AI might be high tech but its real talent is ploughing through grunt-work very quickly. So, at its best, it can free up more time for clinicians to spend with their patients.
Before a patient has radiotherapy for their cancer, the oncologist needs to spend up to three hours marking out around 100 images from a detailed CT scan, literally drawing a line around all of the parts they want to avoid damaging with the high energy beam. With prostate cancer, they need to avoid hitting the nearby bladder for example. It’s a monotonous yet critical job, that can now be done around two and a half times faster, with the NHS’s very first AI radiotherapy planning system. All the tests show it’s just as accurate as a human and of course, it never gets tired. A human always double checks the result by the way.
OSAIRIS was eleven years in the making, and is the brainchild of Dr Raj Jena at Addenbrooke’s and Aditja Nori at Microsoft Research. It started when Raj went to a talk by Microsoft engineers about using computers to check scans. The early algorithms weren’t good enough to be used in a hospital, but things changed around 2015, when Deep Learning came along. It’s a way of teaching computers to process data more like the human brain and it’s much better at spotting complex patterns in pictures. When they started using a Deep Learning algorithm, the computer results started matching human results.
Even when they’d cracked the technology, it took two years for a team at Addenbrooke’s to work through all the health and safety approvals, making it fit for use in any Trust in the country. They literally sat down and thought up everything that might go wrong, before devising a solution to combat the risk. It could be anything from software problems and data security to how well the system might cope with patients with extreme body types.
Osairis works for prostate and neck cancers, but that is just the start. There are 60 NHS radiotherapy centres around the country that could use it straight away. Raj is also working on a model for the chest, which would work for lung and breast cancer and he wants to develop a system for brain scans too. Longer term, it could also be developed to help surgeons plan their operations more quickly.
The big bonus of course, is that time-saving for doctors will help them make a dent in our ever-increasing waiting lists. And the quicker you treat people, the better their chances of survival.
Other AI uses you won’t believe
More efficient trials (twins and more), understanding disease, predicting disease, identifying and targeting treatments.
