Analysis And Design Of Breast Cancer Classification System With Multimodal Features

With the continuous improvement of medical standards,the use of machine learning methods to classify and predict breast cancer has become a research hotspot in recent years.Because breast cancer clinical diagnosis data and ultrasound images are the key basis for diagnosing its type,this article analyzes and features fusion of clinical diagnosis data and ultrasound images of breast cancer patients from the Biendata Competition public data set,and constructs a classification model based on decision tree to predict the type of breast cancer,design and implement a classification system.The main research contents of this article include.(1)Screening of breast cancer classification models.First,three types of machine learning classification algorithms: Decision Tree,Logistics Regression and Support Vector Machine are analyzed.Then, the experimental part uses the breast cancer clinical diagnosis data to train the classification model,extract image features through Re s Net-50,and fuse with the diagnosis data features,train the classifi cation model,compare the performance of the three classification m odels,determine the Decision Tree algorithm to build the model.Fi nally,comparing the performance of the decision tree classification model trained with monomodal and bimodal data,the experimental r esults show the auxiliary role of images in classification.(2)Fusion of salient features and clinical diagnostic data features to construct a breast cancer classification model.First,based on Faster R-CNN to automatically label the salient focus area of ultrasound images.Then,Res Net-50 was used to extract the features of the significant lesions in the image,and the features of the extracted significant lesions were fused with the corresponding diagnostic data features to construct a breast cancer classification model.Finally,the performance of the classification model before and after automatic labeling was compared,and the experimental results prove that the salient features of the image can improve the performance of the classification model.(3)Design and implementation of breast cancer classification system.Firstly,the front-end module of the system is built,which mainly includes the user login and upload information module.Then,the communication mechanism between the classification model and the classification system is established,and the classification model is scheduled to assist users in predicting and classifying breast cancer types;Finally,the forecast type results are shown.In this dissertation,the classic classification algorithm of machine learning is applied to breast cancer classification,and the research is carried out from three aspects: classification model selection,model performance improvement and system implementation.Finally,it provides users with a friendly human-computer interaction window,which has important practical significance.