Have you been diagnosed with a brain tumour? Order your free information pack.

Brain tumour AI helping others

Artificial intelligence (AI) may be the key to fast and consistent renal cell tumour diagnoses, all because of work we fund on brain tumours.

Taking a chance on great early ideas is just one of the ways that The Brain Tumour Charity supports research. Dr Phedias Diamandis, a pathologist in Toronto, Canada, was the recipient of one of our Expanding Theories awards in 2018. His proposal was to use a branch of AI, called Convolutional Neural Networks (CNN), to analyse images of brain tumours and put forward a diagnosis. It’s been just over 18 months since the work started and Dr Diamandis and his team have made great strides towards this future thinking reality with an algorithm called VGG19 CNN.

Originally the program was taught what a brain tumour looked like using one set of images, then with a different set of images they tested the ability to diagnose brain tumours correctly.

A nice thing about this kind of research is that you never quite know where it’ll take you, and in this case Dr Diamandis, and his team at the University Health Network, have proved the adaptability of the algorithm by applying it to renal cell cancer.

In this new publication, published today in the JCO Clinical Cancer Informatics we see how, computer scientist and lead author Kevin Faust, has used VGG19 CNN to distinguish between different types of renal cell cancer, with no further training (aka unsupervised). This makes it one of a rare group of algorithms that have been shown to have clinically relevant outputs using unsupervised image analysis.

The team is still expanding this work looking into brain tumours. Now that they’ve taught the algorithm some of what they know about brain tumours they can start to see what we can learn from it. The benefits of using AI in the diagnostic pathway could be game changing.

Computers can work 24 hours a day, they can retain and compare images from thousands of brain tumours and with the right programming they can see patterns that aren’t discernible to even the most trained human eye.

In real world terms, people in places without highly trained pathologists could have access to the same speed and accuracy of diagnosis as someone in a major centre. Or perhaps the algorithm will be able to accurately diagnose a very rare type of brain tumour, which a pathologist may only see once or twice in their career, and may have missed otherwise. Like I said, game changing.

This just goes to show how new discoveries can have wide reaching, and often unlooked for, effects thanks to collaborative science.