Case study: iCAIRD
Industrial Centre for AI Research in Digital Diagnostics (iCAIRD)
Overview
Led by NHS Greater Glasgow and Clyde, the West of Scotland Innovation Hub, and Canon Medical Research Europe (radiology), iCAIRD sought to enable deidentification of images and unstructured text for research use.
The project had three main objectives:
- Develop AI infrastructure for digital diagnostics and radiology by leveraging existing national initiatives including Health Data Research UK and the national Picture Archiving Communication System, operating within the NHS Safe-haven network through Canon's SHAIP and Clinical Cockpit systems to enable secure AI development without exporting identifiable data.
- Support SME-led AI validation exemplars in three clinical areas: acute stroke medicine, chest x-ray triage, and mammogram interpretation, creating a standardised national network through application programming interfaces that enables training and validation across multiple regional centres.
- Establish an SME Accelerator Programme providing access to expertise, facilitating regulatory approval, supporting commercial development, and promoting job creation within the AI healthcare sector.
Who was involved?
The project was a collaboration between the lead organisations, Philips Healthcare (digital pathology), industrial collaborators, academic partners (Universities of Glasgow, Edinburgh, St Andrews and Aberdeen) and the Grampian Data Safe Haven (DaSH).
Other collaborators include:
- Bering Limited
- Nvidia Ltd
- Kheiron Medical Technologies Ltd
- Glencoe Software Limited
- NHS Grampian R&D
- Deepcognito Ltd
- Greater Glasgow Health Board
The challenge
UK clinical radiology departments were already fully digitised. However, access at scale to clinical radiology images and associated 'real-world' clinical data to enable development of AI applications for use in clinical practice was severely limited, preventing effective adoption of AI.
Contemporary technological capabilities could more effectively harness the vast amounts of data generated throughout healthcare systems, facilitating earlier disease identification, prompt therapeutic intervention, and improved patient outcomes.
The solution
iCAIRD united Scotland's clinical and academic expertise across digital radiology and pathology, sophisticated data storage systems, governance frameworks, and database interoperability, fostering an active community creating clinically meaningful and commercially viable solutions while enhancing collaboration with national programmes coordinated by NSS.
The initiative implemented an ethical framework grounded in HDRUK best practices, ensuring patient information remained securely within NHS systems while enabling industry-researcher-clinician collaboration to develop more effective diagnostic technologies. They established an Accelerator Programme helping SMEs tackle healthcare challenges through regulatory guidance and market entry expertise, generating economic development and patient care improvements.
Partnering with Canon, they constructed Safe Haven Artificial Intelligence Platforms (SHAIP) integrated into NHS data Safe Havens, beginning in Glasgow, expanding to Aberdeen, achieving national coverage. This framework supported distributed AI training while maintaining identifiable patient data within NHS boundaries, enabling smaller companies to collaborate directly with local healthcare professionals.
The team transformed Glasgow's Pathology Department—Europe's largest—from traditional microscopy to digital systems. Collaborating with Edinburgh Parallel Computing Centre, they created a secure National Pathology Research Image Database with Philips' guidance, capable of receiving images from various locations and equipment types.
Initial applications validated AI systems for stroke care, chest X-ray prioritisation, mammography analysis, colorectal cancer data, and gynaecological pathology assessments.
Problems encountered
The project was complex, involving many different streams. This has meant multiple revisions of cohorts, with some projects still ongoing now.
The extracted Digital Imaging and Communications in Medicine (DICOMS) had to be validated, demonstrating that they were clear of patient information. This involved one person having to spend six months examining 900 CT slices, in case patient names were shown. This was because, although CHI numbers and names can be removed, these are sometimes burned into images.
A certain number of results had to be examined by each machine to prove that these machines could be trusted. They used different modalities, such as CT and X-ray scans. It could have been possible to use Python, but the decision was made to make the process human-lead.
Key achievements
- Implemented an ethical framework grounded in HDRUK best practice standards.
- Transformed Glasgow's Pathology Department from traditional microscopy systems to digital platforms.
- Established technical infrastructure for developing and deploying AI applications in digital diagnostics, pathology, and radiology.
- Facilitated partnerships between research-engaged clinicians and innovative small-to-medium enterprises to enhance clinical inquiry and accelerate healthcare problem-solving with greater efficiency.
- Received the Innovative Collaboration Award at Scotland's Life Sciences Awards in 2021.
“Scotland has a rich life sciences ecosystem which can shorten the distance between inventors and their customers. iCAIRD is a national programme which focusses that ecosystem on delivering AI solutions for digital diagnostics. Diversity, inclusivity, creativity & collaboration lie at the heart of the programme.”
More information
I-CAIRD: Industrial Centre for AI Research in Digital Diagnostics (UKRI)
Image credit: Umanoide. Source: Unsplash