APPLICATION OF AUTOMATIC IMAGE SEGMENTATION TO TIBIAE AND VERTEBRAE FROM OVARIECTOMIZED RATS

Automatic contextual segmentation algorithms were developed to objecti vely identify bone compartments in pQCT images of tibiae, femora, and vertebrae. Principal advantages of this approach over existing techniq ues such as histomorphometry are as follows: (a) the algorithms can be implemented in a fast, uniform, nonsubjective manner across many imag es, allowing unbiased comparisons of therapeutic efficacy; (b) much la rger volumes in the region of interest can be analyzed to derive true volumetric parameters for trabecular and cortical bone compartments; a nd (c) pQCT can be used to quantitate bone effects longitudinally in v ivo. An automatic contextual segmentation algorithm was used to analyz e over 600 scans of proximal tibiae, distal femora, and L-4 vertebrae from studies with ovariectomized rats. Accuracy and precision analyses were performed, and correlation to histomorphometry parameters showed that pQCT trabecular bone density correlates to Tb.N with r = 0.93, w hile BV/TV correlates to Tb.N with r = 0.95. In other words, pQCT corr elates as well to histomorphometry as histomorphometry does to itself. We conclude that the developed automatic segmentation algorithm provi des fast, precise, and objective quantitation of bone compartments tha t are highly correlated with histomorphometry measurements. (C) 1997 b y Elsevier Science Inc. All rights reserved.