Predict Heart Attacks with Neural Networks

In the past twelve months machine learning has moved ahead more than it has in the previous six decades with neural networks able to recognise physical objects within photos, this doesn’t sound impressive until you look ahead to how this technology might be used in future.  Facial recognition systems advanced enough that thy can replace locks and key codes to machines able to scan x-rays for abnormalities removing human error.

To this we can now add a new project from Alphabet (Googles parent company) to predict a person’s age and blood pressure from a single non-invasive retina scan which may even be capable of acting as an early warning system for an impending heart attack.

These convolutional networks are able to interpret and analyse an entire image inspired by the way a human brain interprets and understand the content of an image without the need to break the image up into smaller segments. And these intelligent networks are getting better every day.

Prior to the retina scanning concept the Alphabet labs created a DNA sequence scanning tool called DeepVariant to search for small mutations that would be missed entirely by modern methods. The retina scanning system needed a baseline dataset that required scanning in the retina images of nearly three hundred thousand individuals before being turned on to thirteen thousand patients to validate the process. Using just the retinal images Alphabets systems were able to determine age to within three years, gender in almost 100% of cases, smoking status nearly three quarters of the time and blood pressure to better than 90%. The likelihood that a patient was about to suffer a cardiac event was accurate to 70% which for a system that is based entirely off scanning the blood vessels within the eye is an amazing feat.

The concept has yet to be peer reviewed but even though it’s accuracy rate isn’t perfect (yet) with its low degree of invasiveness in the future the technology could be deployed on a wide scale throughout train stations, airports, shopping centres and office buildings to act as an early warning system.

Trending Now

Error: No articles to display

The Mordialloc Chronicle Logo

Part of The Australian Chronicle media network.

The Mordialloc Chronicle is a free community newspaper distributed to 48,000 homes in Mordialloc, Aspendale, Aspendale Gardens, Edithvale, Chelsea, Chelsea Heights, Bonbeach, Patterson Lakes Parkdale, Mentone, Moorabbin Airport and parts of Cheltenham.

© 2017 The Mordialloc Chronicle     23 618 351 018       Designed by NERV

Please publish modules in offcanvas position.