Building open infrastructure to accelerate AI deployment in healthcare

The NHS faces a 7.6% vacancy rate, including a 29% shortfall of radiologists, and a backlog of over 7.5 million appointments. AI has the potential to help address these pressures, but the time taken to move from clinical research to live clinical implementation is slow.

Two competing approaches were emerging in the market. PACS and RIS suppliers were developing costly proprietary AI add-ons that tied trusts into single vendor strategies and focused primarily on radiology. At the other end, innovative AI model creators were building bespoke point-to-point integrations with proprietary infrastructure. While this provided flexibility, it wasn't scalable.

Both approaches locked NHS trusts into single vendors and misused the limited investment available to adopt AI at scale. Legacy technology and strict NHS data sharing rules made it difficult to access the large datasets needed to train algorithms efficiently. Projects often focused on implementing one solution at a time, with no clear path to scaling across modalities or organisations.

The NHS needed infrastructure that could deploy multiple world-leading AI models safely and affordably, navigate strict data sharing standards and deliver value quickly to clinicians and patients without vendor lock-in.

We worked with NVIDIA, world-renowned academic institutions and innovative NHS healthcare providers as part of the Medical Open Network for AI (MONAI) to build a first-of-its-kind AI orchestration engine called MONAI Deploy.

Commissioned by the AI Centre for Value Based Healthcare, we developed the AI Deployment Engine (AIDE) based on MONAI Deploy technology. AIDE provides intermediate steps where researchers and model developers can build confidence in techniques and approaches, allowing for iterative workflow until the AI inference infrastructure is ready to move to clinical environments.

We created a blueprint for information governance, clinical safety assessment and scalable deployment processes. The platform allows organisations to trial models in shadow mode before live deployment, building confidence in clinical teams without disrupting workflows. Once live, it provides capability to analyse model outputs in real-time to ensure they perform as expected.

We then worked with clinical scientists at King's College Hospital to take an AI model for analysing stroke CT scans from ideation to live minimum viable product in three months, demonstrating the platform could deliver at the speed healthcare needs.

Over two years, six other London and south-east trusts deployed AIDE. East Kent Hospitals University NHS Foundation Trust integrated the deployment engine in four weeks.

The platform transformed how NHS organisations deploy and manage AI models. Built on open standards, it enables rapid deployment and scalability even for organisations with less digital maturity. With the same NHS resources and integration investment as proprietary solutions, trusts can now access the full AI radiology marketplace, trial and select the best models, and expand into genomics, digital pathology and clinical coding as their AI maturity grows.

Following the completion of the Innovate UK-funded AIDE project, we formed relationships with best-in-class CE marked AI models. These models, covering chest CT, chest X-ray, musculoskeletal, 2D and 3D mammogram and prostate services, can now be deployed in as little as one week.

Trusts following an open platform architecture model are now at the forefront of innovation, able to realise clinical benefits rapidly rather than being locked out by single provider dependencies.
The programme won the UKIT Award for AI Project of the Year in 2024, recognising the platform's impact on healthcare AI adoption.