Computer-Assisted Densitometric Image Analysis of Digital Subtraction Images: In Vivo Error of the Method and Effect of Thresholding

Abstract

THE AIM OF THE PRESENT STUDY was to assess the in vivo error of the method as well as the effect of thresholding when obtaining and evaluating standardized periapical radiographs for computer-assisted densitometric image analysis (CADIA). Twenty healthy volunteers participated in an experimental gingivitis study in which neither mechanical nor chemical plaque control was performed for 21 days. Two pairs of standardized periapical radiographs were taken at days O (baseline) and 21 (follow-up), one from a maxillary area (15 volunteers) and one from a mandibular molar/premolar area (17 volunteers). Each baseline radiograph was digitized and its image displayed on a monitor. The follow-up radiograph was then superimposed and digitized as well. After gray level correction, subtraction radiographic images were produced. The difference in gray level between the baseline and the follow-up image was calculated within each region of interest (ROI) at each picture point (pixel). In bone ROI, changes in density reflected the amount of change due to methodological errors plus the basic bone remodeling over 3 weeks. For gingival ROI, changes in density reflected the methodological error plus a possible change in soft tissue density during the experimental gingivitis. Within all of the ROI, some pixels indicated a change in gray level. A change in gray level was then thresholded; i.e., only changes >5 and then >10 gray levels were registered and used for calculation of the CADIA values. With a threshold of 5, 44/45 maxillary bone ROI and 60/66 mandibular bone ROI showed a change in density, while 41/45 maxillary gingiva ROI and 26/66 mandibular gingiva ROI indicated a change in density. With a threshold of 10, 16/45 maxillary bone ROI and 12/66 mandibular bone ROI indicated a change in density, while 13/45 maxillary gingiva ROI and 1/66 mandibular gingiva ROI indicated a change. The amounts of changes in density calculated in the various ROI were low even when applying no threshold, ranging from -0.279 to 0.621. Applying a threshold of 5, the CADIA values ranged from -0.234 to 0.727. With a threshold of 10, the changes in density ranged from -0.318 to 0.133. In vivo, CADIA of standardized radiographs indicated change in density due to methodological errors. Application of thresholds may avoid false-positive diagnoses. When applying CADIA in clinical research, the range of change to be expected due to methodological limitations as well as the threshold for true change should be evaluated. These thresholds may differ in various areas of the mouth, i.e., bone or gingival, maxillary/mandibular, anterior/ posterior ROI.