The US Food and Drug Administration (FDA) made a historic decision, for the first time allowing the Artificial Intelligence to
independently conduct a medical examination of people without the participation of a living specialist.
The FDA has granted permission to
IDx , which is developing the
IDx-DR diagnostic system for ophthalmology. As stated in the materials of the manufacturer, the device is running AI.
IDx-DR software from eye photographs recognizes signs of diabetic retinopathy. This is a widespread complication of diabetes, affecting the vessels of the retina of the eyeball. It is observed in 90% of patients with diabetes mellitus and can lead to loss of vision.
The program has an AI algorithm that analyzes photos taken with a special Topcon NW400 device for photographing the retina. A doctor or nurse uploads photos to a cloud server or a local computer — and in less than 1 minute, the program returns a positive or negative diagnostic result. In case of a negative result, the next examination is appointed after 12 months, and in case of a positive patient, he is referred to a specialist.
The standard procedure still involves human participation: a trained operator must take pictures and upload them to the cloud. But it can be any clinic employee, technician or trainee, not an ophthalmologist. Diagnosis of diabetic retinopathy is performed without the participation of a single person who understands this disease. Specialists are not required either at the time of the survey or for subsequent interpretation of the results. If the results are negative, then the ophthalmologist will not even see the pictures taken.
Before issuing permission, the FDA evaluated the results of clinical trials in 900 patients with diabetes from ten clinics. The system showed an accuracy of 87.4% in identifying patients with a minor degree of diabetic retinopathy, and an accuracy of 89.5% for patients with complications requiring treatment.
Recently, the FDA
has given permission to use the program with deep learning to diagnose a stroke, but in that case the program is
an auxiliary tool, so right now the first time that the AI has received a complete blank check for diagnosis.
Apparently, this is just the beginning. The head of the FDA, Scott Gottlieb,
promised that in the future, manufacturers of other “smart” diagnostic devices will receive similar authorization for autonomous diagnostics: “Artificial intelligence and machine learning have great promise for future medicine. The FDA is taking steps to promote innovation and support the use of medical devices based on Artificial Intelligence, said the head of the FDA. - We are moving to the approach of pre-certification of AI software, which needs to continue to learn and improve as it is used. The FDA is developing an effective regulatory approach for analyzing AI algorithms that are used in the diagnosis and treatment of diseases. ”
Of course, in the foreseeable future, the software will not replace real people-specialists, but it will greatly facilitate their work, increase the accuracy of diagnosis and the effectiveness of treatment.
Diabetic retinopathy develops when high blood sugar leads to damage to the blood vessels in the retina. This complication threatens about 30 million people with diabetes in the United States alone, and probably more than one hundred million in other countries. Every year more than 24,000 people lose their sight because of it. But vision loss can be prevented by diagnosing the disease at an early stage. Even in the United States with a relatively advanced health care system, studies show that 50% of patients with diabetes do not undergo an annual examination by an ophthalmologist, so they have no chance to diagnose this complication at an early stage. The automatic IDx-DR system will make diagnostics a more accessible and simple procedure. Thus, it will increase the number of patients who are disciplined in the annual examination.