Meet Dawn, one of the most powerful AI supercomputer in the UK. With more than a thousand top-end Intel graphics processing units (GPUs) operating inside its server stacks, Dawn enables scientists within the University of Cambridge and across the UK to make advances in critical research fields such as clean energy, personalised medicine and climate.
Dawn has been created via a highly innovative long-term co-design partnership between the University of Cambridge, UK Research & Innovation, the UK Atomic Energy Authority and global tech leaders Intel and Dell Technologies. This partnership brings highly valuable technology first-mover status and inward investment into the UK technology sector.
AI could be ‘magic bullet’ that helps save lives of kidney cancer patients
University of Cambridge researchers, supported by the power of the Dawn supercomputer, believe AI is the ‘magic bullet’ needed to save countless lives through an affordable kidney cancer screening programme.
Because kidney cancer – which kills around 5,000 people in the UK every year – is often asymptomatic, most people are not aware they have it until the disease is at an advanced stage, which can limit treatment options.
Computerised Tomography (CT) scans are the most reliable and least invasive way to look for these cancers. However, analysing these images is labour-intensive and therefore expensive. It means screening for those at higher risk of developing the disease is not currently performed within the NHS.
But new research, led by Cambridge University PhD student Bill McGough, suggests that using AI to sift through patients’ scans and identify the presence of cancer could offer a much more affordable approach and make screening possible.
Initial findings from Bill’s research – comparing the performance of AI against that of radiologists looking at the same data, and simulating an AI kidney screening setting – show that a deep learning diagnostic tool reaches the same level of accuracy reading CT scans as radiologists, after being trained using doctors’ medical expertise.
And, as well as reducing costs, speeding up diagnosis, and freeing up radiologists’ time, the AI tool offers safety benefits for patients – by still being able to accurately read scans which are taken using less radiation than would usually be used for diagnostic purposes, and reducing the amount of dye, or ‘contrast medium’, a patient needs to be given to show detail in the CT images.
Bill, whose research is funded by the Cancer Research UK Cambridge Centre and is based in the Early Cancer Institute, said: “Although screening makes sense in the vast majority of diseases, it's ridiculously expensive – hundreds of millions of pounds – and so it isn’t practically possible. So the question we’re asking is: Can we increase the likelihood of screening happening using AI? Can AI extend the capability of the NHS to so something it wouldn’t otherwise have the budget to do? And we think it can.”
Bill’s research also suggests that as well as helping to diagnose kidney cancer, AI could also play a further role in triaging the care of patients. So, based on the risk it identifies in a scan, the AI tool could, for example, recommend a patient see a doctor straight away, or skip the next round of screening – again saving the NHS money and radiologists time.
Bill’s research – and its promising findings so far – would not be possible without the development of artificial intelligence supercomputers, and the support he receives from the Dawn team at Cambridge.
He said: “These computers enable us to perform calculations that we just couldn’t perform on our laptops. We have the data and the code, but it’s the supercomputers that give us the sheer power to be able to train these AI models on enormous amounts of data and develop them, which is a huge task.”
As part of the research, the AI model is fed hundreds of multi-phase CT scans showing a ‘tracer’ substance injected into the arm and making its way around a human body. The tracer ‘lights up’ different parts of the body in sequence and helps the tool to learn how human biology works, including how cancer behaves when there is a tracer going through it.
“It’s about the AI tool learning to recognise biological behaviour so it can predict what is happening in the kidneys,” said Bill. “It’s not learning how to identify cancer specifically; it’s drawing sensible correlations based on the CT scan – it’s like it’s learning the ‘grammar’ of these images.”
Bill hopes that his research will show that cost-effective kidney cancer screening is feasible for around 2 million people a year.
“We think we can unlock screening for millions and ultimately save lives using AI,” he said. “Early detection is one of the most powerful ways in which we can improve cancer survival, and we’re showing that AI could help to detect small cancers, catch the disease early, and improve survival rates.”
