Part-Based Lumbar Vertebrae Tracking in Videofluoroscopy Using Particle Filter

Abstract

Vertebrae tracking in videofluoroscopy is a challenging problem because of the low quality ‎of ‎image ‎sequences, like poor image contrast, ambiguous geometry details, and vertebrae rotation. The aim of this article is to tackle this problem by ‎proposing a ‎method for rigid object tracking based on the ‎fragmentation of the tracked object. The proposed method ‎is based on the particle filter using the calculation of the similarity between the ‎respective‏ ‏fragments of ‎objects instead of the whole objects. The similarity measures used are the Jaccard index, the ‎correlation ‎coefficient, and the Bhattacharyya coefficient. The tracking starts with a semi-automatic initialization. ‎The results show that the fragments-based object tracking method outperforms the classical ‎method ‎‎(without fragmentation) for each of the used similarity measures. The results show that the ‎tracking based on the Jaccard index is more stable and outperforms methods based on ‎other similarity ‎measures.‎