Subscribe now

Health

AI matches humans at diagnosing brain cancer from tumour biopsy images

By Gege Li

8 January 2020

MRI shows brain tumor at right parietal lobe of cerebrum

Artificial intelligence could be used to diagnose brain cancer in the future

Puwadol Jaturawutthichai / Alamy Stock Photo

An artificial intelligence can now diagnose some cancers from brain tumour biopsy images with the same level of accuracy as humans.

The AI analyses high-resolution images of tumours produced using a method called stimulated Raman histology (SRH).

Todd Hollon at the University of Michigan and his colleagues generated more than 2 million SRH images of brain tumours from 415 people with known diagnoses. Each image showed a small region of an excised tumour and was labelled with which type of brain tumour it was out of the 10 most common types. The team fed them all to the AI so it could learn from the images to identify tissue features linked to these specific types of cancer.

The images had either come from biopsies that remove a small sample of a suspected tumour for analysis or from surgeries to remove tumours. Particularly aggressive tumours can be removed entirely, but this rarely works for brain cancer because the tumours are often integrated into the brain itself.

Hollon’s team then put the AI to the test in a clinical trial in which images of tumours from 278 patients with neurological symptoms were randomly assigned to either the AI or to human pathologists to diagnose.

The AI’s diagnoses – which take about 15 seconds – were accurate 94.6 per cent of the time, compared with the human accuracy of 93.9 per cent. Accuracy was checked by comparing the visual diagnoses with one involving more lengthy tests in the lab.

This AI could eventually mean humans could be taken out of the loop when analysing images of tumours for cancer, says Bilal Mateen at King’s College Hospital. “I’m very optimistic [it] could make a huge difference.”

However promising it looks, “until you’ve done a study across multiple centres in real time you can’t really know the extent of the performance”, says Paul Brennan at Cancer Research UK.

Topics:

Sign up to our weekly newsletter

Receive a weekly dose of discovery in your inbox! We'll also keep you up to date with New Scientist events and special offers.

Sign up