Research And Application Of Intelligent Orthodo Ntic Detection Model Based On Deep Learning

The demand for orthodontic treatment in people’s lives is increasing.Cephalometric measurement is a critical step in the orthodontic process.Through cephalometric measurement,we can understand the current facial morphology of patients.The traditional measurement method is carried out by manual marking cephalometric landmarks by expert doctors,requiring doctors and related technicians to invest a lot of time and energy.In this thesis,the deep convolution neural network method is used for head shadow detection,and an automatic detection,measurement and treatment planning process is proposed.The main work is as follows:（1）This thesis studies the detection and measurement of landmarks in head skeleton X-ray images,and proposes a head skeleton X-ray images.It proposes a head skeleton X-ray image measurement and analysis model based on whale optimization algorithm and improvement of Faster R-CNN.In this thesis,the marker points of the X-ray image of the head bone are measured and analyzed.The critical measurement point values of the patient are automatically generated.Firstly,the problem of less information in the coordinates of head shadow marker points is solved by expanding points into faces.Then,the Faster R-CNN detection model of head shadow detection is built.Finally,in the Faster R-CNN framework model,the improved whale optimization algorithm is used to establish an optimization model of learning rate parameters and loss function parameters to search for the best effect parameter settings.（2）The ISBI challenge data and the patient data of a plastic surgery hospital were selected as the training and test data.The image data was input through the network in three modes: original image,high contrast,and low contrast,and the input channels changed from from 3 to 9 channels.Thus it can clearly distinguish the characteristics of different landmarks in cephalometric measurements during the orthodontic process.The detection success rate within the error distance is selected to evaluate the algorithm.The error distance is set at four levels: within 1mm,within2 mm,within 3mm,and 4mm.The experimental test shows that the improved Faster R-CNN has a better effect than the original Faster R-CNN.At the same time,it proposes to use the whale optimization algorithm to search for the best coordinate point in the target frame,which improves the detection success rate within 1mm and makes the position of the marker point more accurate.The final detection results are compared with the existing cephalometric marker detection model,and the improved algorithm in this thesis has good results.（3）The head shadow detection algorithm is applied to the intelligent orthodontic system,and the medical staff can import the head shadow frontal film or lateral film.The system will automatically display the mark point by operating the automatic AI fixed point.At the same time,the corresponding aesthetic measurement indexes are given according to the definition,the aesthetic measurement index report is automatically generated,and the related treatment scheme is given.The doctor can modify and adjust the automatically generated marker points,messages and strategies.This thesis introduces the design and implementation process of the intelligent orthodontic system in detail from the aspects of demand analysis,overall architecture design,database design,sequence diagram,class diagram and implementation effect.In addition,the cephalometric data generated by the intelligent orthodontic system is used for the continuous training of the detection model.