How do we deploy AI tools safely at scale in the NHS: The role of an AI deployment platform

With a shortfall of around 2,000 clinical radiologists in the NHS, demand for radiology services far exceeds available capacity, driving a growing backlog of unreported scans. This unprecedented pressure on the NHS has coincided with a surge of interest in AI technologies, and we’re seeing many NHS Trusts turning to AI-based tools to help alleviate pressures on radiology workflows. But how can we ensure they maximise potential, whilst protecting patient safety? In this blog, we explore the key barriers of implementing AI tools and outline the characteristics required of an AI deployment platform for the NHS.

The roll out of AI diagnostic tools could result in faster diagnosis, such as the early detection of cancer, ultimately improving the odds for timely and effective treatment. As this interest in AI technologies picks up in the healthcare sector, there is increasing focus on deployment platforms to support their adoption, such as the AIDE (AI Deployment Engine) and the AIDF (AI Diagnostic Fund).

In collaboration with NHS England AI Lab, we piloted an NHS AI Deployment Platform (NHS AIDP) that utilised a vendor-neutral, cloud-based platform with built-in AI safety mechanisms. Once an NHS Trust is connected to the centralised platform, it’s easier and faster to access multiple AI products, helping to address key challenges around scalability and safety of AI tools adopted in radiology workflows.

Despite the growing interest, teams on the ground implementing AI tools up and down the country are all facing several key barriers, resulting in slow, fragmented and potentially unsafe deployments at a Trust by Trust level. 

• AI technologies adopted into clinical workflows require robust evaluation and performance monitoring post implementation, to ensure the tools used are safe and performing consistently. However, there is limited detailed guidance of best practice resulting in variable approaches taken by NHS Trusts, who often have limited resources to run large scale validation exercises with ‘ground truth’ data. 

• Digital infrastructure is fragmented across the NHS, making scalable integration of AI tools a challenge. These challenges are compounded by a lack of NHS staff capacity, varying levels of expertise in AI at NHS Trusts and significant variation in key processes to support implementation, such as information governance and change management.

For AI tools to effectively transform healthcare, an underlying AI deployment platform must be more than just a technological solution—it must facilitate safe and scalable AI integration. When considering what makes a good AI deployment platform for the NHS, the following characteristics are essential:

1. Scalable: The platform must be capable of scaling across disjointed NHS Trusts, accounting for both fragmented digital infrastructure and varying technical requirements. 

2. Adaptable: The platform must support a variety of AI tools across different imaging modalities (e.g. CT, X-Ray, MRI) and be capable of evolving as clinical needs and priorities change over time. Flexibility is key to ensuring the platform can integrate new tools and meet emerging challenges without disrupting current established workflows.  

3. Interoperable: A good platform is one that accounts for the fragmentation of the existing digital infrastructure ecosystem within the NHS. The platform must be able to integrate with a variety of healthcare systems (such as Electronic Health Records, Picture Archiving and Communication Systems).

4. AI safety mechanisms: A key feature of a gold standard deployment platform is built-in robust safety mechanisms, to ensure the ongoing performance and safety of AI tools post-implementation. This includes supporting independent continuous monitoring of AI model performance and facilitating the collection of ‘ground truth’ data to assess whether AI tools are performing as expected. These safety measures are vital for maintaining trust and ensuring that AI applications are delivering clinical value without introducing risks.

5. Risk mitigation for AI deployments: A robust platform can help mitigate deployment risks by thoroughly vetting AI vendors before they are integrated into a platform offering, ensuring they meet essential regulatory and safety standards while also meeting clinical needs of the Trust and delivering cost-effective value add. This adds an extra layer of assurance for NHS Trusts, enhancing confidence in the deployment of AI tools and supporting safer, more reliable integration into clinical workflows.

The future of AI in NHS medical imaging depends on creating platforms that are scalable, safe, and adaptable enough to navigate the complexities of the healthcare system. By providing a unified, platform-based approach, the NHS can ensure that AI tools are deployed in a way that maximises their potential to address clinical challenges — while maintaining patient safety and improving outcomes.

To find out more about the role of AI in health and care, visit our industry page or get in touch with our team.