| Liver cancer is a malignant tumor of the liver with high morbidity and mortality.Liver cancer usually evolves from hepatitis and cirrhosis,is easily confused with cirrhosis,and the disease progresses rapidly.Therefore,to improve the diagnosis rate of liver cancer in the background of liver cirrhosis is of great significance to improve the patient’s productivity and reduce the mortality.This article mainly carries out the following aspects of research:(1)Case collection and liver image segmentation.Image data was collected from the Imaging Department of the Second Affiliated Hospital of Xuzhou Medical University.Total MRI images of 120 subjects were collected from July 2017 to December 2019,with 60 images of liver cancer and 60 of cirrhosis.After preprocessing the image by denoising and enhancement,the liver on the MRI image was divided using U-net and FCN networks.The results show that the segmentation accuracy of the FCN segmentation algorithm is 81.45%,and the U-net segmentation algorithm is up to 85.77%.The result accuracy of U-net segmentation algorithm was higher.60 images of liver cancer and 60 cirrhosis images segmented by U-net algorithm were selected as samples for subsequent classification research.(2)Classification study of liver cancer and cirrhosis based on texture characteristics and SVM.In the thesis,the gray-scale symbiosis matrix is used to extract energy,entropy,contrast,correlation,and inverse variance as the texture eigenvalues,and SVM acts as a classifier to study the classification of liver cancer and cirrhosis.Nuclear function is the core of SVM classification recognition,and the choice of kernel function determines the accuracy of the classification.Combinatorial kernel functions are designed based on the Gaussian and Polynomial kernel functions,and the parameters are optimized.The results show that the improved classifier recognition accuracy is 83.7%,80% improved over the classification accuracy of single kernel functions,but the classification recognition needs to be further improved.(3)Study on liver cancer and cirrhosis based on a convolutional neural network.Based on the already extracted texture eigenvalues,use VGG16 convolutional neural network for recognition,compare with Alexnet for deep semantic feature full convolutional neural network,establish a confusion matrix,and evaluate the results.The results show that the overall accuracy of the VGG16 model is 93.3% and the Alexnet model is 96.7%.Comparcomparative analysis of two classical convolutional neural classification results identify Alexnet network.In summary,through the research of this topic,the classification of liver cancer and liver cirrhosis images is initially realized,which can provide doctors with auxiliary diagnosis means and help doctors to detect liver cancer and conduct timely treatment,which is of great significance to reduce the mortality of patients.The paper has 47 drawings and 66 references. |
