Pith. sign in

REVIEW 4 major objections 4 minor 42 references

Benchmarking Ophthalmology Foundation Models for Clinically Significant Age Macular Degeneration Detection

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Across seven fundus-image datasets, a natural-image self-supervised model achieves the best out-of-distribution AUROC for clinically significant AMD detection, beating two retina-pretrained foundation models and a no-pretraining baseline.

desk verdict Useful benchmark and a genuinely new dataset, but the headline claim that in-domain pretraining is unnecessary is confounded and contradicted by the paper's own table. read the letter →

arxiv 2505.05291 v2 pith:J53CQZGJ submitted 2025-05-08 eess.IV cs.AIcs.CVq-bio.TO

Benjamin A. Cohen , Jonathan Fhima , Meishar Meisel , Baskin Meital , Luis Filipe Nakayama , Eran Berkowitz , Joachim A. Behar This is my paper · ORCID
classification eess.IVcs.AIcs.CVq-bio.TO
keywords age-relatedmaculardegenerationfoundationmodelsself-supervisedlearningvisiontransformersdigitalfundusimagesout-of-distributiongeneralizationAMDNetBRAMD
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether self-supervised pretraining on retinal images is necessary for detecting clinically significant (intermediate-to-late) age-related macular degeneration from digital fundus photographs. The authors benchmark six pretrained vision transformers on 70,000 expert-annotated images from seven datasets, fine-tuning all models on the training split of the large AREDS dataset and testing on six external domains. They find that iBOT, a vision transformer pretrained on natural images with self-supervised learning, achieves the highest out-of-distribution AUROCs (area under the ROC curve, a standard discrimination score) of 0.80-0.97, better than retina-pretrained models (0.78-0.96) and a no-pretraining baseline (0.68-0.91). If true, the result means that general natural-image pretraining is sufficient for this task and that building large in-domain ophthalmic foundation models may not be necessary. The paper also releases BRAMD, a new open-access dataset of 587 fundus images from Brazil, and AMDNet, its best model obtained by fine-tuning iBOT with multi-source domain training.

What carries the argument

The argument is carried by a comparison protocol intended to isolate pretraining: six self-supervised pretrained vision transformers are fine-tuned with the same augmentations, learning-rate schedule, loss, and 10 epochs on the same AREDS-train split, so the pretraining recipe is the main variable. The best backbone, iBOT, is then taken through a multi-source leave-one-domain-out training stage that pools all but the target dataset and balances classes with inverse-frequency weights; the resulting model is AMDNet. AUROC (area under the receiver-operating-characteristic curve) is the ordering metric, and Grad-CAM maps are used to check that the final model focuses on the macula.

What would settle it

Run the same seven-dataset benchmark with a retina-pretrained vision transformer that matches iBOT's architecture and pretraining scale (about 307M parameters, roughly one million fundus images); if that model matches or exceeds iBOT's out-of-distribution AUROCs, the claim that in-domain pretraining is unnecessary is refuted. The complementary check is to shrink iBOT to the size of the smaller in-domain model and see whether its advantage disappears.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central finding is that pretraining domain does not decide performance: iBOT, a self-supervised vision transformer pretrained only on natural images, delivers the best average AUROC (0.890) across the seven datasets and the best out-of-distribution AUROC on three of six external domains, while general natural-image models beat both retina-pretrained models in five of six external domains. Applying the same multi-source leave-one-domain-out training to iBOT yields AMDNet, which reaches out-of-distribution AUROCs of 0.842-0.977 on six unseen datasets and outperforms DeepSeeNet by 10.2% on average. The authors read these results as evidence that in-domain pretraining is not necessary for moderate-to-late AMD identification, and that the value of foundation models for this task comes from large-scale self-supervised pretraining on natural images.

Load-bearing premise

The load-bearing premise is that the six models differ only in pretraining domain, but they also differ in model size and pretraining data volume, so the natural-image advantage could be a scale effect rather than a domain effect.

Editorial extensions

If this is right

  • Fine-tuning only on one large dataset can transfer: a model fine-tuned solely on the AREDS training split identifies intermediate-to-late AMD across US, Indian, Israeli, Chinese, and Brazilian fundus images with out-of-distribution AUROCs of 0.80-0.97.
  • Retina-specific pretraining on up to about one million fundus images does not beat natural-image SSL pretraining for this task, since general models outperformed the in-domain models in five of six external domains.
  • The multi-source AMDNet raises the bar for an open AMD detection baseline, with out-of-distribution AUROCs of 0.842-0.977 and an average 10.2-point improvement over DeepSeeNet.
  • BRAMD adds a Brazilian two-device benchmark of 587 images, giving the community a new external test set for AMD models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the result holds for other ophthalmic tasks, the same leave-one-dataset-out protocol could be run on diabetic retinopathy, glaucoma, or OCT data; a natural-image backbone winning there would turn this single-task finding into a general principle about medical imaging pretraining.
  • Because the natural-image winners are 307M-parameter ViT-L models while one in-domain competitor is an 86M-parameter ViT-B, a matched-scale comparison is the natural next experiment before concluding that in-domain pretraining is unnecessary.
  • The false-positive analysis implies that a deployed AMDNet would need comorbidity-aware triage: the conditions most often mistaken for AMD (macular scar, exudation, chorioretinitis, RPE changes, myopia, retinitis pigmentosa) mimic macular disease rather than random noise.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. This paper benchmarks six self-supervised vision transformer foundation models (MAE, Mugs, iBOT, DINOv2, RETFound, and VisionFM) plus a randomly initialized ViT-L baseline, all fine-tuned on the AREDS training split, for binary classification of intermediate-to-late age-related macular degeneration in seven fundus image datasets totaling roughly 70,000 images. The authors report AUROC on AREDS-test and six external out-of-distribution datasets, introduce a new open-access Brazilian dataset BRAMD (n=587), and present AMDNet, which uses an iBOT backbone with a leave-one-domain-out multi-source fine-tuning strategy and is compared against DeepSeeNet. The headline claims are that iBOT, pretrained on natural images, achieves the highest out-of-distribution generalization and that in-domain ophthalmic pretraining is not necessary for this task.

Significance. If the central comparison were unconfounded, the paper would provide a valuable, reproducible benchmark for AMD detection from fundus images. The BRAMD dataset, the use of seven diverse datasets, bootstrap standard deviations, and the public release of AMDNet are concrete contributions. The paper also repeats an important and contested observation in the community: large natural-image SSL models can match or exceed retinal-specific foundation models on a downstream ophthalmic task. However, the strength of the claim is currently undermined by uncontrolled confounds between pretraining domain, model architecture, and pretraining data volume, by an internal inconsistency in the reported per-dataset win counts, and by an undefined aggregation measure in the headline comparison. The scientific contribution is therefore conditional on the authors substantially reframing or re-analyzing the evidence.

major comments (4)
  1. [Abstract; §3, Table 3; §4, Table 4] The central claim that in-domain pretraining is unnecessary is confounded with pretraining-data volume and architecture. On the same ViT-L/16 architecture, MAE pretrained only on ImageNet-1k has a mean OOD AUROC of about 0.800 over the six external datasets, while RETFound, which is initialized from MAE and further pretrained on about 900k fundus images, reaches 0.865, an improvement of about 0.06 that demonstrates the incremental value of in-domain pretraining at fixed architecture. The advantage of iBOT and DINOv2 over RETFound likely reflects pretraining on ImageNet-1k plus ImageNet-22k (about 14M images) rather than the irrelevance of domain, and VisionFM additionally uses a much smaller ViT-B backbone (86M parameters). Without matching architecture and pretraining corpus size, the abstract's statement that the results 'challenge the assumption that in-domain pretraining is necessary' is not supported by the data as presented.
  2. [§4, Results; Table 4] The Results text states that general foundation models performed better in five of six target domains and that only on BRAMD did VisionFM outperform the general foundation models; Table 4 contradicts this. On RFMiD1, VisionFM (0.872) is the top fine-tuned model, above the best general model DINOv2 (0.864). On BRAMD, RETFound (0.960) is the top model, above iBOT (0.952), while VisionFM (0.943) is below iBOT. The correct count of OOD datasets where a general foundation model attains the highest AUROC is four (HYAMD, ADAM, FIVES, STARE), not five, and the 'only in the case of BRAMD' sentence is inaccurate. This internal inconsistency directly weakens the empirical basis for the general-versus-in-domain conclusion.
  3. [§3, Performance Measures; §4, Results, Figure 1a] The aggregation measure underlying the headline comparison is undefined. The Performance Measures paragraph states that Figure 1a reports the median AUROC across the seven domains, while the Figure 1a caption and the Results text describe the same quantity as the average AUROC. The distinction matters: for the two leading models, the mean across the seven datasets is approximately 0.889 for iBOT and 0.888 for DINOv2, whereas the median is 0.900 for both, tying the two models. The authors must specify whether mean or median is used, and they should provide a paired significance test or otherwise quantify the uncertainty of the difference, because the raw 0.002 mean-AUROC margin is within the bootstrap standard deviations reported in Table 4.
  4. [§3, Multi-source domain training; §4, AMDNet] The AMDNet out-of-distribution evaluation is selection-optimistic. The iBOT backbone was chosen after evaluating average AUROC over all seven datasets, including each external target domain that subsequently serves as the left-out test domain in the leave-one-domain-out protocol. This uses held-out information during model selection and likely inflates the reported OOD AUROC of AMDNet. A nested model-selection procedure, or an explicit statement that the reported AMDNet numbers are optimistic upper bounds affected by selection on the test domains, is required before AMDNet's superiority over DeepSeeNet can be assessed fairly.
minor comments (4)
  1. [§4, Results] The sentence 'foundation models outperformed the baseline ViT-L (p<0.05)' does not state which statistical test was used, whether the comparison was paired, or whether any multiple-comparison correction was applied; please provide this information.
  2. [Table 4 caption] The caption says 'The highest out-of-distribution (OOD) results are marked with †,' but the dagger symbols are attached to all AMDNet rows, not to the highest OOD entries among the six foundation models, and the AREDS-test column is in-domain rather than OOD; please revise the caption so the marker convention is unambiguous.
  3. [§2, BRAMD] The BRAMD control group is described as diabetic retinopathy patients, so 'non-AMD' in this dataset does not mean healthy controls; please state explicitly in Table 1 or the dataset description that controls may have other retinal pathology, as this affects interpretation of the non-AMD class.
  4. [Throughout] The model name is written inconsistently as 'Dinov2' in several places (e.g., Table 3 and the Results text) and 'DINOv2' elsewhere; please standardize the capitalization.

Circularity Check

0 steps flagged · score 2.0 of 10

Benchmark is an independent external measurement; self-citations are not load-bearing; AMDNet evaluation has test-set selection bias but no derivation circularity.

full rationale

The paper's central claim is an empirical benchmark rather than a derivation: six SSL-pretrained ViTs are fine-tuned identically on AREDS-train and evaluated on six external datasets (Table 4). The reported AUROC advantage of iBOT over RETFound and VisionFM is a direct external measurement, not an equation that reduces to its inputs, and no parameter is fitted to the external test sets to produce this comparison. The multi-source-domain AMDNet does use the same external datasets to select iBOT as the backbone before leave-one-domain-out evaluation, which introduces optimistic bias in AMDNet's OOD numbers and makes the phrase 'unseen datasets' in Section 3 overly strong. However, this is a statistical protocol issue (selection on test data), not derivation circularity: the AMDNet scores are measured, not algebraically forced by construction, and the iBOT-versus-domain-specific comparison itself remains independent. The self-citations [5,6] motivate the MSD methodology and prior related observations, but the current paper re-implements and evaluates the method, so those citations are not load-bearing. The comparison also confounds pretraining domain with pretraining corpus scale (RETFound is MAE-IN1k plus 900k DFIs, while iBOT is IN-1k plus IN-22k), which threatens the causal conclusion that in-domain pretraining is unnecessary, but this is a confound, not circularity. The Results text states general models performed better in five of six target domains, whereas Table 4 shows domain-specific models win on RFMiD1 and BRAMD, so the correct count is four of six; this internal inconsistency is a factual issue separate from circularity. Overall, no circular step of the specified kinds is exhibited; score 2 reflects only minor non-load-bearing self-citations.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

This is an empirical machine learning benchmark, so there are no invented theoretical entities. The central claim depends on training hyperparameters and label assumptions, which are listed here as free parameters and axioms.

free parameters (4)
  • Fine-tuning learning rate = 1e-5 to 3e-4 (not reported per model)
    A learning rate range is given but the final value for each model is not disclosed; the comparison could depend on this choice.
  • Number of fine-tuning epochs = 10 for foundation models, 60 for baseline ViT-L
    The no-pretraining baseline receives six times more epochs, which may affect the fairness of the comparison.
  • Input resolution = 518x518 pixels
    All images are resized to 518x518, a hand-chosen resolution that may interact differently with models pretrained at 224 or 256 resolution.
  • Multi-source validation balance = 100 to 772 images per included dataset
    Validation set sizes are hand-set to balance domains during AMDNet model selection; this affects which checkpoint is chosen.
assumptions (3)
  • domain assumption All seven datasets reliably discriminate intermediate-to-late AMD from non-AMD under the authors' label mapping.
    Section 2: STARE labels rely on image features rather than comprehensive ophthalmic exams; ADAM controls contain early AMD and other conditions; BRAMD controls are diabetic retinopathy patients. If negatives contain other maculopathies, AUROC comparisons across datasets are affected.
  • domain assumption The fine-tuning protocol on AREDS-train is a fair common protocol for all models.
    Section 3: all models are fine-tuned on AREDS-train for 10 epochs except the baseline for 60 epochs, with a learning rate range of 1e-5 to 3e-4. Exact per-model learning rates are not reported, so protocol differences could favor some models.
  • standard math Bootstrap resampling with 80% of each target set gives valid AUROC estimates.
    Section 3, Performance Measures: 1,000 bootstrap samples with 80% of the data is a standard procedure, but it does not account for patient-level clustering and may be unstable on the smallest sets such as STARE.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Benchmarking Ophthalmology Foundation Models for Clinically Significant Age Macular Degeneration Detection." pith.science (2026). https://pith.science/paper/J53CQZGJ

@misc{pith2026250505291,
  author       = {Pith},
  title        = {Pith review of: Benchmarking Ophthalmology Foundation Models for Clinically Significant Age Macular Degeneration Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J53CQZGJ}},
  note         = {Machine review of arXiv:2505.05291}
}
read the original abstract

Self-supervised learning (SSL) has enabled Vision Transformers (ViTs) to learn robust representations from large-scale natural image datasets, enhancing their generalization across domains. In retinal imaging, foundation models pretrained on either natural or ophthalmic data have shown promise, but the benefits of in-domain pretraining remain uncertain. To investigate this, we benchmark six SSL-pretrained ViTs on seven digital fundus image (DFI) datasets totaling 70,000 expert-annotated images for the task of moderate-to-late age-related macular degeneration (AMD) identification. Our results show that iBOT pretrained on natural images achieves the highest out-of-distribution generalization, with AUROCs of 0.80-0.97, outperforming domain-specific models, which achieved AUROCs of 0.78-0.96 and a baseline ViT-L with no pretraining, which achieved AUROCs of 0.68-0.91. These findings highlight the value of foundation models in improving AMD identification and challenge the assumption that in-domain pretraining is necessary. Furthermore, we release BRAMD, an open-access dataset (n=587) of DFIs with AMD labels from Brazil.

Figures

Figures reproduced from arXiv: 2505.05291 by the authors.

Figure 1
Figure 1. Models performance for AMD identification. (a) Results for a ViT-L model (no pretraining) as well as a set of six foundation models fine-tuned on AREDS-train. The performance is presented as the average AUROC over the AREDS-test and the six external datasets (target domains). (b) The best backbone identified, namely iBOT, is fine-tuned using an all-against-one multi-source domain training approach. For the resulting… view at source ↗
Figure 2
Figure 2. Error analysis. a) AMDNet probability output for the different subgroups for AREDS; b) AUROC per age group for BRAMD. C: number of control images, A: number of AMD images in the age bin; c) False positive rate per comorbidity for RFMiD1. LD: large drusen; GA: geographic atrophy; NVAMD: Neovascular AMD. 0.972 and 0.979 respectively for the 18–70 and 70- 80 groups (Figure 2b). We found the most prevalent comorbidities… view at source ↗
Figure 3
Figure 3. Examples of attention maps of AMDNet on images from BRAMD. We used GradCAM [30] to plot in red the regions used by the model to generate its predictions. p is the probability for AMD given by AMDNet. population sample for glaucoma and coronary heart disease diagnosis, as well as 3-year risk prediction. They found no statistically significant improvement. Taken together, our findings challenge the value of recent con… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 24 canonical work pages

  1. [1]

    Poplin et al., Prediction of cardiovascular risk fac- tors from retinal fundus photographs via deep learning, Nat Biomed Eng 2 (3) (2018) 158–164

    R. Poplin et al., Prediction of cardiovascular risk fac- tors from retinal fundus photographs via deep learning, Nat Biomed Eng 2 (3) (2018) 158–164. doi:10.1038/ s41551-018-0195-0. URL https://doi.org/10.1038/s41551-018-0195-0

    doi:10.1038/s41551-018-0195-0 2018

  2. [2]

    D. S. W. Ting et al., Artificial intelligence and deep learning in ophthalmology, Br J Ophthalmol 103 (2) (2019) 167–175. doi:10.1136/bjophthalmol-2018-313173

    doi:10.1136/bjophthalmol-2018-313173 2019

  3. [3]

    C. Y. Cheung et al., A deep learning model for detection of alzheimer’s disease based on retinal photographs: a retro- spective,multicentrecase-controlstudy,LancetDigitHealth 4 (11) (2022) e806–e815. doi:10.1016/S2589-7500(22) 00169-8

    work page doi:10.1016/s2589-7500(22 2022

  4. [4]

    Srivastava et al., Artificial intelligence and machine learning in ophthalmology: A review, Indian Journal of Ophthalmology 71 (1) (2023) 11–17.doi:10.4103/ijo.IJO_ 1569_22

    O. Srivastava et al., Artificial intelligence and machine learning in ophthalmology: A review, Indian Journal of Ophthalmology 71 (1) (2023) 11–17.doi:10.4103/ijo.IJO_ 1569_22

    work page doi:10.4103/ijo.ijo_ 2023

  5. [5]

    Men et al., Deep learning generalization for diabetic retinopathy staging from fundus images, Physiological Mea- surement 13 (1) (2025).doi:10.1088/1361-6579/ada86a

    Y. Men et al., Deep learning generalization for diabetic retinopathy staging from fundus images, Physiological Mea- surement 13 (1) (2025).doi:10.1088/1361-6579/ada86a

    work page doi:10.1088/1361-6579/ada86a 2025

  6. [6]

    GONet: A Generalizable Deep Learning Model for Glaucoma Detection

    O. Abramovich et al., Gonet: A generalizable deep learning model for glaucoma detection (2025).arXiv:2502.19514. URL https://arxiv.org/abs/2502.19514

    work page Pith review arXiv 2025

  7. [7]

    Krishnan et al., Self-supervised learning in medicine and healthcare, Nat Biomed Eng 6 (12) (2022) 1346–1352

    R. Krishnan et al., Self-supervised learning in medicine and healthcare, Nat Biomed Eng 6 (12) (2022) 1346–1352. doi:10.1038/s41551-022-00914-1. URL https://doi.org/10.1038/s41551-022-00914-1

    doi:10.1038/s41551-022-00914-1 2022

  8. [8]

    Zhou et al., A foundation model for generalizable disease detection from retinal images, Nature 622 (7981) (2023) 156–163

    Y. Zhou et al., A foundation model for generalizable disease detection from retinal images, Nature 622 (7981) (2023) 156–163. doi:10.1038/s41586-023-06555-x

    doi:10.1038/s41586-023-06555-x 2023

Show all 42 references
  1. [9]

    Qiu et al., Development and validation of a multimodal multitask vision foundation model for generalist ophthalmic artificial intelligence, NEJM AI 1 (12) (2024) AIoa2300221

    J. Qiu et al., Development and validation of a multimodal multitask vision foundation model for generalist ophthalmic artificial intelligence, NEJM AI 1 (12) (2024) AIoa2300221. doi:10.1056/AIoa2300221

    2024 doi

  2. [10]

    Shi et al., Eyefound: A multimodal generalist foundation model for ophthalmic imaging (2024).arXiv:2405.11338

    D. Shi et al., Eyefound: A multimodal generalist foundation model for ophthalmic imaging (2024).arXiv:2405.11338. URL https://arxiv.org/abs/2405.11338

    2024 arXiv

  3. [11]

    W. L. Wong et al., Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: a systematic review and meta-analysis, : Preprint submitted to Elsevier Page 8 of 9 Lancet Glob Health 2 (2) (2014) e106–e116.doi:10.1016/ S2214-109X(13)70145-1

    2014

  4. [12]

    D. B. Rein et al., Prevalence of age-related macular degeneration in the us in 2019, JAMA Ophthalmol 140 (12) (2022) 1202–1208. doi:10.1001/jamaophthalmol.2022. 4401. URL https://doi.org/10.1001/jamaophthalmol.2022. 4401

    2022 doi

  5. [13]

    He et al., Masked autoencoders are scalable vision learners, in: 2022 IEEE/CVF Conf

    K. He et al., Masked autoencoders are scalable vision learners, in: 2022 IEEE/CVF Conf. on Comput. Vis. and Pattern Recognit. (CVPR), 2022, pp. 15979–15988. doi: 10.1109/CVPR52688.2022.01553

    2022

  6. [14]

    Zhou et al., Mugs: A multi-granular self-supervised learning framework (2022).arXiv:2203.14415

    P. Zhou et al., Mugs: A multi-granular self-supervised learning framework (2022).arXiv:2203.14415. URL https://arxiv.org/abs/2203.14415

    2022 arXiv

  7. [15]

    Zhou et al., ibot: Image bert pre-training with online tokenizer (2022)

    J. Zhou et al., ibot: Image bert pre-training with online tokenizer (2022). arXiv:2111.07832. URL https://arxiv.org/abs/2111.07832

    2022 arXiv

  8. [16]

    Oquab et al., Dinov2: Learning robust visual features without supervision (2024)

    M. Oquab et al., Dinov2: Learning robust visual features without supervision (2024). arXiv:2304.07193. URL https://arxiv.org/abs/2304.07193

    2024 arXiv

  9. [17]

    F. L. Ferris et al., A simplified severity scale for age-related macular degeneration: Areds report no. 18, Archives of Ophthalmology 123 (11) (2005) 1570–1574. doi:10.1001/ archopht.123.11.1570

    2005

  10. [18]

    A. L. Goldberger et al., Physiobank, physiotoolkit, and physionet, Circulation 101 (23) (2000) e215–e220. doi: 10.1161/01.CIR.101.23.e215

    2000 doi

  11. [19]

    [On- line]

    Age-Related Eye Disease Study (AREDS) Research Group, National Eye Institute (NEI) Age-Related Eye Disease Study (AREDS) dataset, Database of Genotypes and Phe- notypes (dbGaP), study Accession: phs000001.v3.p1. [On- line]. Available:https://www.ncbi.nlm.nih.gov/projects/ gap/...

    2014

  12. [20]

    Lyu et al., Fractal dimension of retinal vasculature as an image quality metric for automated fundus image analysis systems, Scientific Reports 12 (1) (2022) 11868

    X. Lyu et al., Fractal dimension of retinal vasculature as an image quality metric for automated fundus image analysis systems, Scientific Reports 12 (1) (2022) 11868. doi:10.1038/s41598-022-16089-3. URL https://doi.org/10.1038/s41598-022-16089-3

    2022 doi

  13. [21]

    Pachadeet al., 2020, Retinal fundus multi-disease image dataset (rfmid), IEEE Dataport.doi:10.21227/s3g7-st65

    S. Pachadeet al., 2020, Retinal fundus multi-disease image dataset (rfmid), IEEE Dataport.doi:10.21227/s3g7-st65. URL https://dx.doi.org/10.21227/s3g7-st65

    2020 doi

  14. [22]

    Meisel et al., Hyamd high-resolution fundus image dataset for age related macular degeneration (amd) diag- nosis (2025)

    M. Meisel et al., Hyamd high-resolution fundus image dataset for age related macular degeneration (amd) diag- nosis (2025). arXiv:2505.04230. URL https://arxiv.org/abs/2505.04230

    2025 arXiv

  15. [23]

    Fu et al., 2020, Adam: Automatic detection challenge on age-related macular degeneration, IEEE Dataport.doi: 10.21227/dt4f-rt59

    H. Fu et al., 2020, Adam: Automatic detection challenge on age-related macular degeneration, IEEE Dataport.doi: 10.21227/dt4f-rt59. URL https://dx.doi.org/10.21227/dt4f-rt59

    2020 doi

  16. [24]

    Jin et al., Fives: A fundus image dataset for artificial intelligence based vessel segmentation, Scientific Data 9 (1) (2022) 475

    K. Jin et al., Fives: A fundus image dataset for artificial intelligence based vessel segmentation, Scientific Data 9 (1) (2022) 475. doi:10.1038/s41597-022-01564-3. URL https://doi.org/10.1038/s41597-022-01564-3

    2022 doi

  17. [25]

    Goldbaum and A

    M. Goldbaum and A. Hoover, 2004, The STARE (STruc- tured Analysis of the Retina) Project, [Online]. Available: https://cecas.clemson.edu/~ahoover/stare/

    2004

  18. [26]

    Parsania, P

    P. Parsania, P. Virparia, A review: Image interpolation tech- niques for image scaling, International Journal of Innovative Research in Computer and Communication Engineering 02 (2015) 7409–7414. doi:10.15680/IJIRCCE.2014.0212024

    2015

  19. [27]

    Dosovitskiy et al., An image is worth 16x16 words: Transformers for image recognition at scale (2021).arXiv: 2010.11929

    A. Dosovitskiy et al., An image is worth 16x16 words: Transformers for image recognition at scale (2021).arXiv: 2010.11929. URL https://arxiv.org/abs/2010.11929

    2021 arXiv

  20. [28]

    Y. Peng et al., Deepseenet: A deep learning model for automated classification of patient-based age-related mac- ular degeneration severity from color fundus photographs, Ophthalmology 126 (4) (2019) 565–575. doi:10.1016/j. ophtha.2018.11.015

    2019 doi

  21. [29]

    NCBI NLP Group, Deepseenet: Automated grading of age-related macular degeneration (amd) using deep learn- ing, https://github.com/ncbi-nlp/DeepSeeNet, accessed: 2025-02-13 (2024)

    2024

  22. [30]

    R. R. Selvaraju et al., Grad-cam: Visual explanations from deep networks via gradient-based localization, International Journal of Computer Vision 128 (2) (2019) 336–359.doi: 10.1007/s11263-019-01228-7. URL http://dx.doi.org/10.1007/s11263-019-01228-7

    2019 doi

  23. [31]

    Jonasson et al., Prevalence of age-related macular degen- eration in old persons: Age, gene/environment susceptibility reykjavik study, Ophthalmology 118 (5) (2011) 825–830

    F. Jonasson et al., Prevalence of age-related macular degen- eration in old persons: Age, gene/environment susceptibility reykjavik study, Ophthalmology 118 (5) (2011) 825–830. doi:10.1016/j.ophtha.2010.08.044

    2011 doi

  24. [32]

    Xiong et al., How generalizable are foundation models when applied to different demographic groups and settings?, NEJM AI 2 (1) (2025) AIcs2400497

    Z. Xiong et al., How generalizable are foundation models when applied to different demographic groups and settings?, NEJM AI 2 (1) (2025) AIcs2400497. doi:10.1056/AIcs2400497. URL https://ai.nejm.org/doi/full/10.1056/ AIcs2400497

    2025 doi

  25. [33]

    Pead et al., Automated detection of age-related macular degeneration in color fundus photography: a systematic review, Survey of Ophthalmology 64 (4) (2019) 498–511

    E. Pead et al., Automated detection of age-related macular degeneration in color fundus photography: a systematic review, Survey of Ophthalmology 64 (4) (2019) 498–511. doi:10.1016/j.survophthal.2019.02.003

    2019 doi

  26. [34]

    F. L. Ferris et al., Clinical classification of age-related macular degeneration, Ophthalmology 120 (4) (2013) 844–

    2013

  27. [35]

    F. Grassmann et al., A deep learning algorithm for pre- diction of age-related eye disease study severity scale for age-related macular degeneration from color fundus pho- tography, Ophthalmology 125 (9) (2018) 1410–1420. doi: 10.1016/j.ophtha.2018.02.037

    2018 doi

  28. [36]

    P. M. Burlina et al., Automated grading of age-related macular degeneration from color fundus images using deep convolutional neural networks, JAMA Ophthalmol 135 (11) (2017) 1170–1176. doi:10.1001/jamaophthalmol.2017. 3782

    2017 doi

  29. [37]

    S. Y. Wang, The rural shortage of ophthalmic subspecialists, JAMA Ophthalmology 143 (2) (2025) 125. doi:10.1001/ jamaophthalmol.2024.5704

    2025

  30. [38]

    M. R. Mookiah et al., Automated diagnosis of age-related macular degeneration using greyscale features from digital fundus images, Computers in Biology and Medicine 53 (2014) 55–64. doi:10.1016/j.compbiomed.2014.07.015

    2014 doi

  31. [39]

    M. A. Zapata et al., Artificial intelligence to identify retinal fundus images, quality validation, laterality evaluation, macular degeneration, and suspected glaucoma, Clinical Ophthalmology 14 (2020) 419–429. doi:10.2147/OPTH. S235751

    2020 doi

  32. [40]

    Xie et al., Automation of macular degeneration classifi- cation in the areds dataset, using a novel neural network design, Clinical Ophthalmology 17 (2023) 455–469

    L. Xie et al., Automation of macular degeneration classifi- cation in the areds dataset, using a novel neural network design, Clinical Ophthalmology 17 (2023) 455–469. doi: 10.2147/OPTH.S396537

    2023 doi

  33. [41]

    Yan et al., Deep-learning-based prediction of late age-related macular degeneration progression, Nature Ma- chine Intelligence 2 (2) (2020) 141–150

    Q. Yan et al., Deep-learning-based prediction of late age-related macular degeneration progression, Nature Ma- chine Intelligence 2 (2) (2020) 141–150. doi:10.1038/ s42256-020-0154-9. : Preprint submitted to Elsevier Page 9 of 9

    2020

  34. [851]

    doi:10.1016/j.ophtha.2012.10.036

    2012 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.