Jamie Burke. IDHea, Topcon Healthcare –– Earlier this year at ARVO 2025, we proudly introduced the Institute for Digital Health (IDHea), a digital health data-as-a-service platform built to accelerate AI-driven research in ophthalmology. IDHea provides secure, scalable access to large volumes of ophthalmic data, equipping researchers with the infrastructure needed to build advanced AI systems that accelerate earlier detection and improved patient care.

Today, we are excited to announce the next step in this journey: the availability of RETFound1, the pioneering foundation model for retinal images, on the IDHea platform. RETFound, first published in Nature, demonstrated superior generalization across multiple disease tasks compared to single-task models. This added capability brings cutting-edge AI directly to researchers, making it easier than ever to apply foundation models for ophthalmic research.

What are Foundation Models and why they matter

Traditional AI models in ophthalmology are typically trained for narrow, single-purpose tasks such as disease detection or image segmentation using labeled data (Figure 1). While effective, these models often require extensive retraining when applied to new diseases, populations, or clinical contexts, limiting their off-the-shelf generalizability.

Foundation models are changing this paradigm. Trained on massive, diverse and potentially unlabeled datasets, they capture broad visual representations that can be fine-tuned for a wide variety of downstream tasks (Figure 2). This enables researchers to:

• Fine-tune models with fewer labels.2
• Better generalize across different diseases and patient populations.3
• Leverage deep vector embeddings as features for downstream tasks such as biomarker discovery and multimodal research.


This adaptability makes foundation models a cornerstone for the next generation of AI-accelerated research in ophthalmology.

What is RETFound?

Developed by Moorfields Eye Hospital and University College London under the leadership of Professor Pearse Keane, RETFound is the first large-scale foundation model for retinal images. Trained on 1.6 million retinal images, it has shown powerful performance in generalizable ophthalmic and systemic disease detection.

A core strength of RETFound lies in its ability to generate learned, compact and information-rich numerical representations from retinal images, known as deep vector embeddings (or simply, features). These embeddings capture structural and pathological features that can be directly applied or fine-tuned to downstream prognostic and diagnostic tasks such as early detection or screening, disease stratification, and clinical decision support systems.

RETFound and IDHea

With this launch, IDHea now provides researchers with direct access to RETFound through its Databricks infrastructure. This includes:

• Ready-to-use Python notebooks demonstrating how RETFound can be fine-tuned on IDHea’s ophthalmic datasets.
• Pre-computed vector embeddings across our large-scale datasets, enabling researchers to immediately begin downstream analyses without heavy computation.
• Scalable, user-friendly training and validation workflows that simplify experimentation and accelerate time to insight.


This integration empowers researchers to move seamlessly from data to discovery on the IDHea platform, without needing to build complex infrastructures from scratch.

Accelerating Ophthalmic AI Research with IDHea

The combination of IDHea’s integrated data analytics platform and RETFound’s AI capabilities creates a powerful environment for advancing ophthalmic research. With secure, scalable data access and cutting-edge AI models, researchers can explore:

• Improving early detection of eye diseases.
• Supporting large-scale screening in primary care.
• Developing and validating new AI-driven diagnostic tools.

Conclusion

The integration of foundation models into IDHea is more than a technical milestone – it is a step towards AI-accelerated healthcare research where insights can be scaled, shared, and applied to real-world patient pathways. By combining robust data infrastructure with cutting-edge AI models, we aim to enable breakthroughs in disease detection, risk prediction, and ultimately, patient care.

As the ecosystem around foundation models grows, IDHea will continue to serve as a bridge between large-scale health data and the AI tools needed to unlock its full potential. RETFound is just the beginning.

Explore RETFound on IDHea today and kickstart your ophthalmic research with foundation models!

1. Zhou, Y., Chia, M. A., Wagner, S. K., Ayhan, M. S., Williamson, D. J., Struyven, R. R., … & Keane, P. A. (2023). A foundation model for generalizable disease detection from retinal images. Nature, 622(7981), 156-163.
2. Tham, Y. C., Goh, J. H. L., Nderitu, P., Zhou, Y., Ran, A. R., Srinivasan, S., … & Keane, P. A. (2025). Building the world’s first truly global medical foundation model. Nature Medicine, 1-6.
3. Chia, M. A., Antaki, F., Zhou, Y., Turner, A. W., Lee, A. Y., & Keane, P. A. (2024). Foundation models in ophthalmology. British Journal of Ophthalmology, 108(10), 1341-1348.

Not for clinical use. For research purposes only.