Comparison of digitized images with original radiography for semiquantitative assessment of osteoporotic fractures

Abstract

We validated a vertebral fracture assessment (VFA) workstation developed by our group for semiquantitative assessment of vertebral fractures in large-scale, multicenter osteoporosis drug trials. Baseline and follow-up spine radiographs (lateral views) of 50 patients who participated in a clinical trial were digitized and were archived on CD-ROM. Both original radiographs and the digitized images were independently assessed by three experienced radiologists. Prevalent fracture scores of vertebrae were rated in increments of 1 on a 4-point scale. Incident fractures were defined as any worsening of grade on follow-up films. Generally good to excellent agreement among the three readers was found between the two methods, with kappa scores (kappa) from 0.91 to 0.96 for prevalence of fractures, and from 0.80 to 0.90 for incidence of fractures. Reproducibility (intra-reader variability) of each method was comparable. For assessing prevalent fracture, kappa was from 0.87 to 0.96 using radiographs, and from 0.87 to 0.94 using VFA images. For incident fractures, the kappa was from 0.78 to 0.89 using radiographs, and from 0.82 to 0.88 using VFA images. Level-specific agreement between the two approaches was consistent. Overall, there is no difference between readings of digital images and readings of conventional radiographs. The quality of the new VFA for visualization of vertebral fracture is excellent.