pith. sign in

arxiv: 1906.08501 · v1 · pith:VXV5K43Enew · submitted 2019-06-20 · 📡 eess.IV · cs.CV

A Segmentation-Oriented Inter-Class Transfer Method: Application to Retinal Vessel Segmentation

Chengzhi Shi , Jihong Liu , Dali Chen This is my paper

Pith reviewed 2026-05-25 19:32 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords retinal vessel segmentationinter-class transfersemi-supervised clusteringinformation bottleneckpatch-based transfermedical image segmentationtransfer learningdimensionality reduction
0
0 comments X

The pith

A two-stage transfer method using dimensionality reduction and semi-supervised clustering shows that similar images from different classes improve retinal vessel segmentation more than some same-class images.

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

The paper introduces a patch-based two-stage transfer learning approach to address data scarcity in retinal vessel segmentation tasks that require pixel-wise labels. It inserts a dimensionality-reduced layer based on information bottleneck theory to produce a task-specific feature space, followed by semi-supervised clustering to select similar instances across different source databases. The central empirical result is that images from different classes but sharing feature similarities contribute more to performance gains than some images from the same class. Reported accuracies reach 97 percent on DRIVE, 96.8 percent on STARE, and 96.77 percent on HRF, exceeding prior methods and independent human observers on two of the sets.

Core claim

The paper establishes a segmentation-oriented inter-class transfer method in which a dimensionality-reduced layer first creates a task-specific feature space according to information bottleneck principles, after which semi-supervised clustering identifies and transfers patches from other classes that exhibit similarities in that space, with results indicating these cross-class selections yield better segmentation performance than some intra-class selections on retinal vessel datasets.

What carries the argument

The dimensionality-reduced layer inserted according to information bottleneck theory, which produces the feature space enabling semi-supervised clustering to select beneficial inter-class instances for transfer.

If this is right

  • The method attains accuracies of 97 percent, 96.8 percent, and 96.77 percent on DRIVE, STARE, and HRF respectively.
  • Cross-class instances selected by feature similarity outperform some same-class instances for the segmentation task.
  • The approach mitigates data scarcity by drawing useful patches from multiple source databases without restricting to the target class.
  • Reported performance exceeds independent human observers on DRIVE at 96.37 percent and STARE at 93.39 percent.

Where Pith is reading between the lines

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

  • Feature similarity in the reduced space may prove a stronger selection signal than class labels for transfer in pixel-level segmentation problems.
  • The same two-stage process could be tested on other annotation-scarce medical imaging tasks such as lesion or organ boundary segmentation.
  • Ablation experiments that vary the number of clusters or the dimensionality reduction target would clarify how sensitive the gains are to those choices.

Load-bearing premise

The dimensionality-reduced layer produces a feature space in which semi-supervised clustering can identify cross-class similarities that genuinely improve segmentation performance rather than introducing selection artifacts.

What would settle it

Retraining the segmentation model on the target dataset while replacing the inter-class selected instances with an equal number of same-class instances or random patches and checking whether accuracy falls below the reported levels on DRIVE, STARE, or HRF would test the claim.

read the original abstract

Retinal vessel segmentation, as a principal nonintrusive diagnose method for ophthalmology diseases or diabetics, suffers from data scarcity due to requiring pixel-wise labels. In this paper, we proposed a convenient patch-based two-stage transfer method. First, based on the information bottleneck theory, we insert one dimensionality-reduced layer for task-specific feature space. Next, the semi-supervised clustering is conducted to select instances, from different sources databases, possessing similarities in the feature space. Surprisingly, we empirically demonstrate that images from different classes possessing similarities contribute to better performance than some same-class instances. The proposed framework achieved an accuracy of 97%, 96.8%, and 96.77% on DRIVE, STARE, and HRF respectively, outperforming current methods and independent human observers (DRIVE (96.37%) and STARE (93.39%)).

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 / 0 minor

Summary. The manuscript proposes a patch-based two-stage transfer method for retinal vessel segmentation. A dimensionality-reduced layer based on the information bottleneck is inserted to create a task-specific feature space; semi-supervised clustering then selects similar instances across source databases, including cross-class instances. The authors claim this yields better performance than same-class instances and report accuracies of 97%, 96.8%, and 96.77% on DRIVE, STARE, and HRF, outperforming prior methods and human observers (96.37% and 93.39%).

Significance. If the inter-class transfer mechanism is shown to drive the gains via controlled experiments, the work would offer a practical route to mitigate labeled-data scarcity in medical segmentation by exploiting cross-dataset similarities. The information-bottleneck reduction supplies a principled basis for the clustering step.

major comments (3)
  1. [Abstract] Abstract: only accuracy is reported on an imbalanced pixel-labeling task. Without sensitivity, specificity, or AUC, the headline numbers cannot be attributed to improved vessel detection rather than background bias.
  2. [Method] Method section: no values are given for the reduced dimension, clustering algorithm or hyperparameters, number of transferred instances, or selection criterion. These omissions prevent verification that the dimensionality reduction produces a feature space in which cross-class clustering reliably improves segmentation.
  3. [Experiments] Experiments/Results: the central claim that cross-class instances outperform same-class ones lacks any ablation (transfer removed, replaced by intra-class selection, or random selection). Without these controls the performance numbers cannot be attributed to the proposed inter-class transfer step.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and will incorporate revisions to improve clarity and rigor where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: only accuracy is reported on an imbalanced pixel-labeling task. Without sensitivity, specificity, or AUC, the headline numbers cannot be attributed to improved vessel detection rather than background bias.

    Authors: We agree that accuracy alone is insufficient to evaluate performance on an imbalanced task such as retinal vessel segmentation. In the revised manuscript we will report sensitivity, specificity, and AUC in addition to accuracy to better substantiate the improvements in vessel detection. revision: yes

  2. Referee: [Method] Method section: no values are given for the reduced dimension, clustering algorithm or hyperparameters, number of transferred instances, or selection criterion. These omissions prevent verification that the dimensionality reduction produces a feature space in which cross-class clustering reliably improves segmentation.

    Authors: We acknowledge that these implementation details were omitted. The revised manuscript will specify the reduced dimension, the clustering algorithm, all relevant hyperparameters, the number of transferred instances, and the exact selection criterion to enable verification and reproducibility. revision: yes

  3. Referee: [Experiments] Experiments/Results: the central claim that cross-class instances outperform same-class ones lacks any ablation (transfer removed, replaced by intra-class selection, or random selection). Without these controls the performance numbers cannot be attributed to the proposed inter-class transfer step.

    Authors: The manuscript reports an empirical demonstration that cross-class similar instances yield better performance than certain same-class instances via cross-dataset comparisons. To strengthen attribution to the inter-class mechanism, the revision will include explicit ablation experiments (no transfer, intra-class selection, and random selection). revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical method with no derivation chain or fitted predictions

full rationale

The paper describes a two-stage empirical pipeline (information-bottleneck dimensionality reduction followed by semi-supervised clustering for inter-class instance selection) and reports accuracy numbers on DRIVE/STARE/HRF. No equations, uniqueness theorems, ansatzes, or predictive claims appear in the provided text. Performance figures are presented as experimental outcomes rather than quantities derived from or fitted to the same data in a self-referential loop. The central claim therefore cannot reduce to its inputs by construction, satisfying the default expectation of no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review is based on abstract only; no explicit free parameters, axioms, or invented entities are described in the provided text.

axioms (1)
  • domain assumption Information bottleneck theory can be applied to insert a dimensionality-reduced layer that creates a task-specific feature space for segmentation.
    Invoked as the basis for the first stage of the method.

pith-pipeline@v0.9.0 · 5676 in / 1300 out tokens · 25658 ms · 2026-05-25T19:32:29.410652+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

55 extracted references · 55 canonical work pages

  1. [1]

    INTRODUCTION After being processed, retinal fundus images are used to diagnose diseases like diabetic retinopathy, glaucoma, cardiovascular and hypertension. Accurately extracting vessels is still a difficult task for several reasons like the presence of noise, the low contr ast between vasculature and background, the variations in illumination and shape....

    work page

  2. [2]

    PREPROCESSING AND METHODOLOGY 2.1. Preprocessing We work on the green channel of the retinal image since the green channel of the RGB color retinal image presents a higher contrast between the vessels and the ba ckground. In addition, we normalize the green channels of images by using the following formula: 𝐼𝑃𝐺 = 𝐼𝐺−𝜇 𝜎 (1) where μ and σ denote the mean a...

    work page

  3. [4]

    Initialize mean vector 𝜇𝑗 = 1 |𝑆𝑗| ∑ 𝑥𝑥∈𝑆𝑗

    work page

  4. [5]

    Initialize set of clusters 𝐶𝑗 ← ∅(1 ≤ 𝑗 ≤ 𝑘)

    work page

  5. [6]

    First divide labeled samples into corresponding cluster coll ection 𝐶𝑗 ← 𝐶𝑗 ∪ {𝑥}

    work page

  6. [7]

    for all 𝑥𝑖𝜖𝐷\𝑆 (1 ≤ i ≤ m) do

    work page

  7. [8]

    Initialize set of cluster results of patches of each group 𝑟𝑁 ← ∅

    work page

  8. [9]

    for all 𝑥𝑛 𝑖 𝜖𝐷\𝑆 (1 ≤ n ≤ 10) do

    work page

  9. [10]

    Compute distance 𝑑𝑛𝑗 𝑖 = ‖𝑥𝑛 𝑖 − 𝜇𝑗‖2

    work page

  10. [11]

    Find nearest cluster to sample 𝑥𝑛 𝑖 : 𝑟𝑛 = arg 𝑚𝑖𝑛𝑗∈{1,2,..,𝑘} 𝑑𝑛𝑗 𝑖

    work page

  11. [12]

    Put every individual clustered result into a collection 𝑟𝑁 ← 𝑟𝑁 ∪ {𝑟𝑛}

    work page

  12. [13]

    Find the element most often appear in 𝑟𝑁 : r

    work page

  13. [14]

    Divide 𝑥𝑖 into the corresponding cluster collection 𝐶𝑟 ← 𝐶𝑟 ∪ {𝑥𝑖}

    work page

  14. [15]

    for j=1, 2, ..., k do

    work page

  15. [16]

    Update mean vector 𝜇𝑗 = 1 |𝐶𝑗| ∑ 𝑥𝑥∈𝐶𝑗

    work page

  16. [17]

    until mean vector does not update

    work page

  17. [18]

    Databases In the experimental part, we mainly rely on the public dataset DRIVE [11] for evaluating the feasibility and validity of the proposed algorithm

    EXPERIMENT AND RESULTS 3.1. Databases In the experimental part, we mainly rely on the public dataset DRIVE [11] for evaluating the feasibility and validity of the proposed algorithm. Due to the small sample size of DRIVE, we are motivated to use seve ral datasets for transfer, either because they (STARE [12], CHASE [13], HRF [14]) are also fundus image da...

    work page

  18. [19]

    CONCLUSION In this paper, we propose a task -specific transfer method based on the information bottleneck theory to solve the data scarcity in medical image segmentation. Our me thod outperformed state -of-the-art methods with an accuracy of 97%, 96.8% and 96.77% on DRIVE, STARE and HRF respectively, outperforming both current competitor methods and indep...

    work page

  19. [20]

    Expert consensus on the diagnosis of osteoporosis in Chinese Population ,

    Z Z hang, Z Liu, N Li, et al. “Expert consensus on the diagnosis of osteoporosis in Chinese Population ,” Chin J Osteoporos , vol.09(4), pp.1007-1010,2014

    work page 2014

  20. [21]

    Semi-Supervised Discriminative Classification Robust to Sample -Outliers and Feature -Noises,

    Adeli E , Thung K H , An L , et al. "Semi-Supervised Discriminative Classification Robust to Sample -Outliers and Feature -Noises," IEEE Transactions on Pattern Analysis and Machine Intelligence , 41(2), pp. 515-522, 2019

    work page 2019

  21. [22]

    Semi-supervised learning of anatomical manifolds for atlas -based segmentation of medical images,

    M. Borga, T. Andersson and O. D. Leinhard, "Semi-supervised learning of anatomical manifolds for atlas -based segmentation of medical images," 2016 23rd International Conference on Pattern Recognition (ICPR), Cancun, pp. 3146-3149,2016

    work page 2016

  22. [23]

    Label Transfer by Measuring Compactness,

    R. Varga and S. Nedevschi, "Label Transfer by Measuring Compactness," IEEE Transactions on Image Processing, vol. 22, no. 12, pp. 4711-4723, Dec. 2013

    work page 2013

  23. [24]

    Convolutional neural networks based transfer learning for diabetic retinopathy fundus image classification,

    X. Li, T. Pang, B. Xiong, W. Liu, P. Liang and T. Wang, "Convolutional neural networks based transfer learning for diabetic retinopathy fundus image classification," International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP -BMEI), Shanghai, pp. 1-11, 2017

    work page 2017

  24. [25]

    Cross-modality medical image detection and segmentation by transfer learning of shapel priors,

    Y. Zheng, "Cross-modality medical image detection and segmentation by transfer learning of shapel priors," International Symposium on Biomedical Imaging (ISBI), New York, NY, pp. 424-427, 2015

    work page 2015

  25. [26]

    Constrained Deep Transf er Feature Learning and Its Applications,

    Y. Wu and Q. Ji, "Constrained Deep Transf er Feature Learning and Its Applications," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, pp. 5101-5109,2016

    work page 2016

  26. [27]

    Adaptive histogram equalization and its variations,

    S.M. Pizer, E.P. Amburn, “Adaptive histogram equalization and its variations,” Comput. Vis. Graph. Image Process. 39 (3) ,pp. 355–368, 1987

    work page 1987

  27. [28]

    Image Detail and Color Enhancement Based on Channel-Wise Local Gamma Adjustment,

    C Tseng, S Wang, Y Lai, et al. 67.5: “Image Detail and Color Enhancement Based on Channel-Wise Local Gamma Adjustment,” Sid Symposium Digest of Technical Papers, 40(1), pp. 1022–1025, 2009

    work page 2009

  28. [29]

    The information bottleneck method,

    N. Tishby, F. C. Pereira, and W. Bialek, “The information bottleneck method,” in Proc. 37 -th Annu. Allerton Conf. Commun. , Control Comput.,pp. 368–377,1999

    work page 1999

  29. [30]

    Ridge -based vessel segmentation in color images of the retina,

    J. Staal, M. D. Abrà moff, M. Niemeijer, M. A. Viergever, and B. van Ginneken, “Ridge -based vessel segmentation in color images of the retina,” IEEE Trans. Med. Imag., vol. 23, no. 4, pp. 501–509, Apr. 2004

    work page 2004

  30. [31]

    Locating blood vessels in retinal images by piecewise threshold probing of a matched filter response,

    A. Hoover, V. Kouznetsova, and M. Goldbaum, “Locating blood vessels in retinal images by piecewise threshold probing of a matched filter response,” IEEE Trans. Med. Imag., vol. 19, no. 3, pp. 203 –210, Mar. 2000

    work page 2000

  31. [32]

    Measuring retinal vessel tortuosity in 10 -year-old children: Validation of the computer -assisted image analysis of the retina (CAIAR) program,

    C. G. Owen et al., “Measuring retinal vessel tortuosity in 10 -year-old children: Validation of the computer -assisted image analysis of the retina (CAIAR) program,” Invest. Ophthalmol. Vis. Sci ., vol. 50, no.5, pp. 2004–2010, 2009

    work page 2004

  32. [33]

    Robust Vessel Segmentation in Fundus Images,

    A Budai et al, Georg, “Robust Vessel Segmentation in Fundus Images,” International Journal of Biomedical Imaging, vol. 2013, 2013

    work page 2013

  33. [34]

    An Integrated Micro - and Macroarchitectural Analysis of the Drosophila Brain by Computer-Assisted Serial Section Electron Microscopy,

    A Cardona et al. “An Integrated Micro - and Macroarchitectural Analysis of the Drosophila Brain by Computer-Assisted Serial Section Electron Microscopy,” PLoS Biology, 8(10) , e1000502,2010

    work page 2010

  34. [35]

    Manually segmented vascular networks from images of retina with proliferative diabetic and hypertensive retinopathy,

    N Popovic, M Radunovic, B Jelena, et al, “Manually segmented vascular networks from images of retina with proliferative diabetic and hypertensive retinopathy,” Data in Brief, S2352340918302464,2018

    work page 2018

  35. [36]

    VascuSynth: simulating vascular trees for generating volumetric image data with ground -truth segmentation and tree analysis,

    G Hamarneh, P Jassi, “VascuSynth: simulating vascular trees for generating volumetric image data with ground -truth segmentation and tree analysis,” Computerized Medical Imaging & Graphics the Official Journal of the Computerized Medical Imaging Society, 34(8):605, 2010

    work page 2010

  36. [37]

    Evaluation and benchmark for biological image segmentation,

    E. D Gelasca, J Byun, B Obara, et al, “Evaluation and benchmark for biological image segmentation,” IEEE International Conference on Image Processing, pp.1816-1819,2008

    work page 2008

  37. [38]

    Understanding the difficul ty of training deep feedforward neural networks,

    X Glorot, Y Bengio. “Understanding the difficul ty of training deep feedforward neural networks,” Journal of Machine Learning Research, pp. 249-256, 2010

    work page 2010

  38. [39]

    Segmenting Retinal Blood Vessels With Deep Neural Networks

    L Paweł, and K. Krawiec. "Segmenting Retinal Blood Vessels With Deep Neural Networks." IEEE Transactions on Medical Imaging 35.11,2369-2380, 2016

    work page 2016

  39. [40]

    Patch -based fully convolutional neural network with skip connections for retina l blood vessel segmentation,

    Z. Feng, J. Yang and L. Yao, "Patch -based fully convolutional neural network with skip connections for retina l blood vessel segmentation," 2017 IEEE International Conference on Image Processing (ICIP), Beijing, pp. 1742-1746, 2017

    work page 2017

  40. [41]

    Segmentation of retinal blood vessels by combining the detection of centerlines and morphological reconstruction,

    A. Mendonca and A. Campilho, “Segmentation of retinal blood vessels by combining the detection of centerlines and morphological reconstruction,” IEEE Trans. Med. Imag., vol. 25, no. 9, pp. 1200–1213, Aug. 2006

    work page 2006

  41. [42]

    Robust retinal vessel segmentation via locally adaptive derivative frames in orientation scores,

    J. Zhang et al., “Robust retinal vessel segmentation via locally adaptive derivative frames in orientation scores,” IEEE Trans. Med. Imag., vol. 35, no. 12, pp. 2631-2644, Aug. 2016

    work page 2016

  42. [43]

    A cross -modality learning approach for vessel segmentation in retinal images,

    Q. Li et al., “A cross -modality learning approach for vessel segmentation in retinal images,” IEEE Trans. Med. Imag., vol. 35, no. 1, pp. 109-118, Jan. 2016

    work page 2016

  43. [44]

    A fully convolutional neural network based structured prediction approach towards the retinal vessel segmentation,

    A. Dasgupta, and S. Singh, “A fully convolutional neural network based structured prediction approach towards the retinal vessel segmentation,” Proc. ISBI, pp. 18-21, 2017

    work page 2017

  44. [45]

    Joint Segment -level and Pixel -wise Losses for Deep Learning based Retinal Vessel Segmentation,

    Z Yan, X Yang, K T T Cheng. “Joint Segment -level and Pixel -wise Losses for Deep Learning based Retinal Vessel Segmentation,” IEEE Transactions on Biomedical Engineering,65(9), pp.1912 - 1923, 2018

    work page 1912

  45. [46]

    Holistically -Nested Edge Detection,

    S. Xie and Z. Tu, "Holistically -Nested Edge Detection," 2015 IEEE International Conference on Computer Vision (ICCV) , Santiago, pp. 1395-1403, 2015

    work page 2015

  46. [47]

    Retinal blood vessel segmentation using line operators and support vector classification,

    E. Ricci and R. Perfetti, “Retinal blood vessel segmentation using line operators and support vector classification,” IEEE Trans. Med. Imag ., vol. 26, no. 10, pp. 1357–1365, Oct. 2007

    work page 2007

  47. [48]

    Automated Vessel Segmentation Using Infinite Perimeter Active Contour Model with Hybrid Region Information with Application to Retina Images

    Y Zhao, L Rada, K Chen et al, “Automated Vessel Segmentation Using Infinite Perimeter Active Contour Model with Hybrid Region Information with Application to Retina Images.” IEEE Transactions on Medical Imaging, 34(9):1797-1807, 2015

    work page 2015

  48. [49]

    Retinal vessel segmentation using the 2-D Gabor wavelet and supervised classification,

    J. Soares et al., “Retinal vessel segmentation using the 2-D Gabor wavelet and supervised classification,” IEEE Trans. Med. Imag., vol. 25, no. 9, pp. 1214–1222, Sep. 2006

    work page 2006

  49. [50]

    A new supervised method for blood vessel segmentation in retinal images by using gray-level and moment invariants -based features,

    D. Marí n, A. Aquino, M. E. Gegú ndez-Arias, and J. M. Bravo, “A new supervised method for blood vessel segmentation in retinal images by using gray-level and moment invariants -based features,” IEEE Trans. Med. Imag., vol. 30, no. 1, pp. 146–158, Jan. 2011

    work page 2011

  50. [51]

    An ensemble classification-based approach applied to retinal blood vessel segmentation,

    M. M. Fraz et al, “An ensemble classification-based approach applied to retinal blood vessel segmentation,” IEEE Trans. Biomed. Eng., vol. 59, no. 9, pp. 2538–2548, Sep. 2012

    work page 2012

  51. [52]

    General retinal vessel segmentation using regularizationbased multiconcavity modeling,

    B. Lam et al., “General retinal vessel segmentation using regularizationbased multiconcavity modeling,” IEEE Trans. Med. Imag., vol. 29, no. 7, pp. 1369–1381, Mar. 2010

    work page 2010

  52. [53]

    Blood vessel segmentation methodologies in retinal images –a survey,

    M. M. Fraz, P. Remagnino, A. Hoppe, B. Uyyanonvara, A. R. Rudnicka, C. G. Owen, and S. A. Barman, “Blood vessel segmentation methodologies in retinal images –a survey,” Computer M ethods and ProgramsIn Biomedicine, vol. 108, no. 1, pp. 407–433, 2012

    work page 2012

  53. [54]

    Retinal vessel segmentation by improved matched filtering: Evaluation on a new highresolution fundus image database,

    J. Odstrcilik, R. Kolar, A. Budai, J. Hornegger, J. Jan, J. Gazarek, T. Kubena, P. Cernosek, O. Svoboda, and E. Angelopoulou, “Retinal vessel segmentation by improved matched filtering: Evaluation on a new highresolution fundus image database,” IET Image Process., vol. 7, no. 4, pp. 373–383, 2013

    work page 2013

  54. [55]

    Fast vessel segmentation in retinal images using multiscale enhancement and second-order local entropy,

    H. Yu, S. Barriga, C. Agurto, G. Zamora, W. Bauman, and P. Soliz, “Fast vessel segmentation in retinal images using multiscale enhancement and second-order local entropy,” Proc. SPIE, vol. 8315, pp. 83151B-1– 83151B-12, Feb. 2012

    work page 2012

  55. [56]

    Leveraging Multiscale Hessian -Based Enhancement With a Novel Exudate Inpainting Technique for Retina l Vessel Segmentation,

    R. Annunziata, A. Garzelli, L. Ballerini, A. Mecocci and E. Trucco, "Leveraging Multiscale Hessian -Based Enhancement With a Novel Exudate Inpainting Technique for Retina l Vessel Segmentation," in IEEE Journal of Biomedical and Health Informatics , vol. 20, no. 4, pp. 1129-1138, July 2016

    work page 2016