AI-powered Eye2Gene uses multimodal imaging to diagnose inherited retinal disease

Image credit: Heidelberg Engineering

A new study published in Nature Machine Intelligence describes a major advance in precision ophthalmology with the debut of Eye2Gene, an artificial intelligence (AI) model developed to assist in the diagnosis of inherited retinal diseases (IRDs). The system, a result of collaboration between researchers at University College London (UCL) and Heidelberg Engineering, offers multimodal imaging data to predict the gene most likely responsible for a patient’s IRD, potentially restructuring the path to diagnosis and genetic counseling.

Eye2Gene was trained on 58,030 retinal scans from 2,451 genetically confirmed patients with IRDs, and was externally validated on another 775 patients across five global centres. Using an ensemble of 15 convolutional neural networks (CNNs), the system analyses data from spectral-domain OCT (SD-OCT), infrared reflectance (IR) and fundus autofluorescence (FAF) imaging modalities.

Dr Nikolas Pontikos, associate professor at UCL and lead author of the study, said in a statement that Eye2Gene showed promise in tests. “We demonstrate a top-5 prediction accuracy of 83% compared to world-leading experts,” he stated. He noted that Eye2Gene outperformed popular phenotyping-only tools in over 75% of tested cases, exemplifying the benefit of combining imaging and genetic data in complex differential diagnoses.¹

Eye2Gene showed superior performance when interpreting FAF images alone, reaching 76% accuracy, substantially outperforming experienced clinicians, who scored 36% or lower. These results were consistently replicated across five independent clinical centers, including sites in Tokyo, Bonn, São Paulo, Oxford and Liverpool, supporting the model’s robustness and broad applicability across diverse populations and imaging protocols.

The tool is designed for be integrated through Heidelberg Engineering’s HEYEX 2 platform and Heidelberg AppWay. The setup allows real-time gene prediction which allows clinicians to analyse multimodal SPECTRALIS® scans and receive gene-ranking suggestions on-site. This technology has potential to expedite referrals for genetic testing and to support centers that lack specialised IRD expertise.

With coverage of 63 disease-associated genes, encompassing over 90% of European IRD cases, Eye2Gene may help reduce diagnostic delays that currently leave many patients’ IRDs undiagnosed for years. According to Arianna Schoess Vargas, managing director of Heidelberg Engineering, the tool “empowers eye care professionals with new insights into the genetic landscape of IRDs—enhancing diagnosis and ultimately the development of new treatments.”

Eye2Gene is not a replacement for genetic testing, but it can increase the diagnostic yield, improving the likelihood of identifying a causative gene. It may serve as a valuable decision-support tool, by narrowing down potential causative genes, it can help prioritise testing and reduce reliance on genetic panels. The system also shows promise in underserved or lower-resource settings, where expert-level phenotyping may not be readily available.

While Eye2Gene holds considerable promise, it is important to note the broader ethical and logistical challenges around integrating AI into ophthalmology and healthcare overall. Cecilia Lee, MD, MS and her colleagues noted in their article published in Opthamology Science, the concern around data privacy, algorithmic bias, and the need for diverse, representative training data, especially for rare diseases like IRDs. Validation of Eye2Gene’s performance in varied demographics may be necessary, and standardisation of data formats and transparent reporting of model limitations will be essential to ensure responsible, equitable deployment across all healthcare systems.²

Eye2Gene represents a meaningful step closer to bringing AI-assisted diagnosis to clinical routine. Further research is warranted to explore regulatory pathways, include additional genes or syndromic forms of IRDs and integrate the AI with electronic medical record systems for broader adoption.

References:

1.Heidelberg Engineering. Heidelberg Engineering Celebrates Eye2GeneTM AI Breakthrough in Precision Ophthalmology | Heidelberg Engineering Inc.. Corporate US | Corporate Website Heidelberg Engineering Inc. Published July 14, 2025. Accessed July 17, 2025. https://www.heidelbergengineering.com/us/press-releases/heidelberg-engineering-celebrates-eye2gene-ai-breakthrough-in-precision-ophthalmology/

2.Lee CS, Brandt JD, Lee AY. Entering the Exciting Era of Artificial Intelligence and Big Data in Ophthalmology. Ophthalmology Science. 2024;4(2):100469. doi:https://doi.org/10.1016/j.xops.2024.100469