AIMed and NHSx, the digital unit of the UK National Health Service (NHS) co-hosted the very first NHS AI Lab virtual conference last Thursday (24 September). You may now watch the event on demand here. However, if you are unable to devote six hours to digest the entire event, AIMed has kindly prepared some quick summaries for you. We hope that even if you don’t work for the NHS, you will still find some of these points relevant.

  1. NHS artificial intelligence (AI) effort is an agenda on the move

Last summer, NHS received a £250 million boost to set up a new National AI Laboratory. The funding is believed to be part of an ongoing plan to make NHS a world leader in AI. Since then, the 70-year-old public health system had never stopped preparing itself for the great leap.

In the words of Dr. Indra Joshi, Director of AI at the NHSx, it is not just about technology, but also making sure getting the rules right; understanding the deployment process, and what to do with the data gathered to solve problems found both internally within the organization as well as on the frontline of delivering medical and healthcare services.

This is probably why many up and coming effort were announced in this conference; some of them include:

– AI Virtual Hub: A virtual space where interested parties can get involved via various discussion forums being set up to share their knowledge and support the mission of “getting it right” to accelerate the deployment of safe, ethical and effective AI in healthcare.

– AI Skunkworks: Helping the health and care community to take their ideas from scratch to the “MVP level”.

– A multi-agency advice service provided by National Institute for Health and Care Excellence (NICE), Care Quality Commission (CQC), Health Research Authority (HRA), and Medicines and Healthcare products Regulatory Agency (NHRA) will begin soon to offer support, information and advice on regulations and health technology assessment for AI technologies in health and care system.

– A Centre of Data Expertise will be launching next month.

– A new data strategy for health and social care to be released in autumn. It will contain key principles around empowering healthcare staff and system leaders to be confident in handling data through simplified information governance. It will also facilitate the understanding of levels of maturity in different areas so that data can be used to plan and commission health and care services and high-quality research more effectively.

– As part of the Global Digital Partnership, a consortium constituting the World Health Organization (WHO) and government agencies aimed to outline standards and policy framework on interoperability, cybersecurity and so on. NHS will be drafting a White paper on AI policy.

– Matthew Gould, Chief Executive Officer of NHSx said he and his team’s priority for the rest of the financial year is to set up a project to join up care services to make health and case more close-knitted. Specifically, the whole country will have to share care records, put in place an increasing level of digital maturity in social care sector and deploy remote monitoring technologies that can be used out of formal care setting and be extended to individuals’ homes.

  1. The National COVID-19 Chest Imaging Database (NCCID) welcomes more data and collaborations

NCCID is a joint initiative between the NHSx, The British Society of Thoracic Imaging (BSTI), Royal Surrey County NHS Foundation Trust and AI Lab Strategic Partner Faculty. At the dawn of the pandemic, researchers and companies in the UK were working on fragmented chest imaging data to try to understand the novel coronavirus and its impact on the human body. They decided to pivot some of their activities but struggled to find high-quality, representation and accessible imaging data. This became the motivation to start a centralized chest imaging database.

Over the last few months, more than 32,000 x-rays, CT and MRI images from hospitals and patients were collected from 18 NHS Trust (i.e., these Trusts represent 84 individual organizations) and over eight approved data users were allowed to have access to these resources for training and research purposes. In the next few months, NCCID hopes expand their current work and provide more support to the imaging community. Thus, they urged fellow data holders and innovators to come forward and be part of the effort to take any working concept to a probable deployment given the accessibility of a large dataset.

  1. The reason why some UK innovations are leaving the country

NHS announced the 42 winners for the first round of AI in Health and Care Awards on 8 September. Some of them spoke at the conference to share tips and advice with future applicants on how to champion the next round of the Award. Peter Kecskemethy, Chief Executive Officer of Kheiron Medical Technologies suggested allocating a good chunk of time for application and to request practical feedbacks from those who know the solution or project whether the idea has come across clearly.

He added sometimes some fantastic technologies failed to be adopted for the most trivial reasons. For example, he remembered a hospital forgo the adoption of a new system because they could not recall the password and did not know how to reset it. As such, it’s important to understand the human factors behind AI, including who will be the users, how to fit into the workflow, what else are these users doing at the time of using the technology and so on. Bear in mind that doctors will always want to try the technology before using it clinically and the developed product needs to engage with patients.

There are many routes to bring a technology to hospitals and it takes an enormous amount of time and effort to find the most suitable approach. Kecskemethy believes the tech industry carries the idea of bringing their solutions to as many people as possible and administrative burdens tend to be the factors that are stopping them. He thought, traditionally, the amount of paperwork which a company requires to complete in order to take their technologies to the market is probably why some UK innovations chose to leave the country for foreign markets, where they can have easy access to larger patient population and markets. The NHS ought to understand this and streamline processes to avoid losing their hands on some wonderful innovations.

  1. Be clear of what you will like to achieve with AI

Haris Shuaib, AI Transformation Lead and Topol Digital Health Fellow at Guy’s and St. Thomas’ NHS Foundation Trust said AI requires significant clinical buy-in and it’s never an individual effort but an effort from the whole department, hence managing different priorities become crucial. Aligning these priorities and frank discussions on what would be most technically feasible will lead to some early successes.

Review, not only on the outcomes, but also the economic values of the product. If an AI model needs to be reviewed by the team every week, how much time is it actually saving and how much time are the human experts willing to throw in to handhold the model are questions that needed to be considered too.

Nick Barlow, Director of Applied Digital Healthcare at the University Hospitals Birmingham NHS Foundation Trust added the need to have absolute clarity on what one will like to achieve with AI. He said there is no shame to acknowledge that there are plenty of other solutions which do not necessarily require AI. So, avoid the temptation of using AI to solve small problems that only concern part of the organization or part of the system.

As Dr. Rizwan Malik, Divisional Medical Director Diagnostics at Bolton NHS Foundation Trust put it, the starting point is wrong if your first question is “How can I use AI?”. The correct question should be “What are the problems that I want to solve?” and proceed on to search for tools that the organization can rely on.

  1. You won’t know your AI until you deploy it

Shuaib recalled three years ago, he and his team were accessing an algorithm built to look at pediatric burn development through hand x-ray. They thought the solution was a quick win with low clinical risk until they realized a divergent in the results provided by the AI and human consultants. In the end, it turned out that the AI was correct. On one hand, Shuaib and team thought it was great, the AI was working well. On the other hand, they were also caught off-guard by the conflict. They did not have a strategy to deal with the situation when AI and human arrived at different answers.

On another occasion, a company with good publication record and technical results was chosen to provide an AI-driven brain imaging solution to be used at the Accident and Emergency (A&E) department to support overnight registrar who is covering for three hospitals. The algorithm was tested against three human consultants on a dataset of 400 head CT scans. The algorithm did fine but failed to process 25% of the scans because in those images, the A&E patients were not aligned in the CT scanner. Thus, Shuaib said, one will never have full knowledge of their AI models until they run the data through the technology.

  1. AI is part of the solution, not THE solution

Jacques Du Preez, Director of Digital Health and AI at Psephos Biomedica believes it’s acceptable to have some failures and learn through the process before finding a suitable AI solution. Always seek clarifications from the manufacturers and make sure they meet certain requirement needed for a particular organization. For example, in certain cases, gaining a CE trademark does not guarantee a product’s penetration into the NHS. The process does not end with successful deployment of AI because AI is part of the solution, not THE solution, there is a need to walk the journey, to maintain, evaluate, and support it and ensure it keeps up with other ongoing development.


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.