Detecting reduced bone mineral density from dental radiographs using statistical shape models

We describe a novel method of estimating reduced bone mineral density (BMD) from dental panoramic tomograms (DPTs), which show the entire mandible. Careful expert width measurement of the inferior mandibular cortex has been shown to be predictive of BMD in hip and spine osteopenia and osteoporosis. We have implemented a method of automatic measurement of the width by active shape model search, using as training data 132 DPTs of female subjects whose BMD has been established by dual-energy X-ray absorptiometry. We demonstrate that widths measured after fully automatic search are significantly correlated with BMD, and exhibit less variability than manual measurements made by different experts. The correlation is highest towards the lateral region of the mandible, in a position different from that previously employed for manual width measurement. An receiver-operator characterstic (ROC) analysis for identifying osteopenia (T < -1: BMD more than one standard deviation below that of young healthy females) gives an area under curve (AUC) value of 0.64. Using a minimal interaction to initiate active shape model (ASM) search, the measurement can be made at the optimum region of the mandible, resulting in an AUC value of 0.71. Using an independent test set, AUC for detection of osteoporosis (T < -2.5) is 0.81.