FDA approves AI software to detect diabetic retinopathy
IDx’s De Novo software is the first AI-based fully-autonomous diagnostic system for diabetic retinopathy which has been approved by the FDA for commercialization.
Artificial intelligence is helping to solve problems in healthcare. Zocdoc uses AI to help patients verify whether a particular doctor for a particular procedure was covered by their insurance. The doc.ai app helps doctors interpret lab test results so that doctors have one less job on their workload and can focus on diagnosis and care.
Nuance’s Dragon Medical Virtual Assistant for caregivers and patients helps to automate workflows, lower reduce burnout and provide more patient-friendly care. So far, AI is being seen as a technology that will reduce costs, increase efficiency, speed and accuracy in the healthcare industry.
Now, in a landmark move, the FDA has allowed an Artificial Intelligence-based diagnostic system to be marketed for use by diabetics to detect diabetic retinopathy without the need for a clinician’s inference of test results.
The IDx-DR is a software program that will use AI algorithms to analyse images of the retina, taken with a specialised camera called the Topcon NW400.
Patients can be screened by this device for early detection of retinopathy, which is very important for managed care and prevention of sight impairment. However, many diabetics are not diagnosed early enough. Malvina Eydelman, FDA’s director of the Division of Ophthalmic, Ear, Nose and Throat Devices, says, “50 percent of [diabetics] do not see their eye doctor on a yearly basis.” This means they have not been screened adequately for retinopathy.
Diabetic retinopathy is a common eye disease in diabetic patients. High blood sugar levels can damage the blood vessels in the retina, causing them to leak and swell in macular oedema, or stop the progress of blood in macular ischemia. Such changes can lead to impairment in vision.
Traditionally diagnosis of diabetic retinopathy has been done by using an injectable dye in fluorescein angiography, or by an imaging machine that scans the retina during Optical coherence tomography (OCT).
The IDx-DR is installed in the cloud. The doctor uploads digital images of the diabetic patient’s retinas on the cloud server. The software will scan the image, analyse it (if it is of good enough quality) and provide either a ‘Positive’ or ‘Negative’ result for ‘more than mild retinopathy’. If the result is positive, the patient will have to go for further diagnostic evaluation and treatments. The device can, therefore, be used by healthcare providers who aren’t specialised in eye care as well.
The FDA tested 900 retinal images obtained from patients, with the IDx-DR. It found that the device was able to correctly spot the presence of more than mild retinopathy 87.4% of the time and its absence 89.5% of the time.
There are some limitations to the device. Patients who have had eye surgery, laser treatment or injections in the eye in the past cannot be safely screened with the IDx-DR. Patients with some other conditions like macular oedema, blurred vision, retinal vein occlusion etc. should not be screened with the device.
The device is not a one-stop solution for diabetic retinopathy screening. But it does mark a breakthrough of sorts. It is the first fully autonomous tool that is different from the decision-guiding AI devices that the FDA has approved before this.
The IDx-DR is seen as a low- to moderate-risk device that the FDA’s new regulations are trying to speed to market. Many in the industry see it as a reneging of the FDA’s promise to focus on only high-risk devices since it has limited resources.
Image credit: www.eyediagnosis.net