Automatic location of vertebrae in digitized videofluoroscopic images of the lumbar spine

Back pain is a widespread problem, and the disability it engenders continues to grow, despite efforts to contain it. A major problem in the diagnosis and management of back pain is the assessment of the degree to which mechanical factors play a part. Of considerable importance in understanding these mechanical factors is being able to quantify how the human spine actually moves in vivo. Digitized videofluoroscopy is currently the only practical method available for studying spinal motion in vivo at the segmental level. Low-dose, planar motion X-rays of the spine are captured on videotape and subsequently digitized for analysis. Until now, vertebrae in the digitized images were identified and marked manually as a basis for calculating intervertebral kinematics. This paper describes a procedure for automatically identifying the vertebrae in the motion sequences. The process increases objectivity and repeatability, and significantly reduces the manual effort required in locating the vertebrae prior to calculating the kinematics. The technique has been applied to images of a calibrated model and the results are promising. In-plane rotations may be calculated to an accuracy of at least 1 degree. Repeated analysis reveals standard deviations of less than 0.5 degree for intervertebral rotations and less than 0.25 mm for translations.