Accurate Artificial Intelligence-Aided Diagnosis of Periodontitis Based on Radiographic Images

Objectives: This study aimed to use a deep convolutional neural network to calculate alveolar bone level and assign periodontal diagnosis based on the radiographic images.
Methods: An end-to-end deep learning network was developed by integrating three segmentation networks (bone area, cementoenamel junction (CEJ), tooth) and image analysis to assign periodontitis stages from intraoral radiographic images. Among 700 periapical radiographic images, 80% of images were used for training and 20% were used for model validation and testing. U-Net architecture was used as the backbone of the models but the encoder was replaced with ResNet-34. Several image processing techniques such as Gaussian filtering for noise removal, contour detection, and window sliding were employed to improve the output of the segmentation models and extract each individual tooth as well as its corresponding bone area and CEJ line. The percentage of bone loss for each tooth was calculated by measuring the root length and the distance between bone level and CEJ level. The stage of bone level for each tooth and periodontal diagnosis of the whole dentition were assigned based on the the 2018 periodontitis classification. Alveolar bone level measurement and bone level staging were confirmed by three independent and calibrated examiners.
Results: The average dice coefficients for segmenting bone area and tooth were 0.96 and 0.94 respectively and pixel accuracy for CEJ line segmentation was 0.9966. The average Area Under the Receiver Operating Characteristics Curve (AUC-ROC) for periodontal stage assignment was 0.97. The accuracy of case diagnosis is 90%. Paired T-Test confirmed that there is no significant difference in bone level measurement between the model and examiners (p=0.42).
Conclusions: The deep learning network provides an accurate and reliable periodontal diagnosis based on periapical radiographic images. The artificial intelligence-aided diagnosis is helpful when comprehensive periodontal examination is not available.