By John Elder
Dear robots: please save my citizens. This message was at the heart of a speech given last month by UK Prime Minister Theresa May. She pledged millions of pounds – essentially making a bet – on artificial intelligence (AI) to solve Britain’s cancer crisis and to better diagnose other conditions such as diabetes and Alzheimer’s disease.
The subtext here was that instead of trying to fix Britain’s National Health Service (NHS) – which is under-manned, under-equipped and going backwards – the Prime Minister is hoping that AI will cheaply and efficiently take up the slack. There simply isn’t the money to fix the NHS by traditional means – more doctors, specialists, pathologists. A revolution is required. And fast.
For at least 10 years the UK has lagged behind the rest of Europe in terms of cancer diagnosis, treatment and survival rates. According to a 2017 report, the UK is behind on nine out of 10 cancers in terms of survival – including bowel, lung, breast, ovarian, prostate and kidney cancer. The report concluded that if the UK achieved the cancer survival rates of Germany, 35,000 more people would be alive five years after diagnosis. If the UK had the cancer death rates of France, more than 100,000 women's deaths could be prevented over the next decade.
Theresa May recognises that early detection and diagnosis is linked to increased survival rates. As she said in her speech:
“Late diagnosis of otherwise treatable illnesses is one of the biggest causes of avoidable deaths.”
Why turn to AI?
Because the last 18 months has seen major breakthroughs in machine learning – where the diagnostic ability of experimental AI programs has reportedly matched or out-gunned that of medical specialists.
As Mrs May noted:
“The development of smart technologies to analyse great quantities of data quickly and with a higher degree of accuracy than is possible by human beings opens up a whole new field of medical research.”
Probably the most daring part of her plan, May wants AI technology to give the entire nation a check-up. She wants to see algorithms scanning people’s medical records, genetic data, and lifestyle habits to spot cancer. This is what’s coming, and not just in Britain. AI will get to know you and predict when you’re going to get sick – you won’t feel a thing. But what were these breakthroughs that got Mrs Theresa May so excited?
In 2017, the MRC London Institute of Medical Sciences (LMS) announced that computers had learned to predict heart failure. According to one of the researchers, this was the first time computers had managed to interpret heart scans to accurately predict how long patients would live. Here's how it works:
- The machine analyses moving images of a patient’s heart captured during an MRI scan.
- From these images, it builds a ‘virtual 3D heart’ so accurate that it replicates the 30,000+ points involved in a heart contraction during each beat (see video below).
- Researchers feed the system data from hundreds of past patients. The machine links these data with its models, and learns which attributes of a heart, its shape and structure put an individual at a given risk of heart failure.
According to researchers, the computer was able to perform the analysis in seconds. It could simultaneously interpret data from imaging, blood tests and other investigations without any human intervention.
And it’s technology like this that can help doctors find patients at risk of dying in the near future, treat them intensively, and hopefully save them.
Yeah but what about cancer?
There was another piece of news in 2017 that made an even bigger splash than the heart-stopper: Stanford researchers announced that they had created an AI algorithm to diagnose skin cancer that matched the performance of dermatologists.
The researchers began by building on an algorithm developed by Google that was trained to identify 1.28 million images from 1000 object categories (like ‘cats’ and ‘dogs’). Specifically, they needed to teach the algorithm how to differentiate between malignant carcinomas (bad ones) and benign cancers (not-so-bad ones). So they collected 130,000 images of different skin lesions off the internet and fed them to the system, along with a disease label. Immediately, the algorithm was able to identify particular cancers.
Then, it was put to the test against 21 board-certified dermatologists. The dermatologists were shown high-quality images of the most common and deadliest skin cancers and asked whether, based on each image, they would proceed with biopsy or treatment, or reassure the patient. The researchers evaluated success by how well the dermatologists were able to correctly diagnose both cancerous and non-cancerous lesions in over 370 images.
After that, the researchers assessed the algorithm’s performance by looking at its 'sensitivity' and 'specificity'. Sensitivity represented its ability to correctly identify malignant lesions (bad ones) and specificity represented its ability to correctly identify benign lesions (not-so-bad ones). The algorithm was given three key tasks. It had to diagnose and classify images of:
- Keratinocyte carcinomas - the more common type of skin cancer
- Melanomas - the less common, but more deadly type of skin cancer
- Melanomas imaged up-close with a handheld microscope
The result? In all three tasks, the algorithm matched the performance of the dermatologists. The next step is to make the algorithm smartphone compatible, which would allow people to do their own initial skin cancer diagnosis at home.
Fast-forward to February 2018: Cornell University has developed an AI program that can distinguish breast, lung and bladder cancers from images of cells with almost 100% accuracy. In a statement published on the university website, the researchers said the program was able to discriminate lung cancer subtypes with 92% accuracy, and detect biomarkers for bladder and breast cancer with 99% and 91% accuracy respectively.
Hold your horses
The fast pace of research and development has seen some people worried. In March, Nature magazine published an editorial that cautioned it was too early for these new programs to come out of the lab and into mainstream clinical practise. Nature said there were questions of testing not being rigorous enough, and that the fine technical detail of these experiments wasn’t being offered up for peer-review. The editorial concluded that slow and careful research is a better approach. Gathering reliable data and using robust methods may take longer, and will not deliver as many ‘crowd-pleasing’ announcements, but it can prevent deaths and change lives.
It seems, though, that the horse has just about bolted. In April, the MIT Technology Review reported that the FDA had approved an AI-powered diagnostic device for eye disease to go on the market. The software is designed to detect diabetic retinopathy, which occurs when high blood sugar damages blood vessels in the retina. It affects millions of people world-wide. An AI algorithm analyses images of the adult eye taken with a special retinal camera. A doctor uploads the images to a cloud server, and the software delivers a positive or negative result. No specialist required.
MIT noted that the FDA had also recently cleared AI-based software to help detect stroke and was hinting that even more AI medical devices were on their way to the open market.
The take-home message
Healthcare is one of the great burdens on economies everywhere. The recent strides in development of AI diagnostic tools has the potential to ease that burden. It’s exciting, promising and happening so fast that we can barely keep up. Calls for scientific rigor and testing are more than reasonable, but are they being heard? That part isn’t clear.
Of course an algorithm that can detect cancer is EXACTLY the kind of thing that excites us at HeadUp. Is there a better way to use technology, than to assess and improve your health? We don’t think so. And the diagnostic algorithms we’ve talked about in this article take that idea to the next level.
At the same time, we agree with the ideas posed in Nature’s editorial: slower, more thorough testing is important. If we want powerful, evidence-based, and high-quality healthcare at our fingertips, we’ve got to get the tech right. Otherwise, we’ll end up only mildly more rigorous than the Instagram Influencer promoting the use of coconut oil for everything from weight-loss to acne. When it comes to your health, we believe that the tests you run on your body – and any subsequent lifestyle changes – should be thoroughly tested.
This is why we’ve been testing the algorithms that will go into our app for six months now. We also make it a priority to keep up with the science around health, fitness and wearable tech every day, to see how our findings stack up against the scientific literature. Soon, we’ll reveal what we’ve been working on, and you’ll discover what our app can do for you. Until then, you can keep up with our progress on Facebook, Instagram, or Twitter.
John is a Melbourne journalist and writer. For 21 years he was senior writer for the Sunday Age newspaper and Fairfax Media, with a special interest in science, medicine and health research – and in philosophical questions facing modern society. He continues to be a life-issues columnist for The Sunday Age, science and health contributor for The New Daily and is an editorial advisor for the Clueless Economist.
*Please note: header image was found here: https://bit.ly/2twcSnd