Can we teach a computer to recognise signs of deadly disease, just by showing a picture of your face? Dr Syed Shamsul Islam explains how artificial intelligence is improving medical diagnosis.
Tiny differences in the structure of our faces could hide clues to a host of medical conditions that even the most experienced physicians would have no chance of noticing.
These minute variations are undetectable by the naked eye but computers can be trained to recognise them using artificial intelligence techniques.
It’s the same technology Facebook uses to pick out your friends in photos uploaded to the social network or by Netflix and Amazon to suggest to you what to watch or buy next.
That’s where computer scientists like Syed Shamsul Islam, a researcher from ECU’s School of Science, come in.
Dr Islam is applying deep learning, a version of artificial intelligence-based machine learning technology, to 3D scans of faces in order to develop a screening tool for sleep apnoea.
It’s a condition affecting about 4 per cent of Australian adults that costs the economy more than $21 billion a year in lost productivity and medical expenses.
Sleep apnoea has been linked to heart disease and stroke and leads to reduced productivity at work as well as increased risk of car
However less than 20 per cent of those suffering from sleep apnoea ever get diagnosed — and that’s a big problem.
“We’re targeting those undiagnosed patients,” Dr Islam says.
“We can run a very short facial scan that can give the patient’s GP an indication of the likelihood of suffering from sleep apnoea.”
From there patients can be directed to further tests and treatment.
Dr Islam’s research is about lowering the barriers, both cost and time, of diagnosis.
“The gold-standard assessment of sleep apnoea is laboratory-based polysomnography, or sleep tests, which are expensive, timeconsuming, and uncomfortable,” he says.
“The test also requires sophisticated specialist facilities, technical and scientific staff and sleep clinicians, which are commonly not available in all regions
“Historically medical practitioners have used x-ray or CT scans to create images of patient’s heads and faces.
“Unfortunately, those are far less accurate or far more intrusive and expensive than the imaging technology we’re working with.”
X-rays and CT scans involve giving a patient a high dose of radiation and the images lack the minute details of soft-tissue variation that can be seen with Dr Islam’s scanning equipment.
“If we can give patients and medical practitioners the information they need to make better decisions about diagnosis and possible treatment that will hopefully improve people’s quality of life,” he says.
Dr Islam is currently working with colleagues from UWA’s School of Dentistry and School of Computer Science and Software Engineering, as well as Hollywood Private Hospital, to develop the screening tool.
Please leave a comment about your rating so we can better understand how we might improve the page.