Artificial intelligence in medicine is perhaps the disruptive technology about which there is greatest consensus, it’s coming, and the impact will be huge. We haven’t yet identified all the areas where it will make an appearance, but one of the first applications we hope to see is in the (early) detection of illnesses.

WHY ARE WE INTERESTED IN ARTIFICIAL INTELLIGENCE IN MEDICINE? WHAT MIGHT IT BE ABLE TO DO?

Can an algorithm really reach more precise diagnoses than a doctor can? Well, the answer is yes, but not in the way that people currently think it can.

The profile of the doctor of the future will inevitably change. In some tasks, we will be substituted by tools that do the mechanical part of our job, and they will do it infinitely better than we will.

Let’s limit ourselves to the bureaucratic part. The idea that, one day, medical records might edit themselves is a major relief!

Just think about all the time that we could free up to spend with patients – even though the thought might alarm more than one of us. More time with patients implies investing more time, empathy and patience, with them and, for some, that’s a bona fide nightmare.

From what I’ve understood until now, I think the role of the doctor in the near future will be as a subtle interface between patients and AI.

Just as we’re currently the bridge between patients and diagnostic methods, it will also be us who guide interaction between them and AI.

It strikes me as a logical and natural progression and doesn’t scare me in the slightest.

WHO IS DOING WHAT?

 

Diagnostic imaging – MIT

 

Regina Barzilay is a professor at the Massachusetts Institute of Technology who taught computers to learn, via natural language processing, until she was diagnosed with breast cancer.

“Going through it (cancer treatment), I realized that today we have more sophisticated technology to select your shoes on Amazon than to adjust treatments for cancer patients,”

Barzilay’s group now collaborates with Massachusetts General Hospital, applying their knowledge to help improve the diagnosis and treatment of cancer.

Some aspects being investigated are if AI can detect early signs of breast cancer on a mammogram before a human can, and if the analysis of big data from patients can help develop more personalized treatment options. Before being diagnosed with cancer, Barzilay underwent mammograms for more than two years without anything being found.

“Looking back, there was clearly no tumor on the previous mammograms, but was there something in these very complex images that would hint at… a wrong development?” Barzilay asked. “It obviously didn’t just appear. Biological processes are in place to make a successful growth, and it clearly impacts the tissue. So for a human who looks at it, it’s very hard to quantify the change, but a machine can compare millions of these images in a short period of time and find patterns that would be impossible for the human eye to identify.”

The are others pursuing the same approach.

 

PathAI

This research group is made up of a pathologist from Harvard Medical School, (Dr. Andy Beck), the deacon of the Medical Center (Beth Israel), and a software engineer from MIT and Caltech (Aditya Khosla). They also believe in the potential for diagnostic cancer imaging via AI. They are teaching their computers to tell normal cells from cancerous ones. The company, PathAI, was set up after they won a competition to detect breast cancer.

 

Results from the 2016 competition

• Error margin for an expert pathologist: 3.5%
• Error margin for AI: 7.5%
• Combination of the 2 (expert human and AI): decreased the error margin of the expert by 85% (in other words, the combined error of the two drops to 0.5%).

But this equation leaves out the AIs exponential learning curve. Only six months later, the AI beat the expert. Bye-bye human superiority!

Deep learning

 

The true potential of AI lies in deep learning. Machines don’t need you to stand beside them, hold their hand and tell them what is important and what isn’t. This automatic learning, this advantage, is what we term “intelligence”. In truth, it’s the opposite – it’s just a robot that is carrying out orders at high speed.
It’s not a fresh mind. It’s a completely different one! The computer uses its own criteria to look for what it deems to be important. That’s how it ends up analyzing data that we wouldn’t even look at and why it opens the door to new correlations that would never have occurred to our human minds.

 

Deep Patient

 

This is what Joel Dudley’s team in Mount Sinai in New York dedicate themselves to. His Deep Patient system analyses and correlates de-identified data from all over the hospital.
“A physician or researcher focusing on type 2 diabetes, for example, may develop a model focusing on blood glucose or weight to try to predict who may be at risk for disease. But that then ignores all the other information in the health record that could be useful for predicting someone who’s at risk,” Dudley said. “So we use a deep learning approach where we could just pour in all the information we have on 5 million patients in our health system, from any test that’s ever been run on a patient.”
The results of the study were published in Nature. Deep Patient improved the prediction of diseases such as schizophrenia, cancer, and diabetes.

Fuente: Popsci.com

WHAT’S ON THE MARKET?

Investment in the private sector is immense. The AI market in health predicts growth of 40% year on year, reaching 6.6 million dollars in 2021 (Frost & Sullivan).

1. Google Deepmind Health:

This is a collaboration between Google and the NHS (UK health system). It’s based around AI apps for mobile to optimize patient care from the first time they make contact until treatment. The goal is to speed up all non-medical processes and, so far, has shown positive results with nurses reporting that, on average, they have two extra hours to spend with patients.

Fuente: Theverge.com

2. Alphabet:

Alphabet work on an initiative to collect gene information – the Baseline study. They try to use the same algorithms as search engine Google to find out what it is that makes people healthy. They are also experimenting with technology to monitor disorders, for example, a digital contact lens that analyzes blood sugar levels.

3. IBM WatsonPaths:

Alphabet work on an initiative to collect gene information – the Baseline study. They try to use the same algorithms as search engine Google to find out what it is that makes people healthy. They are also experimenting with technology to monitor disorders, for example, a digital contact lens that analyzes blood sugar levels.

Watson is being used in the Alder Hey Children’s Hospital.

There are many similar projects, and you can find out more about them by following these links:

Careskore
Zephyr Health
Oncora Medical
Sentrian
CloudMedX Health

LIVING WITH THE CHANGE

 

“To improve is to change; to be perfect is to change often.” – Winston Churchill

“Intelligence is the ability to adapt to change.” -Stephen Hawking

Although to adapt, you have to accept change. It seems that there are some “old school” academics that are having a harder time acknowledging new paradigms.

Take the stethoscope, for example. It’s a classic artifact that’s still around today, but to get here, since being presented in 1819 by French doctor Laënnec, it had to pass through several decades of ignominy before being accepted by the medical community. The other day, in cardiology, I was looking for the surgical report from a mitral valve replacement. The doctor was a famous cardiac surgeon, but I couldn’t find anything in the database.

Days later, the anesthetist told me that the doctor in question almost never used the computer and that all his reports were written by hand and stored in a file. A classic, well-written and fully comprehensive report finally came to light after losing hours looking in the digital system.

Does that still happen? Are surgeons still allowed to write their reports by hand? Don’t they have a resident or someone who could digitalize the information? This is where worlds collide.

WHAT DO WE NEED TO DO?

• Basic AI training. They say that doctors’ chemistry training is rudimentary (something I can confirm by comparing the average results in biochemistry with the average from other subjects). Even so, a doctor should understand the base molecules of his profession and the basic tools in daily use at a medical center. We need to train our doctors to use these technologies if we want them to be able to work with them.
• AI implementation needs to be followed up by the health system. In order to see if there is a net benefit, and to what extent, where it’s working and where it’s not working, we need to collect data and study results, just like we do in a clinical study.
• There needs to be more communication between the private sector and the general public. Debate should be incentivized, knowledge shared, false perceptions corrected and concerns – like privacy – mitigated.

 

LIMITATIONS?

• It’s just hype. That’s possible. We’ll see how quickly programs develop, how long we need for successful implementation and how accessible they turn out to be.
• Technological barriers. They are being broken down.
• Economic barriers. I have no idea what it costs.
• Data access. The legal implication could slow things down considerably.
• Human communication. The technology is still at an early stage of development. Will that translate into a permanent and definitive limitation? I don’t think so.

 

CRITICISMS OF AI

In general, Hawkins was and Musk is against AI. They fear it and worry that a Terminator-style scenario awaits us. Mark Zuckerberg, in contrast, is totally in favor of AI and has expressed frustration with experts who are limiting AI development with predictions that encourage fear.

AI in medicine will begin with a more innocuous focus – we’re talking about improving an area that costs workers hugely in terms of time and energy. Because of that, handing administrative and bureaucratic organization over to AI seems very desirable.

That’s why I believe it’s important to differentiate between the Terminator style scenarios at a global level and specific application of tech, like Fitbit.