EFFECTS OF REGION OF INTEREST OUTLINE VARIATIONS OF GRAY-SCALE FREQUENCY-DISTRIBUTIONS FOR ALVEOLAR BONE

Normalized cumulative percentage gray-scale distributions from digitiz ed radiographs of dry skulls can be used to evaluate alveolar bone hea lth. With this method, manually defined regions of interest are used t o identify the area of the image to be analyzed. The purpose of this s tudy was to determine if variations in the dimensions of the regions o f interest affect the gray-scale distributions. Two approaches were us ed. The effect of region of interest outline variations on (1) the abi lity to differentiate between various sized bone chips and (2) the abi lity to distinguish between two groups of skulls with the alveolar bon e subjectively rated as normal or diseased were investigated. Four dif ferent regions of interest, which consisted of outline and area variat ions to a basic region, were used. Ninety-one percent of the cumulativ e histograms that represented chips were sufficiently separated from n onchip plots to be distinguished. Bone health was correctly assessed w ith a numeric classifier for 37 of 44 quadrants, for an accuracy of 84 %, a specificity of 79%, and a sensitivity of 90%. Although ROI size a nd shape consistency in longitudinal studies are important in radiomet ric analyses, small variations had minimal impact in this study.