Lumbar Spine Location in Fluoroscopic Images by Evidence Gathering

Low back pain (LBP) is a very common problem and lumbar segmental instability is one of the causes. It is important to investigate lumbar spine movement in order to understand instability better and as an aid to diagnosis. Digital videofluoroscopy provides a method of quantifying the motion of individual vertebrae, but due to the relatively poor image quality, it is difficult and time consuming to locate landmarks manually, from which the kinematics can be calculated. Some semi-automatic approaches have already been developed but these are still time consuming and require some manual interaction. In this paper we apply the Hough transform (HT) to locate the lumbar spinal segments automatically. The HT is a powerful tool in computer vision and it has good performance in noise and partial occlusion. A recent arbitrary shape representation avoids problems inherent with tabular representations in the generalised HT (GHT) by describing shapes using a continuous formulation. The target shape is described by a set of Fourier descriptors, which vote in an accumulator space from which the object parameters of translation (including the x and y direction), rotation and scale can be determined. At present, this algorithm has been applied to the images of lumbar spine, and has been shown to provide satisfactory results. Further work will concentrate on reducing the computational time for real-time application, and on approaches to refine information at the apices, given initialisation by the new HT method.