The AI's Have It: An AI First

The (A) I’s Have It: An AI First

It’s a rare day that goes by without someone mentioning how AI algorithms are going to change the healthcare world. But, last February, when Merdis Wells visited the diabetes clinic at the University Medical Center in New Orleans to have her eyes checked for signs of the most common cause of blindness, diabetic retinopathy, Artificial Intelligence made the call. Specifically, the IDx-DR did. The machine is the first device that makes a screening decision – reporting whether there are signs that a patient’s vision is starting to erode – without a clinician needing to get involved in the interpretation.

The system’s inventor, Dr. Michael Abramoff, is an ophthalmologist at the University of Iowa, who believed a computer algorithm could scan retina images and automatically pick up early signs of diabetic retinopathy. He spent years developing the algorithm, and even longer getting a green light from the FDA, since the doctor, who’s the CEO and founder of IDx, needed to prove to the agency that the technology was safe and effective and that it would work on the very diverse population that gets diabetes. The FDA’s bar was set deliberately high, according to Abramoff, who said of the system, “It’s better than me, and I’m a very experienced retinal specialist.”

The IDx-DR costs about $20,000, with options to rent or lease-to-own the camera. Yet, in short order, it pays for itself by enabling a lot more patient visits to a primary care – or an optometrist’s or optician’s location – where there isn’t an MD administering the test. And, by sorting out those who need to see an ophthalmologist from those who don’t.

ABL Healthcare and Technology InsightsThe fact is, a lot of Healthcare AI is in development. Just this week Researchers at NYU and Princeton relayed that they’ve developed a framework that evaluates clinical notes, and assigns a risk score indicating whether patients will be readmitted within 30 days. The researchers claim that the code and model parameters, which are publicly available on Github, can handily outperform baselines. A valuable model for hospitals aiming to dramatically lower their readmissions.

Of course, the FDA is tasked with ensuring that the software in approved devices remains safe and efficacious – and doesn’t fail. But, when it does, just like cars with faulty airbags, it gets recalled. In fact between 2011 and 2015, there were 627 software recalls, including 12 “high risk” devices such as ventilators and a defibrillator.

So it’s not surprising that the Food & Drug Administration has be cautious in adopting Artificial Intelligence’s use in Healthcare. On his way out the door, FDA Commissioner Dr. Scott Gottlieb has prepared a 20-page exploratory whitepaper on how the Agency could address AI and Machine Learning algorithms. In it, he proposes a new framework that would allow companies to include their plans for anticipated modifications in their AI products’ premarket submissions. Since, by definition, AI software constantly learns and self-improves, should the guidance in this “exploratory whitepaper” become finalized, developers wouldn’t need to run back to the FDA for a new approval every time changes are made in their algorithms.

In the meantime, Merdis Wells, and thousands of diabetics like her, will be able to more easily access the equipment and the operators (not doctors) needed to run it, so they can get the highly recommended annual exams to diagnose whether they just need to manage their blood glucose levels and diet, or if they need a referral to an ophthalmologist to prescribe specific medications, laser treatment, or even surgery. Ideally, this way, they won’t be one of the more than 24,000 people who go blind every year from diabetic retinopathy.

by Mimi Grant, President, Adaptive Business Leaders (ABL) Organization – Round Tables and Events for CEOs of Technology and Healthcare Companies


Photo Credit: