Research On Liver Segmentation Algorithm Based On Convolutional Neural Network

Accurate segmentation of liver medical images is an important prerequisite for assisting doctors in the diagnosis of liver disease patients,forming accurate preoperative planning,intraoperative guidance and postoperative evaluation,and plays an important guiding role in clinical surgery as well.In recent years,although the study of liver segmentation algorithm has achieved great development,it still faces some urgent problems to be solved.Therefore,it is particularly important to study a liver segmentation algorithm that can be clinically applied accurately and with strong applicability.Based on the theory of image processing,pattern recognition and machine learning and aiming at automatic and accurate liver segmentation,this paper focuses on the application of watershed algorithm and deep convolutional neural network in liver segmentation and realizes automatic and accurate segmentation of liver images.In addition,the interactive segmentation of the liver image with complex contour is realized.The main work is summarized as follows:(1)Based on the classic U-net and combined with the watershed algorithm,a liver segmentation algorithm based on watershed correction and U-net is proposed.It mainly solves the problem that the full convolutional neural network easily misses the position information in the liver image data with complex background and blurry liver boundary.This algorithm use U-net will be shallow and deep semantic characteristics of layered study combining advantages,initial liver image segmentation results,and on the basis of combining the characteristics of the watershed algorithm has good response to weak edge.According to the correction conditions The initial segmentation result is judged,and the super-pixel block formed by the watershed is used to correct the initial segmentation result of the liver,and the final liver segmentation result is output.Through the simulation analysis of the algorithm on the 3DIRCADb and Data A datasets.The results show that the algorithm can effectively correct the hole and non-smooth boundaries in the U-net segmentation,and improve the accuracy of liver segmentation.(2)Based on the classic V-net and combined with the geodesic distance interactive segmentation algorithm,An improved interactive liver segmentation algorithm based on geodesic distance and V-net is proposed.It mainly solves the problem that the existing deep convolutional neural network has wrong segmentation when segmenting liver images with complex contours,low contrast and large lesions.This algorithm use three-dimensional segmentation network V-net can effectively take into account the characteristics of spatial context information to segment the liver medical images and obtain the preliminary segmentation results of the liver image.The interactive segmentation method based on geodesic distance has a simple,easy to implement,and responds well to the liver boundary,so in the results of the initial segmentation using this algorithm form artificial hard constraints on the correction,and further the watershed algorithm to form the super pixel piece as sample points is introduced into the algorithm operation,greatly improving the liver segmentation efficiency.The results show that the algorithm can accurately segment the whole liver region and realize the convenient in clinical doctors use less user intervention and shorter time of interactive image segmentation framework liver.(3)Aiming at the interactive liver segmentation method based on geodesic distance proposed in this paper,the GUIDE tool in MATLAB programming software is used to construct the GUI interface,and the visual liver segmentation system based on the interactive liver segmentation method based on geodesic distance is completed.