A retinal camera with AI screening is a camera that photographs the back of the eye and uses an AI model to check the photo for diabetic eye disease on the spot. The back of the eye is called the retina, and a photograph of it is called a fundus image. The disease it checks for is called diabetic retinopathy. It is damage to the small blood vessels of the retina caused by diabetes, and it can lead to blindness if it is not treated in time.
Diabetic retinopathy often causes no symptoms until it is advanced. So people with diabetes need their retinas checked regularly, usually once a year. India has more than 10 crore people with diabetes, and far too few eye specialists to examine them all. Most eye specialists work in cities.
The camera is handheld, and some models attach to a smartphone. A health worker holds it close to the patient's eye and takes two or three photos of each retina. The AI model runs on the camera itself, so it needs no internet connection. Running a model on the device in this way is called edge AI. In under a minute, the model gives one of two results. Either the photos show no disease that needs a specialist, or they show disease that needs referral to one.
Take, for example, a primary health centre near Thrissur, in Kerala. On its diabetes clinic day, a health worker screens 30 patients. One of them is a 58-year-old tea shop owner who has had diabetes for 12 years. The model flags signs of retinopathy in his left eye. The health worker gives him a referral slip for the district eye hospital, with an appointment that week. There, an eye specialist examines him and decides on treatment, which may use ophthalmic surgical equipment. The other 29 patients are asked to come back next year. Each result and its photos are saved to the patient's record, in the hospital EHR or in the health centre's own software.
The model sorts patients by whether they need a specialist, which is a kind of triage. It does not decide on treatment. Because it gives a medical result, it is regulated as software as a medical device (SaMD). It is set so that it rarely misses disease. As a result, it sometimes refers a person whose eyes turn out to be healthy.
The camera is part of medical imaging, but it is different from AI radiology. Radiology AI reads images from large machines in hospitals, and sends its results to a radiologist. A retinal camera makes its own image and gives its result on the spot, in places where there is no eye specialist at all.