Machine learning and artificial intelligence have never been as red hot as they are now, as companies apply them for use in everything from autonomous vehicles to game systems to contact center software. However, not every industry has embraced the benefits of machine learning equally. Healthcare in particular is lagging behind significantly -- unfortunate since the use of machine learning and AI can positively impact society in a big way by enabling faster diagnosis.
I know some, but not all, of you are aware that in late 2016 my wife, Christine, was diagnosed with something called "common variable immune disorder," or CVID, after being sick for the better part of three years. The diagnosis came only after she had gone from specialist to specialist, having had vials of blood drawn and having been poked and prodded only to hear, "We don't know what you have but it isn't X" (X = the doctor's area of specialty). One specialist even ruled out CVID, but a different doctor later said she did indeed have the disorder.
There is no cure for CVID, but Christine is now going through a treatment in which she gets a four-plus-hour IV infusion of immunoglobulin every 28 days to boost her immune system. An accurate diagnosis early in the process would obviously have improved her quality of life much faster.
Diagnostic Blinders
In June, we attended a conference hosted by the Immune Deficiency Foundation, and there we met others who have similar illnesses to Christine's. (As a side note, I had no idea that immunity illnesses are so numerous -- there are dozens.) We attended the event to learn more about her illness and see how other people cope with it. Everyone we talked to had a similar story of being sick from anywhere from two up to 10 years before getting a proper diagnosis.
During a session on understanding immunological testing, Dr. Manish Butte from UCLA Medical Center went shared the diagnostics and testing results of a number of case studies. In every case, something odd popped up in the datasets, but as the anomaly didn't pertain to the primary diagnosis for which the doctor was testing, it had been ignored. Because immunity problems are rare, nobody starts the process looking for one.
Me, being the curious sort of person I am, asked Butte why the doctors had ignored the anomalous data. Since the anomalous data didn't contribute to the diagnosis for which they were testing, and because doctors rarely look outside their fields of view, they ignore that sort of data, he explained. I then probed a bit deeper as to how the process works. If the patient is getting multiple tests from multiple doctors, why doesn't someone try and connect those dots? He was brutally honest in his response, as many doctors tend to be. Many doctors are not all that aware of many immunity conditions. In addition, doctors are great at finding common problems, but have a problem finding uncommon ones -- and most immunity problems fall into the latter category.
For me, the logical next question was about whether the medical industry has ever considered using machine learning algorithms to connect the dots. This seems like a perfect use case for AI, as it can look through all collected datasets and diagnose a problem much faster than any one doctor. In Butte's opinion, we are probably a decade or so away from using AI in this way.
Resistance to AI
Technical constraints, such as electronic health records not being sharable content, are one reason. While these records are in digital formats, they're often stored on local hard drives that can't be accessed remotely. In fact, as my wife went through her testing, we found some doctors couldn't even access their own records if those were in a different office. In many cases, she would need to have the office print the records so she could bring them to the new doctor for rescanning. I know that seems nuts, but that's the state of the medical industry. (And we live in the Boston area, home of many great advances in medical science -- but apparently not in sharing records.)
Another issue with the use of AI in healthcare has to do with resistance from the medical industry itself. Earlier this year I attended Nvidia's GPU Technology Conference, which featured a track on the use of AI in healthcare. In one of the session, I heard about how AI could be used to speed up diagnoses in radiology. MRIs often show very small spots that are difficult for the human eye to pick up but that AI would "see" instantly. Graphical processing unit (GPU) technology, which has advanced by leaps and bounds over the past five years, is now low cost and extremely powerful -- and can be a great complement to doctors.
After the session I chatted with the speaker, a doctor from Stanford Medical Center, about why AI isn't being used more for applications like this. Many clinicians fear AI, seeing it as a threat to their professions, he told me. I'd heard the same sentiment from Butte at the IDF conference; many old school doctors refuse to let machines diagnose their patients.
I'm no doctor, but in my opinion professionals in healthcare should view machine learning as another tool that can help diagnose patients faster -- and thus lead to better quality care. Aren't MRI machines, X-rays, heart monitors, and other medical equipment merely tools to aid doctors? How is AI any different?
The fact is, massive amounts of data are being generated, and with the Internet of Things being on the verge of exploding, there will be more. Instead of solely focusing on medical insurance reform, perhaps our government could focus on figuring out how to use technology to improve healthcare. We live in a world in which everything is connected, and that's allowing us to have self-driving cars, find spouses using data (I met Christine on e-Harmony), and have music recommended to us. Shouldn't the same technology that does all of these things be used so people don't need to be sick for years before someone tells them what's wrong?
If you're interested in Christine's story, you can view it below. And, if you'd like to make a donation to the good folks at the Immune Deficiency Foundation, you can do that here.
Follow Zeus Kerravala on Twitter and Google+!
@zkerravala
Zeus Kerravala on Google+
