Research Of Liver Segmentation In CT Image Based On Deep Learning

The mortality of liver cancer is very high,and it's seriously endangering human health.China is one of the high incidence of liver cancer.In recent years,with the development of medical imaging technology,computer tomography(CT)has been widely used in the diagnosis of liver diseases,and has become the first choice for the diagnosis of liver disease.Taking use of computer image processing technology,combined with medical imaging diagnosis technology,can make early diagnosis of liver disease,threedimensional modeling and quantitative analysis,which enables doctors to master enough data before surgery to make preoperative planning to improve the success rate of surgery and develop a reasonable effective treatment regimen.Accurate and reliable segmentation of liver from the abdomen CT image is the first step of early diagnosis,three-dimensional modeling and estimation of the size of the liver and the condition of disease.It is also a critical step that directly affects the accuracy and accuracy of subsequent treatment.Clinically,radiologists with rich experience make the segmentation manually,but this process is very time consuming,and by the subjective factors,different radiologists often will get different results.Therefore,in order to reduce the workload of doctors,and improve work efficiency,but also to obtain more objective and accurate segmentation results,it is necessary to introduce computer-aided diagnosis technology.Traditional methods of liver segmentation depend on some of the shallow features of the image,such as gray scale,statistical structure,and texture.These features can be obtained directly from the image,or extracted by a manually designed extraction operator.These features are less robust,less representative,and susceptible to noise.Actually,the abstract and deep features are more representative.Deep learning can extract the deep abstract features of data,and applying it to the segmentation of the liver can improve the accuracy and robustness of segmentation.This paper presents a method of abdominal CT segmentation based on fully convolution neural network.The Alex Net is modified to a fully convolutional neural network,and it was trained by lots of labeled data.In order to fix gradient vanish caused by traditional active function,Re LU was used as active function.For avoiding overfitting,Dropout was imported.However,the results is fairly rough for lacking of detail information in high-level output,we come up with a new architecture for segmenting liver from CT images,which combines low-level features with high-level features,to get more accurate segmentation.Experiments show that our algorithm has better robustness and accuracy,and it is more efficient than Patch-based method.