With the appearance of the artificial intelligence (AI) system ChatGPT—a software program that can answer questions as if you’re chatting with a very intelligent and well-informed friend—AI went from science fiction to science fact. But does it play a role in the science of medicine? Can you and your doctor use AI to help you achieve better health?
To answer that question, Bottom Line Health spoke with Michael Abramoff, MD, PhD, an ophthalmologist and computer science expert at the University of Iowa, who led the creation of the first autonomous (self-contained) AI diagnostic systems cleared by the U.S. Food and Drug Administration.
BLH: Let’s start by defining terms. What is AI, exactly?
In health care, artificial intelligence is a type of software that solves a cognitively complex task that, so far, had been done only by doctors, like making a diagnosis.
There are different types of AI, which has been in development since the 1950s. Some AI uses mainly “rule-based” systems or algorithms—a step-by-step procedure designed to solve a specific problem. For example, in making a diagnosis, a medical algorithm might proceed: “Take a throat culture. Was it positive? If yes, check for the following symptoms.”
Recently, AI has used machine learning: The software is trained with massive amounts of data, forming neural networks similar to those used in the brain. If, for example, the software is trained on thousands of scans of breasts with and without breast cancer, it can make an accurate diagnosis of when a woman has diseased breast tissue—a diagnosis which may be more accurate than that of a evaluating the scan.
BLH: What do you see as the role of AI in health care?
I make the distinction between what I call “glamour AI” and “impact AI.” Glamour AI is AI that is exciting and gets everyone talking, but doesn’t improve clinical outcomes for the patient.
Impact AI, on the other hand, is focused on clinical outcomes for patients and may help them lead healthier, longer lives. And impact AI has the potential to provide these better patient outcomes at higher quality and lower cost. It also addresses potential bias against particular groups of patients, like low-income patients or African-Americans. Additionally, impact AI is well-integrated with the health-care system, including regulators like the FDA and insurers like Medicare and Aetna.
BLH: What are some examples of impact AI in health care?
Currently, the most impactful is so-called autonomous AI, which makes an accurate diagnosis by itself or helps a doctor to make a more accurate diagnosis. The FDA has already signed off on 700 Artificial Intelligence and Machine Learning-Enabled Medical Devices—mostly in the diagnostic specialty of radiology, but also in cardiology, ophthalmology, gastroenterology, urology, neurology, and many other medical specialties.
Diabetic retinopathy. Nearly 40 million Americans have diabetes, and all of them are at risk for diabetic retinal disease, an eye condition that is the leading cause of vision loss and blindness in America. The key to preventing vision loss and blindness from diabetic retinopathy is detecting the problem before the onset of symptoms like blurry vision and floaters. But while people with diabetes may regularly see a primary care physician or endocrinologist for their condition, they rarely see an eye doctor.
Now, an FDA-de novo cleared autonomous AI can diagnose diabetic retinal disease without an eye care provider—or even a doctor. (Any qualified technician with a minimal amount of training can operate the device.) Using machine-learning derived from millions of retina images of eyes, the AI uses a camera designed to take images of the retina and diagnose a patient, providing results in less than a minute. If no diabetic retinal disease is detected, the test can be repeated a year later. If it is detected, the patient is referred to an ophthalmologist or retina specialist for further treatment.
The device has now been used on more than 100,000 patients. In a study of the device, published in NPJ Digital Medicine in March 2023, it was estimated that widespread use of the device could prevent 27,000 cases of vision loss in the U.S. over a five-year period.
This autonomous AI does not need a specialist to make the diagnosis—and specialists can spend their time on patients that need their expertise. This provides increased and faster access to needed health care, and lowers costs. If you have diabetes, you and your doctor can find out more about the device—LumineticsCore—at www.digitaldiagnostics.com. LumineticsCore has been cleared by the FDA, and can be reimbursed by insurance. Other interesting developments include:
Skin cancer. A study presented at the European Academy of Dermatology and Venereology Congress in October 2023 showed that AI software had a high success rate in identifying skin cancer and precancerous lesions. Involving more than 22,000 patients over 2.5 years, the software identified 100 percent of melanoma cases (59/59), 99.5 percent of all skin cancers (189/190), and 92.5 percent of precancerous lesions (541/585). The FDA has given breakthrough designation status to Sklip (www.sklip.ai), an AI system that identifies skin cancers with more than 95 percent accuracy. You can use it with a smartphone.
Breast cancer. Nearly 300,000 cases of breast cancer are diagnosed in American women each year. In a study published in Lancet Oncology in August 2023, Swedish scientists found that using AI software—which had data from millions of mammograms—detected 20 percent more breast cancers than standard mammography readings by radiologists. The first FDA-
cleared system for 3D mammography is ProFound AI. You and your doctor can find out more about it at www.icadmed.com/breast-health.
BLH: Can an AI chatbot like ChatGPT be a personal health assistant?
Like any search engine, an AI chatbot (also called generative AI) can help you become a more informed patient, and may help guide you in asking your doctor questions about your condition. But take any answer from generative AI with a grain of salt. A chatbot may generate fabricated research citations about your health or condition.
It is also unable provide foolproof solutions. In medicine, there are always risks and benefits to any decision. Your decision-making must be based on what you want, what the doctor’s experience tells them is best for you, what science shows, and what has been proven to work and not to work.
Finally, ask yourself this question: Does the company that created and markets the chatbot assume liability if the chatbot is wrong? Companies such as LumineticsCore that market the most advanced AI solutions in healthcare—autonomous diagnosis—do take liability if the AI is wrong. But a chatbot does not.
Many Americans are afraid of AI because they’re afraid of machines becoming smarter than humans and making big, catastrophic mistakes. Many of those fears focus on health care. A recent survey from the Pew Research Center showed that 60 percent of Americans say they would feel uncomfortable if their own health-care provider relied on AI to diagnose disease and recommend treatments. But among those who have studied AI for more than a decade, there is great enthusiasm for the benefits this technology can bring—and minimal fear of the risks. Among them are Mark Minevich, author of Our Planet Powered by AI, co-chair of the AI for the Planet Alliance, an organization working with UN agencies, chair of the executive committee at the AI for Good Foundation, and president of Going Global Ventures.
AI can unlock many possibilities in health, climate, infrastructure, transportation, energy, food and security, says Minevich. It can help solve the greatest problems humanity is facing. AI itself is not malicious or dangerous, he says. It is a tool—a cognitive agent—that is there to help us. The next step, he says, is for the technology to become trusted, and for guardrails to be instituted that prevent abuses like creating misinformation or exposing private information.
In terms of medicine, AI should be widely used for diagnosis, personalized medicine (individualized treatment plans based on factors like genetics), drug development, disease monitoring during infectious incidents like pandemics, and prevention strategies.
“AI will not replace but complement health-care providers,” says Minevich, “creating the opportunity for a healthier, longer life. The AI-powered future will be the age of healing.”
