This talk will describe new technology to automatically annotate skeletal structures in radiographic images, aiming to rapidly transform image data into useful medical information.

 Musculoskeletal diseases affect millions of people globally, posing a major cost to healthcare systems worldwide. In clinical practise and research into musculoskeletal diseases, 2D X-ray images are the imaging modality of choice due to wide availability, speed of acquisition and low cost. However, the vast amount of imaging data currently collected is underutilised in terms of the rich information that it holds, such as details on the overall shape and appearance of the bones and joints of interest. Manually analysing this information is tedious and prone to inconsistencies, and automatically analysing it is challenging since skeletal structures are complex objects with significant radiographic variation due to anatomical differences between individuals and as a consequence of disease.

 BoneFinder is a software system that tackles this challenge. It fully automatically and accurately annotates skeletal structures in radiographic images, facilitating automated methods to better understand, diagnose and monitor diseases of the bones and joints. The system is based on state-of-the-art computer vision algorithms combining recent developments in machine learning with statistical shape modelling. Based on pre-annotated training data, BoneFinder learns what to look for in an image and uses this acquired knowledge to find and annotate similar bones in new unseen images. The automatically generated annotations can then be used for a range of applications such as segmentation and quantitative shape analysis, with a view to diagnosing and monitoring disease. Current applications of BoneFinder include studies of the hip, knee, hands, and head.