A major barrier to progress in AI in medical imaging is the lack of standardised and accessible imaging data for the development, evaluation and validation of AI algorithms.
The development of potential AI software solutions requires high-quality, labelled, curated and validated data.
NCIMI supports a network of NHS Trusts in providing data from their systems to support AI development. We have invested in the expertise to ensure that data is of high quality, consistent, and annotated and curated as needed.
The de-identified data is held within our central databank to support specific research project activity, and access is governed through our Data Access Committee, using FAIR principles – findable, accessible, interoperable and reusable – for scientific data management and stewardship.
We can support AI development, training and validation across various stages of development.
AI software development requires large patient data sets from diverse, representative populations of patients.
To address this need, we support our industry partners in developing new software solutions that are co-developed with NHS stakeholders and by accessing data from our diverse Hospital network.
Our broad range of NHS Trusts of different sizes, expertise and patient catchment populations throughout England and Scotland ensure that diverse real world data can be accessed to support development of robust solutions.
NCIMI is committed to equality and diversity and ensures the diversity of data acquired for healthcare AI development reflects diversity in ethnicity, socio-economic background and that solutions generated are broadly applicable.
NCIMI Data Bank
To best protect the interests of patients and the NHS, NCIMI has established a research databank, so that data once collected, and validated, may be used by multiple academic and commercial companies to develop, test and validate algorithms.
The data bank:
- Avoids duplication of effort in data collection and processing, by ensuring that data collected for one study can then be re-used for others without imposing any additional burden upon patients, or the NHS.
- Protects the interests of patients and the NHS, by ensuring that any processing of and access to data takes place under controlled, audited conditions and that responsibility for this resides with a single lead organisation.
- Supports the creation of a fair, sustainable ecosystem in which researchers and companies have access to the data that they need to create and validate new algorithms for care, and hospitals and health professionals are able to maximise the value of this data.
Data collection for NCIMI is undertaken only within the framework of specific approved studies, overseen by our Data Access Committee. For each study, the Big Data Institute provides the single point of data integration and supply. We receive and check data from our NHS partners, and make it available to commercial companies and academics for specific research and development purposes.
If you are interested in accessing existing data in the Databank, or working with NCIMI for novel data collection please get in touch.
Reader studies and real-world evaluation
NCIMI is able to provide evaluation of developed algorithms against data held in NCIMI’s databank, and in a real-world environment via its network of hospitals. It is also able to provide algorithm evaluation compared to a variety of different levels of clinical experience and expertise, by using its partner RAIQC’s cloud-based platform for reader comparisons.
If you are interested in working with NCIMI to validate and assess clinical performance of your algorithm, please get in touch.
- 1: Unmet need
- 2: Prototyping
- 3: Internal Testing
- 4: External Validation
- 5: Market Evaluation
Following early-stage R&D there may be a beta-AI solution. These projects likely involve greater data volume and possibly more sites, to develop robust, diverse training data sets
Diabetes & imaging
Early diagnosis and treatment planning for Endometriosis
NCIMI supports external testing by providing ‘unseen’ real-world data sets; evaluation against current readers in clinical practice; and real-time evaluation in clinical practice
No projects at this stage found