Research On Automatic Recognition Technology Of Lymph Node Ultrasound Image Based On Deep Learning

Lymph nodes are peripheral immune organs that are distributed throughout the body.As an important component of the human immune system,they filter lymphatic fluid and produce lymphocytes and antibodies that help eliminate foreign cells and bacteria and participate in the body’s immune response.When malignant tumors occur in the lymph nodes,there is a risk of cancerous transformation.Timely diagnosis and treatment can cure most malignant nodules.Currently,ultrasound examination is the most commonly used method for lymph node diagnosis.Computer-aided diagnosis based on ultrasound imaging has important clinical significance,as it can effectively help doctors shorten examination time and reduce the impact of subjective factors on misdiagnosis.This article focuses on the research of deep learning-based automatic identification technology for lymph node ultrasound images,and has completed the work from data preprocessing to lymph node ultrasound image segmentation,classification,and subsequent automated measurement,achieving good results.The main work and achievements are as follows:To begin with,the construction of the ultrasound lymph node dataset was completed.After obtaining the original ultrasound dataset,the lymph node dataset was annotated under the guidance of professional doctors.However,the benign and malignant lymph node dataset is imbalanced and the data volume is small,making it difficult to fully support subsequent deep learning network training.To solve this problem,we proposed an improved image data generation algorithm to supplement the medical dataset with a small amount of data.It was proven to be very effective through experiments,and received recognition from professional doctors.This makes generated data a good supplement to real data,effectively increasing the overall lymph node data volume and improving the quality of the dataset.Then,A deep learning network model for lymph node ultrasound imaging features was studied.We studied a new deep learning network model for lymph node ultrasound imaging,R2AB-Net segmentation model.The model addresses the segmentation difficulty of ultrasound imaging data due to unclear edges,overly close overall regions,and difficulty in extracting features effectively,by adopting innovative BDM modules and Res2Att modules to achieve better boundary segmentation and feature extraction under multi-head attention mechanism.At the same time,we designed a composite edge loss function to improve the training effect.Through multiple sets of comparative and ablation experiments,we found the best model structure and parameters,and verified the effectiveness of the network in lymph node segmentation tasks.Further more,a lymph node segmentation and classification integrated model was designed.Combining the results of the segmentation network with the important benign and malignant classification task in lymph node examination,we designed an integrated segmentation and classification model and verified the reliability of the algorithm through comparative experiments.Finally,the lymph node automated measurement method was studied.We designed methods for predicting the automatic measurement of lymph nodes,and integrated these research results into a web system interface along with the lymph node classification and segmentation research,in order to assist medical staff in diagnosis and treatment more easily.