Study On Image-assisted Diagnosis Of Periodontitis Based On Convolutional Neural Network

BackgroundPeriodontitis is a microbe-related,host-mediated inflammatory disease that leads to the loss of periodontal support tissue(gingival,periodontal membrane,alveolar bone,cementum),causing pathological absorption of alveolar bone,and is the main cause of tooth loss.Between 20% and 50% of the world’s population suffers from varying degrees of periodontitis and is closely associated with the development and development of multiple systemic diseases.The early detection and diagnosis of periodontitis is crucial for the effective treatment of diseases,but the current traditional diagnostic methods mainly rely on periodontal clinical examination and manual interpretation of oral X-ray images,and faced with the disadvantages of strong technical sensitivity,examination time-consuming and difficult popularization at the grass-roots level.Convolutional neural network(CNN)is a multi-layer artificial neural network with convolutional structure,which shows high accuracy in computer vision,image recognition and feature extraction,and has been widely used in image segmentation and auxiliary diagnosis of medical imaging.There are also relevant studies in the oral field,but these studies have small sample sizes,no multicenter testing,and lack further clinical data to verify the reliability of this method.Therefore,this study aims to develop a convolutional neural network-based intelligent detection model for periodontitis to assist dentists in the rapid and accurate diagnosis of periodontitis,and to evaluate the accuracy and potential use of this method in early screening of periodontitis.MethodsTo detect and collect the panoramic radiographs(PARs),periapical radiographs(PERs)and related clinical data from periodontitis cases and healthy controls in the Department of Stomatology of the Second Affiliated Hospital and Affiliated Hospital of Jiangxi University of Traditional Chinese Medicine,Comprehensive convolutional neural network model was designed in two steps.The first step was to train PAR using deep CNN algorithm,establish PAR-CNN model,obtain PAR-CNN prediction score,and preliminarily determine potential periodontitis cases.The second step trains the PER from the same individual using the deep CNN algorithm and support vector machine(SVM)algorithm to establish the PER-CNN+SVM model,obtain the PER-CNN+SVM prediction score,and confirm periodontitis cases.This study applied these two steps to make an accurate diagnosis of periodontitis.Model performance was evaluated by calculating sensitivity,specificity,receiver operating characteristics(ROC)curve,area under the ROC curve(AUC),confusion matrix and95% confidence intervals(CIs).Heat map clustering analysis revealed a correlation between PER-CNN prediction score and clinical characteristics(including sex,age,smoking,hypertension,diabetes,and hereditary periodontitis disease)and the severity of periodontitis.ResultsIn this study,2,275 panoramic radiographs and 11,910 periapical radiographs(PERs)were collected from 3194 individuals.When 1924 PAR(training set 1,276,validation set 376,testing set 272)from Diagnosis and Treatment Center of the Second Affiliated Hospital of Nanchang University were used to train and evaluate the CNN model,the AUC of PAR-CNN model performance was 0.843(85%CIs,80.3%-87.8%),and the AUC of PAR-CNN model performance was 0.793(95%CIs,74.7%-83.4%).Similarly,using 11,910(training set 9,190,testing set 2,720)from Diagnosis and Treatment Center of the Second Affiliated Hospital of Nanchang University were used to train and evaluate the CNN + SVM model,the AUC of PER-CNN+SVM model performance was 0.977(95%CIs,92.3%-99.7%).Finally,the comprehensive modeling performance of the two models is as follows: sensitivity90.4%,specificity 93.1%,and accuracy 91.5%.PER-CNN prediction scores were significantly associated with age and smoking by heat map clustering analysis.ConclusionsThis study presents an image-assisted diagnosis model based on convolutional neural network,and the results show that the model can quickly and accurately diagnose periodontitis by identifying panoramic radiograph and periapical radiograph.Using convolutional neural network algorithm to assist oral clinicians in diagnose periodontitis can greatly reduce the diagnostic workload of oral clinicians.