Research On Periodontal Image Recognition And Auxiliary Diagnosis Technology Based On Machine Learning

There are more than 80% periodontitis in oral disease,which comes from the statistics of the World Health Organization.However,due to the large population base in China,insufficient attention is paid to periodontal diseases and patients do not seek treatment at an early stage,and the diagnosis is often severe.In addition to clinical manifestations,the combination of image data is used to diagnose oral diseases,where panoramic periodontal images are one of the important methods.With the popularity of applications combining machine learning and medical imaging in clinical diagnosis.This accurate data calculation makes up for the lack of clinical experience to a large extent.This paper is based on the methods of machine learning with the periodontal panoramic image as basic data that to complete the auxiliary diagnosis of periodontitis.This paper contains the following contents mainly:Data collection and collation: Making a preliminary classification of the collected sample data.Because the primary age group of periodontitis patients is adults,children’s periodontal images were screened out,and the elderly were not suitable for periodontitis studies due to the large number of missing teeth.Finally,931 images were identified from the 1247 periodontal images of the original sample data for this study.The FDI tooth number labelling of the periodontal panoramic images was done according to the needs of the experiment.Semantic segmentation of periodontal image based on Res-U-Net: Excellent embodiment of U-Net model in medical image.This paper chooses U-Net model for semantic segmentation,which is based on its excellent embodiment in medical images.To achieve better segmentation,residual structure is added,and the Res-U-Net model is used for semantic segmentation finally.Similarly,in order to improve the correctness of the experimental data,the original data were further equalized,and horizontal flip,vertical flip,horizontal and vertical flips and rotation operations were carried out to increase the number of sample data participating in the experiment.Separation of individual tooth images and calculation of PBL and periodontitis stage determination: On the basis of semantic segmentation,a boundary frame was added for the independent teeth in the periodontal panoramic image,and the individual teeth in the whole mouth were separated and saved.There are 931 panoramic periodontal images that isolate 28,859 individual teeth.on the basis of semantic segmentation,add boundary boxes for individual teeth in the periodontal panoramic image,and complete the separation and preservation of individual teeth in the whole teeth.A total of 28859 independent teeth were separated from 931 periodontal panoramic images.Select one tooth from the teeth separated from each periodontal panoramic image to label CEJ(cemento enamel junction),Bone Level(alveolar ridge top boundary),APEX(root tip)Root,of which single tooth is labeled with 5 points,and multiple teeth are labeled with 6 points.According to the proposed Base-Coord algorithm,the %PBL of the mesial and distal planes of each tooth is calculated.The maximum value of the mesial and distal planes of a single tooth is used as the basis for the diagnosis of periodontitis,and the periodontitis stage of the patient’s teeth is determined.In summary,this paper has mainly accomplished three important tasks: screening and classification of sample data from raw periodontal panoramic images,securing appropriate data for the experiments,and completing the labeling of FDI tooth numbers;The sorted image data is expanded to complete semantic segmentation based on the Res-U-Net model.A bounding box is added to the panoramic periodontal image and label file,individual teeth are cut on this basis,individual teeth are selected for keypoint labeling,the PBL is accurately calculated based on the Base-Coord algorithm,and the stage of diagnosis and disease of periodontitis is determined.