Using chest imaging AI to support COVID-19 research and development
The National COVID-19 Chest Imaging Database (NCCID) was established to fulfil the NHS AI Lab’s mission of enabling the safe adoption of AI technologies, with an immediate focus on combatting the pandemic.
A joint initiative between NHSX, the British Society of Thoracic Imaging (BSTI), Royal Surrey NHS Foundation Trust and Faculty, the NCCID is designed to enable the development of software that helps doctors and researchers to:
understand the impact and progression of the disease
assess the severity of the condition in individual patients
identify factors that may complicate recovery
prioritise patients whose condition is most likely to deteriorate
Since we began this initiative at the start of the pandemic, we have made fast progress in pulling data from hospital sites and enabling researchers to access the data to develop AI technologies that can support the care of those hospitalised with COVID-19.
We are now supported by the efforts of 20 NHS Trusts who are contributing data to the NCCID. As a result of their collaboration, the database now includes over 40,000 image studies for over 12,000 patients, making the NCCID one of the largest initiatives of its kind in the UK.
Medical images such as X-rays play a central role in detecting diseases, predicting patient outcomes and informing the type of care needed. Over 10 research groups, including a number of UK university interdisciplinary groups of clinicians, mathematicians, engineers and computer scientists, are using the NCCID. They are accessing the database to investigate technologies that could use the images to diagnose COVID-19 and predict what care will be needed for those patients with confirmed COVID-19. We have published NCCID project summaries to give more information about the work.
As the NCCID continues to grow, we would like to invite acute trusts, researchers and technology companies to get involved with the project. In particular you can be involved with:
contributing data from hospital sites
using the NCCID as a tool for research
developing a platform for AI imaging research
Read more about each of those three aspects below.
Contributing data from hospital sites
If you work in a radiology department at a hospital site and are interested in contributing data to the NCCID, please reach out directly to email@example.com. All data is collected in a de-identified form in line with robust information governance procedures and legal and ethical arrangements (see more below).
Any data you are able to provide is incredibly valuable for supporting research into the COVID-19 disease. More data means we can increase the sample size for various AI models and enriches the robustness of our AI model development and testing. Importantly, increased data leads to improved geographic, demographic and clinical representation of patients in the NCCID, enabling technologies to be scaled to the UK population.
Using the NCCID as a tool for COVID-19 research
If you are involved in research or technology development and are interested in using the NCCID, you can apply for access to the NCCID training dataset. You will need to fill in an application and provide an accompanying project plan, along with a contractual agreement.
The application approval process is supported by the efforts of our Data Access Committee, which includes technical, clinical, ethical, and patient advisors. They oversee a rigorous Data Access Request process to ensure that those using the data are doing so in both a secure and impactful way.
Developing a platform for AI imaging research
Whilst we have made swift progress in building the database, our focus will now be on longer term improvements to enhance the quality and quantity of the data collected, and ensure that this is used impactfully.
In addition to onboarding more sites and users, we want your suggestions and proposals, so please do reach out to firstname.lastname@example.org.
We are exploring a number of methods to partially automate data collection from hospital sites, which will aim to minimise efforts required to contribute data.
We are working with several other database initiatives to link our datasets and provide a more comprehensive and granular resource for our data users.
Long term impact of COVID-19
We will work with research studies focused on the long term impact of COVID-19, such as PHOSP-COVID, to ensure that the NCCID can support these efforts as a valuable resource.
Validating the performance of AI models
We have set aside a portion of the NCCID data to test AI model performance on an unseen dataset to examine their efficacy, accuracy and safety.
The NCCID has been a successful collaboration between healthcare institutions, academic researchers and commercial companies, all of whom quickly came together to respond to the challenges presented by the pandemic.
We would like to thank all those who have supported this initiative and contributed to its achievements so far. In particular, we would like to thank all the participating hospital sites, our NCCID partners, the National Consortium of Intelligent Medical Imaging (NCIMI), the Health Informatics Centre (HIC) at the University of Dundee, and several other research institutions that have helped guide the strategic direction of the NCCID.
We look forward to continuing to build on this initiative in future.
NCCID data collection information
The NCCID is underpinned by the collection of completely de-identified patient data at a national scale. This has been achievable due to the support provided by the Scientific Computing Team at Royal Surrey NHS Foundation Trust, who applied the same de-identification methods to the NCCID that have been used on previous data collection projects, such as OMI-DB (a national database containing 3,000,000 mammography images).
Our data collection methods have been enabled by robust information governance procedures and legal and ethical arrangements. We are supported by the provisions of the COPI Notice issued by the UK Government, have received formal ethical approval from the Health Research Authority’s Research Ethics Committee, and have a Data Sharing Agreement and Data Protection Impact Assessment in place with each participating hospital site.