pith. sign in

arxiv: 2606.15129 · v2 · pith:SLO2OJLAnew · submitted 2026-06-13 · 💻 cs.CV · cs.AI

EyeMVP: OCT-Informed Fundus Representation Learning via Paired CFP--OCT Pretraining

Zhuo Deng , Ruiheng Zhang , Ziheng Zhang , Weihao Gao , Yitong Li , Qian Wang , Lei Shao , Jiaoyue Dong
show 18 more authors
Zhixi Zeng Lijian Fang Haibo Wang Xiaobin Lin Tao Liu Zhicheng Du Zhengwei Zhang Lin Yang Zheng Gong Xinyu Zhao Zhenquan Wu Fang Li Zhiguang Zhou Guoming Zhang Sun Jing Han Lv Wenbin We Lan Ma
This is my paper

Pith reviewed 2026-06-30 10:20 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords cross-modal learningretinal foundation modelcolor fundus photographyoptical coherence tomographyrepresentation learningmacular edemamyopic macular schisisfundus image analysis
0
0 comments X

The pith

Paired CFP-OCT pretraining transfers depth information into color fundus photography features for stronger single-modality diagnosis.

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

The paper presents EyeMVP, a foundation model pretrained on 674,893 paired CFP-OCT triples to learn CFP representations enriched by OCT structural details. Cross-modal masked reconstruction supplies the supervision while source-constrained cross-attention and CFP-derived masks handle the geometric mismatch between en-face photos and cross-sectional scans. At test time only CFP is required, yet the model matches or exceeds prior retinal foundation models on 15 dataset-level tasks in full-data and few-shot regimes, with particular gains on macular edema and myopic macular schisis. A reader study further shows the model outperforming junior and intermediate ophthalmologists on these conditions.

Core claim

EyeMVP shows that cross-modal masked reconstruction from paired CFP-OCT data can enrich CFP features with OCT-associated supervision, producing representations that support accurate CFP-only inference on tasks where depth information matters.

What carries the argument

Cross-modal masked reconstruction paired with source-constrained cross-attention and CFP-derived structural masks that accommodate non-aligned CFP and OCT geometry.

If this is right

  • The model attains AUROCs of 0.923 for macular edema and 0.867 for myopic macular schisis using only CFP at inference.
  • Consistent performance gains appear on macular and optic-nerve tasks across both classification and segmentation.
  • EyeMVP matches or exceeds representative retinal foundation models in 15 dataset-level settings under full-data and few-shot regimes.
  • In the reader study the model surpasses junior and intermediate ophthalmologists on macular edema and all groups on myopic macular schisis.

Where Pith is reading between the lines

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

  • The same paired-pretraining pattern could be tested on other accessible-versus-informative modality pairs in medical imaging.
  • Large multi-hospital paired datasets appear sufficient to bootstrap single-modality models that retain cross-modal benefits.
  • The few-shot gains suggest the learned representations may reduce labeled data needs in new clinical sites.

Load-bearing premise

The cross-modal masked reconstruction and source-constrained cross-attention successfully transfer OCT-derived structural information into CFP features despite non-aligned geometry.

What would settle it

An ablation that removes the OCT pretraining component and shows no drop in AUROC on macular edema or myopic macular schisis classification would falsify the transfer claim.

Figures

Figures reproduced from arXiv: 2606.15129 by Fang Li, Guoming Zhang, Haibo Wang, Han Lv, Jiaoyue Dong, Lan Ma, Lei Shao, Lijian Fang, Lin Yang, Qian Wang, Ruiheng Zhang, Sun Jing, Tao Liu, Weihao Gao, Wenbin We, Xiaobin Lin, Xinyu Zhao, Yitong Li, Zheng Gong, Zhengwei Zhang, Zhenquan Wu, Zhicheng Du, Zhiguang Zhou, Zhixi Zeng, Zhuo Deng, Ziheng Zhang.

Figure 1
Figure 1. Figure 1: Overview of the EyeMVP pretraining framework. Each pretraining sample contains a CFP image, a paired OCT B-scan, and a CFP-derived structural mask (CFP-Seg). Dirichlet sampling dynamically assigns visible token ratios across modalities, and modality-specific input projections map the visible tokens into a shared ViT encoder. Stage I uses three intra-modal decoders to reconstruct CFP, OCT, and CFP-Seg targe… view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison of full-data segmentation results. Columns show the input CFP image, predictions from RETFound, VisionFM, EyeCLIP, EyeMVP, and the ground truth annotation. Rows correspond to retinal vessel segmentation, optic disc/cup segmentation, hard exudate segmentation, and hemorrhage segmentation. EyeMVP produces segmentation masks that are generally closer to the ground truth, particularly fo… view at source ↗
Figure 3
Figure 3. Figure 3: t-SNE visualization of frozen CFP encoder embeddings for RETFound, EyeCLIP, VisionFM, and EyeMVP (Ours). Each point cor￾responds to the encoder output embedding of one CFP image projected to two dimensions; colors denote ground-truth class labels and are not used during projection. Rows correspond to four tasks: DR (six severity grades), AMD (normal vs. AMD), myopic macular schisis (normal vs. schisis), an… view at source ↗
Figure 4
Figure 4. Figure 4: , where all OCT tokens are masked and only partial CFP tokens are provided. Compared with the Stage-I-only variant, EyeMVP produces reconstructions with clearer OCT￾like retinal contours and layer-related contrast. These visual￾izations illustrate the pretraining objective and are not intended to represent diagnostic OCT synthesis. Removing the CFP-Seg branch decreases AUROC by 4.91 percentage points on OR… view at source ↗
read the original abstract

Color fundus photography (CFP) is the mainstay of large-scale retinal screening, but its diagnostic capacity is limited by the lack of depth-resolved structure, which optical coherence tomography (OCT) provides yet is less accessible at population scale. We present EyeMVP, a cross-modal retinal foundation model that uses paired CFP--OCT pretraining to learn OCT-informed CFP representations while requiring only CFP at inference. Pretrained on 674,893 same-eye same-day CFP--OCT triples from 112,642 patients across eight hospitals, EyeMVP uses cross-modal masked reconstruction to enrich CFP features with OCT-associated supervision, and combines source-constrained cross-attention with CFP-derived structural masks to accommodate the non-aligned geometry of en-face CFP and cross-sectional OCT. Across 15 dataset-level settings spanning classification and segmentation, under both full-data and few-shot regimes, EyeMVP performs on par with or better than representative retinal foundation models, with consistent gains on macular and optic-nerve tasks; it attains AUROCs of 0.923 for macular edema and 0.867 for myopic macular schisis, two conditions poorly resolved in CFP. In an exploratory reader study, EyeMVP surpasses junior and intermediate ophthalmologists but not seniors on macular edema, while exceeding all groups on myopic macular schisis. These results indicate that cross-modal reconstruction can enrich CFP representations with OCT-associated supervision, offering a practical route to stronger CFP-based screening.

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 presents EyeMVP, a cross-modal retinal foundation model pretrained on 674,893 same-eye same-day CFP-OCT triples from 112,642 patients. It employs cross-modal masked reconstruction and source-constrained cross-attention (guided by CFP-derived structural masks) to learn OCT-informed CFP representations usable at inference with only CFP input. The model is evaluated across 15 dataset-level settings for classification and segmentation under full-data and few-shot regimes, reporting AUROCs of 0.923 for macular edema and 0.867 for myopic macular schisis, with performance on par with or better than existing retinal foundation models and gains on macular/optic-nerve tasks.

Significance. If the pretraining demonstrably transfers depth-resolved structural cues from OCT into CFP features despite non-aligned geometries, the work would provide a practical route to stronger population-scale CFP screening for conditions poorly resolved in en-face imaging alone. The scale of the paired pretraining corpus and the reader-study comparison with ophthalmologists are notable strengths that would support clinical relevance if the transfer mechanism is validated.

major comments (3)
  1. [Abstract and §3] Abstract and §3 (Pretraining Method): The central claim that cross-modal masked reconstruction plus source-constrained cross-attention successfully transfers OCT depth cues into CFP features rests on the attention mechanism accommodating non-aligned en-face vs. cross-sectional geometry; no alignment metrics, attention-map visualizations, or ablation isolating the cross-attention component versus data-scale effects are referenced, leaving open the possibility that reported gains are explained by pretraining volume alone.
  2. [§4 and Tables] §4 (Experiments) and Table 2/3 (if present): Performance numbers (AUROC 0.923 macular edema, 0.867 myopic macular schisis) are stated without reported validation-split details, number of random seeds, confidence intervals, or statistical tests against baselines; this prevents assessment of whether the consistent gains on macular tasks are robust or sensitive to split choice.
  3. [§3.2] §3.2 (Source-Constrained Cross-Attention): The description of CFP-derived structural masks guiding attention must include a concrete demonstration (e.g., correlation with OCT thickness maps or reconstruction error stratified by retinal layer) that depth-resolved information is actually encoded; without this, the method's ability to bridge the geometric mismatch remains an unverified assumption.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'on par with or better than representative retinal foundation models' should be accompanied by the exact list of compared models and their pretraining data sources for immediate clarity.
  2. [§5] §5 (Reader Study): Clarify the number of cases, reader experience definitions (junior/intermediate/senior), and whether the study was conducted on held-out data with the same distribution as the quantitative benchmarks.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify how to strengthen the presentation of our method and results. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (Pretraining Method): The central claim that cross-modal masked reconstruction plus source-constrained cross-attention successfully transfers OCT depth cues into CFP features rests on the attention mechanism accommodating non-aligned en-face vs. cross-sectional geometry; no alignment metrics, attention-map visualizations, or ablation isolating the cross-attention component versus data-scale effects are referenced, leaving open the possibility that reported gains are explained by pretraining volume alone.

    Authors: We agree that visualizations and an ablation isolating the cross-attention would provide stronger support for the transfer mechanism. The current results show consistent gains on macular and optic-nerve tasks relative to other foundation models pretrained at comparable scale, but this does not fully rule out data-volume effects. In revision we will add (i) attention-map visualizations illustrating how source-constrained cross-attention aligns CFP regions with OCT-derived structure and (ii) an ablation comparing the full model against a masked-reconstruction baseline without the cross-attention module. Alignment metrics are not applicable in the strict pixel sense because the modalities have fundamentally different geometries; the method is explicitly designed to operate without such alignment. revision: yes

  2. Referee: [§4 and Tables] §4 (Experiments) and Table 2/3 (if present): Performance numbers (AUROC 0.923 macular edema, 0.867 myopic macular schisis) are stated without reported validation-split details, number of random seeds, confidence intervals, or statistical tests against baselines; this prevents assessment of whether the consistent gains on macular tasks are robust or sensitive to split choice.

    Authors: The observation is correct; the manuscript reports single-split point estimates. In the revised version we will (i) specify the exact train/validation/test splits used for each of the 15 settings, (ii) report means and standard deviations (or 95% confidence intervals) over at least three random seeds, and (iii) add statistical comparisons (DeLong tests for AUROC, paired t-tests for other metrics) against the strongest baselines to quantify whether the observed macular-task gains are statistically significant. revision: yes

  3. Referee: [§3.2] §3.2 (Source-Constrained Cross-Attention): The description of CFP-derived structural masks guiding attention must include a concrete demonstration (e.g., correlation with OCT thickness maps or reconstruction error stratified by retinal layer) that depth-resolved information is actually encoded; without this, the method's ability to bridge the geometric mismatch remains an unverified assumption.

    Authors: We accept that a direct empirical link between the masks/attention and depth-resolved OCT quantities would strengthen the claim. While downstream task gains on conditions that require depth information provide indirect evidence, they do not constitute the requested demonstration. In revision we will add an analysis correlating attention weights (or per-layer reconstruction error) with OCT thickness maps and retinal-layer segmentations on a held-out paired subset, thereby verifying that depth-resolved structure is being encoded. revision: yes

Circularity Check

0 steps flagged

No significant circularity; standard cross-modal pretraining with empirical claims

full rationale

The paper describes a pretraining method on paired CFP-OCT data using cross-modal masked reconstruction and source-constrained cross-attention to learn CFP representations informed by OCT. No equations, fitted parameters renamed as predictions, self-definitional loops, or load-bearing self-citations appear in the abstract or method outline. Performance numbers (e.g., AUROC 0.923) are presented as empirical results from large-scale pretraining rather than reductions by construction. The derivation chain is self-contained against external benchmarks with no patterns from the enumerated circularity types.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no equations, so no free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.1-grok · 5885 in / 970 out tokens · 32180 ms · 2026-06-30T10:20:42.135738+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

40 extracted references · 4 canonical work pages · 3 internal anchors

  1. [1]

    Trends in prevalence of blindness and distance and near vision impairment over 30 years: an analysis for the global burden of disease study,

    R. Bourne, J. D. Steinmetz, S. Flaxman, P. S. Briant, H. R. Taylor, S. Resnikoff, R. J. Casson, A. Abdoli, E. Abu-Gharbieh, A. Afshin, et al., “Trends in prevalence of blindness and distance and near vision impairment over 30 years: an analysis for the global burden of disease study,”The Lancet global health, vol. 9, no. 2, pp. e130–e143, 2021

    2021

  2. [2]

    Mohammadpour,Diagnostics in ocular imaging: cornea, retina, glaucoma and orbit

    M. Mohammadpour,Diagnostics in ocular imaging: cornea, retina, glaucoma and orbit. Cham: Springer Nature, 2020

    2020

  3. [3]

    Self-supervised representation learning: Introduction, advances, and challenges,

    L. Ericsson, H. Gouk, C. C. Loy, and T. M. Hospedales, “Self-supervised representation learning: Introduction, advances, and challenges,”IEEE Signal Processing Magazine, vol. 39, no. 3, pp. 42–62, 2022

    2022

  4. [4]

    Contrastive representation learning: A framework and review,

    P. H. Le-Khac, G. Healy, and A. F. Smeaton, “Contrastive representation learning: A framework and review,”Ieee Access, vol. 8, pp. 193907– 193934, 2020. 12 IEEE TMI

    2020

  5. [5]

    Ophglm: An ophthalmology large language-and- vision assistant,

    Z. Deng, W. Gao, C. Chen, Z. Niu, Z. Gong, R. Zhang, Z. Cao, F. Li, Z. Ma, W. Wei,et al., “Ophglm: An ophthalmology large language-and- vision assistant,”Artificial Intelligence in Medicine, vol. 157, p. 103001, 2024

    2024

  6. [6]

    A foundation model for generalizable disease detection from retinal images,

    Y . Zhou, M. A. Chia, S. K. Wagner, M. S. Ayhan, D. J. Williamson, R. R. Struyven, T. Liu, M. Xu, M. G. Lozano, P. Woodward-Court, et al., “A foundation model for generalizable disease detection from retinal images,”Nature, vol. 622, no. 7981, pp. 156–163, 2023

    2023

  7. [7]

    Development and validation of a multimodal multitask vision foundation model for generalist ophthalmic artificial intelligence,

    J. Qiu, J. Wu, H. Wei, P. Shi, M. Zhang, Y . Sun, L. Li, H. Liu, H. Liu, S. Hou,et al., “Development and validation of a multimodal multitask vision foundation model for generalist ophthalmic artificial intelligence,” NEJM AI, vol. 1, no. 12, p. AIoa2300221, 2024

    2024

  8. [8]

    A multimodal visual–language foundation model for computational ophthalmology,

    D. Shi, W. Zhang, J. Yang, S. Huang, X. Chen, P. Xu, K. Jin, S. Lin, J. Wei, M. Yusufu,et al., “A multimodal visual–language foundation model for computational ophthalmology,”npj Digital Medicine, vol. 8, no. 1, p. 381, 2025

    2025

  9. [9]

    Eyefound: a multimodal generalist foundation model for ophthalmic imaging,

    D. Shi, W. Zhang, X. Chen, Y . Liu, J. Yang, S. Huang, Y . C. Tham, Y . Zheng, and M. He, “Eyefound: a multimodal generalist foundation model for ophthalmic imaging,”arXiv preprint arXiv:2405.11338, 2024

    work page arXiv 2024

  10. [10]

    Multimodal foundation model and benchmark for comprehensive reti- nal oct image analysis,

    J. Morano, B. Fazekas, E. S ¨ukei, R. Fecso, T. Emre, M. Gumpinger, G. Faustmann, M. Oghbaie, U. Schmidt-Erfurth, and H. Bogunovi ´c, “Multimodal foundation model and benchmark for comprehensive reti- nal oct image analysis,”NPJ Digital Medicine, vol. 8, no. 1, p. 576, 2025

    2025

  11. [11]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark,et al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning, pp. 8748–8763, PmLR, 2021

    2021

  12. [12]

    Masked au- toencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. Girshick, “Masked au- toencoders are scalable vision learners,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 16000– 16009, 2022

    2022

  13. [13]

    Multimae: Multi- modal multi-task masked autoencoders,

    R. Bachmann, D. Mizrahi, A. Atanov, and A. Zamir, “Multimae: Multi- modal multi-task masked autoencoders,” inEuropean Conference on Computer Vision, pp. 348–367, Springer, 2022

    2022

  14. [14]

    The war on diabetic retinopathy: where are we now?,

    T. Y . Wong and C. Sabanayagam, “The war on diabetic retinopathy: where are we now?,”Asia-Pacific Journal of Ophthalmology, vol. 8, no. 6, pp. 448–456, 2019

    2019

  15. [15]

    Pre- dicting optical coherence tomography-derived diabetic macular edema grades from fundus photographs using deep learning,

    A. V . Varadarajan, P. Bavishi, P. Raumviboonsuk, P. Chotcomwongse, S. Venugopalan, A. Narayanaswamy, J. Cuadros, K. Kanai, G. Bresnick, M. Tadarati, S. Silpa-Archa, J. Limwattanayingyong, V . Nganthavee, J. Ledsam, P. A. Keane, G. S. Corrado, L. Peng, and D. R. Webster, “Pre- dicting optical coherence tomography-derived diabetic macular edema grades from...

    2018

  16. [16]

    From machine to machine: An oct-trained deep learning algorithm for objective quan- tification of glaucomatous damage in fundus photographs,

    F. A. Medeiros, A. A. Jammal, and A. C. Thompson, “From machine to machine: An oct-trained deep learning algorithm for objective quan- tification of glaucomatous damage in fundus photographs,” 2018

    2018

  17. [17]

    Multieye: Dataset and benchmark for oct-enhanced retinal disease recognition from fundus images,

    L. Wang, C. Qi, C. Ou, L. An, M. Jin, X. Kong, and X. Li, “Multieye: Dataset and benchmark for oct-enhanced retinal disease recognition from fundus images,”IEEE Transactions on Medical Imaging, 2024

    2024

  18. [18]

    A foundation language-image model of the retina (flair): Encoding expert knowledge in text supervision,

    J. Silva-Rodriguez, H. Chakor, R. Kobbi, J. Dolz, and I. B. Ayed, “A foundation language-image model of the retina (flair): Encoding expert knowledge in text supervision,”Medical Image Analysis, vol. 99, p. 103357, 2025

    2025

  19. [19]

    Crossmae: Cross-modality masked autoencoders for region- aware audio-visual pre-training,

    Y . Guo, S. Sun, S. Ma, K. Zheng, X. Bao, S. Ma, W. Zou, and Y . Zheng, “Crossmae: Cross-modality masked autoencoders for region- aware audio-visual pre-training,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 26721– 26731, 2024

    2024

  20. [20]

    Multi-modal masked autoencoders for medical vision-and-language pre-training,

    Z. Chen, Y . Du, J. Hu, Y . Liu, G. Li, X. Wan, and T.-H. Chang, “Multi-modal masked autoencoders for medical vision-and-language pre-training,” 2022

    2022

  21. [21]

    Acquire continuous and precise score for fundus image quality assessment: Fthnet and fqs dataset,

    Z. Gong, Z. Deng, R. Gan, Z. Niu, L. Chen, C. Huang, J. Liang, W. Gao, F. Li, S. Zhang,et al., “Acquire continuous and precise score for fundus image quality assessment: Fthnet and fqs dataset,”Scientific Reports, vol. 15, no. 1, p. 40524, 2025

    2025

  22. [22]

    U-net: Convolutional networks for biomedical image segmentation,

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

    2015

  23. [23]

    Unet++: Redesigning skip connections to exploit multiscale features in image segmentation,

    Z. Zhou, M. M. R. Siddiquee, N. Tajbakhsh, and J. Liang, “Unet++: Redesigning skip connections to exploit multiscale features in image segmentation,”IEEE transactions on medical imaging, vol. 39, no. 6, pp. 1856–1867, 2019

    2019

  24. [24]

    Attention U-Net: Learning Where to Look for the Pancreas

    O. Oktay, J. Schlemper, L. L. Folgoc, M. Lee, M. Heinrich, K. Misawa, K. Mori, S. McDonagh, N. Y . Hammerla, B. Kainz,et al., “Atten- tion u-net: Learning where to look for the pancreas,”arXiv preprint arXiv:1804.03999, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  25. [25]

    Ftsegnet: A novel transformer-based fundus tumor segmentation model guided by pre-trained classification results,

    Z. Deng, Z. Gong, W. Gao, J. Yang, L. Shao, F. Li, W. Wei, and L. Ma, “Ftsegnet: A novel transformer-based fundus tumor segmentation model guided by pre-trained classification results,” in2024 IEEE International Symposium on Biomedical Imaging (ISBI), pp. 1–5, IEEE, 2024

    2024

  26. [26]

    A fundus image dataset for ai-based artery-vein vessel segmentation,

    Z. Deng, W. Gao, Z. Gong, R. Gan, L. Chen, S. Zhang, and L. Ma, “A fundus image dataset for ai-based artery-vein vessel segmentation,” Scientific Data, vol. 12, no. 1, p. 1298, 2025

    2025

  27. [27]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly,et al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  28. [28]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE conference on computer vision and pattern recognition, pp. 248–255, Ieee, 2009

    2009

  29. [29]

    Decoupled Weight Decay Regularization

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” arXiv preprint arXiv:1711.05101, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  30. [30]

    Automated analysis of retinal images for detection of referable diabetic retinopathy,

    M. D. Abr `amoff, J. C. Folk, D. P. Han, J. D. Walker, D. F. Williams, S. R. Russell, P. Massin, B. Cochener, P. Gain, L. Tang,et al., “Automated analysis of retinal images for detection of referable diabetic retinopathy,” JAMA ophthalmology, vol. 131, no. 3, pp. 351–357, 2013

    2013

  31. [31]

    Indian diabetic retinopathy image dataset (idrid): a database for diabetic retinopathy screening research,

    P. Porwal, S. Pachade, R. Kamble, M. Kokare, G. Deshmukh, V . Sa- hasrabuddhe, and F. Meriaudeau, “Indian diabetic retinopathy image dataset (idrid): a database for diabetic retinopathy screening research,” Data, vol. 3, no. 3, p. 25, 2018

    2018

  32. [32]

    Adam: Automatic detection challenge on age- related macular degeneration,

    H. Fu, F. Li, J. I. Orlando, H. Bogunovi ´c, X. Sun, J. Liao, Y . Xu, S. Zhang, and X. Zhang, “Adam: Automatic detection challenge on age- related macular degeneration,” 2020

    2020

  33. [33]

    Dataset: Origa dataset,

    J. L. F. Yin, “Dataset: Origa dataset,” 2024

    2024

  34. [34]

    Refuge challenge: A unified framework for evaluating automated methods for glaucoma assessment from fundus photographs,

    J. I. Orlando, H. Fu, J. B. Breda, K. Van Keer, D. R. Bathula, A. Diaz- Pinto, R. Fang, P.-A. Heng, J. Kim, J. Lee,et al., “Refuge challenge: A unified framework for evaluating automated methods for glaucoma assessment from fundus photographs,”Medical image analysis, vol. 59, p. 101570, 2020

    2020

  35. [35]

    Palm: Pathologic myopia challenge,

    H. Fu, F. Li, J. I. Orlando, H. Bogunovi ´c, X. Sun, J. Liao, Y . Xu, S. Zhang, and X. Zhang, “Palm: Pathologic myopia challenge,” 2019

    2019

  36. [36]

    International competition on ocular disease intelligent recognition.,

    “International competition on ocular disease intelligent recognition.,” 2019

    2019

  37. [37]

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

    J. Staal, M. Abramoff, M. Niemeijer, M. Viergever, and B. van Gin- neken, “Ridge-based vessel segmentation in color images of the retina,” IEEE Transactions on Medical Imaging, vol. 23, no. 4, pp. 501–509, 2004

    2004

  38. [38]

    Teleophta: Machine learning and image processing methods for teleophthalmology,

    E. Decenciere, G. Cazuguel, X. Zhang, G. Thibault, J.-C. Klein, F. Meyer, B. Marcotegui, G. Quellec, M. Lamard, R. Danno,et al., “Teleophta: Machine learning and image processing methods for teleophthalmology,”Irbm, vol. 34, no. 2, pp. 196–203, 2013

    2013

  39. [39]

    A convnet for the 2020s,

    Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 11976–11986, 2022

    2022

  40. [40]

    Visualizing data using t-SNE,

    L. van der Maaten and G. Hinton, “Visualizing data using t-SNE,” Journal of Machine Learning Research, vol. 9, no. 86, pp. 2579–2605, 2008

    2008