Macquarie University and Macquarie Medical Imaging has joined partners GE Healthcare and Fujitsu Australia to research the ways artificial intelligence can help diagnose and monitor brain aneurysms on scans faster and more efficiently.

The university will provide clinical expertise for the development and testing of the technology, which is provided by GE Healthcare, while Fujitsu will lead the initiative.

Fujitsu will apply AI methods to images of the brain generated by GE’s Revolution CT scanner, and use a specifically-trained algorithm to look for abnormalities and aneurysms, with the aim of creating an AI assistant to automatically flag potential aneurysms and allow for accurate follow-ups.

“Brain aneurysms can often be difficult to detect, even for expert radiologists, so by augmenting our abilities using AI, any potential aneurysms can be automatically flagged, reducing the 'miss rate' and better overcoming the shortcomings of humans variable performance over time,” John Magnussen, professor of radiology at Macquarie University, told HealthcareITNews.

Magnussen explained generating a large enough data set of reliable, well curated normal and abnormal cases is the most challenging part of training AI to look for indicators that a brain aneurysm, because as with all machine learning, “if you put garbage in, you will get garbage out”.

He said that, compared with tagging autonomous driving datasets, there are relatively few qualified people around the world who can accurately enough categorise the medical images to determine what is truly normal and abnormal.

“As has been happening in other areas of professional expertise, we are likely to see a gradual 'cherry-picking' of problems in medical imaging being solved by AI until there are systems which can reliably augment or even partially replace the experts,” Magnussen said. 

“This will put downward pressure on prices, particularly in single payer health systems, as the perceived cost of making a diagnosis is reduced, further forcing the adoption of such systems.”

He explained the next step in this area of research is to look for other well constrained and yet clinically relevant problems to begin creating new datasets prior to expanding the AI methods and generating solutions that can translate into the clinic.

“Historically, AI research has been a very academic undertaking, yet with the rapid increase in computational power that companies like Google and IBM can bring to bear, more and more problems that were considered 'too hard' are being solved,” Dr. Magnussen noted.

He said by bringing together the academic skills of the universities with expertise in bringing products to market that private enterprise can deliver, scientists can much more rapidly deliver solutions to the “customer”.

“In this case, governments are often the 'customer' as they are paying for the provision of health care,” he noted.




White papers