Medical CT Image Segmentation Based On Deep Learning

US President Barack Obama,in his State of the Union Address in 2015,put forward the "Precision Medical Program",which opened a new era of medical research.Accurate medical treatment focuses not on "medical treatment" but on "accuracy".Undoubtedly,medical imaging technology,especially medical CT image segmentation technology,will provide a reliable guarantee for "accuracy".Of course,there are many kinds of medical images,including ultrasound,CT,MRI,PET,SPECT and so on.Medical CT image,commonly known as medical CT image,sometimes referred to as image.This paper refers to medical image,especially medical CT image;medical image segmentation algorithm,especially medical CT image segmentation algorithm.Of course,the traditional medical CT images because the human body contains biological tissue information,image acquisition time and space complexity(spiral CT),so not only for the CT image segmentation technology is difficult,but also on the quality of CT image requirements.Especially,under the abnormal circumstances such as serious pathological changes of human tissues,prone to false segmentation of medical CT images.Medical CT image segmentation technology based on deep learning integrates the advantages of traditional image segmentation technology.It has high efficiency even in abnormal circumstances,and saves the time and space cost of the algorithm.According to the deep convolution neural network(DCNN),optimizing or improving the existing neural network model can effectively segment the human tissue of medical CT images,making it more robust and generalization.The research work of this paper is based on the project of National Natural Science Youth Fund(item number: 61401049),which was completed in Shenzhen Anke High-tech Co.,Ltd.Nanjing Branch(joint training for two years).On the basis of extensive research and in-depth reading of domestic and foreign literature,this paper summarizes the basic theory of commonly used medical CT image segmentation technology and the tools for making segmentation labels,and completes the lung and bone labeling algorithm.By summarizing the commonly used segmentation algorithms of traditional medical CT images,including two-dimensional and three-dimensional,several commonly used network models and structures of medical CT images are studied and analyzed.Full-convolution network(U-Net),including two-dimensional and three-dimensional U-Net,is emphatically studied.In view of the relatively simple structure of U-Net network,which results in rough feature extraction of CT images,a 2.5D ResUnet network is studied by combining residual network(ResNet)to improve U-Net network.For this reason,a combination of traditional algorithm and deep learning method is adopted.The traditional algorithm is used to label and modify medical tissue,and the deep convolution neural network is used to segment human biological tissue in medical CT images.Based on this,research on human biological tissue segmentation based on medical CT images was carried out,including lung,skeletal and cerebral hemorrhage.With the help of post-processing technology after three-dimensional morphology segmentation of human biological tissue,more accurate segmentation of lung,skeletal and cerebral hemorrhage was achieved.The Dice coefficients were 98.39%,91.40%,89.69% respectively,which can meet the clinical needs of medicine.