Research On Intelligent Diagnosis Method Of Lymph Nodes Based On CT Images

A computer tomography(CT)scan for each patient can get about 500 image data sets,lymph nodes are tissues that occupy the cornerstone of the human immune system,the labeling of lymph node swollen tissues depends on the manual labeling of the radiologist,and the accuracy of the labeling depends on the doctor’s theoretical basic knowledge and practical knowledge,and it takes a long time.With the advent of deep learning,image recognition and target detection have begun to become research hotspots.Applying deep learning to the detection of medical CT images can greatly reduce the burden on doctors,save patients’ waiting time,and receive treatment earlier.This topic is mainly about the research of intelligent diagnosis methods for enlarged lymph nodes,including segmentation and classification algorithms.At present,the target recognition algorithm for enlarged lymph nodes mainly uses 2D,2.5D and 3D images for network training.The existing problems are:(1)When using 3D data for training,the hardware requirements are higher;(2)When using separate 2D data for training,the connection between the upper and lower layers is ignored,which will lead to the loss of information;(3)Insufficient number of data sets makes it impossible to train the model well.In response to the above problems,Continuous 2D images are used as the input of the enlarged lymph node segmentation model to ensure that the inter-layer information of CT slices is not lost,multi-resolution images are used as the input of the enlarged lymph node classification model,multi-scale features are fused with each other,and background noise can be suppressed.This paper designs a lymph node segmentation network based on ResNet-101,extracts the texture and gray features of lymph nodes,and compares them with the U-Net algorithm,it is found that better results can be obtained.Aiming at the problem of lymph node classification,this paper adopts a target detection algorithm Cascade R-CNN based on ResNet-50,through RPN and cascaded detectors,the adjustment of the detector threshold can increase the recall rate of enlarged lymph nodes,suppress false positives,and reduce the phenomenon of over-fitting.