Detection of Wrist Fractures in X-Ray Images

The commonest diagnostic error in Accident and Emergency (A&E) units is that of missing fractures visible in X-ray images, usually because the doctors are inexperienced or not sufficiently expert. The most commonly missed are wrist fractures [7,11]. We are developing a fully-automated system for analysing X-rays of the wrist to identify fractures, with the goal of providing prompts to doctors to minimise the number of fractures that are missed. The system automatically locates the outline of the bones (the radius and ulna), then uses shape and texture features to classify abnormalities. The system has been trained and tested on a set of 409 clinical posteroanterior (PA) radiographs of the wrist gathered from a local A&E unit, 199 of which contain fractures. When using the manual shape annotations the system achieves classification performance of 95.5% (area under the Receiver Operating Characteristic (ROC) curve in cross validation experiments). In fully automatic mode the performance is 88.6%. Overall the system demonstrates the potential to reduce diagnostic mistakes in A&E.