Automatic Lumbar Vertebrae Segmentation in Fluoroscopic Images via Optimised Concurrent Hough Transform

Low back pain is a very common problem in the industrialised countries and its associated cost is enormous. Diagnosis of the underlying causes can be extremely difficult. Many studies have focused on mechanical disorders of the spine. Digital videofluoroscopy (DVF) was widely used to obtain images for motion studies. This can provide motion sequences of the lumbar spine, but the images obtained often suffer due to noise, exacerbated by the very low radiation dosage. Thus determining vertebrae position within the image sequence presents a considerable challenge. In this paper, we show how our new approach can automatically detect the positions and borders of vertebrae concurrently, relieving many of the problems experienced in other approaches. First, we use phase congruency to relieve difficulty associated with threshold selection in edge detection of the illumination variant DVF images. Then, our new Hough transform approach is applied to determine the moving vertebrae, concurrently. We include optimisation via a genetic algorithm as without it the extraction of moving multiple vertebrae is computationally daunting. Our results show that this new approach can indeed provide extractions of position and rotation which appear to be of sufficient quality to aid therapy and diagnosis of spinal disorders.