The Optimam Mammography Image Database (OMI-DB) has been created to support research involving medical imaging with the aim of optimising the use of existing and adoption of new X-ray imaging technologies, including digital breast tomosynthesis (DBT), for detecting breast cancers and improving early detection in the NHS Screening Programme. Research into digital medical imaging requires large-scale collection and anonymisation of images and data. This demand has led to the formation of a centralised flexible image repository which prospectively collects images and associated data from multiple sites throughout the UK, securely. Collection, annotation and storage of the images is almost fully automated and is very adaptable allowing for quick and easy expansion to disparate imaging sites.

Images from the extensive database and associated software are available for licensing. Cancer Research UK is also seeking collaborative partners interested in using and helping to develop the available tools.

Database
This very valuable database of mammograms contains both processed and unprocessed medical images, typically for women who have had a screen detected breast cancer and contain two views of each breast (medio-lateral oblique and cranio-caudal). The images are annotated with the location and appearance of the cancers, and where applicable, expert determined ground truths describing features of interest are added. Tables 1 and 2 provide a summary of the images and their classifications currently present in the database. The size of the database will continue to grow according to predictions in Figure 2 (overleaf), with images being collected from three main sites.

Tools have been developed to:

• identify patient cases of interest from the National Breast Screening System (NBSS)
• anonymise collected images
• automatically retrieve corresponding images from the site’s PACS together with any associated radiological, clinical and pathological data from local databases that may be of interest
• allow remote expert radiologists to annotate the images with interesting clinical features and provide descriptors of these features.

MedXViewer
MedXViewer (Medical Extensible Viewer) is a separate bespoke application designed to allow workstation independent, PACS-less viewing and interaction with mammography images. Regions of interest can be identified by a user and any associated information about a mark, an image or a study can be added. The flexible software design allows the application to be easily extended to support other imaging modalities.

Applications
The inclusion of unprocessed images enables a multitude of potential research applications and is essential for studies on for example, different systems, radiation doses, beam qualities, image processing and computer aided detection on clinical performance.

The database can be accessed by a web portal and to date has mainly been used to run remote observer studies in which observers at remote sites inspect selected images, and locate and rate regions they suspect to be cancer with likelihood of malignancy. Such studies can be used to measure the performance of radiologists, provide a training infrastructure for radiologists, as well as optimise design, choice and use of new technology in cancer screening.

The availability of associated data and expertly determined ground truths facilitates other research applications such as radiomics where large amounts of quantitative features from non-invasive medical images can provide information about tumour phenotype. This has the potential to help guide and monitor response to therapy.