Fluorangiography (FAG) is one of the most important diagnostic examinations to check the health of the retina and optic nerve. Using an innovative model based on Artificial Intelligence (AI) algorithms, FAG videos were generated from photos of the ocular fundus and textual descriptions of the retina.
The results of this method were presented in a study by Wu and colleagues, published in June 2025 in JAMA Ophthalmology.
It is an advanced application of 'text-to-video' generative AI techniques, already used in neurosurgery and endoscopy.
Fluorangiography: the clinical test
Fluorangiography was chronologically the first diagnostic examination available in the retinal field. Introduced into clinical practice in the 1960s, it is considered a milestone in the study of ocular fundus disorders. Even today, it is still considered the basic investigation in the management of maculopathies.
For the examination, the fluorescein, a water-soluble, yellow-orange organic dye with a molecular weight of 376.67. Injected intravenously, 70-80% binds to serum proteins while the remainder remains free in the plasma. Once it reaches the retinal circulation, under normal conditions it remains intravascular and circulates with the bloodstream. In fact, neither the serum protein-bound form nor the free form are able to cross the inner blood-retinal barrier, formed by the intercellular tight junctions of endothelial cells.
The dye remains, therefore, confined in normal conditionsin the lumen of the capillaries. Instead, dye diffusion occurs only in the presence of pathological conditions that cause a rupture of the inner blood-retinal barrier. At the choroidal level, the situation is different. The major vessels are impermeable to both forms, whereas the choriocapillary vessels have a very thin wall composed of fenestrations, through which free fluorescein can diffuse into the extravascular spaces, but cannot cross the Bruch's membrane - retinal pigment epithelium complex (outer blood-retinal barrier) under normal conditions.
The fundamental characteristics of Fluorangiography are related to these propertieswhich are the key to understanding and interpreting the images obtained:
- the fluorescein molecule has a low molecular weight and is found in high free amounts in plasma: it therefore diffuses very easily under pathological conditions;
- The emission wavelength of fluorescein can be easily detected by imaging systems, even when the dye is present in modest amounts, as in the case of clinically inconspicuous lesions (high sensitivity);
- the light used to excite Fluorescein is unable to pass through the retinal pigment epithelium, as well as turbid fluids and blood.
FAG's progress
Angiography, which uses fluorescein as a contrast agent, therefore remains a key examination for visualising blood vessels in the retina and, thanks to developments in diagnostic imaging, has simplified patient data collection and clinical image sharing.
Static images, however, retain a subjective nature due to the variability of qualitative interpretations, whereas in the study by Wu and colleagues, a method is proposed to obtain a standardised quantitative output.
From Text to Video
The process of video generation through AI involves a series of steps:
- Data Collection: acquisition of ocular fundus images.
- AI processingAnalysis of visual and textual data.
- Video Generationdynamically representing the acquired data.
Transformation from Subjective to Objective
In traditional procedures, assessments of retinal anomalies are subject to human interpretation, which involves margins of error.
Generative AI is changing this landscape with:
- Automated image analysis.
- Reducing human error
- Generation of objective and detailed reports.
Advantages of Objective Measurement
Objective measurement has many advantages, including:
- Greater precision
- Faster and more accurate diagnoses.
- Reducing discrepanciesfewer errors of interpretation.
These advantages translate into more targeted and effective treatments.
FAG-generated videos
Areas of non-perfusion, leakage and microaneurysms are examples of the rectal anomalies for which a dynamic video can be developed using this AI-driven model.
A series of objective and subjective verifications are used to confirm alignment and correspondence with the description of the lesion or pathology.
Text-to-video generation also has high potential in the creation of medical data for sharing between clinical centres and medical education, while fully respecting patient privacy.
Improving Diagnostic Accuracy
Diagnostic accuracy is crucial for the effective treatment of patients. Generative AI, thanks to deep learning and the volume of data it can process, allows for more accurate representations.
With more accurate and complete data:
- They improve clinical decision-making.
- Misdiagnoses are reduced.
- It increases patient confidence.
Future of Fluorangiography with AI
The evolution of generative AI promises to further redefine fluorangiography and healthcare in general.
Expected Innovations in the Health Sector
New technologies are emerging to improve the quality of care even more.
Future expectations:
- Greater integration with medical devices.
- Development of more advanced algorithms.
- Expansion of global accessibility.
These innovations make the health sector more dynamic and resilient.
Challenges and Opportunities for Medical Professionals
With new technologies come new challenges and opportunities for health professionals.
Opportunities to consider:
- Ongoing training on new technologies.
- Interdisciplinary collaboration.
- Development of new skills.
Conclusions
Advances coming from AI applications in ophthalmology should be seen as an opportunity to provide better services to patients by integrating the potential of new technologies with specialist expertise.
Facing these challenges with flexibility and open-mindedness is the only strategy that can ensure that doctors can make the most of the great potential of these innovations.
- Wu X, Wang L, Chen R, et al. Generation of Fundus Fluorescein Angiography Videos for Health Care Data Sharing. JAMA Ophthalmol. Published online June 26, 2025. doi:10.1001/jamaophthalmol.2025.1419
- Feng X, Xu K, Luo MJ, et al. Latest developments of generative artificial intelligence and applications in ophthalmology. Asia Pac J Ophthalmol (Phila). 2024 Jul-Aug;13(4):100090. doi: 10.1016/j.apjo.2024.100090. Epub 2024 Aug 14. PMID: 39128549.
English 
Italian 
Spanish