Research On Automatic Segmentation Algorithm Of Lymph Nodes Based On Deep Learning

Medical imaging has always been an important clinical reference information.With the development of medical imaging,the quality and quantity of pictures are constantly improving,and reading medical images occupies a considerable part of medical resources.In recent years,with the development of computer technology and the wave of artificial intelligence,computer-assisted diagnosis and treatment technology emerges at the right moment.It can not only accelerate the speed of detection and diagnosis,but also avoid the subjective deviation of manual reading.As an important part of computer aided diagnosis and treatment,medical image segmentation technology plays a key role in clinical diagnosis and treatment.Automatic segmentation of involved lymph nodes in PET/CT images of lymphoma patients is one of the important applications of medical image segmentation technology.However,at present,both the traditional medical image segmentation algorithm and the medical image segmentation algorithm based on deep learning are used to segment one kind of image,and the collaborative segmentation processing of PET images and CT images cannot be realized.To solve this problem,this paper studies the cooperative segmentation of PET/CT images.The objective of this study is to achieve automatic segmentation of involved lymph nodes in PET/CT images of lymphoma patients.For the two types of data sources involved in lymph node segmentation:PET images with anatomical information and CT images with metabolic degree information,we proposed a double-path fully convolutional neural network--W-net,and added the group convolutional structure to reduce the parameter complexity of the network.Three evaluation indexes were set to evaluate the segmentation results.Through training and testing,it was proved that this network has better segmentation ability for lymph nodes than the commonly used U-net.In order to make the two-path full convolutional neural network converge rapidly and stably,a pre-segmentation algorithm 3D-FCM was designed in this paper.3D image data was taken as the segmentation object to extract the soft tissue area in the CT image first.Based on the fuzzy theory,the algorithm clusters the pixels in the three-dimensional space and increases the dimension of the feature space to enhance the clustering effect.Taking 3D-FCM as the front end of W-net,the convergence speed and accuracy of the network are improved,so that the segmentation effect is also improved.