We are developing a new generation of AI models designed to extract clinically relevant insights from retinal imaging, supporting earlier investigation, more informed clinical decisions, and more accessible assessment of systemic health.Our long-term goal is to help transform a fast, safe, and routinely available image of the eye into a non-invasive clinical decision-support signal.
Researching AI Models for Non-Invasive Clinical Decision Support
A New Window Into Systemic Health
The retina provides a uniquely accessible view of the body’s microvascular system. Subtle changes in retinal structure, vascular patterns, and tissue characteristics may reflect biological processes extending beyond the eye, including microvascular stress, metabolic imbalance, oxidative stress, and other systemic changes.
Our vision is to develop AI models capable of recognizing these complex patterns and translating them into probabilistic, clinically meaningful insights.
Retinal imaging offers several important advantages: it is non-invasive, fast, repeatable, and already used extensively across ophthalmology and screening workflows. This creates the possibility of generating additional clinical value from an existing and widely scalable form of medical imaging.
The intended role of these models is not to replace clinicians or laboratory testing. Their purpose is to support clinical decisions, helping identify when further investigation may be valuable, prioritizing confirmatory testing, and potentially enabling earlier recognition of clinically relevant biological changes.
A direct view of microvascular health
The retina is one of the few places where blood vessels and microvascular structures can be observed directly and non-invasively.
Accessible and repeatable
Retinal imaging can be performed quickly and repeatedly without invasive procedures, making it suitable for screening and longitudinal monitoring.
AI-readable biological complexity
Advanced models can analyse image patterns and relationships that may be too subtle or multidimensional for conventional visual assessment.
Research Built on Exceptional Scientific and Clinical Expertise
Our R&D program brings together leading expertise in artificial intelligence, computer vision, biomedical image analysis, systems modelling, ophthalmology, and clinical research.
The active scientific background behind the program includes professors, clinical directors, senior medical researchers, and AI specialists from the University of Debrecen, Pázmány Péter Catholic University, the University of Pécs Clinical Centre, and HUN-REN SZTAKI.
This multidisciplinary foundation allows us to approach the challenge from every critical direction: biological plausibility, clinical relevance, image quality, model development, statistical validity, and real-world translation.
Curehub Ocular AI
Curehub Ocular AI is our flagship research program exploring whether retinal images contain reproducible patterns associated with systemic microvascular and metabolic states.
Can AI identify biologically meaningful systemic signals within retinal images, and distinguish those signals from age, image quality, device differences, existing ocular conditions, and other confounding factors?
What we are developing
We are building and evaluating AI models that learn high-dimensional representations of retinal images and test whether structured microvascular and metabolic signals can be recovered consistently.
Our work includes
- •Retinal-image preparation, harmonisation, and quality control
- •AI-based representation learning and image analysis
- •Controlled signal-recovery experiments
- •Negative controls and artifact detection
- •Cross-device and cross-dataset validation
- •Explainability and uncertainty analysis
- •Preparation for validation with real clinical and laboratory data
What the research is designed to deliver
The immediate objective is a scientifically defensible technical and validation framework showing whether the underlying approach is sufficiently robust to progress toward clinical-data studies.
A successful research outcome will establish
- •A meaningful signal can be detected
- •Which retinal features and model architectures carry the strongest information
- •Whether results remain stable across datasets, devices, and patient groups
- •Which confounding factors must be controlled
- •What clinical data are required for real-world biological validation
The current research is a feasibility and validation program that creates the foundation for future clinical models.
From Retinal Patterns to Non-Invasive Biomarker Research
The next phase is to connect our retinal AI models with real clinical evidence.
This requires carefully governed datasets in which retinal images can be studied alongside laboratory measurements and relevant clinical information. These paired datasets will allow us to determine whether image-derived signals correlate with real biological states and whether they provide information beyond conventional risk factors.
The initial biological focus includes pathways connected with microvascular health, one-carbon metabolism, redox balance, and oxidative stress.
Our Development Path
- 01
Technical Feasibility
Develop robust retinal-image representations and demonstrate that the research pipeline can recover controlled signals without relying on image artifacts or dataset shortcuts.
- 02
Clinical-Data Validation
Evaluate the models using retinal images paired with blood biomarkers, clinical metadata, and longitudinal or genomic information.
- 03
Independent Validation
Test model performance across new clinical sites, devices, populations, and patient groups that were not involved in model development.
- 04
Clinical Decision Support
Develop probabilistic, uncertainty-aware outputs that help clinicians identify patients who could benefit from confirmatory laboratory testing or further investigation.
Building the Evidence for Non-Invasive Clinical AI
Our work is bringing together retinal imaging, artificial intelligence, clinical science, and biological validation to explore a new frontier in non-invasive decision support.