The extraordinary progress of Artificial Intelligence (AI)-based imaging technologies has led to a growing interest in its application to the early diagnosis and subsequent management of retinal diseases, in particular the screening and monitoring of geographic atrophy.
Geographical atrophy is an advanced stage of Age-Related Macular Degeneration (AMD), which is a major global health problem and one of the most frequent causes of legal blindness in developed countries.
Diagnosed cases of geographic atrophy are now about 5 million worldwide and could double to 10 million by 2040.
Geographical atrophy
Geographic atrophy is a retinal pathology with a complex management, partly due to the incomplete knowledge of its aetiology and pathogenesis.
This severe form of maculopathy is characterised by the progressive death of retinal pigment epithelium (EPR) cells and macular photoreceptors and the loss of portions of the choriocapillaris. These changes result in atrophic areas that are clearly detectable on retinal imaging and, in cases where the area of the central fovea is also affected, lead to severe impairment of visual function. This macular damage begins in the form of small areolae that later develop into larger areas; a person with early-stage geographic atrophy may experience problems with reading or night vision. If the disease progresses to advanced stages, permanent blind spots (scotomas) will develop in the centre of the visual field.
The loss of vision caused by geographic atrophy severely compromises the independence and quality of life of sufferers, making it difficult for them to manage their daily activities independently.
Treatment
On the therapeutic side, intravitreal pegcetacoplan, an antagonist of complement factor C, which inhibits its cleavage into C3a and C3b and has been shown to slow down disease progression, was recently approved.
This very important therapeutic innovation - in whose phase 3 studies the IRCCS Bietti Foundation in Rome participated - increases the need for early diagnosis and constant monitoring to maximise the benefits of the new treatment and minimise visual loss. In this regard, the use of the most advanced retinal imaging becomes crucial: OCT (Optical Coherence Tomography), Fundus Autofluorescence (FAF) and Colour Fundus Photography (CFP). It is precisely the latter diagnostic technique that constitutes a common, simple and widely used, low-cost method for screening and monitoring geographic atrophy.
AI and retinal imaging
Artificial Intelligence (AI) algorithms have shown enormous potential in retinal diagnostics due to their ability to analyse huge amounts of data, acquired through different imaging techniques, which makes it possible to detect slight retinal changes that sometimes cannot be detected by the human observer.
Through deep learning, AI models 'learn' to recognise profiles and features of geographic atrophy from these extensive image databases and can subsequently apply the acquired knowledge to new images for an automated diagnosis of geographic atrophy. Deep learning models using FAF and OCT have shown good performance in identifying geographic atrophy, but have limited applicability and high costs. Early AI models using CFP were, on the other hand, marred by poor sensitivity and limited explicability.
A study by the ophthalmology team of the University of Udine, Explainable artificial intelligence model for the detection of geographic atrophy using colour retinal photography, proposed a very accurate and user-friendly AI model for the diagnosis of geographic atrophy using CFP images.
540 colour photographs of the fundus were collected and divided into three groups: 300 to train the machine, 120 for validation and 120 to test the performance of the algorithm. The experimental model demonstrated a sensitivity of 100%, a specificity of 97.5% and a diagnostic accuracy of 98.4%.
This model already represents a very good result and the AI algorithms developed in the future will effectively enable automated screening of geographic atrophy, based on the use of colour images of the fundus, which are easy to acquire and transmit.
This optimisation of the management of geographic atrophy, which is possible thanks to artificial intelligence, will allow the patient to take full advantage of the latest technological innovations, which, correctly and effectively used, can truly preserve them from serious visual loss.
See also:
- AI & Oculomics - Oculista Italiano
- Artificial intelligence and OCT in diabetic macular oedema - Oculista Italiano
- Diabetic retinopathy and ultra wide-field imaging - Oculista Italiano
- Sarao V, Veritti D, De Nardin A, et al. Explainable artificial intelligence model for the detection of geographic atrophy using colour retinal photographs. BMJ Open Ophthalmol. 2023 Dec 6;8(1):e001411. doi: 10.1136/bmjophth-2023-001411.
- Holzinger A, Biemann C, Pattichis CS, et al. On the importance of Explainable AI (XAI) for trustworthiness in Biomedicine and Healthcare. Wires Data Mining Knowl Discov 2022;12:e1452.
- Keenan TD, Dharssi S, Peng Y, et al. A deep learning approach for automated detection of geographic atrophy from color fundus photographs. Ophthalmology 2019;126:1533–40.