Progressive Breast Cancer Diagnosis Model Based On Multi-classifier And Multi-factor Fusion

Smart healthcare is a cross field of medicine and artificial intelligence.It is one of the hot research fields all around the world in recent years.Machine learning can effectively alleviate the contradiction between the lack of medical resources and the growing demand of patients.Breast cancer is one of the biggest threats to women today.As a heterogeneous tumor,breast cancer diagnosis involves many complicated factors such as demographic information,immunology,biochemistry,etc.,which the judgment basis is complex and diverse.In the actual clinical diagnosis of breast cancer,due to the different stages of the disease,different examination methods are used for different patients at the same time.The above situation has brought great challenge to the construction of comprehensive diagnosis model of breast cancer.Based on the study of multiple classifiers and fusion methods,this paper constructs a progressive integrated diagnosis model based on multi-classifier and multi-factor fusion.The main tasks are as follows:(1)Data features are extracted and selected for the problems of breast cancer involving complex physical and chemical indicators that have various expressions.Combined with relevant medical literature and existing data,three typical medical examinations in the early,middle and late stages of breast cancer diagnosis were selected.Use LASSO and random forest algorithm to select the key indicators,and refer to professional medical data to confirm its medical interpretability.(2)According to the variety of breast cancer data,this paper analyzes the six classification models of random forest,decision tree,k-nearest neighbor,support vector machine,logical regression,LSTM.The relevant characteristics of their model advantages and disadvantages,key parameters and so on were explored.By comparing the classification performance of each model on blood routines,blood tumor markers,and immunohistochemical data containing demographic information.The results show that the optimal classifier of complete blood count is RF which accuracy rate is 77.59%,the optimal classifier of blood tumor markers is KNN which accuracy rate is 82.92% and the optimal classifier of immunohistochemistry is SVM which accuracy rate is 84.18%.(3)Aiming at the characteristics of different data distribution and evaluation methods among breast cancer examination data,multi-classifier fusion strategy is adopted.The experiment compares and analyzes three kinds of fusion algorithms: voting method,average fusion method and fusion algorithm based on multi-criteria decision making(MCDM).The results show that the accuracy of MCDM fusion algorithm in different data sets is higher than the other two strategies.(4)Aiming at the problems of diverse types of examinations(factors)involving breast cancer diagnosis,different examination items and time for different patients,this paper proposes a progressive two-level fusion comprehensive diagnosis model.The two-level fusion structure of classifier-level fusion and progressive factor-level fusion is designed to realize the flexible increase and decrease of classifier and factor types and improve the scalability of the model.At the same time,the incremental mechanism makes the model still have a high accuracy when a large number of data features are missing which means the model has good robustness.Based on real data from a top three hospital in Shanghai,the results show that the accuracy of this model is 91% under full feature input.It is more than 10% higher than the model without a progressive mechanism when a large number of features are missing.(5)Construction of a progressive breast cancer diagnosis system.While providing diagnostic functions,it also provides corresponding data visualization services to help doctors diagnose patients more conveniently.