AI: Project InnerEye
Deep-learning algorithms are helping oncologists get patients into treatment more quickly.
From Charles Babbage to Alan Turing, Cambridge could be considered the birthplace of modern computing. Often dubbed ‘the father of computing’, Babbage was a Cambridge mathematician who is considered to have invented the first mechanical computer in 1821 – the Difference Machine. Studying in Cambridge 100 years later, Turing laid the foundation for what we now know as artificial intelligence (AI) through his ‘Turing Test’, by which a machine could be designated as intelligent.
In the 1980s, William Tunstall-Pedoe, a computer science student at Cambridge University, was inspired by the idea of using technology to easily access the vast amount of information in the university library. So, he set about creating a machine that could understand speech and have virtually unlimited information within its reach.
In 2012, Tunstall-Pedoe unveiled Evi – an AI-powered app that allowed users to ask a question using normal language, and get an answer back in real time. It was a huge success and, later that year, the Cambridge-based company was bought by Amazon and integrated into what became the Alexa platform. Cambridge mathematical and engineering excellence also underpin the algorithms that power AI globally, and Cambridge-developed technology is incorporated in Apple’s Siri and SwiftKey’s predictive text, which is used in millions of mobile phones.
In an example of applying Cambridge mathematical and computational expertise to health sciences, a team lead by CUH oncologist and university researcher Dr Raj Jena has been working with Cambridge-based Microsoft Research Labs for the past 10 years, on a machine-learning tool to speed up preparation for radiotherapy treatment.
Project InnerEye can reduce by up to 90% the amount of time an experienced doctor must spend processing scans before a patient can receive potentially life-saving radiotherapy. Starting radiotherapy promptly improves cancer survival rates and reduces anxiety in newly diagnosed patients. Before any radiotherapy can take place, however, the oncologist must spend maybe one or two hours per patient making sure the radiation will be delivered to the correct part of the body, without damaging any healthy tissue.
Using deep-learning algorithms, the InnerEye technology can carry out this preparation as well as an expert clinician, in just a few minutes, freeing up the doctor’s time and enabling them to get patients onto treatment more quickly. The clinical implementation paper was published in the Journal of the American Medical Association and the tool has been made available as an open source toolkit by Microsoft.
Cambridge University Hospitals (CUH) NHS Foundation Trust is leading the implementation of machine learning into routine clinical practice. Having demonstrated that InnerEye can be used to accurately differentiate between a cancerous tumour and healthy tissue on scans, the software is now being used to speed up cancer waiting times. InnerEye is taking on new life in two ways to ease facilitation into routine use.
First, through the work of Raj Jena’s data science team in the CRUK RadNet research programme, CUH has taken ownership of the innovation by training its own radiotherapy models with its own patient data using the Microsoft toolkit. With the help of Dr Afzal Chaudhry, its chief medical information officer, CUH is preparing the cloud-based infrastructure needed to deploy the tools, and Sonya Sireau and her colleagues in the Department of Clinical Engineering and Innovation are helping to obtain in-house approval of the AI tool as a medical device. From 2022, CUH intends that all patients with head and neck cancer, or prostate cancer, will benefit from InnerEye software carrying out the segmentation of their scans before radiotherapy, reducing the time they have to wait for treatment.
Second, Raj Jena has been awarded nearly £0.5m from the NHS AI accelerator award to pay for a detailed evaluation of the tool for device registration (both in Cambridge and at University Hospitals Birmingham), funds to pay an external agency to help with CE marking of the tool, and stakeholder and public engagement and an expanded access programme, to help with deployment of the tool in other hospitals. The project involves CUH working with several other hospitals and organisations, and with patients. It will perform extensive evaluation and risk assessment of an AI-powered tool to ensure it is safe for clinical implementation. The workflow acceleration achieved by the tool enables oncologists to focus on looking after a newly diagnosed cancer patient.
This AI accelerator work is being delivered through partners across the city of Cambridge, which brings together patients, health, university and industry expertise – as well as partners such as the University of Birmingham – with a view to system-wide adoption of an open-source, machine-learning device, built within the NHS and deployed across all 60 NHS radiotherapy centres.