Lymph Node Metastasis Prediction Of Papillary Thyroid Carcinoma Based On Artificial Intelligence

In recent years,as ultrasound imaging technology has become more mature,more and more patients with thyroid carcinoma have been detected.Papillary thyroid carcinoma is the most common thyroid carcinoma and it is very prone to lymph node metastasis.Whether metastasis occurs is highly related to the stage,risk,and recurrence of the carcinoma after surgery,and thus greatly affects the decision-making of treatment options.However,it is currently impossible to make accurate preoperative diagnosis of lymph node metastasis through ultrasonography,and the accuracy of diagnosis is greatly affected by the level of doctors.Therefore,for the sake of insurance,preventive lymph node dissection and even excision operation which are traumatic are generally used.The treatment options for removing the thyroid gland present a serious problem of overdiagnosis and overtreatment.In order to solve the above problems,this thesis cooperates with the ultrasound department of a famous hospital in Beijing,and uses artificial intelligence technology to achieve efficient prediction of lymph node metastasis of papillary thyroid carcinoma based on clinical real ultrasound images and pathological data.The main work is as follows:1.A new dataset of papillary thyroid carcinoma patients was constructed,and the content of data labeling was determined together with doctors.For this purpose,the labeling tool was designed and developed,and all the labeling work was completed by professional doctors.2.In terms of function realization,the prediction tasks are completed by using machine learning methods such as Genetic Algorithm and Support Vector Machine,and deep learning methods such as ResNet structure and Stochastic Gradient Descent method.Various improvements are made to the ResNet structure to suit this research scenario and dataset and finally SPP-ResNet model is constructed.Besides,ablation experiments are designed to demonstrate the effectiveness of the operation.3.The current clinically known medical knowledge of the disease is extracted as priori features to input model to assist and the effectiveness of this operation is deeply analyzed.Based on the multi-task learning method and the ResNet structure,MTL-ResNet model is constructed to pre-extract the priori features simultaneously and automatically.The experimental results show that the SPP-ResNet model finally constructed in this thesis can efficiently predict lymph node metastasis and can be properly used as a primary screening tool to assist doctors in clinical diagnosis.The model is also efficient,convenient,free from time and space constraints and the level of doctors.At the same time,the research of artificial priori features improves model performance and enhances model interpretability and credibility.