Automated tracking in digitized videofluoroscopy sequences for spine kinematic analysis

Spine kinematic analysis provides useful information to aid understanding of the segmental motion of the vertebrae. Digitized videofluoroscopy (DVF) is the existing practical modality to image spine motion for kinematic data acquisition. However, obtaining kinematic parameters from DVF sequence requires manual landmarking which is a laborious process and can be subjective and error prone.This work develops an automated spine motion tracking algorithm for DVF sequences within a Bayesian framework. By utilizing the anatomical relationships between vertebrae, a dynamic Bayesian network with a particle filter at each node is constructed. The proposed algorithm overcomes the dimensionality problem in a regular particle filter and has more efficient and robust performance. It can provide results of about 1° and 2 pixels (0.2mm) variability in rotation and translation estimation, respectively, during repeated initialization analysis on sequences from simulation and in vivo healthy human subject studies.