rationale

The paper introduces the Local-Sensitive Connectivity Filter (LS-CF) as an unsupervised post-processing heuristic that computes pixel-level vessel continuity with a local tolerance to bridge Frangi response discontinuities. No equations, fitted parameters, or derivations are presented that reduce the method or its reported accuracy gains to quantities defined internally by construction. Performance is evaluated through direct comparisons to external state-of-the-art methods and baselines on independent multimodal datasets (OSIRIX, IOSTAR, DRIVE, STARE, CHASE-DB), with no self-citation load-bearing steps, ansatz smuggling, or renaming of known results as novel derivations. The central claims rest on empirical outperformance rather than any closed logical loop, rendering the approach self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that retinal vessels exhibit locally continuous structures that a tolerance heuristic can safely bridge; no free parameters or invented entities are explicitly quantified in the abstract.

free parameters (1)
  • local tolerance threshold
    Heuristic parameter controlling how large a discontinuity the filter will bridge; value not stated in abstract.
axioms (1)
  • domain assumption Vessel structures in retinal images are locally continuous and can be recovered by pixel-level connectivity checks with tolerance.
    Invoked to justify filling discontinuities in the Frangi response.

pith-pipeline@v0.9.0 · 5781 in / 1352 out tokens · 39742 ms · 2026-05-21T03:57:48.460663+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages · 2 internal anchors

  1. [1]

    A Comprehensive Study of Retinal Vessel Classification Methods in Fundus Images

    Miri, M.; Amini, Z; Rabbani, H.; Kafieh, R. A Comprehensive Study of Retinal Vessel Classification Methods in Fundus Images. J. Med. Signals Sens. 2017, 7,59-70. [PubMed] Lim, LS; Ling, LH.; Ong, PG.; Foulds, W.; Tai, E.S.; Wong, T.Y. Dynamic Responses in Retinal Vessel Caliber With Flicker Light Stimulation and Risk of Diabetic Retinopathy and Its Progre...

    work page 2017

  2. [2]

    ELEMENT: Multi-modal retinal vessel segmentation based on a coupled region growing and machine learning approach

    Chaudhuri, S.; Chatterjee, S.; Katz, N.; Nelson, M.; Goldbaum, M. ELEMENT: Multi-modal retinal vessel segmentation based on a coupled region growing and machine learning approach. IEEE J. Biomed. Health Inform. 2020, 24, 3507-3519. Rodrigues, E.O.; Conci, A.; Liatsis, P. Morphological classifiers. Pattern Recognit. 2018, 84, 82-96. [CrossRef] Porcino, T.;...

    work page 2020

  3. [3]

    Exercise-induced alterations of retinal vessel diameters and cardiovascular risk reduction in obesity

    [CrossRef] [PubMed] Hanssen, H.; Nickel, T; Drexel, V.; Hertel, G.; Emslander, 1; Sisic, Z.; Lorang, D.; Schuster, T.; Kotliar, K.E.; Pressler, A.; etal. Exercise-induced alterations of retinal vessel diameters and cardiovascular risk reduction in obesity. Atherosclerosis 2011, 216,433-439. [CrossRef] Stergiopulos, N.; Young, D.F; Rogge, T.R. Computer sim...

    work page 2011

  4. [4]

    Computational Methods for Liver Vessel Segmentation in Medical Imaging: A Review

    Ciecholewski, M; Kassjariski, M. Computational Methods for Liver Vessel Segmentation in Medical Imaging: A Review. Sensors 2021, 21,

    work page 2021

  5. [5]

    Retinal vessel segmentation by improved matched filtering: Evaluation on a new high-resolution fundus image database

    [CrossRef] [PubMed] Odstrcilik, J.; Kolar, R.; Budai, A.; Hornegger, J.; Jan, J.; Gazarek, J.; Kubena, T.; Cernosek, P; Svoboda, O.; Angelopoulou, E. Retinal vessel segmentation by improved matched filtering: Evaluation on a new high-resolution fundus image database. IEEE Trans. Inf. Technol. Biomed. 2013, 7, 373-383. [CrossRef] Jiang, H.; He, B.; Fang, D...

    work page 2013

  6. [6]

    Frangi-Net: A Neural Network Approach to Vessel Segmentation

    Mo, J.; Zhang, L. Multi-level deep supervised networks for retinal vessel segmentation. Int. J. Comput. Assist. Radiol. Surg. 2017, 12,2181-2193. [CrossRef] [PubMed] Lupascu, C.A.; Tegolo, D.; Trucco, E. FABC: Retinal Vessel Segmentation Using AdaBoost. IEEE Trans. Inf. Technol. Biomed. 2010, 14, 1267-1274. [CrossRef] [PubMed] Fu, W,; Breininger, K.; Wiir...

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  7. [7]

    Rodrigues, E.O

    Available online: https:/ /github.com /Oyatsumi/ConnectivityFilter (accessed on 10 August 2022). Rodrigues, E.O. Combining Minkowski and Chebyshev: New distance proposal and survey of distance metrics using k-nearest neighbours classifier. Pattern Recognit. Lett. 2018, 110,66-71. [CrossRef] Rodrigues, E.O. An efficient and locality-oriented Hausdorff dist...

    work page 2022

  8. [8]

    Ensemble of Deep Convolutional Neural Networks for Learning to Detect Retinal Vessels in Fundus Images

    Al-Rawi, M.; Karajeh, H. Genetic algorithm matched filter optimization for automated detection of blood vessels from digital retinal images. Comput. Methods Programs Biomed. 2007, 87, 248-253. [CrossRef] Soares, J.V.B.; Leandro, JJ.G.; Cesar, R.M.; Jelinek, H.E; Cree, ML]. Retinal vessel segmentation using the 2-D Gabor wavelet and supervised classificati...

    work page internal anchor Pith review Pith/arXiv arXiv 2007

  9. [9]

    Segmenting Retinal Blood Vessels With Deep Neural Networks

    Liskowski, P.; Krawiec, K. Segmenting Retinal Blood Vessels With Deep Neural Networks. IEEE Trans. Med. Imaging 2016, 35,2369-2380. [CrossRef] Fan, Z.; Rong, Y.; Lu,J.; Mo, J.; Li, F.; Cai, X.; Yang, T. Automated blood vessel segmentation in fundus image based on integral channel features and random forests. In Proceedings of the 12th World Congress on In...

    work page 2016

  10. [10]

    Pixel-wise losses for deep learning based retinal vessel segmentation

    Yan, Z.; Yang, X.; Cheng, K.T. Pixel-wise losses for deep learning based retinal vessel segmentation. IEEE Trans. Biomed. Eng. 2018, 66, 1912-1923. [CrossRef] Welikala, R.A.; Foster, PJ.; Whincup, PH.; Rudnicka, A.R.; Owen, C.G.; Strachan, D.P; Barman, S.A. Automated arteriole and venule classification using deep learning for retinal images from the UK Bi...

    work page 2018

  11. [11]

    A Deep Convolutional Encoder- Decoder Architecture for Retinal Blood Vessels Segmentation

    74 Adeyinka, A.A.; Adebiyi, M.O.; Akande, N.O.; Ogundokun, R.O.; Kayode, A.A.; Oladele, TO. A Deep Convolutional Encoder- Decoder Architecture for Retinal Blood Vessels Segmentation. In Lecture Notes in Computer Science; Springer: Cham, Switzerland, 2019; pp. 180-189. Shin, S.Y.; Lee, S.; Yun, LD.; Lee, K.M. Deep Vessel Segmentation by Learning Graphical ...

    work page 2019

  12. [12]

    A visual attention guided unsupervised feature learning for robust vessel delineation in retinal images

    Srinidhi, C.L.; Aparna, P; Rajan, J. A visual attention guided unsupervised feature learning for robust vessel delineation in retinal images. Biomed. Signal Process. Control. 2018, 44, 110-126. [CrossRef] Samant, P; Bansal, A.; Agarwal, R. A hybrid filteringbased retinal blood vessel segmentation algorithm. In Computer Vision and Machine Intelligence in M...

    work page 2018

  13. [13]

    Supervised Segmentation of Un-annotated Retinal Fundus Images by Synthesis

    Zhao, H; Li, H.; Maurer-Stroh, S.; Guo, Y.; Deng, Q.; Cheng, L. Supervised Segmentation of Un-annotated Retinal Fundus Images by Synthesis. [EEE Trans. Med. Imaging 2018, 38, 46-56. [CrossRef]

    work page 2018

  14. [14]

    Segmentation of Coronary Angiograms Using Gabor Filters and Boltzmann Univariate Marginal Distribution Algorithm

    Cervantes-Sanchez, E;; Cruz-Aceves, L; Hemandez-Aguirre, A.; Avina-Cervantes, J.G ; Solorio-Meza, S.; Ornelas-Rodriguez, M; Torres-Cisneros, M. Segmentation of Coronary Angiograms Using Gabor Filters and Boltzmann Univariate Marginal Distribution Algorithm. Comput. Intell. Neurosci. 2016, 2016, 2420962. [CrossRef] Fatemi, M.J.R ; Mirhassani, S.M.; Yousefi...

    work page 2016

  15. [15]

    Hybrid Retinal Image Registration

    Chanwimaluang, T.; Fan, G.; Fransen, SR. Hybrid Retinal Image Registration. [EEE Trans. Inf. Technol. Biomed. 2006, 10, 129-142. [CrossRef] Rodrigues, E.O;; Liatsis, P; Satoru, L.; Conci, A. Fractal triangular search: A metaheuristic for image content search IET Image Processing 2018, 12, 1475-1484. [CrossRef] Rodrigues, E.O.; Conci, A.; Morais, EE.C.; Pe...

    work page 2006