The Auxiliary Analysis Of Machine Learning In The Diagnosis Of Benign Thyroid Nodules And Malignant Thyroid Nodules

Thyroid nodular disease is a common clinical disease.There are a variety of clinical thyroid diseases,which can be quickly diagnosed as benign thyroid nodules or malignant thyroid nodules by the physical features of thyroid nodules.At present,the commonly methods for distinguishing benign thyroid nodules and malignant thyroid nodules are imaging diagnostics or chemical diagnostics,but they have a high rate of misdiagnosis.Therefore,based on the data of thyroid nodules with different physical characteristics,this paper proposes a simple and intuitive method to identify the benign thyroid nodules or malignant thyroid nodules,so as to improve the efficiency of imaging diagnosis.This paper first preprocesses the collected data.In particular,it adopts the latest missing data processing method to fill in the missing data.Then,the relationship between thyroid nodules properties and other characteristic variables was analyzed by means of histograms and box charts,and the effects of characteristic variables on benign thyroid nodules and malignant thyroid nodules were obvious.With the help of the correlation test between characteristic variables and dependent variables,we find out the feature with a large correlation between them,and then do a random forest algorithm analysis to obtain several characteristic variables affecting the thyroid nodules properties.Finally,based on the first two methods,a logistic regression model of the complete data set is established,and a logistic regression model with missing data sets is established by using the missing mechanism model and the EM algorithm.The results of machine learning analysis showed that the thyroid nodule size,aspect ratio,edge and boundary conditions,thyroid internal structure and calcification have a significant impact on discriminating the benign thyroid nodules and malignant thyroid nodules.For actual data,the validity of the estimation method used in this paper can be illustrated.The established logistic regression model is more reasonable.