A new cloud data platform to enable ground-breaking healthcare research for Moorfields Eye Hospital

The Challenge

Help Moorfields Eye Hospital and its strategic partners take forward their pioneering healthcare research.

• Moorfields Eye Hospital is the leading provider of eye health services in the UK and a world-class centre of excellence for ophthalmic research. Moorfields wanted to use its millions of ophthalmic images to support its ground-breaking research, while carefully following information governance policies and respecting patient confidentiality. 
• Moorfields have around forty different imaging systems, which produce medical images in various proprietary formats. Some files were stored locally or on shared drives, and other data was stored across multiple clinical and patient systems. Producing a dataset required researchers to look across multiple systems to find candidate images, manual de-identification of demographic and outcomes data drawn from clinical records, and lengthy Information Governance sign-off. The process was laborious and manual, taking up to 9 months to complete. 

Our Solution

We built a bespoke, cloud-based data platform that automates the ingestion, curation, and transformation of data, and produces an interactive catalogue of scans using Looker. It also standardises images in open-source formats and separates them from their patient-identifiable data. Our solution enables researchers to identify potential study cohorts quickly, and to receive tailored, de-identified datasets in days, not months. 

We designed and built a centralised image store, with metadata stored in BigQuery. To date we have catalogued over 20 million images and automatically ingest around 50,000 new scans every week, transforming them into standard formats which are easy to use and suitable for the application of AI models.

We also pull in metadata about the images from hospital patient systems, allowing authorised researchers to request, for example, “The complete ophthalmic record (images, injections, operations, visual acuities) for patients over 50 who have been assessed to have diabetic retinopathy with no clinical notes mentioning ‘Stargardt ’”.

We export data using custom SQL queries according to requirements and substantially reduce processing time through heavy parallelisation using Apache Beam. 

We designed our solution with information governance at its heart, focussing on privacy protection and traceability. We worked closely with Moorfields Information Governance to establish strict requirements for working with patient data and designed our solution to fully meet national standards.

Data we store on Google Cloud Platform is separated into distinct service perimeters with different levels of anonymisation. Data can only pass from one service perimeter to another through a specific secured pipeline, with increasing pseudonymisation automatically applied at each stage. Other transfers are restricted, so deanonymisation cannot happen accidentally.

Our traceability solution also makes it possible for any file to be traced back to its source system, allowing its fields to be checked against the original image in a quality assurance step for a sample of images in each dataset. 

Our technical approach to data governance is compliant with stringent anonymity and security requirements, both giving Moorfields the needed confidence in data for research study results, and giving patients the confidence that their data is handled securely and their privacy will be protected. 

The Result

We have supported pioneering healthcare research projects, while our data platform is enabling the study of significantly larger and more detailed datasets than ever before.   

Enabling breakthrough research  

The central database of de-identified images and clinical data from across Moorfields Eye Hospital’s sites has been used to generate curated datasets for a variety of Trust-approved research projects, some of which have already led to the publication of breakthrough findings.  

The study to predict the progression of exAMD, for example, was able to create an AI model capable of performing as well as – or better than – clinicians at predicting whether an eye will convert to this condition within a six-month period of initial assessments.  

Pearse A Keane, Consultant Ophthalmologist at Moorfields Eye Hospital NHS Foundation Trust and Professor of Artificial Medical Intelligence at University College London, explains: 

 “If you looked at the typical types of medical studies before, in ophthalmology, you’d often see a study with 300 patients regarded as a large study. Now, it’s tens of thousands, or hundreds of thousands, or even millions of patients.  It wouldn’t have been possible to do even a fraction of the things that we’ve been able to achieve without the help of Softwire.”