Changing the conversation around AI

AI and Data, Opinion

31 January 2023

AI in lung cancer diagnosis

December 1, 2016

Ahead of his keynote session on the AI, Data and Analytics Stage at Digital Health Rewired in March 2023, Dominic Cushman, director of AI, imaging and deployment at the NHS Transformation Directorate, explains the need to focus on supporting NHS organisations to deploy AI technologies.

Artificial intelligence has been hailed as a utopia to solve many healthcare problems, but so far we have lacked much evidence to demonstrate its effectiveness or created frictionless mechanisms to allow NHS organisations to adopt AI technologies.

We’ve had the pandemic, and now we face the challenges of waiting lists, backlogs and workforce recruitment. People are asking how we can apply technology such as AI to address these issues.

One of the things we’re really good at in the UK is research and coming out with papers and models for how AI could help solve challenges in healthcare. However, we are still to crack the puzzle of how to translate this research into solutions that can be used by the NHS today.

Currently, we’re testing and evaluating 77 projects, through the AI in Health and Care Awards to see which technologies can benefit the NHS. The next step is to take this work and scale it up so hospitals can start implementing these technologies to improve patient care. For the NHS AI Lab that means moving the conversation onto prioritising our activities towards deployment, rather than research.

At the AI Lab, we’re working on how to deploy AI based on existing evidence rather than the hype. In many instances, algorithms are the end-user case of the wider digital transformation programmes within NHS England (NHSE).

There’s a lot we need to get right first like basic digitisation and getting the infrastructure in place, so these models can be integrated into clinical pathways in an appropriate and safe manner. That needs to be wrapped around the work we’re delivering on AI ethics, clinical workforce training and ongoing patient safety programmes.

The programme of work at the AI Lab has been an enabler around how we can support these technologies and provide assurances. We have a robust regulatory system, which requires several different agencies to bring a product to market.

The AI Lab has funded the development of a multi-agency advice service (MAAS) to give innovators and health and care providers a one-stop shop for support, information and guidance on the regulation and evaluation of AI technologies. The MAAS is currently in beta testing stage and we plan to launch it as a service in Summer 2023.

I believe we should focus on deployment (machine learning operations) in a couple of areas, such as radiology and emergency medicine, where AI technologies are mature, and we can think about how to integrate these algorithms into the clinical pathway to support decision-making and enable clinicians to help patients.

By starting with addressing one use case for deployment, such as radiology AI solutions, we hope to create blueprints and case studies which can be used to solve AI adoption problems from IG compliance to infrastructure to standards to scalability in other areas. That plays nicely into the work we’ve all been doing since the NHS Long Term Plan on building up imaging networks and community diagnostic hubs. We’re looking at how we can provide the appropriate level of algorithm support in those areas quickly, rather than piecemeal.

We have some good work that’s already underway. In 2022 we worked with the NHS Shared Business Services to launch the AI Software in Neuroscience for Stroke Decision Making Support Framework Agreement, allowing people to buy AI technology on the stroke framework. The use of AI has the potential to speed up the diagnostic process, which is crucial in stroke where significant harm and long-term disability can be caused if left undiagnosed and untreated.

One of the technologies we funded through the AI Awards is Oxford University spinoff Brainomix. The startup provides a set of tools that uses AI methods to interpret acute stroke brain scans and helps doctors make decisions about treatment and the need for specialist transfer of patients.

Brainomix also provides a platform for doctors to share information between hospitals in real-time. This is now being deployed in nine stroke networks, which is around 50% of stroke services in England.

Another example is the deployment of startup Adience’s AI platform which is being used for the NHSE Targeted Lung Health Check programme. It provides the Veye chest system that supports early lung cancer detection by acting as a ‘second reader’ to help relieve pressure on radiologists.

We’re also supporting work in diabetic ophthalmology to make sure we don’t have any failure points around ethical decisions. The AI Lab is funding a programme of work to ensure that AI technologies that detect diabetic retinopathy work for all, by validating the performance of AI retinal image analysis systems that will be used in the NHS Diabetic Eye Screening programme in different subgroups of the population.

AI technologies are one piece of the digital transformation picture, but sometimes they can be at odds with what clinicians need. With many technologies on the market, we need to make sure we are working with the Royal Colleges and clinical staff to understand where AI can help and provide them with the support and assurances they need to use it with confidence. We just published a second report with Health Education England (HEE) identifying workforce training needs to provide confidence in the use of AI.

Ultimately the aim of using AI is to improve patient care and outcomes whilst freeing up clinicians’ time and making their jobs easier so they can spend more time with patients. These technologies have the potential to lead to faster diagnostics and more targeted interventions at a population health level. AI can improve the health of nation long-term, but we need to deploy it safely and with the right evidence.