UK can lead in Radiology AI. Here’s how…

Hugh Harvey
Sep 17, 2017 · 6 min read

The UK has a rich history of leading the way in radiology — after all, Brits discovered the processes behind CT and MR scanners. However, it’s been a while since we discovered anything quite so ground-breaking…. I say it’s time we lead the way in radiology artificial intelligence.

Artificial intelligence software as applied to radiology and medical imaging has the potential to vastly improve quality of care and timely diagnosis for patients, and reduce the burden on a stretched radiology workforce. Automation of repetitive measurement tasks, clinically validated quantitation of biomarkers, autonomous triaging of emergent findings… the list goes on.

However, for researchers and companies to develop software to recognise radiological pathology from all parts of the body, they need to train algorithms using huge amounts of medical imaging data. These data already exist on the PACS (Picture Archive and Communication System) archives in every NHS hospital in the UK, but they are essentially stored redundantly with no benefit to the scientific community. It’s like sitting on an oil field, but not allowing anyone to drill into it!

Access to NHS data is difficult, and mired in controversy, as highlighted by the Google DeepMind deal. Quite rightly, patients are concerned about their confidentiality. However, it is unavoidable that artificial intelligence requires data on which to train. In my experience, radiology artificial intelligence imaging start-ups struggle most at data access. In a data-starved field, how are they to progress and ultimately produce the software that will benefit patients the most? Some manage to pen deals with local hospitals or academic institutes, others get access to small training sets from other researchers or competitions such as Kaggle. But it’s not enough. The time has come for the NHS to embrace the coming technological wave, and be part of creating the future.

America is ahead… for now

The dominant industry vendors in radiology have already placed their stakes and formed strategic partnerships with academic institutions across the United States. From Massachusetts General to Harvard, big thinking and big ideas have taken hold, and capital investment plowed into medical imaging research. GE, IBM Watson, Philips and many others have created deals to allow data access in return for funding research. The American College of Radiologists (ACR) has announced it’s own Data Science Institute. Even in Europe, where Siemens dominates, we can see EU/industry partnerships forming. Below is just a small example of how some of these partnerships look. Sadly, the UK is woefully under-represented.

One of the only UK academic institutions to have a big-name partnership is KCL.

We need a BRAIN

I have previously proposed a British Radiology Artificial Intelligence Network — BRAIN. The idea is simple… create access to anonymised NHS medical imaging data through a central data science institute.

Essentially the idea behind BRAIN is to create an approved access route to anonymised data, by creating the infrastructure for a national NHS imaging data science hub. This would sit across existing Academic Health Science Networks (AHSNs). BRAIN would have access to and govern curation of anonymised NHS medical imaging data from across the country, and act as an incubator for Artificial Intelligence and Machine Learning (ML) small to medium companies (SMEs) wishing to train and validate image perception algorithms. Researchers and companies wishing to train and validate would apply through BRAIN for fast-track access and a world-class environment in which to grow and scale fruitful research. Annotated images and structured data would be stored and shared, creating the world’s largest curated imaging dataset.

There is already precedent in the UK for this type of infrastructure. The Clinical Practice Research Datalink (CPRD) allows researchers to apply for access to anonymised primary care records, and the UK BioBank project holds even larger clinical datasets. Clearly, a similar project for imaging data is sorely needed.

BRAIN would not only centralise access to NHS imaging data, it would also be able to set clinical standards for radiology AI, and work alongside the Royal College of Radiologists (RCR) and Medicines and Healthcare products Regulatory Agency (MHRA) on validation and regulatory approval processes for marketisation of algorithms into clinical workflows. Instead of each individual researcher or start-up having to negotiate access, apply for funding for clinical trials, and pay through the nose for regulatory compliance processes, it could all be done through BRAIN. Let the researchers concentrate on the algorithms, and let BRAIN nurture and guide through the rest. The regulatory landscape in particular can be a nightmare for fledgling companies. Why trudge around hospitals begging for access and drawing up time-limited contracts? Why not bake-in regulatory systems right from the start, and work alongside the governing bodies? Why run a trial in one hospital site, when through BRAIN, you can run a trial on national imaging data?

The business model behind BRAIN is simple. In essence it would be funded by government/industry partnership providing capital investment, relying on income either by pay-for-access or share of equity/IP for any products realised, supported by research grant funding. In this manner NHS imaging data would be governed under the remit of the NHS, and profits from successful research fed back into the national health and research system.

The additional benefit to the UK is that of creating a world leading and entirely new industry within life sciences, enabled by leveraging of NHS data (centrally controlled and secure) for the ultimate benefit of NHS patients. Software created could then be exported, creating global revenue streams, and cementing the UK as the world leader in this field.

Building a national infrastructure

Recently Professor Sir John Bell authored a report on a new Life Sciences strategy for the UK.

It outlines strategic goals, including collaboration between the NHS and industry for the benefit of UK patients and the creation of 2–3 entirely new industries over the next few years. I believe one of those industries should be the development of AI as applied to medical imaging.

To quote the strategy report:

“This opportunity requires a partnership with the NHS to provide a steady flow of well-characterised samples in combination with good longitudinal data, as these two characteristics will inevitably allow the creation of the most competitive algorithms both in the immediate future and over time. No other system has the scale to provide such important data that, when captured, could produce a globally dominant commercial offering in this diagnostic space.”

There is a huge alignment between the ideas behind BRAIN and the Life Sciences report recommendations, namely:

  • Building a national ‘front door’ to patient data.
  • Investing in infrastructure to support SMEs and research.
  • NHS and industry collaborations.
  • Supporting clinical trials for algorithm validation.

The report suggests the government set up HARP — Health Advanced Research Program — to undertake large research infrastructure projects and ‘moonshots’. This to me sounds like exactly the kind of funding source BRAIN would fit under.

I highly recommend that anyone reading this article has a good look at the report. While it is only an advisory report, it is clear that the only way the UK can excel as a global healthcare leader is by large-scale thinking of this nature.

Final thoughts

I have recently been co-opted to sit on the national Radiology Informatics Committee at the RCR, who have recently drawn up recommendations to the House of Lords Artificial Intelligence review committee. These stated that the RCR is developing a specification stipulating how NHS imaging data can be anonymised at source before passing through the firewalls of hospitals, as well as the need for safeguarding patient privacy by releasing data in a completely anonymised form to a central, independently regulated national database. This is a crucial first step towards opening up NHS imaging data in way that is controlled, anonymised and secure, and absolutely vital if the UK is to lead the way in radiology artificial intelligence.

If you are as excited as I am about the future of artificial intelligence in radiology, and want to discuss these ideas, please do get in touch. I’m on Twitter @drhughharvey

If you enjoyed this article, it would really help if you hit recommend and shared it.

About the author:

Dr Harvey is a board certified radiologist and clinical academic, trained in the NHS and Europe’s leading cancer research institute, the ICR, where he was twice awarded Science Writer of the Year. He has worked at Babylon Health, heading up the regulatory affairs team, gaining world-first CE marking for an AI-supported triage service, and is now a consultant radiologist, Royal College of Radiologists informatics committee member, and advisor to AI start-up companies, including Kheiron Medical.

Hugh Harvey

Written by

Doctor² (radiologist & academic) MBBSs BSc(Hons) FRCR MD(Res). Clinical AI, machine learning in radiology imaging and research.