Medical Image Classification And Segmentation Based On Convolutional Neural Network

As we all know,malignant tumor is a global disease which is difficult to treat and has become one of the major diseases that threaten human life.Early accurate diagnosis,localization,classification and segmentation of tumors is very important to subsequent treatment.With the development of imaging technology,medical images have become an indispensable part of disease screening,early diagnosis,treatment selection and prognosis evaluation.However,due to the current limitations of medical resources,the classification and segmentation task is also a huge burden for doctors.In addition,manual segmentation also has certain subjectivity,which is not conducive to the continuous development of cancer researchIn recent years,with the development of machine learning,convolutional neural net-works have become an effective way of computer processing images.The use of existing resources and related deep learning techniques to process tumor images is of great sig-nificance for effectively reducing the workload of doctors,improving the efficiency of medical treatment,and alleviating the shortage of medical resourcesThis paper proposes a classification and segmentation method based on convolu-tional neural networks for medical images.For the classification of medical images,gated multi-modal convolutional neural network and its compressed sub-model after distillation learning are proposed.This model can extract all information from three directions of the sagittal axis,vertical axis and coronal axis.And the ifusion method can help model to obtain effective classification result of benign and malignant tumors.For medical image segmentation,this paper proposes two automatic segmentation methods based on various innovative structures,one is pseudo 3D-Unet,and the other is parallel dense convolu-tional neural network.Both networks can process three-dimensional medical images with different slice numbers and obtain better segmentation results.Finally,in this paper,the gated multi-modal convolutional neural network and the compressed sub-model after distillation learning are fully tested on the public dataset BraTS 2015.The pseudo 3D-UNet was evaluated on the chest CT image datasets used in real radiotherapy.The performance of the parallel dense convolutional neural network was evaluated on the BraTS 2015 and BraTS 2017 public datasets.The results of all experiments fully verified the effectiveness and superiority of the proposed methods.