| With the development of society and the improvement of people’s living standards,people’s oral health awareness has also been greatly improved.In addition to basic oral problems such as teeth and periodontal problems,the arrangement and occlusion of teeth and the beauty and coordination of face and face have also been greatly improved.It has also attracted more and more people’s attention,and the orthodontic industry is in a stage of rapid development.In traditional orthodontic clinical diagnosis and treatment,there are a lot of manual operations,such as plaster model making,cephalometric analysis,bracket bonding and positioning,etc.These operations not only take up valuable clinical time of orthodontists,but also the accuracy and efficiency of clinical diagnosis and treatment may be affected by the existence of human factors.In recent years,the rapid development of digital technology and artificial intelligence has also gradually changed the clinical diagnosis and treatment of orthodontics.From the acquisition and storage of case data,to the tools and methods of diagnosis and analysis,as well as the simulation and implementation of treatment plans,they have become more efficient,comprehensive and accurate.Orthodontic clinics have entered the digital era of precision.The digital dental model is an important part of the digital diagnosis and treatment work.The simulation of the orthodontic plan,the indirect bonding of the brackets,the design of the personalized appliance,and the production of the orthognathic surgical guide are all inseparable from the three-dimensional information of the patient’s dentition.Tooth segmentation is the most critical step in model processing before the digital model is put into application.The result of tooth segmentation will directly affect the effect and accuracy of subsequent orthodontic diagnosis and treatment related to the digital jaw model.At present,the methods of tooth segmentation mainly include segmentation based on human-computer interaction and automatic segmentation based on artificial intelligence deep learning network.Tooth segmentation based on human-computer interaction is the most used method in clinical practice.It is based on various virtual orthodontic systems or orthodontic auxiliary software.The adjacent areas of teeth and gingival contours are marked in the software through manual operations.After the calculation,the segmentation of the teeth on the digital model is completed.However,due to the limited observation field setting in the virtual orthodontic system or auxiliary software,the logical setting of plane cutting,and the incomplete reference of tooth parameters in human-computer interaction,the shape of the segmentation plane of the teeth will be damaged to varying degrees.There is a large difference between the results of manual segmentation and the actual anatomical morphology.Fully automatic tooth segmentation based on deep learning belongs to a branch of 3D image segmentation.Although it can automatically complete the task of tooth segmentation without relying on manual operations,the medical 3D image itself has a huge amount of data and a complex learning network,and the complex individual differences faced by tooth segmentation will make the realization of fully automatic tooth segmentation difficult.Some existing studies have successfully built network models for automatic tooth segmentation using different processing methods and data types,but the accuracy and effect of their segmentation cannot well meet the clinical requirements.Therefore,in the clinical application of the existing digital dental and jaw models,inaccurate design and operation of clinical proximal deglazing and unreasonable digital tooth arrangement results often occur due to inaccurate tooth segmentation and destruction of the natural tooth shape.The precise advantages of digital orthodontics have not been fully exerted.Based on the above background,this study focuses on the problem of tooth segmentation in digital orthodontics.This study proposes improvements to the existing manual tooth segmentation methods based on human-computer interaction and automatic tooth segmentation methods based on deep learning,and applies the improved methods to the measurement of dental and jaw models and compares the results.By improving the segmentation method,the accuracy of tooth segmentation can be further improved,thereby improving the accuracy and efficiency of orthodontic clinical digital diagnosis and treatment,and providing a certain method reference for optimizing the accuracy of model analysis and orthodontic plan simulation.This study proposes an improvement to the tooth segmentation method of human-computer interaction.The long axis parameters of the crown and the tooth including the root are added to the interactive operation,and the dentition is aligned before and after orthodontic treatment.The application and comparison of the conventional dental and jaw model and the crown-root composite model were carried out.Through the quantitative measurement results of the segmentation results and the visual deviation analysis results,it was concluded that the improved human-computer interaction segmentation method could obtain more refined results compared with the traditional method.The results of tooth segmentation have less damage to the tooth shape than the original method.In order to make full use of the feature that mesh data has a strong ability to describe 3D images,this study also improved the Mesh segmentation network based on deep learning.By taking the original surface properties of multiple tooth models as input,using the graph constraint learning module to learn multi-scale local context features hierarchically,and concatenating geometric features from local to global through a dense fusion strategy,we finally successfully built a Fully automated deep learning network for tooth segmentation.The DSC coefficient of the network is 0.88,the sensitivity is 0.90,the positive prediction rate is 0.91,and the segmentation accuracy is improved compared with the traditional Mesh segmentation network.In the construction of deep learning network,through the improvement of feature change module,graph constraint learning module,data pooling and fusion concatenation,this paper extracts the semantic features from local to global relatively more completely,so that the generalization ability and accuracy of the whole model are improved.Moreover,the research also paid attention to the importance of the labeling of the dividing line of the adjacent area of the teeth and the labeling of the tooth surface below the interproximal space at the receding gingival papilla when the training set model was labeled.Subsequently,this study applied the improved human-computer interaction tooth segmentation method and the deep learning automatic tooth segmentation method to the measurement of dental and jaw models,compared the differences in the measurement results of different segmentation methods,and analyzed the source of the differences.When the measurement object is a model with neatly arranged teeth,the improved human-computer interaction and deep learning segmentation methods show no difference in the measurement results.When the measurement object is a model with crowded teeth,the improved human-computer interaction of the long axis parameters of the teeth is considered The segmentation method yields more accurate and stable measurements.To sum up,it can be seen that the improved tooth segmentation methods based on human-computer interaction and deep learning in this study have achieved better segmentation results than the methods before the improvement.In the measurement application of the improved method,the shape and arrangement of teeth will affect the The measurement results have a certain impact.In clinical practice,an appropriate tooth segmentation method should be selected according to the actual data type,tooth shape and tooth arrangement,so as to carry out digital orthodontic diagnosis and treatment more accurately. |
