Research On Segmentation Method Of Liver Tumor Medical Image Based On Convolutional Neural Network

Liver cancer is one of the common cancers.Doctors generally diagnose the condition by analyzing the CT images of the patient's abdomen.Due to the doctor's subjective judgment and professional level,the diagnosis may sometimes lead to deviations.Therefore,a doctor can automatically and accurately segment the liver and the lesion(tumor area)method is particularly important.At present,there are two main methods for deep learning application in medical image segmentation.The first is a segmentation method based on two-dimensional convolutional neural network(CNN).The segmentation accuracy of this method is higher than that of traditional segmentation methods.However,because the biological organs are presented in three dimensions,The two-dimensional segmentation model is easy to ignore the inter-slice information of the cross section in the CT image.The second is a segmentation method based on a three-dimensional CNN.This method will weaken the feature extraction of two-dimensional image slices.In addition,the number of network layers of the 3D segmentation model is not deep enough.These factors will reduce the segmentation accuracy to a certain extent.In view of the above problems,this paper proposes a segmentation method based on hybrid CNN,which uses two convolutional neural networks with different functions to complete the segmentation twice.The main work is as follows.First of all,for the information learning within the layer of the data set,this paper designs a segmentation method based on a two-dimensional CNN,which uses the residual structure and the encoding and decoding structure of the "U" network.This design increases the depth of the network and speeds up the convergence of the network to obtain the first segmented liver area.Then,the cube structure is obtained after the first segmentation result is processed,and then preprocessed with the original data set and sent to the three-dimensional CNN.Finally,for the inter-layer information learning of the data set,this paper designs a segmentation method based on a three-dimensional CNN,which uses residual ideas and dilated convolution theory,This structure can learn the deep features of the image,and get the liver and liver tumor regions for the second segmentation.The main experimental data in this paper is the liver tumor data set Li TS.It is verified by experiments that the segmentation effects of the two-dimensional CNN and thethree-dimensional CNN proposed in this paper are higher than the comparison network of the same dimension.And the segmentation accuracy based on the hybrid CNN is higher than the method that only uses a separate two-dimensional network or three-dimensional network.The segmentation coefficients of liver and liver tumors are 0.962 and 0.634 respectively after the hybrid network is adjusted.