EVALUATION OF THE CORTICAL STRUCTURE IN HIGH-RESOLUTION CT IMAGES OF LUMBAR VERTEBRAE BY ANALYZING LOW BONE-MINERAL DENSITY CLUSTERS AND CORTICAL PROFILES
Ma. Haidekker et al., EVALUATION OF THE CORTICAL STRUCTURE IN HIGH-RESOLUTION CT IMAGES OF LUMBAR VERTEBRAE BY ANALYZING LOW BONE-MINERAL DENSITY CLUSTERS AND CORTICAL PROFILES, British journal of radiology, 70(840), 1997, pp. 1222-1228
The structural classification of trabecular bone is of considerable cl
inical importance for the diagnosis of osteoporosis. Assessment of the
cortical bone mineral density (BMD) and analysis of cortical structur
e is necessary because the cortex is also affected by osteoporosis. We
describe a method for the automatic classification of the cortex from
its structure on high resolution (HR) CT images. The method is based
on the distribution of areas with low BMD in the cortex. Two different
approaches are presented; one uses the grey scale profile of the cort
ical ridge, and the other one evaluates the distribution of connected
regions (clusters) of low BMD, i.e. areas that lie below a certain thr
eshold value. In HRCT images from 22 lumbar vertebrae, the number of i
ntersections of the cortical intensity profile with a horizontal line
was determined at various positions of this threshold line. At a certa
in position, there was a local maximum in the number of intersections
which was 38% higher in the osteoporotic cases. The maximum shows a mo
derate correlation with the cortical BMD of r(ni) = -0.72 (p < 0.0001)
. The number n(c) of connected regions (clusters) with pixel values be
low a certain threshold value was determined in relation to the thresh
old value T. The resulting function n(c)(T) shows a relative maximum s
lightly below the average grey scale value of the respective image. Th
e curve depends on the degree of osteoporosis: the height of the maxim
um (i.e. the maximal number of clusters n(cmax)) allows distinction to
be made between different degrees of osteoporosis, and n(cmax) shows
a correlation with the cortical BMD of r(nc) = -0.84 (p < 0.0001).