Radiometric Classification of Alveolar Bone Health

Abstract

A radiograph-based classifier for discrimination between normal and diseased alveolar bone was developed. The alveolar bone health of dry mandibles was rated, by consensus of two trained observers, as normal or diseased. Bitewing radiographs of mandibular quadrants were digitized and regions of interest (ROIs) of the alveolar bone between the first and second permanent molars outlined. Gray-scale histograms for the ROIs were computed, and a classifier based on these histograms was developed. Two observers (independently) used only the classifier to rate the bone as being normal or diseased. These ratings were compared with the original ratings of trained observers who viewed and scored the actual alveolar bone. The sensitivity (the ability to detect diseased alveolar bone), specificity (the ability to detect normal alveolar bone), and accuracy were 0.90 or greater, indicating good agreement with subjective visual assessments of dry mandibles. These results suggest that this procedure may be applicable for evaluations of bone health in humans.