Remember those colourfully-lit weighing machines on railway platforms that would take a coin and give out a small cardboard ticket with the person’s weight and fortune? The Chief Medical Officer at Madurai’s Aravind Eye Hospital (AEH), Dr Ramasamy Kim, uses this analogy to explain how people could get a preliminary eye check-up done instead of visiting an ophthalmologist in the future.
“In one look, the machine will predict the condition of the retina and advise on the next step of action in seconds,” he says. All thanks to artificial intelligence (AI) in ophthalmology. He has been working on the technology with Google since 2013.
“The impact,” he says, “will be seen in patient care and diabetes management.” He imagines a scenario where the scope of algorithms on smartphones will not require the consumer to even go near a machine. “People can take selfies of the eye on their phone cameras and have instant access to ophthalmic care.”
The two organisations have been working on an automated tool that could detect diabetic retinopathy (DR), the second leading cause of blindness. DR is a condition where lesions develop in the retina of the eye of those living with long-standing diabetes. It puts them at risk of losing vision, if left untreated.
In April this year, USA’s Food and Drug Administration validated AI as a significant DR screening tool. “We now have to publish our research paper and await the certification before we can start implementing it in our routine work,” says Dr Kim. The licensing of the AI algorithm for use is expected by the year end, and is foreseen as a tremendous boost to eye care.
As head of the Retina department, Dr Kim gets over 600 patients daily in his out-patient department, of whom many spend time and resources in travelling to the hospital and waiting for their turn. There are many more left out, who do not come to the hospital simply because of lack of awareness. Eye doctors recommend a mandatory annual examination for all people living with diabetes and every person above 40 years. Dr Kim says in a country like ours, where the patient volume is high, AI will make the diagnosis quicker, and also rev up treatment to avoid preventable loss of vision.
“If the captured image of the eye shows negative for DR, then the person will be advised to rescreen after 12 months. And in case of a positive result, the person would be asked to see an ophthalmologist for further evaluation and immediate treatment,” explains Dr Kim.
When a person can avoid hospital visits up to the stage of detection of DR, it may appear healthcare is lending itself to the risk of machine calculations instead of relying on human knowledge and experience. But Dr Kim argues in favour of using technology effectively and efficiently in times when computers are available everywhere and to everybody but healthcare is not.
The World Health Organization estimates 71 million Indians live with diabetes and at least 20% of them suffer from DR. Of these, 20% are not even aware of their eye condition because they haven’t been in to a doctor for examination. Those who come to an ophthalmologist get their retinal images graded manually in what is today a time-consuming process taking from few hours to few days.
- Google has already débuted an algorithm to 97% accuracy that can identify a person’s age, sex, ethnicity and smoking status and predict the five-year risk of a heart attack, all on the basis of retinal imagery. The AI for DR has been found to be 87% sensitive and 90% specific for detecting more than mild diabetic retinopathy.
AEH, however, has been working on a semi-automated format since 2003, tying up with diabetic clinics across Tamil Nadu. When a patient walks into the clinic, a technician takes pictures of his/her retina with a fundus camera, and along with an online questionnaire answered by the patient, emails it to trained Aravind staff who use a software to grade the image for DR. By the time patients are done with their diabetes check-up, they are also informed about the status of their eye and the necessary follow-up action.
A ravind also has an established network of 71 vision centres across rural Tamil Nadu, that are supervised by trained technicians who take snapshots with retinal cameras of the inside of the eye of every person who walks in, and sends the digital reports to Aravind’s doctors, who then call in a diagnosis and course of treatment. Now, with the Google algorithm in place, the process of collating information and grading retinal images will be standardised and faster.
Dr Kim has spent the last four years working on 10,000 retinal images, drawing every lesion, distinctive spots, bleeding in the retina due to diabetes that could occur in various permutations and combinations, to help Google develop the algorithm that would recognise the signs of the disease early. From June this year, Aravind Hospital started supplementing its DR grading process with the Google AI in 10 of its rural tele-consultation centres. “The results are accurate,” says Dr Kim, who is now working on 60,000 retinal images for matching the grading results from the machine and the human eye in order to fine-tune the algorithm. He says the algorithms pick up problems that trained people sometimes can’t and different ophthalmologists can end up giving different opinions looking at the same image of the eye.
Google has created a database of 1,28,000 images from different sight centres around the globe, including two more in India — Sankara Nethralaya in Chennai and Narayana Nethralaya in Bengaluru. Dr Kim is one of the experts evaluating the data.
- One of the first examples of AI being used in science was a project called Dendral in the 1960s, which helped organic chemists identify unknown organic molecules.
“The day is not far when AI will go solo, because there are two benefits of machine learning. More people will be able to check their eyes at a much lower cost, time, and effort, and doctors will be able to treat more patients who are at risk,” says Dr Kim, allaying fears of doctors left with less work. “The AI, in fact, will throw up huge numbers and accurately spot the vision status and detect the multiple problems or vision-killing conditions. It only means I will get many more patients to treat and reduce the several rounds of redundant tests.”
Though the challenge of a machine may lie in any false negative and deprive a patient of consultation, Dr Kim says AI is superior to anything that he has seen in DR screening. “The machine is able to see something beyond the human eye,” he says, “and as a doctor, my only interest is in getting a greater number of patients at an early detected stage for successful treatment.”
