Hand (particularly carpal) bone fracture is very popular among ageing groups and athletes as approximately 190 cases reported for each 100,000 people.  More than 15% of clinical examinations fail to detect the fracture mainly due to insufficient information provided by low resolution X-ray images.

X-rays often used in the diagnosis of trauma are simple, cheap and universally available at point of care in the Emergency department.  However, in several areas, they are difficult to interpret due to either multiple layers being overlaid, e.g. the skull and face, or there is no possibility of a right imaging angle to the fracture line (scaphoid and foot). In such circumstances, diagnosis may be delayed leading to more radical treatment, or more resource demanding imaging such as magnetic resonance images, is used.  However, the information regarding the presence or absence of a fracture is almost certainly present in the original X-ray but not identifiable. Therefore, the fundamental aim of this project is to design a learning system which can learn from a limited number of matched MRI scans and X-ray images on how to detect the hidden diagnostic information, such as a scaphoid fracture, from widely available X-ray images only. Scaphoid fractures are chosen as a demonstration system due to having more access to the related paired images.  However, the system will have multiple uses throughout musculoskeletal radiology and into other body systems.

The outcome of this socially and clinically impactful research will alleviate the need for MRI scans in areas where plain X-rays are difficult to interpret or senior staff are not available, reducing the need for further imaging and leading to more rapid and accurate diagnosis.