UK Scientists Speed up Brain Cancer Diagnosis with AI

A technique combining a blood test with artificial intelligence (AI), developed by the UK company ClinSpec Diagnostics, could help to prioritize which patients need to be scanned for brain cancer.

A team led by researchers at the University of Strathclyde and the University of Edinburgh, UK, trialed the technology on blood samples from 400 people suspected of having brain tumors. The researchers used an existing technique called infrared spectroscopy to screen 20,000 chemicals in their blood, and then used AI to identify the chemicals that signal a brain tumor. The test correctly identified 82% of the patients that would go on to be diagnosed with brain cancer.

Patients flagged with this brain cancer test can be prioritized for confirmatory brain scans, and their diagnosis might take just two weeks. In current practice, it’s difficult to diagnose tumors from patients’ symptoms, and the process can take up to two months, with multiple visits to the doctor. 

“A headache could be a sign of a brain tumor, but it is more likely to be something else and it’s not practical to send lots of people for a brain scan just in case it’s a tumor,” stated Paul Brennan, Senior Clinical Lecturer at the University of Edinburgh, who co-led the study. 

The blood test is being developed by Brennan’s collaborator, the UK company ClinSpec Diagnostics. While other groups are working on cancer tests using infrared spectroscopy and AI, ClinSpec’s test is the most advanced, according to Brennan.

“It’s not enough to say ‘I’ll take a load of blood samples and ‘machine-learn’ them and see if I see something interesting,’” Brennan told me. “We’ve been absolutely focused on the specific problem of taking symptomatic patients. There’s no one else publishing in this space where they’ve tried to develop a product instead of undertaking a research exercise.”

The 400 patients in the trial were due to be scanned for brain cancer, so they were already more likely to have brain tumors than people first visiting the doctor. The next step for the scientists is testing the technology in 600 patients that are earlier in the stages of diagnosis. ClinSpec aims to be ready to commercialize the technology within the next 18 months.

 “The barriers are trying to convince people that it does work because a lot of what clinicians have to deal with on a daily basis is poorly performing tests,” Brennan said.

More and more companies are turning to AI to assist with cancer diagnosis, such as Swiss company SOPHIA Genetics. According to Brennan, AI is a useful tool but needs to be applied with care.

“I’d be hesitant to say that AI is the answer to everything,” he told me. “The answer is to have a robust understanding of the problem and to design datasets with sufficient detail so that when you then apply your machine learning algorithm, it gives you a useful answer. The danger is that people are just applying an algorithm and a process without understanding what happens next.”

Image from Shutterstock