A few months back, my mother needed a 24-hour heart trace. This involved a 50-mile round trip to the hospital to have the cardio-band fitted (and another the following day to return the equipment), only to discover the band hadn’t worked and we needed to repeat the whole process. Amazingly, this could be done today without leaving the house by using a band linked to an Apple watch.
AliveCor’s KardiaBand can record an EKG, use AI to decide if the rhythm is abnormal, and transmit the data to a cardiologist. It’s a miracle of miniaturization and the use of artificial intelligence in medical diagnosis. Use of AI is also being researched in cardiac imaging and to diagnose lung cancer and breast cancer.
AI is predicted to be one of the hottest tech trends in 2018, and the use of AI in healthcare is revolutionizing the world of medical diagnosis.
Investment in AI for healthcare
According to the CB Insights AI 100 for 2018, healthcare is the fastest growing field in the AI sector, with eight companies featuring in the top 100. For example, unicorn company Flatiron Health uses assisted big data in the fight against cancer, while AiCure uses facial recognition to determine adherence to medication regimes.
This has led to an explosion in the job market for AI specialists, as well as researchers and healthcare professionals looking to capitalize on the advent of this new field.
But the field where using AI and big data is really taking off is in pattern recognition, especially in medical imaging. This is because AI and pattern recognition partner naturally with how doctors diagnose.
How do doctors diagnose?
In his article AI vs MD, Siddhartha Mukherjee explains how radiologists diagnose conditions. Research has shown that they do this by recognition, rather than by eliminating possibilities. Think of it in the way we recognize animals.
“You recognize a rhinoceros in its totality—as a pattern,” he writes. “The same was true for radiologists. They weren’t cogitating, recollecting, differentiating; they were seeing a commonplace object.”
And this holds good for other areas of diagnosis as well, such as recognizing the waveforms of an EKG or a set of physical symptoms, at an almost subconscious level.
How can computers be taught to diagnose?
Since diagnosis involves learning through repeated practice, machines can be taught to learn using big data. Experts can now create an algorithm using neural network strategy. In the same way that the brain develops neural pathways through repeated behavior or learning, so the computer ‘learns’ – ‘deep learning’
A diagnosis algorithm usually starts with initial data set (the validation set), e.g. multiple images of mammograms. The computer is given details about each image, such as which images contained tumors, which went on to develop cancer and so on. Over time, the computer learns to recognize patterns rather than classify them according to a set of rules.
VisualDX has developed a program that can help with a range of doctor tasks, such diagnosis. The recent incorporation of machine learning using Apple’s Core ML into the app gives it hugely enhanced capacity to aid doctors. Founder and CEO Dr. Art Papier gave me an example: “They can now take their smartphone and point it a rash and the software will classify the rash, then embed in our application. It can also help them think about diagnostic possibilities.”
However, relying on an algorithm for diagnosis is not as straightforward as it might sound.
What are potential dangers of using AI in diagnosis?
“The data that you use to ‘train’ these systems can be biased,” he explained to me. “And that bias can then become part of the algorithm that’s used in the future. Obviously, that is a huge potential problem.”
This raises another concern. Although machines can create an algorithm and reach a conclusion, we don’t know exactly how that algorithm functions. Joel Selanikio emphasized that we don’t know exactly how it’s weighting different factors, or even if it’s using factors we don’t want it to use – the so-called ‘black box’.
“A lot of work is being done right now to try and find ways to reduce the black box nature of these AI algorithms, to know exactly what’s going on.” said he said.
And at this stage, algorithms don’t take into account other factors which could affect the chances of developing a problem like a heart condition or tumor, such as lifestyle choices, stress, or family history.
There’s certainly the potential for increasing levels of monitoring, driven by the development of wearable tech, but this could be a mixed blessing, Do we really want to be given an early diagnosis of a condition if there’s not yet an effective treatment? Would such intense monitoring lead to increased worry for patients or more unnecessary biopsies?
However, AI and big data in medical diagnosis is here to stay, so what of the future?
What does the future hold?
“I think that some of the first commercializations you’re going to see have to do with things like imaging, whether that’s dermatological, pathological, or radiological imaging. These are some of the things that are being worked on hardest,” says Selanikio.
He also believes that the use of AI will move outside the healthcare system into everyday life, such as in the growth of wearables, apps for lifestyle choices.
Papier also sees AI extending outside healthcare settings. “We’ll see more and more apps that are developed for patients or for doctors.
And we’ll also see intelligence embedded in the electronic health record.” he adds
Will AI ultimately replace doctors?
Of course, we’re not talking robot doctors here. But it’s pretty certain that some of the their tasks will be taken over by machines, as well as other healthcare professionals like radiologists.
Papier doesn’t believe doctors will ever be totally replaced, but AI will make the diagnosis process faster and more efficient. He prefers the term ‘augmented intelligence’ rather than artificial intelligence to make a differential diagnosis – one that speaks to ‘guidance and aided thinking‘ rather than a replacement for the doctor’s diagnosis.
“We’re going to move to an age where there are information tools that doctors and other medical professionals can use directly in the role of taking care of patients,” he says.
And Selanikio agrees.
“Do I think that we’re never going to have doctors or surgeons?” he asks. “I think that’s so far in the future it’s probably foolish to make predictions. But do I think that some of the things that doctors do will be done by machines and that some of those things will be done outside of the traditional healthcare system? Absolutely!”