© 2018 by Periodic Publishing

  • Hugo Creeth

Br-"AI"-n Scans

Artificial Intelligence is everywhere in the media it seems these days, it is making our lives easier from “Alexa” to “Google” we are very much surrounded by the consumer AI. It isn’t just in these areas that AI is making leaps forward at an astonishing rate.


Just this last week an AI that is capable of spotting brain haemorrhages on an X-ray scan has been unveiled. It is thought that it may help better diagnose strokes, head injuries and ruptured blood vessels. The software has been shown to identify signs of bleeds on the brain with similar accuracy to trained radiologists.



It is unlikely that computers will replace doctors any time soon, this is one of the areas that software giants are making progress. The ability for software to identify and recognise moles on images of skin that are likely to be cancerous as well as damage to the backs of eyes caused by diabetes, is exponentially improving to near human capability.


The developers behind the new software are based University of California, San Francisco, utilise multiple X-rays taken of the head to create detailed images.


Currently if someone goes to hospital with dizziness or confusion, indicating potential brain damage, they will often receive a CT scan of the head. Esther Yuh the main developer points out that small bleeds are often very hard to detect on a CT scan.


Yuh’s team have trained the AI software, called PatchFCN, on close to 4400 head CT scans where a diagnosis was known. They proceeded to then test a new set of 200 randomly selected images. The result was that PatchFCN performed similarly to 4 radiologists.


The software is being considered to be used alongside doctors to speed up their work load, whilst maintaining accuracy of diagnosis. This accuracy would be maintained by images always being checked by a radiologist. This way it would be possible to validate the software in a clinical setting and show that it has a use in improving performance and output as well as patient outcomes in the real world.


Journal reference: PNAS, DOI: 10.1073/pnas.1908021116

This site was designed with the
.com
website builder. Create your website today.
Start Now