Logo iCAIRD
Case Study
9 Min Read

Unlocking the data: implementing FHIR Facade in the AI sector

Rien Wertheim

Subscribe to our newsletter

Subscribe

The Industrial Centre for Artificial Intelligence Research in Digital Diagnostics (iCAIRD) is a Scotland-wide collaboration of industry, the NHS, and academia, working to develop and validate diagnostic AI technologies, and accelerate pathways to deployment in the NHS and wider UK clinical sector.

iCAIRD facilitates collaboration between clinicians, academics, and companies on large-scale interoperable databases and projects, with research hubs in Glasgow and Aberdeen. It has adopted established standards including DICOM for images, and FHIR for non-imaging data, including clinical notes.

Firely was engaged as a FHIR consultant for the iCAIRD hub in NHS Grampian. We delivered the FHIR server and provided professional services to help Grampian implement the FHIR Facade approach from day one, to unlock non-FHIR information and expose it as FHIR for research and algorithm development.

The ‘how’ and the ‘why’

As an early research project, iCAIRD seeks to define how and if AI can be used, identify challenges to be resolved, and deliver a platform to enable that work to happen. Funded by government and industry, it enables joined-up academic and commercial partnership: processing the high volume of data gathered in NHS clinics to help solve healthcare challenges, while fully protecting patient identities.


iCAIRD exemplar projects in Grampian have included partnering with SME Kheiron Medical Technologies Ltd to evaluate their Mia® AI algorithm for use in the Scottish Breast Screening Programme. Led by Professor Lesley Anderson, University of Aberdeen, and Dr Gerald Lip, NHS Grampian, the first phase of the study used a three-year historical dataset (58,209 cases) from a UK regional screening programme, with initial results demonstrating the importance of evaluating the local performance of the algorithm before deployment into clinical practice. Mia is now being evaluated prospectively in The GEMINI Project, to determine the safety and efficacy it can bring to the breast cancer screening workflow through opportunities of clinical and operational benefit.


Non-imaging FHIR data

iCAIRD’s two eco-systems are radiology and pathology, with the Grampian centre focusing on radiology and making use of the Grampian Data Safe Haven (DaSH), a secure, virtual healthcare data analysis and storage environment.

Led by Canon Medical Research Europe (Canon), one of the iCAIRD workstreams is developing a technology to optimize diagnosis and time-critical treatment for patients with acute ischemic stroke. This project aims to deliver an AI algorithm for examining images, combined into a clinical cockpit which pulls relevant and critical non-imaging FHIR data, such as recent GP notes, to support rapid and safe clinical decision-making.

“iCAIRD has more than 30 current research projects across radiology and pathology, and 30 active partner companies. Most importantly, it’s a way of allowing real development with actual data.”

Moragh Boyle, iCAIRD Grampian Project Manager

The SHAIP of things to come 

Central to the iCAIRD programme and delivery of the radiology exemplar projects, Canon is leading the development and deployment of a Safe-Haven Artificial Intelligence Platform (SHAIP) working with the existing data Safe Havens in Glasgow and Grampian. This hyper-converged environment uses Firely Server, among other mechanisms, to ingest structured and unstructured patient information from clinical systems and letters. SHAIP also ingests images direct from the hospital radiology picture archiving system (PACS) and is the first research environment in Scotland to be directly connected to the national radiology imaging archive.  

PACS – hospital radiology picture archiving system

SHAIP de-identifies the patient data in preparation for export to a secure research workspace, ensuring personal information is not disclosed, using a combination of technical security protocols and stringent governance processes.

Canon’s platform design includes tools for clinicians and radiologists to annotate images, providing ground truth for machine learning projects. The platform also has an advanced graphical processing capability to allow data and AI scientists to develop and validate clinical algorithms. Machine learning algorithms trained on SHAIP can be deployed into a clinical cockpit for evaluation, with data delivered in FHIR format via the Firely Server.

Deploying SHAIP in two NHS sites enables federated learning, so algorithms can be trained in SHAIP on a local population before being securely transferred to the other region’s SHAIP instance for evaluation and potential re-training. This further protects data and patient privacy, by bringing researchers and algorithms to the data rather than moving the data to a centralized store.

FHIR Facade from Firely

Firely met iCAIRD’s requirement for a known coding language – C# – due to the timescale and resources available, and delivered on Canon’s preference for a FHIR interface to ingest non-imaging data into SHAIP.

Firely Server, implemented as a FHIR Facade, is a fast and powerful shortcut to connect systems to FHIR clients, without the need to create a copy of existing data in FHIR. Instead of using a repository of FHIR resources, data is retrieved from back-end repositories and converted on demand.

‘A real enabler’ 

Firely provided responsive and adaptable expertise, attainable technology, a fast turnaround, and a cost-effective solution, designed from the ground up to unlock iCAIRD’s non-FHIR information for the relevant use cases and work packages.

The Grampian iCAIRD FHIR Facade points to the ODIN data provisioning platform built by Colin McSkimming, the Senior Analyst Developer seconded to iCAIRD from NHS Grampian. ODIN ingests unstructured data from clinical documents and makes this data available for interrogation in real time.

FHIR can also be used to build a research cohort by sending thousands of sequential patient data requests. Firely Server performed exceptionally well here, especially in its speedy delivery of unstructured data.

Two models, one licence 

Using FHIR Facade allows iCAIRD projects to point to multiple data sources in real time, with the ability to access data from non-FHIR back-end systems or expose only selected data as FHIR objects.

With Firely Server available as a full FHIR server or as Facade, the implementer can decide which to configure, but Firely technical consultant Frank Olthuijsen sounds a note of caution: “Using Facade is a great way to avoid duplicating source data, but maintenance can be a lot of work. The more resource types and search parameters you need to map, the higher the maintenance burden in the Facade code. There is a tipping point, after which a full FHIR server makes more sense.”

Understanding the data 

FHIR is all about standardizing data, which means that mapping source data to an appropriate data item is a potential point of failure.

Frank Olthuijsen points out that the code must be precise, and you must know what data means in your source system to reliably map it to FHIR: “FHIR makes things a whole lot easier, but you need to be vigilant about data quality issues.”

In the case of iCAIRD, data comes from a multiplicity of clinical sources: some structured, some not structured, and some which can be incomplete or contradict each other, such as medication records and referrals.

Frank and Colin worked as a team to structure the data, build and configure the FHIR Facade, write the necessary code, and implement support for the resource types and search parameters that were required.

Future potential

Other data available in ODIN and not yet presented in FHIR is clearly relevant for other use cases. NHS Grampian is already using FHIR resources for allergy intolerance, patient diagnostic reports, medication statements, service requests, and AI retrieval of text from PDF documents.
The data journey unlocked by iCAIRD and expedited by FHIR and Firely may be only just beginning.

“We are absolutely sold on FHIR as the way we want to deliver data. It’s making things easy for all kinds of future projects – like finding the hard way up a mountain then throwing down a rope. And having Firely on board as consultants was a real enabler for us.”

Colin McSkimming, Senior Analyst Developer

To learn more about Firely Server or using FHIR Facade, check out our website. Firely also offers comprehensive training courses on HL7 FHIR, including the Firely Server Workshop which helps you build a proof of concept in 2 online sessions. Our support team also provides professional services to help you with the integration of your back-end system. Feel free to reach out to us for more information.

Recommendations for you

Explore more topics

Post a comment

Your email address will not be published. Required fields are marked *