HIC led research projects

PICTURES: is a £3.8M MRC-funded project led by Dundee in conjunction with the University of Edinburgh, Abertay University and NHS Scotland. The programme is making use of the approximately 30 million clinical images (e.g. MRI, CT) collected since 2006 to develop Artificial Intelligence (AI) methods to search for such ‘warning signs’ in the images which predict the development of diseases. This will allow doctors in the future to make use of this information in routine care, greatly enhancing the clinical utility of routine scans. See the architecture of the PICTURES platform.

SATRE: This DARE UK Driver Project to develop a Standardised Architecture for TREs. It is a collaboration between University of Dundee, Alan Turing Institute, UCL, Ulster University and Research Data Scotland. By engaging with the wide UK TRE community from academia, industry, health service and government we will build a specification and design to meet the needs of safe and secure data research within the UK. Follow the Medium blog for more details. The project is fully open science and our github has all the work in progress.

CO-CONNECT: The COVID -Curated and Open aNalysis aNd rEsearCh plaTform (otherwise known as CO-CONNECT) is a £4M MRC-NIHR funded UK wide research programme supporting researchers to find and access COVID-19 data at pace whilst ensuring people’s information is kept private and secure. CO-CONNECT has embedded a cohort discovery tool within the HDR Innovation Gateway and worked with >20 different organisations who manage data to connect to the tool. Researchers can find out about what data is available but without the data leaving the control of the organisation who manages it.

Sustainability Development Fund: The SDF from Research Data Scotland has helped support three themes in HIC:  Developing a Scottish national laboratory dataset, an exemplar federated data analytics, disclosure control of machine learning workflows from TREs.

GRAIMATTER: TREs have historically supported only traditional statistical data analysis. There is an increasing need to also facilitate the training of artificial intelligence (AI) models. The size and complexity of AI models presents significant challenges for the output checking process. Models may be susceptible to external hacking: complicated methods to reverse engineer the learning process to find out about the data used for training, with more potential to lead to re-identification than conventional statistical methods. With input from public representatives, GRAIMatter has assessed a range of tools and methods to support TREs to assess output from AI methods for potentially identifiable information. A Green Paper has been prepared to report on the project’s findings. See a version for a general audience.

TREEHOOSE: has built on HIC’s experience of migrating a TRE hosting anonymised patient data over to ‘public cloud’ – meaning the data is accessible over a secure internet connection rather than in a local data centre – for the benefit of other operators. The project has included a new capability of ‘enclave computing’ to TREs, which will add a layer of software encryption to protect intellectual property and code in addition to the data itself. It released an open-source software to streamline building and operating TREs on public cloud infrastructure whilst maintaining security and trust. Secure enclaves go beyond the traditional TRE infrastructure by adding additional barriers to prevent software algorithms from leaking data.

Phenomics Portal: the National Phenomics Resource project has developed the Phenomics Portal. This reference catalogue of human diseases is integrated into the Health Data Research Innovation Gateway enabling users to search for datasets which may contain their phenotype of interest. A phenotype is an observable and measurable piece of information that is relevant to health or healthcare. For example, it can be a disease (e.g. type 2 diabetes), a blood pressure measurement, a blood sugar value or a prescription of antibiotics. Phenotyping algorithms are special tools that enable researchers to extract phenotypes from complex, and often messy data that get generated during routine interactions within the healthcare system. They identify and extract data from medical records using clinical codes which are the building block of how information is recorded in healthcare (for example ICD-10). Using these specialized algorithms, researchers and clinicians can maximize the value of patient data contained in medical records and answer important questions that can improve health and healthcare.

Alleviate: Alleviate is the Advanced Pain Discovery Platform (APDP) Data Hub. The hub is transforming UK pain datasets to be Findable, Accessible, Interoperable and Reusable (FAIR) and is providing expert data engineering, to enhance responsible, timely and trustworthy analysis by researchers and innovators, with the aim to improve lives. The hub is one of the Health Data Research UK data hubs.