Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again

by Eric J. Topol

developed open-source algorithms that can accurately predict how samples would fluoresce without the need for any fluorescent preparation. They trained the DNN by matching fluorescent labeled images with unlabeled ones, repeating the process millions of times. This method, known as in silico labeling as well as augmented microscopy was called “a new epoch in cell biology.”67 Their assertion was quickly followed by another report of label-free microscopy by scientists from the Allen Institute.

The extensive body of work done by Segal and Elinav is summarized in their book The Personalized Diet. Cumulatively, they’ve studied more than 2,000 people and summed up their revelations about nutrition science as “we realized we had stumbled across a shocking realization: Everything was personal.”

New technology, like an ingestible electronic capsule that monitors our gut microbiome by sensing different gases, may someday prove useful for one key dimension of data input.

For the time being, however, we remain a long way from achieving scientifically validated individualized eating, but this path will likely lead us to a better outcome than the one we’ve been

Setting the table with our diet, we’ve got the perfect launch for the next chapter, which is how AI—well beyond the ability to inform an individualized, tailored diet—will promote consumer health and how our virtual assistant can take on medical coaching responsibilities.

While many AI apps have been developed to promote health or better management of a chronic condition, all are very narrow in their capabilities.

ResApp Health uses a smartphone microphone to listen to a person’s breathing. The machine learning algorithm can purportedly diagnose several different lung conditions—acute or chronic asthma, pneumonia, and chronic obstructive lung disease—with high (~90 percent) accuracy.

There are also many AI chatbots (some that work through Alexa and Google Home) and smartphone apps that perform varied functions like checking symptoms, promoting medication adherence, and answering health-related questions. These include Ada, Florence, Buoy, HealthTap, Your.MD,

as good as we can be at taking care of our own health. We cannot actualize the full potential of deep medicine unless we have something like a virtual medical assistant helping us out. No human, whether doctor or patient, will be able to process all the data.

“No amount of algorithmic finesse or computing power can squeeze out information that is not present.”

The main point here is that quality of the input data is essential,

Each person is dynamic, constantly evolving, for better or worse, in some way, so whatever data are collated, we need to acknowledge that there are key constraints for interpretability. Setting the labels or ground truths for the neural networks could be exceedingly difficult.

Curating all the information and differentiating the quality from more than 2 million biomedical articles that are published each year does not, at least at this juncture, have an AI path to automation. AI extraction from text is a work in progress, certainly improving, and will be essential to support medical coaching.

There are major nonscientific challenges as well—the biggest is having all of a person’s data. The notion that the electronic health record is the hallowed knowledge resource for each patient is a major problem. As we’ve seen, that couldn’t be further from the truth. The EHR is a narrow, incomplete, error-laden view of an individual’s health. This represents the quintessential bottleneck for the virtual medical assistant of the future.

While many have pronounced the end of privacy, that simply will not work for medical data. The privacy and security of your data relies on its decentralization from massive server farms, the principal target of cyberthieves, to the smallest number possible—one person or a family unit being ideal—stored in a private cloud or blockchain platform.

We’ve seen how every EHR has abundant mistakes that are perpetuated from one doctor visit to another, no less that we all have many different EHRs from all our different clinical encounters. And even if they were accurate, remember, too, that EHRs were designed for billing purposes, not to be a comprehensive resource of information about the individual.

I’ve argued, along with colleagues, that owning your medical data should be a civil right.43

I think a pinnacle of the future of healthcare will be building the virtual medical coach to promote self-driving healthy humans.

can go. I’ve tried to emphasize the need for holistic data and backup doctors and human experts. The virtual medical coach will ultimately prove to be a real boon for consumers, even though it’s years

medicine, business interests have overtaken medical care. Clinicians are squeezed for maximal productivity and profits. We spend less and less time with patients, and that time is compromised without human-to-human bonding. The medical profession has long been mired in inefficiency,

One of the most important potential outgrowths of AI in medicine is the gift of time. More than half of all doctors have burnout, a staggering proportion (more than one in four in young physicians) suffer frank depression.

empathy is being built into virtual humans manufactured by the most advanced robotics companies, but even their AI experts admit there will always be a gap, the inability to “imbue such a machine with humanness”—that ineffable presence the Japanese call sonzai-kan.

practice environment. As David Scales, a medical resident, noted, practitioners lack the time to care for patients as the doctors hope and patients deserve, with physicians blaming “the time pressure created by a billing system that promotes quantity of patients seen over quality, the lack of control over the chaotic work environment, and endless time spent on administrative tasks.”

need patients to be free to be storytellers because, even as AI manages to synthesize notes and labs and imaging into something actionable, it will never be able to tell a patient’s story the way a patient would.

deficiencies. I wholeheartedly agree with Verghese’s observation: “For the past two decades I’ve felt that in the United States we touch our patients less and less: the physical exam, the skilled bedside examination, has diminished to where it is pure farce.”

“The significance of the intimate personal relationship between physician and patient cannot be too strongly emphasized, for in an extraordinarily large number of cases both diagnosis and treatment are directly dependent on it.”

and the guts of deep learning neural networks. Much of their efforts in patient care will be supported by algorithms, and they need to understand all the liabilities, to recognize bias, errors, false output, and dissociation from common sense.

DEEP MEDICINE We’re still in the earliest days of AI in medicine. The field is long on computer algorithmic validation and promises but very short on real-world, clinical proof of effectiveness. But with the pace we’ve seen in just the past few years, with machines outperforming humans on specific, narrow tasks and likely to accelerate and broaden, it is inevitable that narrow AI will take hold.

All of this is surrounded by the caveats that individuals must own and control their medical data,

to me, those are the secondary gains of deep medicine. It’s our chance, perhaps the ultimate one, to bring back real medicine: Presence. Empathy. Trust. Caring. Being Human.