Research On The Segmentation Of Tooth CT Image Data

As the development of modern society,medical technology is being much more important,which results in larger amount of images of CT and MRI,making it impossible to manually perform medical images segmentation.As a result,applying computer technology into automatically segment medical images draws significant attention from researchers.Including traditional technology and neural-network-based semiautomatic segmentation methods and 3D convolution neural network segmentation means,this paper treats CT images of teeth as starting points and finally implements a medical image segmentation algorithm evaluation platform.To our knowledge,neural network,with outstanding performance,has been widely applied into practice.However,the big amount of training data it requires makes it improper for medical images,which fails to provide sufficient amount of data and is too time-consuming for manual annotation.All the reasons mentioned above exacerbate overfitting,make the model can be efficient for the data different from training set and thus delay the development of deep learning for medical images.Combining with curve evolution theory,this paper utilizes level set algorithm to converge the initialization into image margins,which outputs as the annotation to achieve semiautomatic annotation.In this paper,the principle of level set algorithm and the theory of curve evolution are used to make the initialization curve converge to the image boundary and output as an image label.To realize semi-automatic labeling,a neural network is trained and applied to complete the segmentation;aiming at the problem that medical images are difficult to locate during training,a 3D convolution network for segmentation is proposed,which uses the distance data between CT image sequence sets to construct 3D information to achieve the segmentation of medical images in 3D space;in order to evaluate our method,an evaluation platform is designed to compare our method with popular algorithms and perform analysis.Based on the data provided by West China Hospital,we propose a novel segmentation method and verify the effectiveness and advantages of the algorithm through extensive experiments and data analysis.The main content of this article is as follows.1.Using the edge contour generated by the Level Set algorithm and the Curve Evolution theory,the obtained edge contour is mixed with the manually labeled labels to participate in the training of the neural network,which reduces the manual operation without reducing the accuracy of network segmentation.2.This paper proposes a neural network segmentation algorithm that fuses 3D features.Analyzing the advantages of 3D features on medical images,it builds a 3D convolution neural network model,uses residual ideas to repair the disappearance of gradients that occur as the number of network layers increases and establishes a bridge channel to supplement high-dimensional information,which solves the problem that local information of 2D neural networks cannot be accurately located.3.Design and implement a medical image segmentation algorithm platform.The platform can compare the accuracy of input network models and generate comparison charts of effects.It also has a variety of popular medical image segmentation algorithms built in and allows users to upload their own network models to participate in the comparison.