pith. sign in

arxiv: 2410.21160 · v2 · submitted 2024-10-28 · 📡 eess.IV · cs.CV

KaLDeX: Kalman Filter based Linear Deformable Cross Attention for Retina Vessel Segmentation

Zhihao Zhao , Yinzheng Zhao , Junjie Yang , Quanmin Liang , Daniel Zapp , Kai Huang , M.Ali Nasseri This is my paper

Pith reviewed 2026-05-23 19:16 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords retinal vessel segmentationKalman filterdeformable convolutioncross attentionUNet++topological losspersistent homologyOCTA
0
0 comments X

The pith

KaLDeX adds a Kalman filter based linear deformable cross attention module to UNet++ to segment thin retinal vessels more accurately.

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

The paper introduces KaLDeX, a network for retinal vessel segmentation that inserts a Kalman filter based linear deformable cross attention module into the UNet++ architecture. Standard CNN downsampling loses fine vessel details, so the module uses Kalman-guided deformable convolution to focus on overlooked thin structures and cross-attention to combine those details with higher-level features. A persistent homology based topological loss further enforces vessel continuity. Experiments on DRIVE, CHASE_DB1, STARE, and OCTA-500 datasets report higher accuracy than prior leading methods.

Core claim

The KaLDeX network integrates a Kalman filter based linear deformable convolution module with cross-attention inside UNet++ and adds a persistent homology topological loss. This combination lets the model adaptively sample thin vessels that standard convolutions miss and aggregate multi-scale vascular features, producing higher segmentation accuracy across the tested retinal fundus and OCTA datasets.

What carries the argument

Kalman filter based linear deformable cross attention (LDCA) module, which uses Kalman filtering to guide adaptive sampling in deformable convolution and cross-attention to fuse detailed and high-level features.

If this is right

  • Reports average accuracy of 97.25% on DRIVE, 97.77% on CHASE_DB1, 97.85% on STARE, 98.89% on 3 mm OCTA-500, and 98.21% on 6 mm OCTA-500.
  • Outperforms prior best models on all five evaluated vessel segmentation datasets.
  • The LD module focuses sampling on thin vessels while the CA module improves global structure awareness.
  • The topological loss based on persistent homology constrains segmentation to preserve vessel continuity.

Where Pith is reading between the lines

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

  • The LDCA design may transfer to other medical segmentation tasks that involve fine linear structures such as nerves or airways.
  • Kalman filtering inside the attention block offers a route to incorporate prediction uncertainty directly into feature sampling.
  • Replacing UNet++ with other encoder-decoder backbones could test whether the module's benefit is architecture-specific.

Load-bearing premise

The measured accuracy gains come from the LDCA module rather than from unstated differences in preprocessing, training protocols, or dataset-specific tuning.

What would settle it

An ablation experiment that removes only the LDCA module from KaLDeX, retrains on the same splits of DRIVE, and checks whether accuracy falls to the level of a plain UNet++ baseline.

Figures

Figures reproduced from arXiv: 2410.21160 by Daniel Zapp, Junjie Yang, Kai Huang, M.Ali Nasseri, Quanmin Liang, Yinzheng Zhao, Zhihao Zhao.

Figure 1
Figure 1. Figure 1: The first row shows the ophthalmic images and their segmentation [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our proposed KaLDeX. (a) is the overall framework of our proposed structure. (b) showcases the key module (LDCA), which primarily consists of two sub-modules: LD(b1) and CA(b2). LD is primarily focused on capturing vascular structures through deformable convolutions and optimizing the coordinates of deformable convolutions using KF. CA aggregates the output feature maps of LD with the feature m… view at source ↗
Figure 3
Figure 3. Figure 3: LD is primarily focused on capturing vascular structures through LD [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: In figure (a), the gray background illustrates the vascular area, while [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: CA aggregates the output feature maps of LD with the feature maps [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of segmentation results. The first and third rows depict the segmentation results of fundus images and OCTA, respectively, while the second [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The results of testing on multiple datasets using di [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The first and second rows show fundus images, highlighting the con [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
read the original abstract

Background and Objective: In the realm of ophthalmic imaging, accurate vascular segmentation is paramount for diagnosing and managing various eye diseases. Contemporary deep learning-based vascular segmentation models rival human accuracy but still face substantial challenges in accurately segmenting minuscule blood vessels in neural network applications. Due to the necessity of multiple downsampling operations in the CNN models, fine details from high-resolution images are inevitably lost. The objective of this study is to design a structure to capture the delicate and small blood vessels. Methods: To address these issues, we propose a novel network (KaLDeX) for vascular segmentation leveraging a Kalman filter based linear deformable cross attention (LDCA) module, integrated within a UNet++ framework. Our approach is based on two key components: Kalman filter (KF) based linear deformable convolution (LD) and cross-attention (CA) modules. The LD module is designed to adaptively adjust the focus on thin vessels that might be overlooked in standard convolution. The CA module improves the global understanding of vascular structures by aggregating the detailed features from the LD module with the high level features from the UNet++ architecture. Finally, we adopt a topological loss function based on persistent homology to constrain the topological continuity of the segmentation. Results: The proposed method is evaluated on retinal fundus image datasets (DRIVE, CHASE_BD1, and STARE) as well as the 3mm and 6mm of the OCTA-500 dataset, achieving an average accuracy (ACC) of 97.25%, 97.77%, 97.85%, 98.89%, and 98.21%, respectively. Conclusions: Empirical evidence shows that our method outperforms the current best models on different vessel segmentation datasets. Our source code is available at: https://github.com/AIEyeSystem/KalDeX.

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 paper proposes KaLDeX, a UNet++-based architecture for retinal vessel segmentation that incorporates a Kalman filter-based linear deformable cross-attention (LDCA) module (combining linear deformable convolution and cross-attention) together with a persistent-homology topological loss. It evaluates the method on DRIVE, CHASE_DB1, STARE, and OCTA-500 (3 mm and 6 mm), reporting average accuracies of 97.25 %, 97.77 %, 97.85 %, 98.89 %, and 98.21 % respectively, and claims these results outperform prior state-of-the-art models.

Significance. If the reported gains can be shown to arise specifically from the Kalman-filter LDCA component rather than from the topological loss, training schedule, or preprocessing, the work would offer a concrete mechanism for recovering fine-scale vessels that are typically lost under repeated down-sampling. The public release of source code supports reproducibility and is a clear strength.

major comments (3)
  1. [Abstract / Results] Abstract and Results: the reported accuracies (97.25 % DRIVE, 97.77 % CHASE_DB1, etc.) are presented without error bars, standard deviations across runs, or statistical tests against the cited baselines. This prevents any assessment of whether the numerical improvements are reliable or merely within run-to-run variation.
  2. [Methods / Experiments] Methods and Experiments: no component-wise ablation is supplied that isolates the contribution of the Kalman-filter LDCA module from the persistent-homology topological loss or from the plain UNet++ backbone under identical training protocols and preprocessing. Because the central claim attributes the performance lift to the LDCA module, the absence of such controlled comparisons renders the attribution unverifiable.
  3. [Experiments] Experiments: the manuscript does not indicate whether the same data augmentation, optimizer schedule, and post-processing steps were used for all re-implemented baselines; any uncontrolled differences could account for the observed margins on DRIVE, CHASE_DB1, STARE, and OCTA-500.
minor comments (2)
  1. [Abstract] Abstract: the dataset name is written “CHASE_BD1”; the conventional spelling is CHASE_DB1.
  2. [Methods] Notation: the distinction between the linear deformable convolution (LD) and the full cross-attention (CA) sub-modules inside LDCA is not always maintained consistently in the text and figures.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and commit to revisions that will strengthen the empirical support and clarity of the manuscript.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results: the reported accuracies (97.25 % DRIVE, 97.77 % CHASE_DB1, etc.) are presented without error bars, standard deviations across runs, or statistical tests against the cited baselines. This prevents any assessment of whether the numerical improvements are reliable or merely within run-to-run variation.

    Authors: We agree that variability measures and statistical tests are necessary for reliable assessment of improvements. In the revised manuscript we will report mean and standard deviation of accuracy (and other metrics) over at least five independent training runs with different random seeds, and we will add paired statistical tests (e.g., Wilcoxon signed-rank) against each cited baseline. revision: yes

  2. Referee: [Methods / Experiments] Methods and Experiments: no component-wise ablation is supplied that isolates the contribution of the Kalman-filter LDCA module from the persistent-homology topological loss or from the plain UNet++ backbone under identical training protocols and preprocessing. Because the central claim attributes the performance lift to the LDCA module, the absence of such controlled comparisons renders the attribution unverifiable.

    Authors: We concur that controlled ablations are required to substantiate the attribution to the LDCA module. The revised manuscript will include a component-wise ablation table that evaluates (i) plain UNet++, (ii) UNet++ plus topological loss, and (iii) the full KaLDeX model, all trained with identical protocols, preprocessing, and hyperparameters. revision: yes

  3. Referee: [Experiments] Experiments: the manuscript does not indicate whether the same data augmentation, optimizer schedule, and post-processing steps were used for all re-implemented baselines; any uncontrolled differences could account for the observed margins on DRIVE, CHASE_DB1, STARE, and OCTA-500.

    Authors: All baselines were re-implemented under the exact same data-augmentation pipeline, optimizer, learning-rate schedule, batch size, and post-processing steps used for KaLDeX. The revised manuscript will explicitly state this protocol equivalence in the Experiments section and will list the precise augmentation and post-processing operations applied uniformly. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical claims rest on external public datasets

full rationale

The paper proposes KaLDeX (Kalman-filter LDCA inside UNet++ plus topological loss) and reports measured accuracies on the public DRIVE, CHASE_DB1, STARE and OCTA-500 benchmarks. No equations, parameters, or performance figures are defined in terms of themselves or obtained by fitting on the test sets and then relabeled as predictions. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work appear in the provided text. The central claim therefore remains an ordinary empirical comparison against external data and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on standard deep-learning assumptions (gradient descent finds useful minima, cross-entropy plus topological loss is a suitable surrogate) and on the untested premise that the Kalman-filter adaptation improves localization of sub-pixel vessels without introducing new artifacts. No new physical entities are postulated.

axioms (1)
  • domain assumption Gradient-based optimization on the combined segmentation and topological loss yields a network whose outputs generalize to unseen retinal images.
    Invoked implicitly when claiming superior performance on held-out test sets.

pith-pipeline@v0.9.0 · 5889 in / 1193 out tokens · 26505 ms · 2026-05-23T19:16:02.779719+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

62 extracted references · 62 canonical work pages

  1. [1]

    S. You, E. Bas, D. Erdogmus, J. Kalpathy-Cramer, Principal curved based retinal vessel segmentation towards diagnosis of retinal diseases, in: 2011 IEEE First International Conference on Healthcare Informatics, Imaging and Systems Biology, IEEE, 2011, pp. 331–337

    work page 2011

  2. [2]

    Y . Zhao, Z. Zhao, J. Yang, L. Li, M. A. Nasseri, D. Zapp, Ai-based fully automatic analysis of retinal vascular morphology in pediatric high my- opia, BMC ophthalmology 24 (1) (2024) 415

    work page 2024

  3. [3]

    C. D. REGAN, C. S. FOSTER, Retinal vascular diseases: clinical presen- tation and diagnosis, International Ophthalmology Clinics 26 (2) (1986) 25–53

    work page 1986

  4. [4]

    Z. Zhao, S. Faghihroohi, J. Yang, K. Huang, N. Navab, M. Maier, M. A. Nasseri, Unobtrusive biometric data de-identification of fundus images using latent space disentanglement, Biomedical Optics Express 14 (10) (2023) 5466–5483

    work page 2023

  5. [5]

    T. H. Rim, A. W. J. Teo, H. H. S. Yang, C. Y . Cheung, T. Y . Wong, Retinal vascular signs and cerebrovascular diseases, Journal of Neuro- ophthalmology 40 (1) (2020) 44–59

    work page 2020

  6. [6]

    Uysal, G

    E. Uysal, G. E. G ¨uraksin, Computer-aided retinal vessel segmentation in retinal images: convolutional neural networks, Multimedia Tools and Applications 80 (2021) 3505–3528

    work page 2021

  7. [7]

    Q. Qin, Y . Chen, A review of retinal vessel segmentation for fundus image analysis, Engineering Applications of Artificial Intelligence 128 (2024) 107454

    work page 2024

  8. [8]

    C. Chen, J. H. Chuah, R. Ali, Y . Wang, Retinal vessel segmentation using deep learning: a review, IEEE Access 9 (2021) 111985–112004

    work page 2021

  9. [9]

    Z. Zhao, Y . Zhao, J. Yang, K. Huang, N. Navab, M. A. Nasseri, Kldd: Kalman filter based linear deformable di ffusion model in retinal image segmentation, arXiv preprint arXiv:2410.02808 (2024)

    work page arXiv 2024

  10. [10]

    J. V . Soares, J. J. Leandro, R. M. Cesar, H. F. Jelinek, M. J. Cree, Retinal vessel segmentation using the 2-d gabor wavelet and supervised classifi- cation, IEEE Transactions on medical Imaging 25 (9) (2006) 1214–1222

    work page 2006

  11. [11]

    Z. Zhao, J. Yang, S. Faghihroohi, K. Huang, M. Maier, N. Navab, M. A. Nasseri, Label-preserving data augmentation in latent space for diabetic retinopathy recognition, in: International Conference on Medical Im- age Computing and Computer-Assisted Intervention, Springer, 2023, pp. 284–294

    work page 2023

  12. [12]

    X. Li, Y . Jiang, M. Li, S. Yin, Lightweight attention convolutional neu- ral network for retinal vessel image segmentation, IEEE Transactions on Industrial Informatics 17 (3) (2020) 1958–1967

    work page 2020

  13. [13]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Con- ference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18, Springer, 2015, pp. 234–241

    work page 2015

  14. [14]

    J. Hu, H. Wang, S. Gao, M. Bao, T. Liu, Y . Wang, J. Zhang, S-unet: A bridge-style u-net framework with a saliency mechanism for retinal vessel segmentation, IEEE Access 7 (2019) 174167–174177

    work page 2019

  15. [15]

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, J. Liang, Unet ++: A nested u-net architecture for medical image segmentation, in: Deep Learning in Medical Image Analysis and Multimodal Learning for Clini- cal Decision Support: 4th International Workshop, DLMIA 2018, and 8th International Workshop, ML-CDS 2018, Held in Conjunction with MIC- CAI 2018, Grana...

    work page 2018

  16. [16]

    X. Li, W. Qian, D. Xu, C. Liu, Image segmentation based on improved unet, in: Journal of Physics: Conference Series, V ol. 1815, IOP Publish- ing, 2021, p. 012018

    work page 2021

  17. [17]

    Hirahara, E

    D. Hirahara, E. Takaya, M. Kadowaki, Y . Kobayashi, T. Ueda, E ffect of the pixel interpolation method for downsampling medical images on deep learning accuracy, Journal of Computer and Communications 9 (11) (2021) 150–156

    work page 2021

  18. [18]

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, Y . Wei, Deformable con- volutional networks, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 764–773

    work page 2017

  19. [19]

    Q. Jin, Z. Meng, T. D. Pham, Q. Chen, L. Wei, R. Su, Dunet: A de- formable network for retinal vessel segmentation, Knowledge-Based Sys- tems 178 (2019) 149–162

    work page 2019

  20. [20]

    Y . Qi, Y . He, X. Qi, Y . Zhang, G. Yang, Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 6070–6079

    work page 2023

  21. [21]

    S. Shit, J. C. Paetzold, A. Sekuboyina, I. Ezhov, A. Unger, A. Zhylka, J. P. Pluim, U. Bauer, B. H. Menze, cldice-a novel topology-preserving loss function for tubular structure segmentation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 16560–16569

    work page 2021

  22. [22]

    J. R. Clough, N. Byrne, I. Oksuz, V . A. Zimmer, J. A. Schnabel, A. P. King, A topological loss function for deep-learning based image segmen- tation using persistent homology, IEEE transactions on pattern analysis and machine intelligence 44 (12) (2020) 8766–8778

    work page 2020

  23. [23]

    Cervantes, J

    J. Cervantes, J. Cervantes, F. Garc ´ıa-Lamont, A. Yee-Rendon, J. E. Cabr- era, L. D. Jalili, A comprehensive survey on segmentation techniques for retinal vessel segmentation, Neurocomputing 556 (2023) 126626

    work page 2023

  24. [24]

    S. Y . Shin, S. Lee, I. D. Yun, K. M. Lee, Deep vessel segmentation by learning graphical connectivity, Medical image analysis 58 (2019) 101556

    work page 2019

  25. [25]

    Siddique, S

    N. Siddique, S. Paheding, C. P. Elkin, V . Devabhaktuni, U-net and its variants for medical image segmentation: A review of theory and appli- cations, Ieee Access 9 (2021) 82031–82057

    work page 2021

  26. [26]

    T. A. Soomro, A. J. Afifi, L. Zheng, S. Soomro, J. Gao, O. Hellwich, M. Paul, Deep learning models for retinal blood vessels segmentation: a review, IEEE Access 7 (2019) 71696–71717

    work page 2019

  27. [27]

    N. Wang, K. Li, G. Zhang, Z. Zhu, P. Wang, Improvement of retinal vessel segmentation method based on u-net, Electronics 12 (2) (2023) 262

    work page 2023

  28. [28]

    F. I. Diakogiannis, F. Waldner, P. Caccetta, C. Wu, Resunet-a: A deep learning framework for semantic segmentation of remotely sensed data, ISPRS Journal of Photogrammetry and Remote Sensing 162 (2020) 94– 114

    work page 2020

  29. [29]

    Mahapatra, A

    D. Mahapatra, A. Poellinger, M. Reyes, Interpretability-guided inductive bias for deep learning based medical image, Medical image analysis 81 (2022) 102551

    work page 2022

  30. [30]

    P. K. Gadosey, Y . Li, E. A. Agyekum, T. Zhang, Z. Liu, P. T. Yamak, F. Essaf, Sd-unet: Stripping down u-net for segmentation of biomedical images on platforms with low computational budgets, Diagnostics 10 (2) (2020) 110

    work page 2020

  31. [31]

    J. Ba, B. Frey, Adaptive dropout for training deep neural networks, Ad- vances in neural information processing systems 26 (2013)

    work page 2013

  32. [32]

    W. Liu, H. Yang, T. Tian, Z. Cao, X. Pan, W. Xu, Y . Jin, F. Gao, Full-resolution network and dual-threshold iteration for retinal vessel and coronary angiograph segmentation, IEEE Journal of Biomedical and Health Informatics 26 (9) (2022) 4623–4634

    work page 2022

  33. [33]

    N. Shen, Z. Wang, J. Li, H. Gao, W. Lu, P. Hu, L. Feng, Multi-organ segmentation network for abdominal ct images based on spatial atten- tion and deformable convolution, Expert Systems with Applications 211 (2023) 118625

    work page 2023

  34. [34]

    S. Dong, Z. Pan, Y . Fu, Q. Yang, Y . Gao, T. Yu, Y . Shi, C. Zhuo, Deu-net 2.0: Enhanced deformable u-net for 3d cardiac cine mri segmentation, Medical Image Analysis 78 (2022) 102389

    work page 2022

  35. [35]

    G. Tsechpenakis, Deformable model-based medical image segmentation, in: Multi Modality State-of-the-Art Medical Image Segmentation and Registration Methodologies: V olume 1, Springer, 2011, pp. 33–67

    work page 2011

  36. [36]

    X. Yang, Z. Li, Y . Guo, D. Zhou, Dcu-net: A deformable convolutional neural network based on cascade u-net for retinal vessel segmentation, Multimedia Tools and Applications 81 (11) (2022) 15593–15607

    work page 2022

  37. [37]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, 12 Ł. Kaiser, I. Polosukhin, Attention is all you need, Advances in neural information processing systems 30 (2017)

    work page 2017

  38. [38]

    R. Hou, H. Chang, B. Ma, S. Shan, X. Chen, Cross attention network for few-shot classification, Advances in neural information processing sys- tems 32 (2019)

    work page 2019

  39. [39]

    Y . Yuan, L. Zhang, L. Wang, H. Huang, Multi-level attention network for retinal vessel segmentation, IEEE Journal of Biomedical and Health Informatics 26 (1) (2021) 312–323

    work page 2021

  40. [40]

    Y . Lv, H. Ma, J. Li, S. Liu, Attention guided u-net with atrous convolution for accurate retinal vessels segmentation, IEEE Access 8 (2020) 32826– 32839

    work page 2020

  41. [41]

    C. Dong, S. Xu, D. Dai, Y . Zhang, C. Zhang, Z. Li, A novel multi- attention, multi-scale 3d deep network for coronary artery segmentation, Medical Image Analysis 85 (2023) 102745

    work page 2023

  42. [42]

    N. Mu, Z. Lyu, M. Rezaeitaleshmahalleh, J. Tang, J. Jiang, An atten- tion residual u-net with di fferential preprocessing and geometric post- processing: Learning how to segment vasculature including intracranial aneurysms, Medical image analysis 84 (2023) 102697

    work page 2023

  43. [43]

    Cheng, S

    J. Cheng, S. Tian, L. Yu, C. Gao, X. Kang, X. Ma, W. Wu, S. Liu, H. Lu, Resganet: Residual group attention network for medical image classifica- tion and segmentation, Medical Image Analysis 76 (2022) 102313

    work page 2022

  44. [44]

    L. Xia, H. Zhang, Y . Wu, R. Song, Y . Ma, L. Mou, J. Liu, Y . Xie, M. Ma, Y . Zhao, 3d vessel-like structure segmentation in medical images by an edge-reinforced network, Medical Image Analysis 82 (2022) 102581

    work page 2022

  45. [45]

    M. R. K. Mookiah, S. Hogg, T. J. MacGillivray, V . Prathiba, R. Pradeepa, V . Mohan, R. M. Anjana, A. S. Doney, C. N. Palmer, E. Trucco, A review of machine learning methods for retinal blood vessel segmentation and artery/vein classification, Medical Image Analysis 68 (2021) 101905

    work page 2021

  46. [46]

    C. Guo, M. Szemenyei, Y . Yi, W. Wang, B. Chen, C. Fan, Sa-unet: Spatial attention u-net for retinal vessel segmentation, in: 2020 25th international conference on pattern recognition (ICPR), IEEE, 2021, pp. 1236–1242

    work page 2020

  47. [47]

    Y .-f. Zhu, X. Xu, X.-d. Zhang, M.-s. Jiang, Ccs-unet: a cross-channel spa- tial attention model for accurate retinal vessel segmentation, Biomedical Optics Express 14 (9) (2023) 4739–4758

    work page 2023

  48. [48]

    Pahwa, D

    E. Pahwa, D. Mehta, S. Kapadia, D. Jain, A. Luthra, Medskip: Medical report generation using skip connections and integrated attention, in: Pro- ceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 3409–3415

    work page 2021

  49. [49]

    Agarwal, C

    A. Agarwal, C. Arora, Attention attention everywhere: Monocular depth prediction with skip attention, in: Proceedings of the IEEE /CVF Winter Conference on Applications of Computer Vision, 2023, pp. 5861–5870

    work page 2023

  50. [50]

    X. Mao, Y . Zhao, B. Chen, Y . Ma, Z. Gu, S. Gu, J. Yang, J. Cheng, J. Liu, Deep learning with skip connection attention for choroid layer segmenta- tion in oct images, in: 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), IEEE, 2020, pp. 1641–1645

    work page 2020

  51. [51]

    Welch, G

    G. Welch, G. Bishop, et al., An introduction to the kalman filter (1995)

    work page 1995

  52. [52]

    Wong, C.-M

    C.-C. Wong, C.-M. V ong, Persistent homology based graph convolu- tion network for fine-grained 3d shape segmentation, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 7098–7107

    work page 2021

  53. [53]

    Wasserman, Topological data analysis, Annual Review of Statistics and Its Application 5 (2018) 501–532

    L. Wasserman, Topological data analysis, Annual Review of Statistics and Its Application 5 (2018) 501–532

    work page 2018

  54. [54]

    Br ¨uel-Gabrielsson, B

    R. Br ¨uel-Gabrielsson, B. J. Nelson, A. Dwaraknath, P. Skraba, L. J. Guibas, G. Carlsson, A topology layer for machine learning (2020). arXiv:1905.12200

    work page arXiv 2020

  55. [55]

    Staal, M

    J. Staal, M. D. Abr `amoff, M. Niemeijer, M. A. Viergever, B. Van Gin- neken, Ridge-based vessel segmentation in color images of the retina, IEEE transactions on medical imaging 23 (4) (2004) 501–509

    work page 2004

  56. [56]

    K. U. Research., Chase db1 database, [EB /OL], http://blogs. kingston.ac.uk/retinal/chasedb1/ Accessed Jan. 2011

    work page 2011

  57. [57]

    Hoover, V

    A. Hoover, V . Kouznetsova, M. Goldbaum, Locating blood vessels in reti- nal images by piecewise threshold probing of a matched filter response, IEEE Transactions on Medical imaging 19 (3) (2000) 203–210

    work page 2000

  58. [58]

    K. Jin, X. Huang, J. Zhou, Y . Li, Y . Yan, Y . Sun, Q. Zhang, Y . Wang, J. Ye, Fives: A fundus image dataset for artificial intelligence based vessel segmentation, Scientific Data 9 (1) (2022) 475

    work page 2022

  59. [59]

    M. Li, K. Huang, Q. Xu, J. Yang, Y . Zhang, K. Xie, S. Yuan, Q. Liu, Q. Chen, Octa-500: a retinal dataset for optical coherence tomography angiography study, Medical Image Analysis (2024) 103092

    work page 2024

  60. [60]

    M. M. Rahman, R. Marculescu, G-cascade: E fficient cascaded graph con- volutional decoding for 2d medical image segmentation, in: Proceedings of the IEEE /CVF Winter Conference on Applications of Computer Vi- sion, 2024, pp. 7728–7737

    work page 2024

  61. [61]

    Y . He, V . Nath, D. Yang, Y . Tang, A. Myronenko, D. Xu, Swinunetr- v2: Stronger swin transformers with stagewise convolutions for 3d med- ical image segmentation, in: International Conference on Medical Im- age Computing and Computer-Assisted Intervention, Springer, 2023, pp. 416–426

    work page 2023

  62. [62]

    W. Wang, J. Zhong, H. Wu, Z. Wen, J. Qin, Rvseg-net: An e fficient fea- ture pyramid cascade network for retinal vessel segmentation, in: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceed- ings, Part V 23, Springer, 2020, pp. 796–805. 13

    work page 2020