Automatic region of interest determination in dual photon absorptiometry of the lumbar spine

In a heuristic approach, the authors developed an algorithm for automatic region-of-interest (ROI) determination in bone mineral density (BMD) measurements of the lumbar spine. First, the algorithm detects the boundaries of the spine utilizing simple smoothing and gradient operators followed by a dynamic programming technique. Second, it selects L2, L3, and L4 from the spine by examining the BMD values along lines that are orthogonal to the local direction of the spine. The algorithm was tested in studies of a spine phantom, normal subjects (30), and patients (94). In all but two patient studies of severely affected spines the contours were detected correctly. The ROI determination was performed satisfactorily in all studies of the phantom, normals, and patients that had no spinal fractures. The coefficient of variation (CV) in the phantom studies was equal to 0.7%. In duplicate studies of 15 normal subjects and nine patients, the CV was equal to 1.0 and 2.7%, respectively. Compared to manual determination of the ROI, the precision of BMD measurements was clearly improved by the automatic procedure