Multi-feature Fusion Classification Algorithm For Hepatocellular Carcinoma Based On Non-local Neural Network

Hepatocellular carcinoma(HCC)is ranked as the sixth most common cancer in the world and the third leading cause of cancer death.With the development of medical imaging technology,computer-aided diagnosis technology can fully analyze the nuclear magnetic resonance imaging of hepatocellular carcinoma.Through the analysis and synthesis of a large amount of data,and the rapid calculation of existing highperformance computers,the pathological benignity and malignancy of patients with hepatocellular carcinoma can be automatically classified,which not only reduces the workload of doctors,but also improves the efficiency of disease diagnosis and the survival rate of patients with hepatocellular carcinoma.In this paper,the magnetic resonance imaging data of hepatocellular carcinoma in a hospital in Guangdong Province was used as the research object.Fusion algorithm of Location non-local channel and Channel nonlocal features is completed.Local depth feature and fusion algorithm of local depth feature and non-local depth feature is designed.The main research work of this paper is as follows:(1)The extraction and classification experiments of local features of hepatocellular carcinoma images is completed.Based on the analysis of the magnetic resonance imaging characteristics of hepatocellular carcinoma,the convolutional neural network was used to extract the depth features of the axial,coronal and dysplastic images of hepatocellular carcinoma,and the extracted features were used as the local features of this paper,and used for the classification of benign and malignant hepatocellular carcinoma.Compared with the traditional texture feature classification results,the deep learning method has better classification effect.(2)The non-local features of the magnetic resonance imaging of hepatocellular carcinoma and the extraction of non-local features of the channel and related classification experiments were completed.Based on the idea of non-local mean,two kinds of non-local depth features were extracted from hepatocellular carcinoma(HCC)images by using location non-local module and channel non-local module.The spatial location dependence and channel dependence of HCC feature maps were captured.A comparative experiment of two kinds of non-local features in benign and malignant classification of HCC was completed.The experimental results show that the non-local features are more accurate than the local features,and are more suitable for the classification of benign and malignant hepatocellular carcinoma.(3)The design and implementation of fusion algorithm of location non-local feature and channel non-local feature are completed.Based on the separate experiments of the local location and the non-local features of the channel,two non-local features are fused for classification.The experimental results show that the classification effect of the fusion of the two non-local features is better than the single location non-local feature or the channel non-local feature.(4)A classification algorithm combining local and non-local features is designed.Based on the separate classification experiments of local and non-local features,the two features are fused.The algorithm comparison analysis of multi-view feature fusion method and multi-feature fusion method is completed.The experimental results show that the fusion of local and non-local features is effective in the classification of benign and malignant hepatocellular carcinoma.