In the last 5 years, large-scale databases of medical images have enabled developing crucial AI technologies that interpret images automatically and help clinicians in making diagnosis and planning a treatment course. However, existing databases are limited in their capacity. They do not contain data measured by the acquisition devices but only images, which are in reality a processed output of the machine. This limits the extend of the research and developments that can be achieved using these databases. For example, it becomes very difficult to develop tools that will improve the acquisition itself so that better images can be acquired for each patient.