Splines and Active Shape Model for segmentation of pelvic x-ray images

X-ray segmentation is a challenging problem, particularly when dealing with complex bone structures such as the human pelvis. However, initial diagnosis of traumatic pelvic injuries is typically made via x-ray images, due to low cost and lack of disruption to the patient. Diagnosis is a complex task, as fractures in different regions of the pelvis display different characteristics. An effective decision-making system for such injuries therefore requires accurate segmentation of the key pelvic structures, so specific features may be searched for in the fracture detection process. Previous work in x-ray image segmentation has focused on the use of deformable model techniques, such as active shape model (ASM). However, the low resolution and blurred lines of x-ray images can cause the model to converge to false edges. This paper combines standard Active Shape Model with the use of cubic spline interpolation to regulate deformation, taking advantage of the natural curvature of the pelvic structure. This extended algorithm is compared with standard ASM on both normal and injured cases, and is shown to improve performance even in the presence of fracture.