Background: Diabetes mellitus has a dramatically huge impact on healthcare systems and on patients¿ quality of life. Diabetic eye disease is the leading cause of vision loss in working-age adults and diabetic retinopathy (DR) represents the most common microvascular complication of the disease. Significant advancements have been made in the last decades regarding prevention and early diagnosis. However, about one third of diabetic patients are affected by ocular complications, one third of which are vision-threatening. The constantly increasing prevalence of diabetes and its growing burden on public health demand the creation of a mass screening programme, for the ophthalmologist to be able to fully invest resources and care on patients with DR.
Purpose: the present study aims to build a powerful and effective screening tool for DR, through the development of a portable, smart, non-mydriatic fundus camera with an in-built deep learning detection system. The main task of such a device would be to identify patients with signs of ocular complications and to refer them to an ophthalmology department. The ideal use case scenarios are represented by pharmacies and general practitioners.
Design: observational. Duration 12 months.
Methods: Open access retinal image databases will be used for the training of a deep-learning algorithm aimed at differentiating between healthy and affected retinas. 250 patients with type I and II diabetes examined in the Ophthalmology Clinic of Policlinico Umberto I will undergo a fundus photo through a non-mydriatic fundus camera. The obtained images will be validated by an independent ophthalmologist in order to assess the accuracy of the algorithm.
Results: the initial evaluation by expert ophthalmologists will allow to validate the accuracy of the device and its further use as a mass screening tool for diabetic retinopathy.
Several research groups and companies have explored the potential of a portable non mydriatic fundus camera in the detection of diabetic retinopathy, especially in the last decade. A considerable number of devices have been developed through the years, ranging from table-mounted, expensive retinographs to small clips meant to be attached and used in conjunction with a smartphone. Every device of this kind shares the same basic components: an illumination source, a system of lenses, mirrors and splitters aimed at focusing the light on the retina through the undilated pupil, an image sensor responsible for the acquisition. Fully fledged retinographs are usually expensive, bulky and their use is typically limited to hospitals. Smartphone-based fundus cameras rely on the smartphone as a display console for retinal images, and illumination and imaging optics are integrated to it externally. However, they are tied to one or few models, lack a universal application and tend to become obsolete within a few years. The latest attempt at developing screening devices for diabetic retinopathy is represented by clip-on devices, extremely portable and inexpensive, but still not accurate enough for the aforementioned purpose. The present study aims to develop a portable, wearable nonmydriatic fundus camera, similar to a virtual reality headset, to be used for the mass screening of diabetic retinopathy. The development of the device will be integrated with a deep-learning software able to detect referable DR. Such device should be able to identify early signs of DR, in order to find patients in need for treatment and refer them to an ophthalmology department. Ideally, the device will be enforced to general practitioners, pharmacies and territorial diabetology clinics, all linked to a central ophthalmology unit at the Policlinico Umberto I. The centralization of the ophthalmological screening and the accurate selection of patients will allow for a much more effective resource allocation.