Abdominal Ct Image Segmentation Based On Deep Learning

With the rapid development of deep learning,more and more cross-disciplinary disciplines are combining with each other in order to solve the existing problems more effectively,and medical imaging technology using deep learning methods plays an indispensable role in medical diagnosis.As an important research direction of medical imaging technology,medical image segmentation plays an important role in pathological analysis,clinical diagnosis and the formulation of post-surgical programs.Among them,the human abdominal CT image is the basis for diagnosis of abdominal organ disease,and the liver,as the largest organ in the abdomen,plays a vital role in human life and health,and the abnormal liver function is closely related to many diseases,such as liver cancer,liver benign tumors and some other liver infectious diseases.Therefore,accurate and rapid automatic segmentation of the liver can help doctors better understand the condition for further diagnosis and treatment options,which is of great research value for clinical diagnosis.CT is widely used in clinical examinations for its high density resolution and high signal-to-noise ratio compared to MRI images.Compared with the previous methods of manual and semi-automatic segmentation,the deep learning-based method proposed in this paper not only greatly reduces the split time compared to other methods in CT images,but also better than them in segmentation performance and generalization ability to a certain extent.In order to achieve the precise segmentation of the liver in the abdominal image,it is studied mainly through several aspects:(1)A network model based on attention module is proposed for liver segmentation.First,the 3D CT image is pre-processed and the 2D image is obtained,and then fed into the network model.The proposed model is based on the improvements made on UNet,in which the attention mechanism is used to enhance the weight ratio of the model to the target area,while suppressing the interference of similar tissue organs around it;In addition,in order to reduce the loss of feature information caused by multiple convolutions and downsamplings,the empty space convolution pooled pyramid module is combined on the existing basis,so that the model can obtain multi-scale information without adding parameters,fully excavate the image features at different scales,better obtain the shallow features in CT images,and further improve the network segmentation performance.(2)A liver segmentation algorithm based on feature fusion is proposed.Considering that2 D network only uses the slice information of CT image,and does not combine the spatial information in CT image,in order to fully exploit the correlation between slices,the 3D UNet network is proposed to combine spatial information and channel information on the jump connection based model.Channel attention is used to learn the correlation between channels,and spatial attention is used to capture attention information on different scales.Secondly,taking into account the changes in liver size and shape in CT images,in-depth supervision was added to learn feature representations at different levels,thus enhancing the learning ability of the network.(3)A liver segmentation algorithm based on convolutional attention module is proposed.Although the 2D network training speed is fast,it does not combine the spatial information in the CT image.On this basis,the 3D network uses the 3D mass information of CT image to solve the problem that the 2D network model does not learn the correlation between slices,but there is a problem of long training time.In view of the above existence,this chapter proposes a2.5D network model based on convolutional attention module for liver segmentation,with the aim of balancing training time and combining spatial information in CT images,so that the model can be segmentation more accurately and efficiently.