Debunking the robodoc: the pivotal role of human-machine collaboration in the future of diagnostics

It’s no secret that the future of nearly every industry will involve innovative applications of data. In the healthcare industry, specifically, large-scale, data-driven decision making has the potential to generate as much as $100 billion of value by improving the efficiency of research and clinical trials and building new tools for physicians, consumers, insurers, and regulators that improve and personalize the patient experience.

Artificial intelligence (AI) tools are poised to play a significant role in what may ultimately be a transformative era for the industry — specifically those featuring “deep learning” capabilities. Research from growth consulting firm Frost & Sullivan suggests that the healthcare AI market will experience a compound annual growth rate of 40 percent through 2021, at which point it will account for over $6.5 billion of earned revenues.

Impressive progress in AI-based diagnostics

A variety of research confirms that this transformation is indeed already underway. In 2016, researchers from Beth Israel Deaconess Medical Center and Harvard Medical School used a deep learning algorithm to assess whether a cluster of lymph node cells contained cancer. Their algorithm achieved a diagnostic success rate of 92 percent, slightly below the 96 percent success rate achieved by human diagnosticians.

In a more experimental vein, Israeli researchers created an AI-based device designed to recognize — and differentiate among — 17 different disease conditions, including chronic kidney failure, pulmonary arterial hypertension, and lung cancer based only on samples of patients’ breath. Using an artificially intelligent nanoarray built with molecularly modified gold nanoparticles, the team assessed breath samples from 1,404 patients, correctly diagnosing 86 percent of them.

Where we really stand

With studies like these emerging more frequently, it can be challenging to discern what it all really means for the industry — both in the immediate future and far beyond.

Considering that diagnostic errors contribute to roughly 10 percent of patient deaths and between 6 percent and 17 percent of all hospital complications, AI-driven technologies in the mold of those outlined above have the potential to make substantial contributions to the healthcare industry and the patients it serves. Not only are algorithms quickly approaching human-level diagnostic success rates, they are doing so on timelines of which humans are quite literally incapable. In healthcare — where time is a precious resource — this efficiency is tremendously valuable.

But as promising as many of these explorations of AI-based diagnostics may be, it’s critical to place them in their proper context. For one, AI is unlikely to replace your doctor anytime soon. While its diagnostic success rates are undoubtedly impressive — and will only get better with subsequent refinements — they don’t always tell the whole story. In short, contrary to what the headlines may have you believe, when it comes to healthcare diagnostics, statistical accuracy is not the only metric about which we need to be concerned. In fact, this is one of the major reasons why, even as these technologies mature, human diagnosticians needn’t fear being replaced by a robot.

Understanding the diagnostic process

One of the primary reasons why physicians and diagnosticians are insulated from automation-based job loss is the nature of differential diagnosis itself, a stepwise process which has dominated medical practice for decades — and will continue to do so for the foreseeable future. After gathering facts about a patient’s specific condition and general background and generating a list of potential etiologies, a diagnostician will commonly order a slate of tests in order to narrow the list of possible culprits.

According to Saatchi & Saatchi Wellness’ Group Medical Director Dr. Kavin Shah, “The screening process will often involve laboratory tests that are designed to detect particular markers of a specific disease. For example, the prostate-specific antigen (PSA) test for prostate cancer measures blood concentrations of PSA, a protein produced by the prostate gland.”

As Dr. Shah continues, however, “Many medical evaluations and tests may be thought of as part of the screening process, as well.” From blood pressure tests and routine EKGs to mammograms and questionnaires about personal behavior and risk factors, there are a wide variety of non-lab tests that help diagnosticians hone in on a patient’s specific ailment.

Ultimately, none of these is definitive in isolation. “They raise a heightened suspicion of disease,” says Dr. Shah, “but they aren’t diagnostic. A definitive diagnosis generally requires more extensive, more reliable, and frequently more invasive evaluations.” In other words, outside of unusually simple cases, physicians rely on multiple tests — and multiple kinds of tests — to arrive at their final diagnosis.

Forging a human-machine partnership

Diagnostic certainty dramatically increases once a test is repeated and, especially, once supplemental tests are run. In the end, this is where the true potential of AI-driven diagnostic tools lies. Getting a second opinion matters, and AI can serve as that second opinion.

The more tests we run, the more precise our diagnoses become, and deep learning algorithms like the one developed by the Beth Israel/Harvard team can help medical professionals perform more tests more efficiently than ever before. It will remain a human medical professional’s job to dictate which tests need to be run and when but forging a collaborative partnership with AI-based tools will empower us to amplify the productivity of the diagnostic processes we’ve been executing on our own for decades. In short, when it comes to healthcare diagnostics, AI has the potential to bring about a change in scale, not in kind.

And, according to deep learning pioneer Sebastian Thrun, this — not robot doctors — has always been the plan. “I’m interested in magnifying human ability,” he told The New Yorker. “Just as machines made human muscles a thousand times stronger, machines will make the human brain a thousand times more powerful.”

For patients across the world, this can only be taken as good news.