Research On Cardiac Image Segmentation Model Based On Deep Learning

With the development of computer processing power,artificial intelligence technology is becoming more and more mature and used in medical field to assist doctors in diagnosis and treatment.MRI is an effective method for disease detection,which plays a very important role in the treatment of heart disease.The segmentation of different tissues in cardiac MRI images is a key step in the treatment of heart diseases.However,the task of segmentation of medical images is difficult because of the limitation of medical imaging technology and the characteristics of medical images themselves,such as blurred boundaries,uneven grayscale and image overlap.Therefore,according to the characteristics of cardiac MRI images and clinical practical needs,we propose an automatic segmentation method of cardiac MRI images based on deep learning.The main works are as follows:1.Taking the heart as the research object,a new neural network LU-Net is proposed in this paper for the problems that the current MRI images can not complete the automatic segmentation task and the U-Net is not ideal in heart image segmentation.It is based on UNet and solves the problem that U-Net has blurred boundaries and inaccurate segmentation during heart segmentation.LU-Net adopts the measures of fusion SE-Net module,multiscale input and replacement upsampling,which greatly improves the performance of neural network and realizes better segmentation of cardiac MRI images.2.This paper designs and implements a cardiac MRI image automatic segmentation framework,which applies the proposed LU-Net to the cardiac MRI image automatic segmentation task.The horizontal framework can be divided into two major stages of neural network training and model application.The vertical framework can be divided into three parts: data preprocessing,neural network training and verification,neural network application.This paper introduces each part in turn along the vertical structure of the framework.Firstly,the original cardiac MRI images were preprocessed,including multiple operations such as normalization,image enhancement,median filtering denoising,edge detection,gamma correction,etc.Secondly,the preprocessed data were fed into a neural network for training and validation,and the optimal LU-Net neural network model was preserved.Finally,in the application of the neural network,the optimal model was directly loaded to reconstruct the LU-Net neural network to automatically segment the input image.3.In order to prove that the LU-Net neural network can obtain better segmentation results on cardiac MRI images,we implement FCN,Seg Net,U-Net,IU-Net and LU-Net in the cardiac MRI image automatic segmentation framework and apply them to segmentation tasks for experimental comparison.During the neural network training process,LU-Net achieved higher accuracy than other methods while ensuring fast convergence.We use the optimal models of each neural network to perform cardiac MRI image segmentation tasks and use five commonly used evaluation criteria to measure the segmentation effect.LU-Net surpasses other neural networks in all evaluation criteria.The experiments prove that the LUNet proposed in this paper has higher accuracy and the best segmentation effect.