Segmentation Of Hepatic Veins Based On Deep Convolution Neural Network

The small size and the extremely complicated branching structure of the blood vessels,so the automatic segmentation of blood vessels is still a huge challenge for medical image segmentation.Traditional blood vessel segmentation algorithms include threshold method,region growing method and morphological method.These algorithms require artificial participation,and have problems with blood vessel segmentation with inconspicuous contrast or disease,and the segmentation effect is not good.Medical image segmentation based on neural network does not need to artificially design feature quantity,and can automatically learn image features from a large number of sample images to achieve image segmentation.At present,the segmentation of fundus blood vessels,cerebrovascular vessels and pulmonaryblood vessels in medical images is realized,but the contrast between blood vessels and organs in the liver is not obvious,and the segmentation is difficult,so few people use them for liver vein segmentation.Therefore,this paper Try to learn the characteristics of blood vessels automatically by deep learning,and segment the liver veins in CT images.The two-dimensional convolutional neural network based vessel segmentation algorithm proposed in this paper has a good segmentation effect on hepatic veins,but the two-dimensional convolutional neural network loses the three-dimensional information of CT images and affects the segmentation accuracy.In order to improve the accuracy of vascular segmentation,a three-dimensional convolutional neural network(3D-CNN)was designed based on two-dimensional convolutional neural network,which was applied to the segmentation of hepatic vesselsThe blood vessel segmentation algorithm based on deep convolutional neural network proposed in this paper.Try to use different sizes of data blocks as input,and acquire the characteristics of images through multi-layer convolutional neural networks,finally realize automatic segmentation of blood vessels.Specific studies on hepatic venous vascular segmentation from the following modules:(1)Firstly,the gradation transformation is used to increase the brightness and contrast of the CT image.In the cumbersome process of initial sample marking,the threshold method and the 3D Labeling method are used to reduce the workload and save a large part of the time and effort.Finally,under the guidance of a professional physician,combined with manual labeling of blood vessel samples,the collection of blood vessel gold standards was completed.(2)A blood vessel segmentation algorithm based on two-dimensional convolutional neural network.By inputting image blocks of different sizes,the structure and parameters of the network are continuously adjusted,and the data processing is processed by the deep learning framework TensorFlow developed by Google.Finally,the model obtained from the training of different size image blocks is tested.It is concluded that when the input image block is 25*25,the Dice coefficient,accuracy rate and recall rate of the experimental results of blood vessel segmentation are better,and DSC can reach 0.9031,the accuracy rate was 0.8161,and the recall rate was 0.9989.(3)The algorithm of deep convolutional neural network proposed in Chapter three is limited to two-dimensional.Considering the particularity of CT image,a blood vessel segmentation algorithm based on three-dimensional convolutional neural network.Finally,the Dice similarity coefficient of the algorithm based on three-dimensional convolutional neural network for hepatic vein segmentation was increased to 0.9243,the accuracy rate was 0.8611,and the recall rate was 0.9976.The experimental results show the feasibility of the algorithm of the three-dimensional convolutional neural network,and better experimental results than the two-dimensional input data can be obtained.