Medicine and machine learning

Artificial intelligence is already playing a vital role in disease diagnostics, with machines sifting through data to identify symptoms, risk factors and more.

Losing your sense of smell and taste are key signposts of Covid-19. They are both on the official list of symptoms which warrant a test for the disease.

But this crucial diagnostic link didn’t come from doctors, epidemiologists or lab researchers. It came from computers.

Artificial intelligence (AI) models crunched data from some 2.5 million people who gave daily health updates via an app in order to determine which symptoms were most commonly associated with a positive test result for Covid-19. The analysis showed that the loss of smell and taste – known as anosmia – is a more accurate indicator of the disease than a fever.

Digging further into the data, researchers at King’s College London, Massachusetts General Hospital and health science company ZOE built a mathematical model to predict whether someone was indeed suffering from Covid-19 based on their age, gender and the presence or absence of key symptoms. The model provided diagnosis with an accuracy of nearly 80 per cent, the researchers said.

The approach lays the foundations for dealing with future pandemics. With widespread use of AI-enabled health apps, doctors can more quickly identify all the symptoms of any new disease but also give out a preliminary diagnosis at very early stages. Add in an effective, automated track and trace system (perhaps via the same app), and there's a strong possibility that we will be able to better control future pandemics.

And that’s only the tip of the iceberg. The potential of AI in diagnostics goes far beyond pandemics. After looking at thousands of scans, machines have learned to identify breast cancer with accuracy comparable to that of experienced human radiologists. A broad research overview, recently published in the UK medical journal The Lancet, has reached similar conclusions about the accuracy of AI diagnostics across a range of different diseases.

Such techniques could also open up the possibility of diagnosis in places where there are few or no doctors – particularly in remote locations and in developing countries. They could also speed up cancer diagnosis, meaning that treatment can be started sooner, improving the patients’ prospects and reducing the risk of the disease spreading. That’s crucial. If lung cancer is detected when it is still localised within the lungs, the five-year survival rate in the US is 56 per cent. That drops to just 5 per cent if the diagnosis is not made until the tumours have spread to other organs.

Speedier diagnosis can be achieved not just by increasing testing capacity but by using machine learning to develop tests that can detect cancers at an earlier stage – the latter being the focus of California-based Grail Inc. It is using next-generation sequencing technology to try and detect tiny DNA and RNA fragments in the blood which are produced by cancerous cells and that are often present before more visible symptoms develop.

For diseases where diagnosis requires the analysis of several different factors, AI’s ability to process vast quantity of data can be particularly useful. Take dementia for example. The World Health Organization (WHO) estimates that there are nearly 10 million new registered cases of the disease every year, yet identifying sufferers requires a complex analysis of symptoms. This involves brain scans, blood tests, genetic history, the way people talk and walk and results of special cognitive tests. AI can combine that data for diagnosis, as well as use it for prediction of who might be at risk in the future.

While predictive AI may be in its early stages, some machine learning technology is already in use in diagnostics. ScreenPoint Medical, for example, has developed Transpara – an AI programme that analyses 2D and 3D mammograms to detect potentially cancerous lesions, and scores them based on the risk. The system is used by radiologists to speed up diagnosis and prioritise further investigation.

The world-renowned Moorfields eye hospital in London, meanwhile, has partnered with Google's DeepMind Health to develop a system that can diagnose 50 different eye diseases, while Siemens Healthineers’ AI-Rad Companion product brings machine learning insight to chest X-rays and lung CT scans to help in the fight against Covid-19 . Governments are also getting involved – the UK, for example, is investing GBP250 million into AI for the NHS.

AI, then, is well on the way to playing a key role in diagnostics – working in tandem with trained medical practitioners to make diagnosis more accurate, faster and more accessible.

About

Mega

Mega seeks to energise and enrich the debate over how to create a better-functioning economy and society.

Megatrends are the powerful socio-economic, environmental and technological forces that shape our planet. The digitisation of the economy, the rapid expansion of cities and the depletion of the Earth’s natural resources are just some of the structural trends transforming the way countries are governed, companies are run and people live their lives.

Photo of Mega