Tumor Classification Of Liver CT Images Based On Deep Learning

Liver cancer has become one of the main causes of cancer death worldwide due to its high morbidity and high mortality.The detection and treatment of early liver cancer is an important way to effectively reduce liver cancer mortality.In recent years,medical image analysis has played an increasingly important role in the diagnosis of liver diseases.Medical image classification is an important research direction in medical image analysis.It is used in the pathological analysis of liver diseases,clinical diagnosis,surgical dynamic planning and computer-aided Medical treatment and other aspects have wide application value and research significance.In the research of medical image classification in recent years,deep learning has shown better performance and higher efficiency than traditional machine learning methods.Deep learning makes it easier to build classification models,integrate feature extraction and classification recognition,and reduces the complex feature extraction process.Therefore,this article combines deep learning and medical image recognition technology to carry out research on CT image liver tumor detection and recognition.The specific research contents are as follows:1)Clinically,individual liver tumors are quite different,and the gray-scale contrast with the surrounding tissues is low,and their recognition accuracy is often not high.Deep learning model training requires a lot of data to drive,and in reality,there are often few labeled data that can be used for training.Deep learning models often cannot perform effective feature learning,and will cause serious overfitting problems.In response to the above problems,this paper proposes a liver tumor classification method based on feature fusion adversarial learning network.This method first performs autoencoder training,and then embeds the trained autoencoder encoder into the ASENet model as a feature extractor,where the encoding The device is only used for the extraction of deep features.Finally,the extracted features are passed into the classification model to be fused with the corresponding convolutional layer features and the corresponding deep features are confronted with learning training,so as to achieve effective classification of liver tumors.2)Only a small part of CT images that include liver tumors can be labeled by doctors,and the large amount of liver tumor data usually cannot be effectively used.The small amount of labeled liver tumor data in the study usually cannot meet the learning and training needs of deep models.Therefore,this paper proposes a liver tumor classification method based on semi-supervised multi-scale deep feature integration network.The method consists of a U-Net model and a SENet classification model.The U-Net model can use a large number of clinically unlabeled liver tumor data for unsupervised training,and the learned features can be used to assist the classification model in supervised classification training..At the same time,the classification model can use the multi-scale depth features from the U-Net model for deep feature integration classification training,thereby improving the classification effect of liver tumors.Experiments show that the proposed method can effectively improve the classification effect of liver tumors.