Auxiliary Diagnosis For Thyroid Nodules Based On Historical Diagnostic Report And Feedforward Neural Network

Thyroid nodules are the most common clinical disease of the thyroid,and the incidence of thyroid nodules has been increasing in recent years,which seriously affects human health.It is of great practical significance to carry out research on the auxiliary diagnosis of thyroid nodules.Ultrasound examination is a necessary examination method for the clinical diagnosis of thyroid nodules.With the maturity of medical informatization,a lot of test data of ultrasound diagnosis from patients with thyroid nodules have been effectively saved.These diagnostic text data are very important to be referred for auxiliary diagnosis.How to accurately and efficiently extract valuable structured data from non-standard ultrasound diagnostic text is a necessary prerequisite for research about auxiliary diagnosis of thyroid nodules.In addition,traditional clinical diagnosis of thyroid nodules is often subject to doctors’ professional knowledge,work experience and cognitive abilities,resulting in that patients cannot receive timely and accurate treatment,which has always been a key problem facing ultrasound diagnosis for thyroid nodules.The new generation of artificial intelligence technology provides new ideas for intelligent auxiliary diagnosis in the modern medical field.Based on the above background and problems,this paper analyzes the research status of structured processing for medical text and the auxiliary diagnosis for thyroid disease and proposes a set of solutions of structured processing for non-standard medical diagnostic texts.At the same time,the report data of ultrasound diagnosis for thyroid nodules in a large top three hospital in Hefei,Anhui Province are collected,and five kinds of ultrasonic features are extracted from the report data of ultrasound diagnosis for thyroid nodules by constructing a feature extraction framework based on a custom professional dictionary.On the basis,two relatively mature feedforward neural networks,BPNN and RBFNN,are used to construct the thyroid nodules auxiliary diagnosis model,to conduct the research of auxiliary diagnosis for thyroid nodules from the perspectives of the prediction for benign and malignant properties of thyroid nodules and the prediction for doctor’s diagnosis.In addition,the "gold standard" and interval distance measurement are introduced into the prediction model of doctor’s diagnosis,and a new accuracy formula is defined to measure the diagnosis accuracy.Finally,5569 pieces of diagnostic data of thyroid nodule are used as experimental data set to conduct the experiments of predictions for benign andmalignant properties of thyroid nodules.The accuracy,ROC curve and AUC value are used to verify the prediction results of the model.And the results show that the RBFNN-based prediction model for benign and malignant properties of thyroid nodules has a higher accuracy than that of the BPNN-based prediction model.Then,3868 pieces of diagnostic data of thyroid nodule from five ultrasound doctors with a large number of cases are used as the experimental data set to conduct the experiments of predictions for doctor’s diagnosis.The prediction accuracy is compared with the doctor’s own diagnosis accuracy,and the results show that the accuracy of the RBFNN-based prediction model for doctor’s diagnosis is 5.5～11%higher than that of doctors,which means the RBFNN-based prediction model can effectively assist the doctor in diagnosis for thyroid nodules.