Addenbrooke’s and Microsoft Research Cambridge create world first AI technology to speed up radiography preparations
Addenbrooke’s is becoming the first hospital in the world to use artificial intelligence technology from Microsoft to cut waiting times for cancer patients and potentially improve survival rates.
The hospital has been collaborating with Microsoft Research Cambridge for eight years on Project InnerEye, which helps to automate radiotherapy preparations, saving clinicians hours of laborious work.
Addenbrooke’s oncologist Dr Raj Jena, who is also a University of Cambridge researcher and InnerEye co-lead, said: “I look after patients with cancers of the brain and spine. We do this by using precise radiotherapy treatment to kill off the tumours and save lives.
“Planning radiotherapy treatment takes many hours, carefully marking up scans in order to direct the radiation to the tumour whilst shielding healthy structures around the tumour itself.
“My team at Addenbrooke’s Hospital and the University of Cambridge have been collaborating with Microsoft Research to train computers to perform this task more quickly.
“The results are incredibly exciting. The research has shown that a computer can perform in just a few minutes what would normally take me several hours to do.
“This is important because it means we can start a patient on life-saving radiotherapy much more quickly and we know that the sooner we can start treatment the better the chances for the patients.”
Up to half of the people in the UK will be diagnosed with cancer at some point in their lives, and about 40 of those successfully treated undergo radiotherapy, in which high-intensity radiation beams are targeted on hard, cancerous tumours to damage their DNA, while avoiding surrounding healthy organs.
The work begins with a 3D computed tomography (CT) imaging scan. This creates stacks of 2D images, dozens deep, and each must be examined and marked up by a radiation oncologist, clinical oncologist or specialist technician in a process called contouring.
In each image, a contour line must be drawn manually around the tumours and key healthy organs in the target area using dedicated computer software.
This can take hours for a single patient in more complex cases - but research published by the team in JAMA Network Open shows that InnerEye’s machine learning models can accurately carry out the this ‘image segmentation’ process 13 times faster, saving 90 per cent of the work for clinicians.
“Using AI to help doctors with complex but routine tasks is an absolute game changer,” said Dr Jena. “It gives back clinicians the gift of time and it means that we can focus on the parts that only humans can do, which is actually caring for our patients.
“The deep learning software tools from Project InnerEye have now been released as open source code and we look forward to using these tools routinely to speed up the time taken to start radiotherapy treatment here at Addenbrooke’s.”
The machine learning models developed using the tool need to be tested and validated in each individual healthcare setting, following the lead of Cambridge University Hospitals (CUH), which runs Addenbrooke’s.
Aditya Nori, head of health intelligence at Microsoft Research, said: “This is the first time, we believe, that an NHS trust has implemented its own deep learning solution trained on their own data, so it can be used on their patients.
“It paves the way for more NHS trusts to take advantage of open-source AI tools to help reduce cancer treatment times.”
Microsoft points out that one of the barriers to the uptake of machine learning in clinical use across different hospitals is that most models are only trained on a dataset from a single institution and focus on a single task.
The Cambridge researchers overcame this by developing generic models, trained on anonymised data from eight clinical centres across Europe, Australia, New Zealand, North and South America. A dataset of 519 pelvic 3D CT planning scans and 242 head and neck scans were used.
Dr Yvonne Rimmer, oncologist at Addenbrooke’s, added: “There is no doubt that InnerEye is saving me time. It’s very good at understanding where tumours and healthy organs are. It’s speeding up the process so I can concentrate on looking at a patient’s diagnostic images and tailoring treatment to them.
“But it’s important for patients to know that the AI is helping me do my job; it’s not replacing me in the process. I double check everything the AI does and can change it if I need to. The key thing is that most of the time, I don’t need to change anything.”
Clinical use of the machine learning models is subject to regulatory approval.
Javier Alvarez-Valle, principal research manager at Microsoft Research Cambridge, said: “The AI works in the background, so clinical oncologists just open up the scans on their computer and they can see what their AI model has highlighted. The clinical oncologist then decides what to do with that information.
“AI models trained with InnerEye will be hosted in Microsoft’s Azure cloud, so all the data is securely held in the UK and only available to the medical staff who need to use it.”
Microsoft said the image segmentation model is a “convolutional neural network based on a 3D U-Net architecture, with approximately 39 million trainable parameters”.
The work has won the approval of Health and Social Care Secretary Matt Hancock, who said: “New innovations like this can make all the difference to patients and I am proud to see we are once again leading the way in new cancer treatments.
“Helping people receive treatment faster is incredibly important and will not only improve recovery rates but will save clinicians precious time so they can focus on caring for patients.
“Embracing new technologies will help save lives and is vital for the sustainability of the NHS, and our NHS Long Term Plan will continue to deliver the best possible care for patients so that we can offer faster, more personalised and effective cancer treatment for all.”
Sign up for our weekly newsletter for the latest Cambridge science direct to your inbox every Friday