Automated comparison of dual-photon absorptiometric studies of the lumbar spine

Abstract

An automated image comparison procedure was developed to optimize the precision of bone mineral density measurements by dual-photon absorptiometry. Changed acquisition conditions cause differences between two images to be compared. Alignment of one image with respect to the other is performed by a transformation that involves a rotation, a horizontal or vertical shift, and a correction for the soft tissue level. The best possible transformation is found in a stepwise search, guided by initial estimations of its parameters. After optimum transformation of one image, the region of interest of the other image is applied to both of them. Duplicate measurements of 9 patients and 15 normal subjects were performed; automated analysis yielded improved precision with respect to manual analysis. The coefficient of variation (CV) was also computed. The CV for automated analysis was 2.00% for patients and 1.04% for normal subjects compared to 3.55 and 1.93%, respectively, for manual analysis. For phantoms, the precision was 2.67% for manual analysis and 0.49% for automated analysis.