Multiscale image analysis for the quantitative evaluation of periapical lesion healings

摘要

Intraoral radiographs are used routinely for the diagnosis of periapical lesions and the evaluation of subsequent lesion treatments. However, a reliable and reproductive evaluation is often difficult due to the subjectiveness of visual interpretation and various imaging factors in routine dental radiography. To aid radiographic evaluation of periapical lesion healing, we have developed a multiscale image analysis method based on the change of bone features. Fifty pairs of pre- and post-treatment intraoral radiographs were retrospectively obtained and classified into successful (25 cases) or failed (25 cases) treatments based on the presence of periapical radiolucency. To accurately segment the bone images, an optimal scale was determined at which measured bone thickness reached 95% of the maximum values. Trabecular bone images were segmented in multiple scales to calculate intensity and skeleton features. Texture features were calculated within the user-traced lesion area. Based on relative differences of these features between pre- and post-treatment, several treatment efficacy models were derived and tested by leave one out cross validation. Intensity, skeleton, and texture features were significantly changed after successful treatment (Wilcoxon's paired test, p<0.05). Treatment efficacy of periapical lesions was classified at the sensitivity of 96% and the specificity of 92% (logistic regression analysis). We found that healing lesions show not only brighter but also larger, thicker, more branched trabecular bones, while cases of persistent disease have bones with more uniform density. Our model may aid the early intervention for re-treatment of failed lesion healings.