| Summary: | 碩士 === 靜宜大學 === 資訊工程學系 === 106 === Proximal caries are easily neglected when examining in naked eyes, as they tend to occur at slit between teeth. Thus, X-ray images become very helpful for diagnosing. Because vast amount of dental radiographs are needed to be examined each day, they may be read incorrectly or inconsistently. Thus, automatic caries detection can assist dentists for better diagnosis.
Our proposed proximal caries detection method includes four major steps. In the first step, it applies thresholding to roughly separate the background from tooth and uses horizontal projection to find the approximate gum line. The parts below the gum line are then removed and the rest parts become the ROI for caries detection. The caries indicator-mark- pixels within the ROI, if existed, are then restored by neighborhood processing. In the second step, the top-hat bottom-hat filter is applied to enhance the contrast of tooth boundary and between caries pixels and enamel pixels. Canny edge filter is applied to retrieve coarse tooth boundaries as well as the boundaries between enamel and cementum, then boundary-point continuation and direction are used to fine tune the boundary in the third step. Finally, the average pixel value of ROI is used as the threshold for coarse tooth decay detection. Each preliminary detected decay point is re-examined by comparing its pixel value with the average of its near neighborhood within the enamel.
We conducted an experiment using 32 dental radiograph images with proximal caries on one or both sides of lower enamel to test the effectiveness of our proposed method. The detection results demonstrated that our method can detect all proximal caries marked or visually detectable in the test images with eight false alarms.
|
|---|
