From diagnostics to drug development, artificial intelligence (AI) today has become a valuable extension of the medical field. A new addition to its long list of uses is a hi-tech screening tool developed by Google and an international team of researchers for detecting diabetic retinopathy a diabetic complication in the eye.
A study conducted at two eye care centres in India — Aravind Eye Hospital, Madurai and Sankara Nethralaya, Chennai — which screened over 3,000 patients with diabetes, has shown that the AI’s performance exceeded the conventionally used manual grading method used to identify diabetic retinopathy . The AI had a specificity and sensitivity of around 90%. The results were published in JAMA Ophthalmology.
A specialised retinal fundus camera was used to take photos of the eye. “Usually when we need to evaluate the retina, we dilate the pupil to allow more light to enter the eye and illuminate the back of the eye. But in this fundus photography it is not necessary as a coherent beam of light can enter the small gap (Pupil) and take an image in just two to three minutes,” explains Dr. Rajiv Raman from Sankara Nethralaya, Chennai and one of the authors of the paper.
He adds that it is very easy to operate and the cost of the camera has also significantly reduced in the recent past. Tamil Nadu and Kerala governments already have over 150 of these cameras currently in use.
Once the images are taken, it is fed into the computer and the AI tool screens it for diabetic retinopathy. A previous paper published by the team in 2016 in JAMA explains how the AI tool was shown over 120,000 images of the retina and taught to identify what each lesion meant. According to the International Clinical Diabetic Retinopathy scale, the AI tool was taught to grade the severity (none, mild, moderate, severe or proliferative) and give an instant report along with the recommendations.
“Diabetic patients are normally asymptomatic when the eye is concerned until the late stages or advanced stage when treatment is difficult or not so effective. So it is important to find the patient at an early stage and help prevent loss of vision,” explains Dr. Kim Ramasamy from Aravind Eye Hospital in Madurai.
“What we have deployed here is an opportunistic screening. We placed the camera at the diabetologist’s clinic and trained their technician to take a picture of the back of the eye (retina) and upload it to the AI tool. In about two minutes, the patient can get their eye test report along with the other regular diabetic test reports. Based on this report, the diabetologist can further refer the patient to an ophthalmologist if needed.” he adds.
Dr. Ramasamy explains how the team has been testing this AI at even small Health Centres in Tamil Nadu. “When about 100 patients are screened, about 20 will have any level of diabetic retinopathy and only four to five might need intervention. But to track down this small number we absolutely need to screen all the diabetics, and we currently don’t have the facilities now. It would be great if we can have these opportunistic screenings at offices, railway stations or other public places,” adds Dr. Ramasamy.
Detecting breast cancer
The corresponding author of the paper Lily Peng adds in an email to The Hindu: “Beyond diabetic retinopathy we are also working on a number of other research projects using AI to tackle healthcare problems. Earlier this year, we showed in a research paper that AI models can help detect breast cancer in mammography images more accurately than doctors. Our research is still in the early stages, but it shows that AI can be a path forward to improve screenings for breast cancer and boost the chances of survival.”
Ms Peng is a Product Manager at Google Health, California