Research on Feature Extraction of Tumor Image Based on Convolutional Neural Network

Medical images play a very important role in making the right diagnosis for the doctor and in the patient's treatment process. Using intelligent algorithms makes it possible to quickly distinguish the lesions of medical images, and it is especially important to extract features from images. Many studies have integrated various algorithms into medical images. For medical image feature extraction, a large amount of data is analyzed to obtain processing results, helping doctors to make more accurate case diagnosis. In view of this, this paper takes tumor images as the research object, and first performs local binary pattern feature extraction of the tumor image by rotation invariance. As the image shifts and the rotation changes, the image is stationary relative to the coordinate system. The method can accurately describe the texture features of the shallow layer of the tumor image, thereby enhancing the robustness of the image region description. Focusing on image feature extraction based on convolutional neural network (CNN), the basic framework of CNN is built. In order to break the limitations of machine vision and human vision, the research is extended to multi-channel input CNN for image feature extraction. Two convolution models of Xception and Dense Net are built to improve the accuracy of the CNN algorithm. It can be seen from the experimental results that the CNN algorithm shows high accuracy in tumor image feature extraction. In this paper, the CNN algorithm is compared with several classical algorithms in the local binary mode. The CNN algorithm has more accurate feature extraction ability for tumor CT images on a larger data basis. Furthermore, the advantages of CNN algorithms in this field are demonstrated.