Computer aided long bone fracture detection

Abstract

We have developed a method of automatically detecting fractures in long bones. While bone fractures are a relatively common occurence, their presence can often be missed during x-ray diagnosis, resulting in ineffective patient management. Detection of fractures in long bones is an important orthopaedic and radiologic problem, so we propose a computer aided detection system to help reduce the miss rate. Our fracture detection algorithm consists of a number of steps. The first is extraction of edges from the x-ray image using a non-linear anisotropic diffusion method - the affine morphological scale space - that smoothes the image without losing critical information about the boundary locations within the image. The second is a modified Hough transform with automatic peak detection, which is used to determine parameters for the straight lines that best approximate the edges of the long bones. A composite of the magnitude and direction of the gradient is then created using the calculated line parameters. This allows abnormal regions, including fractures, to be highlighted. Experiments on a library of images show that this method consistently detects mid-shaft long bone fractures.