Research On Medical Image Classification And Segmentation Based On Convolutional Neural Network

In recent years,with the popularity of artificial intelligence,deep learning has become the preferred research method in almost all fields,especially in the field of medical images.In the medical field,the use of biomedical imaging in the diagnosis of patients is an indispensable means of treatment.Early researchers used machine learning to process medical images.Although some achievements have been made,there has been no great breakthrough.The emerging deep learning method can learn the hidden disease characteristics from the large data of medical images,and its unique and powerful feature learning ability has rapidly become a hot research topic in the field of medical images.The research of medical image can promote the development of medical cause and improve the accuracy and reliability of medical facilities.Doctors can use computer-aided system to assist diagnosis of patients,reduce the burden of doctors and improve the efficiency of diagnosis.It has very important significance and value for the development of medical field.There are several tasks in the field of medical image,including classification,recognition,location,detection and segmentation,etc.Of which classification and segmentation are two similar tasks,the former is image-level classification,and the latter is pixel-level classification.Image segmentation is a kind of intensive classification task based on pixels,which needs to recognize every object in the image.The ultimate goal of traceable segmentation is classification.The main work of this paper is based on the classification and segmentation of medical images:1.In the task of medical image classification,using the classical pre-trained deep learning model,the new normalization method Switchable Normalization and Targeted dropout algorithm are applied to the model.Then the long-term and short-term memory model and attention mechanism are introduced to enhance the technology in TTA test.Finally,the trained model is integrated and classified.2.In the task of medical image segmentation,a new 3D Inception network structure and a 3D ResNet network structure are designed by using the idea of full convolution neural network on two-dimensional image segmentation task and combining with the main structure of two-dimensional image classification depth learning model.The innovative use of 3D Squeeze and Excation structure and the use of hollow convolution structure in some layers make it suitable for three-dimensional medical image segmentation.In this thesis,ICIAR2018 data set and Brats2015 data set are used to verify the effectiveness of the model designed in this paper.The results show that the model designed in this paper achieves very good results in two different tasks of medical image.The accuracy of 0.92 is achieved on ICIAR2018 data set and 0.86 is achieved on Brats2015 data set,which is obviously superior to other published methods.