Research On Multi-organ Segmentation Algorithm Based On Convolutional Neural Network

Image segmentation is a technology and process to extract interested objects from an original image.Medical image segmentation is a fast developing field in recent years,and the most important part is organ segmentation.In this paper,a fully automatic method for the segmentation of four organs(liver,spleen and both kidneys)were discussed.Based on the background and significance about organ segmentation,the first of several common image segmentation methods are discussed,such as image segmentation based on graph theory,image segmentation based on active contour model,image segmentation based on threshold,image segmentation based on region and image segmentation based on deep learning.However,after a careful study of the advantages and disadvantages of these methods,we found that these methods don't usually be used alone in modern practical applications.Instead,people used to use a hybrid model method in order to achieve more precisely image segmentation results.In clinical,CT images with high resolution can clearly show the structure of abdominal organs,and thus provide a great reference value for disease diagnosis,preoperative planning and postoperative treatment.Therefore,image segmentation has a very important clinical value for CT images.Our study of multi-organ segmentation is based on the CT images.In this paper,we proposed a fully automatic method employing deep fully convolutional neural networks(CNNs)for the segmentation of multiple organs.Firstly,a 3D CNN is trained to localize and delineate the targeted organs automatically with a probability prediction map.The learned probability map provides both subject-specific spatial priors and initialization for subsequent fine segmentation.Then,for the further adjustment of the multi-organ segmentation,image intensity models,probability priors and a disjoint region constraint are incorporated into an unified energy functional.Finally,a novel time-implicit multi-phase level-set algorithm is utilized to efficiently optimize the proposed energy functional model.Our method has been used on 140 abdominal CT scans for the segmentation of four organs(liver,spleen and both kidneys).With respect to the ground truth,averageDice overlap ratios for the liver,spleen and both kidneys are 96.0,94.2 and 95.4%,respectively.The achieved accuracy compares well to state-ofthe-art methods with much higher efficiency.The results demonstrated its potential in clinical usage with high effectiveness,robustness and efficiency.