The Predictive Power of AI in Healthcare

DZone 's Guide to

The Predictive Power of AI in Healthcare

It sounds crazy — but algorithms have been trained to accurately predict the lifespan of a human simply by looking at images of their organs.

· AI Zone ·
Free Resource

I’ve written a few times recently about the growing ability of AI to make accurate health-related predictions based upon medical data. For instance, a team from the University of Nottingham recently used machine learning to better predict cardiovascular risk.

They are far from alone, however. For instance, a recent study by the University of Adelaide was able to predict the lifespan of a human simply by looking at images of their organs.

The researchers trained their algorithms on images of 48 patients’ chests, with the predictions comparable to those of clinicians.

“Predicting the future of a patient is useful because it may enable doctors to tailor treatments to the individual,” the authors say. “The accurate assessment of biological age and the prediction of a patient’s longevity has so far been limited by doctors’ inability to look inside the body and measure the health of each organ.”

The team hopes to apply this technique to a range of other medical conditions, including the onset of heart attacks, with their next step being to increase the number of images used to train the system.

Predicting Diabetes

It’s a challenge also being undertaken by researchers from Boston University who have published a paper exploring how machine learning can effectively predict heart disease and diabetes. They found that they could predict hospitalization due to each disease roughly a year in advance with an accuracy of 82%, thus giving care providers not only a chance to plan appropriate resources but also to try and intervene and head off the hospitalization.

The team has also worked with surgical teams in Boston to predict readmission rates within 30 days of an operation, and they’re confident that they can help to direct the appropriate care to prevent this from happening.

The system was trained not on medical images but the anonymized electronic health records of patients. The results surpassed those of existing risk scoring systems such as the Framingham Heart Study by around 25%, suggesting significant improvements can be made when this is rolled out in the field. The results showed that the data contained in our patient records is a much better indicator of readmission than the factors used in the Framingham model.

The authors estimate that over four million readmissions could be prevented each year in the U.S. alone, costing over $30 billion, suggesting a huge potential for improvement by using algorithms such as that developed by the Boston team.

Bringing to Market

One company aiming to take this approach out of the lab and onto the market is Dutch startup Aidence. The company, who use machine learning to detect lung cancer from medical scans, recently raised $2.5 million in funding to help bring their technology to market in 2017.

The company says it is currently working to adapt its software to identify other pathologies on chest CT as well as MR imaging of joints and the brain. It also plans applications for triage and workflow optimization, screening programs and routine clinical practice.

“Our ultimate goal is to reach a level of diagnostic accuracy that matches the collective knowledge of all human experts — and then bring it to every hospital in the world,” they say.

Suffice to say, the progress made by all of these projects is undoubtedly fascinating, but we are at the very start of this journey. If we’re capable of making predictions with >80% accuracy from narrow data such as imaging and medical records alone, just imagine what we can do when personal data from mobile apps, wearable devices, and genome sequencing is taking into account.

We could move from hospital based work towards nudge based systems that help us to lead a healthier lifestyle, thus keeping us well rather than fixing us when we’re sick. It’s a shift towards prevention that all health systems around the world need to make, and the secure liberation of data will be crucial to ensuring it happens.

As I argued recently, whilst there is a growing acceptance of this fact, we need to move beyond talking and towards action. As Victor Hugo famously said, there is nothing as powerful as an idea whose time has come, and the time is now for this to become less about talking and more about doing.

ai ,algorithms ,healthcare ,machine learning ,predictive analytics

Published at DZone with permission of Adi Gaskell , DZone MVB. See the original article here.

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}