The Research On Diagnosis Algorithm Of Lymph Node Metastasis In Central Region Of Thyroid Based On Transfer Learning

In recent years,the incidence of papillary thyroid cancer has increased rapidly,and about 20% to 80% of patients have cervical lymph node metastasis,which seriously affects the prognosis of patients with papillary thyroid cancer.Because the central area is the first stop for cervical lymph node metastasis,timely and effective screening of thyroid papillary cancer lymph nodes in this area is of great significance for the prediction of patient prognosis.However,at present,there is still no practical diagnostic plan for clinical preoperative examination,and it is impossible to provide accurate preoperative guidance to doctors.In order to avoid the occurrence of postoperative cancerous lymph node metastasis,it is clinically advocated to prevent lymph nodes in the central area cleaning,and this preventive measure may bring hidden dangers of other diseases.Therefore,it is necessary to research and develop diagnostic techniques for the lymph nodes of thyroid papillary carcinoma in the central area,which can not only improve the survival probability of patients,but also guide the clinical diagnosis of doctors and the formulation of preoperative plans.For this problem,this paper analyzes the deficiencies of the current clinical diagnosis schemes,and sets up a control group to convert the problem into a two-class classification of central lymph nodes,and then proposes two algorithms which contain a central region lymph node classification algorithm based on transfer learning and a central region lymph node classification algorithm based on features engineering.The main work and conclusions of this article are:First,use the powerful ability of convolutional neural networks in image processing to classify central lymph nodes.Aiming at the problem that labeled sample data is difficult to obtain in the field of medical images,transfer learning was introduced during training,and experiments were conducted on three different types of convolutional neural network structures.Later,after an in-depth investigation in the field of medical image analysis at home and abroad,feature engineering was used to solve the classification problem of lymph nodes in the central area.This method not only uses traditional methods to extract the manual features of image data,but also uses the network model based on transfer learning as a feature extractor to extract the deep semantic features of the image.Finally,combining two features and different feature selection algorithms to compare the performance of different classifiers,it is found that the sensitivity and negative predictive value(NPV)of the support vector machine are higher and the deviation is minimum,indicating that the missed diagnosis rate is the lowest and more stable,but the specificity and positive predictive value(PPV)of the random forest are the highest,which means that this model has a lower misdiagnosis rate.Finally,this paper uses 209 cases of central lymph node image data to verify the two proposed classification algorithms.Experiments show that compared with clinical diagnosis methods,the proposed algorithms have greatly improved accuracy and sensitivity,and the algorithms have high applicability and are suitable for clinical auxiliary diagnosis.