An Automatic Diagnose Method for Radiographic Bone Loss and Periodontitis Staging Using Deep Learning

In this study, a deep learning hybrid framework was developed to automatically stage periodontitis in dental panoramic radiographs. The framework was proposed to automatically quantify the periodontal bone loss and classify periodontitis for each individual tooth into four stages according to the criteria that was proposed at the 2017 World Workshop. Radiographic bone level (or CEJ level) was detected using deep learning with a simple structure of the entire jaw in panoramic radiographs. Next, the percent ratio analysis of the radiographic bone loss combined the tooth long-axis with periodontal bone and CEJ levels. The percentage ratios can be used to automatically classify periodontal bone loss. Additionally, the number of missing teeth was quantified by detecting the position of the missing teeth in the panoramic radiographs. A multi-device study was also performed to verify the generality of the developed method. The mean absolute difference (MAD) between periodontitis stages by the automatic method and by the radiologists was 0.31 overall for all the teeth in the whole jaw. The MADs for the images from the multiple devices were 0.25, 0.34, and 0.35 for devices 1, 2, and 3, respectively. The developed method had a high accuracy, reliability, and generality when automatically diagnosing periodontal bone loss and the staging of periodontitis by the multi-device study.