Benchmarking Ophthalmology Foundation Models for Clinically Significant Age Macular Degeneration Detection

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.

HYAMD High-Resolution Fundus Image Dataset for age related macular degeneration (AMD) Diagnosis

The Hillel Yaffe Age Related Macular Degeneration (HYAMD) dataset is a longitudinal collection of 1,560 Digital Fundus Images (DFIs) from 325 patients examined at the Hillel Yaffe Medical Center (Hadera, Israel) between 2021 and 2024. The dataset includes an AMD cohort of 147 patients (aged 54-94) with varying stages of AMD and a control group of 190 diabetic retinopathy (DR) patients (aged 24-92). AMD diagnoses were based on comprehensive clinical ophthalmic evaluations, supported by Optical Coherence Tomography (OCT) and OCT angiography. Non-AMD DFIs were sourced from DR patients without concurrent AMD, diagnosed using macular OCT, fluorescein angiography, and widefield imaging. HYAMD provides gold-standard annotations, ensuring AMD labels were assigned following a full clinical assessment. Images were captured with a DRI OCT Triton (Topcon) camera, offering a 45 deg field of view and 1960 x 1934 pixel resolution. To the best of our knowledge, HYAMD is the first open-access retinal dataset from an Israeli sample, designed to support AMD identification using machine learning models.

Enhancing Retinal Vessel Segmentation Generalization via Layout-Aware Generative Modelling

Generalization in medical segmentation models is challenging due to limited annotated datasets and imaging variability. To address this, we propose Retinal Layout-Aware Diffusion (RLAD), a novel diffusion-based framework for generating controllable layout-aware images. RLAD conditions image generation on multiple key layout components extracted from real images, ensuring high structural fidelity while enabling diversity in other components. Applied to retinal fundus imaging, we augmented the training datasets by synthesizing paired retinal images and vessel segmentations conditioned on extracted blood vessels from real images, while varying other layout components such as lesions and the optic disc. Experiments demonstrated that RLAD-generated data improved generalization in retinal vessel segmentation by up to 8.1%. Furthermore, we present REYIA, a comprehensive dataset comprising 586 manually segmented retinal images. To foster reproducibility and drive innovation, both our code and dataset will be made publicly accessible.

GONet: A Generalizable Deep Learning Model for Glaucoma Detection

Glaucomatous optic neuropathy (GON) is a prevalent ocular disease that can lead to irreversible vision loss if not detected early and treated. The traditional diagnostic approach for GON involves a set of ophthalmic examinations, which are time-consuming and require a visit to an ophthalmologist. Recent deep learning models for automating GON detection from digital fundus images (DFI) have shown promise but often suffer from limited generalizability across different ethnicities, disease groups and examination settings. To address these limitations, we introduce GONet, a robust deep learning model developed using seven independent datasets, including over 119,000 DFIs with gold-standard annotations and from patients of diverse geographic backgrounds. GONet consists of a DINOv2 pre-trained self-supervised vision transformers fine-tuned using a multisource domain strategy. GONet demonstrated high out-of-distribution generalizability, with an AUC of 0.85-0.99 in target domains. GONet performance was similar or superior to state-of-the-art works and was significantly superior to the cup-to-disc ratio, by up to 21.6%. GONet is available at [URL provided on publication]. We also contribute a new dataset consisting of 768 DFI with GON labels as open access.

SHDB-AF: a Japanese Holter ECG database of atrial fibrillation

Atrial fibrillation (AF) is a common atrial arrhythmia that impairs quality of life and causes embolic stroke, heart failure and other complications. Recent advancements in machine learning (ML) and deep learning (DL) have shown potential for enhancing diagnostic accuracy. It is essential for DL models to be robust and generalizable across variations in ethnicity, age, sex, and other factors. Although a number of ECG database have been made available to the research community, none includes a Japanese population sample. Saitama Heart Database Atrial Fibrillation (SHDB-AF) is a novel open-sourced Holter ECG database from Japan, containing data from 100 unique patients with paroxysmal AF. Each record in SHDB-AF is 24 hours long and sampled at 200 Hz, totaling 24 million seconds of ECG data.

SleepPPG-Net2: Deep learning generalization for sleep staging from photoplethysmography

Background: Sleep staging is a fundamental component in the diagnosis of sleep disorders and the management of sleep health. Traditionally, this analysis is conducted in clinical settings and involves a time-consuming scoring procedure. Recent data-driven algorithms for sleep staging, using the photoplethysmogram (PPG) time series, have shown high performance on local test sets but lower performance on external datasets due to data drift. Methods: This study aimed to develop a generalizable deep learning model for the task of four class (wake, light, deep, and rapid eye movement (REM)) sleep staging from raw PPG physiological time-series. Six sleep datasets, totaling 2,574 patients recordings, were used. In order to create a more generalizable representation, we developed and evaluated a deep learning model called SleepPPG-Net2, which employs a multi-source domain training approach.SleepPPG-Net2 was benchmarked against two state-of-the-art models. Results: SleepPPG-Net2 showed consistently higher performance over benchmark approaches, with generalization performance (Cohen's kappa) improving by up to 19%. Performance disparities were observed in relation to age, sex, and sleep apnea severity. Conclusion: SleepPPG-Net2 sets a new standard for staging sleep from raw PPG time-series.

RawECGNet: Deep Learning Generalization for Atrial Fibrillation Detection from the Raw ECG

Introduction: Deep learning models for detecting episodes of atrial fibrillation (AF) using rhythm information in long-term, ambulatory ECG recordings have shown high performance. However, the rhythm-based approach does not take advantage of the morphological information conveyed by the different ECG waveforms, particularly the f-waves. As a result, the performance of such models may be inherently limited. Methods: To address this limitation, we have developed a deep learning model, named RawECGNet, to detect episodes of AF and atrial flutter (AFl) using the raw, single-lead ECG. We compare the generalization performance of RawECGNet on two external data sets that account for distribution shifts in geography, ethnicity, and lead position. RawECGNet is further benchmarked against a state-of-the-art deep learning model, named ArNet2, which utilizes rhythm information as input. Results: Using RawECGNet, the results for the different leads in the external test sets in terms of the F1 score were 0.91--0.94 in RBDB and 0.93 in SHDB, compared to 0.89--0.91 in RBDB and 0.91 in SHDB for ArNet2. The results highlight RawECGNet as a high-performance, generalizable algorithm for detection of AF and AFl episodes, exploiting information on both rhythm and morphology.

DRStageNet: Deep Learning for Diabetic Retinopathy Staging from Fundus Images

Diabetic retinopathy (DR) is a prevalent complication of diabetes associated with a significant risk of vision loss. Timely identification is critical to curb vision impairment. Algorithms for DR staging from digital fundus images (DFIs) have been recently proposed. However, models often fail to generalize due to distribution shifts between the source domain on which the model was trained and the target domain where it is deployed. A common and particularly challenging shift is often encountered when the source- and target-domain supports do not fully overlap. In this research, we introduce DRStageNet, a deep learning model designed to mitigate this challenge. We used seven publicly available datasets, comprising a total of 93,534 DFIs that cover a variety of patient demographics, ethnicities, geographic origins and comorbidities. We fine-tune DINOv2, a pretrained model of self-supervised vision transformer, and implement a multi-source domain fine-tuning strategy to enhance generalization performance. We benchmark and demonstrate the superiority of our method to two state-of-the-art benchmarks, including a recently published foundation model. We adapted the grad-rollout method to our regression task in order to provide high-resolution explainability heatmaps. The error analysis showed that 59\% of the main errors had incorrect reference labels. DRStageNet is accessible at URL [upon acceptance of the manuscript].

Generalization in medical AI: a perspective on developing scalable models

Over the past few years, research has witnessed the advancement of deep learning models trained on large datasets, some even encompassing millions of examples. While these impressive performance on their hidden test sets, they often underperform when assessed on external datasets. Recognizing the critical role of generalization in medical AI development, many prestigious journals now require reporting results both on the local hidden test set as well as on external datasets before considering a study for publication. Effectively, the field of medical AI has transitioned from the traditional usage of a single dataset that is split into train and test to a more comprehensive framework using multiple datasets, some of which are used for model development (source domain) and others for testing (target domains). However, this new experimental setting does not necessarily resolve the challenge of generalization. This is because of the variability encountered in intended use and specificities across hospital cultures making the idea of universally generalizable systems a myth. On the other hand, the systematic, and a fortiori recurrent re-calibration, of models at the individual hospital level, although ideal, may be overoptimistic given the legal, regulatory and technical challenges that are involved. Re-calibration using transfer learning may not even be possible in some instances where reference labels of target domains are not available. In this perspective we establish a hierarchical three-level scale system reflecting the generalization level of a medical AI algorithm. This scale better reflects the diversity of real-world medical scenarios per which target domain data for re-calibration of models may or not be available and if it is, may or not have reference labels systematically available.

End-to-end Risk Prediction of Atrial Fibrillation from the 12-Lead ECG by Deep Neural Networks

Background: Atrial fibrillation (AF) is one of the most common cardiac arrhythmias that affects millions of people each year worldwide and it is closely linked to increased risk of cardiovascular diseases such as stroke and heart failure. Machine learning methods have shown promising results in evaluating the risk of developing atrial fibrillation from the electrocardiogram. We aim to develop and evaluate one such algorithm on a large CODE dataset collected in Brazil. Results: The deep neural network model identified patients without indication of AF in the presented ECG but who will develop AF in the future with an AUC score of 0.845. From our survival model, we obtain that patients in the high-risk group (i.e. with the probability of a future AF case being greater than 0.7) are 50% more likely to develop AF within 40 weeks, while patients belonging to the minimal-risk group (i.e. with the probability of a future AF case being less than or equal to 0.1) have more than 85% chance of remaining AF free up until after seven years. Conclusion: We developed and validated a model for AF risk prediction. If applied in clinical practice, the model possesses the potential of providing valuable and useful information in decision-making and patient management processes.