Health

Paging Dr. AI

Our bodies are made of blood, water, and data. Could machines understand them better than human doctors? by Jordan E. Rosenfeld

September 14, 2017

When 59-year-old Mimi Carroll was diagnosed with breast cancer in 2012, it was a complete shock. The California photographer was an active person who ate healthy. How could she be sick? 

Carroll’s only symptom was a single swollen lymph node. While her paternal grandmother had died of the disease, not one of her five sisters or mother had any evidence of cancer. More unsettling, her type of breast cancer has a 40% chance of recurrence after treatment. 

For Carroll, living with that uncertainty has been difficult. “If somebody could tell me, look, cancer comes back in the seventh year, I’d want to know,” she says. Thanks to artificial intelligence, that reality may be a possibility for patients like her in the future.

Doctors are beginning to use AI to assess the state of a patient’s health, even how long they’ll live, by the state of their organs and other data points. Creepy as it sounds, AI technology could go a long way toward helping patients change their lifestyles and increase their health and longevity. 

It will no longer be a question of whether a physician should be using AI, but whether it is ethical if a doctor doesn’t.

A research team at the University of Adelaide is developing a form of AI known as “deep learning” that could radically change the early diagnosis of serious illness and offer better medical intervention, according to Scientific Reports. For people like Carroll, this could change lives in its ability to predict future health or sickness.

Deep learning” is loosely modeled after neural networks in the brain. It can sift through reams of visual and audio data to find hidden patterns the human mind is too simple to find, and it can predict outcomes. It learns as it goes too, essentially becoming more intelligent.

Under the leadership of Luke Oakden-Rayner, a radiologist and Ph.D. student, the scientists used this AI to analyze the medical imaging of 48 patients' chests. The computer-based analysis was able to predict which patients would die within five years, with 69% accuracy, comparable to manual predictions by clinicians.

In clinical data, how long you’re going to live is very strongly associated with how old you already are, Oakden-Rayner says. Excluding age as a measure, their AI was as good as a human doctor at these predictions, and it will only continue to get better with additional data.

This could be especially helpful to patients with chronic illnesses, which come with unsettling gray areas. For people like Carroll, while waiting between treatments to find out if her cancer has successfully been sent into remission, she felt “like a sitting duck” and begged her oncologist to give her alternatives to pursue while she waited. “I’m a planner, an organizer, all of my life,” she says. 

Her oncologist suggested that while she didn’t cause her cancer, the disease is influenced by lifestyle habits, such as stress, diet, even menopause, so she took action in those parts of her life, which likely helped her overall well-being. 

This is not to suggest that anyone can ever prevent diseases through lifestyle interventions alone, but at the very least, with predictive AI, we might be able to get a head start. As it becomes more integrated into health care systems, AI technology could collect and consolidate data on patients from the moment they’re born, allowing physicians to get ahead of diseases that often present in middle to older age.

“If you’re starting to have memory problems at 60, and you’re trying to figure out if it’s dementia, wouldn’t it be nice to have 20 years of neurocognitive testing before that?” says Joel Dudley, a bioinformatics expert and associate professor of genetics and genomics at Mount Sinai in New York.

AI will also have the power to organize disparate data on a patient’s health that is currently siloed. “You’d be surprised how rare it is that all that data [on a patient] is put together and considered as a whole,” Dudley says. 

Say you get your blood measured as part of a dermatology exam — a procedure that literally taps into information about your body, with potentially relevant data about your heart, your immune system, and whether you’re at risk for diabetes or other health issues. But it’s unlikely your dermatologist will look at that information, just as it’s unlikely that your endocrinologist will look through your dermatology appointments because of a focus on his or her specialization.”

“AI can continually gather the bits of health data about you and monitor it to extract new insights and intelligence out of it,” Dudley explains. 

As for AI growing too smart, Mark Lambrecht, director of the Health and Life Sciences Global Practice at SAS, North Carolina, says it’s nothing to worry about. “[AI] can only say something about the data it has been trained on, and it will speculate and do a very good job there.”

What it comes down to is that human patients deserve the best possible care. Lambrecht feels that “soon it will no longer be a question of whether a physician should be using AI, but whether it is ethical if a doctor doesn’t rely on the support of deep learning and other advanced analytics in making diagnoses and treatment decisions.”

Those decisions have very real life-or-death stakes, after all. For now, Carroll is doing well — she’s been cancer-free for a year — but her continued health will remain in an unsettling gray area for as long as the majority of her body’s data points remain unexplored. 

When people should be alarmed, if at all, says Dudley, is at the moment AI begins to predict your health just by looking at your Instagram or Tinder profile, which very well could be possible in the not-too-distant future. If an app can predict you are likely to develop early onset Alzheimer’s or heart disease, maybe someone won’t want to date you, hire you, or insure you.

Dudley believes that these potential harms can be protected through legislation and that “the benefits will outweigh the risk.” But as Facebook’s recent scrapping of an AI program that taught itself to communicate in a new language shows, there’s still a lot to learn. The bigger question is whether AI will replace physicians, though Dudley believes that no matter how much AI will reign superior in diagnosing health problems, we should rest assured that nothing — no matter how powerful — will replace human contact.

Photo by John Thys/Getty Images.

Recently on GOOD Health