Artificial intelligence in medicine: The good, the bad, and the scary

February 8, 2020

At Glaucoma 360 in San Francisco, Terri-Diann Pickering, MD, told attendees that artificial intelligence is a hot topic, with room to expand in the future.

Artificial intelligence (AI) is a hot topic today, but as with all new technologies, it has a long way to go.

A trend in medicine that is pushing the rise of AI is a doctor shortage in the United States, Europe, and China that is intensifying with the aging of the population and the need for more cost-effective treatment, according to Terri-Diann Pickering, MD, speaking during Glaucoma 360’s annual Glaucoma Symposium.

AI already has many applications in medicine: Microsoft uses AI to diagnose cervical cancer faster in India; and AI has been found to be as good as physicians for diagnosing skin cancer and better than radiologists for detecting certain breast cancers. So, this market is growing rapidly from $2.1 billion to $36.1 billion by 2025, according to Dr. Pickering, clinical instructor, Glaucoma Center of San Francisco.

Driving innovation

Four medical specialties are driving the surge in medical AI: dermatology, radiology, ophthalmology, and pathology. In some cases, AI performs better than doctors, which raises the concern about whether the robots are going to be replacing the doctors.

“Is AI going to create good jobs, make our lives easier, or be the killer of jobs?” she asked.

In the financial sector, Dr. Pickering noted that AI is superseding well-paying Wall Street jobs.

“Currently, the New York Stock Exchange is devoid of humans and runs primarily by AI,” she said. “And 6.14 million people working in finance and insurance are expected to lose their jobs not because they are being replaced by the machines but because they are not trained to work with algorithms.”

AI in Ophthalmology

The first autonomous AI device in Ophthalmology is IDx-DR (IDs Technologies), which is capable of making a screening decision without a doctor and refers patients to doctors.

The company worked with the FDA for 8 years to fine-tune the product. The device was studied in 900 patients at 10 primary care offices before it received FDA approval in 2018. The device is now being used in primary care offices and in grocery stores.

The company has purchased liability insurance in the event of any injuries. Dr. Pickering noted that AI also may be able to predict treatment outcomes and analyze a single optical coherence tomography (OCT) scan to predict which patients might benefit from anti-vascular endothelial growth factor treatment.

“This is a step forward in precision medicine,” she said. “This would save health care and patient and doctor time.”

There is also an Augmented Reality contact lens (Mojo Vision). This technology places micro-displays inside the eye on the retina.

The company hopes to help low-vision patients. Something else in the pipeline is robot-assisted surgery, The Steady Hand developed at John Hopkins. It eliminates tremors, but these is less degrees of motion, but it is 10 times more precise than a human hand and less potential to touch the retina. However, the downside is the slowness of the device, which has been approved for retinal surgery in Europe.

Robotic system

The Intraoperative Robotic Interventional Surgical System from UCLA has the goal of developing a fully automated cataract surgery robot and an assistant vitreoretinal surgery robot.

“This is intended to help not replace the surgeon,” Dr. Pickering said.

For glaucoma, nothing is yet approved, but some technologies are promising, such as a study looking at forecasting future Humphrey Visual Fields using a single visual field, which can help with personalized treatment and prognosis. An automated AI algorithm has been found to be better at diagnosing glaucoma using OCT and visual fields in general ophthalmology practice and equal to glaucoma specialists. Using fundus photos, an algorithm has been found to be highly accurate in identifying glaucomatous optic nerves.

Also, AI has identified 49 previously unknown genome sites that was coupled with other known sites to develop a potential risk score and can identify people with up to 15 times increased risk of developing glaucoma at an earlier age. When viewing AI through rose-tinted glasses, Dr. Pickering concluded, “We will have a host of new tools to predict glaucoma risk, diagnose glaucoma, detect progression, identify optimal treatment using precision medicine, and forecast prognosis.”

There is a lot of promise but there also is the potential for risk because the systems have not been tested vigorously. Some systems have failed in health care centers and a system developed in one hospital may not work in another. Software can discriminate against minorities and they can make odd predictions that have nothing to do with disease. The products have little evidence to support them and the start-ups publish press releases with no research in peer-reviewed journal.

No products have been tested in randomized clinical trials, Dr. Pickering noted. “The industry needs careful oversight but the devices do not need review and approval,” she said and reported that over the past decade many 80,000 deaths and 1.7 injuries have been linked to these devices. These products were cleared for sale because they were associated with moderate risk without testing as long as they are similar to pre-existing devices. So for right now, the take-home message seems to advocate a great deal of caution.