Research On Computer Aided Diagnosis Of Breast Cancer Based On Unbalanced Data

In recent years,breast cancer has become one of the leading causes of female death.Dynamic Contrast Enhancement MRI(DCE-MRI)has been widely used by virtue of its advantages in describing the shape of the lesion and the information of motion enhancement at the same time.However,in order to accurately diagnose benign and malignant breast cancer and molecular classification,patients need to undergo invasive pathological surgery.Computer Aided Diagnosis(CAD)is an important method for early detection and diagnosis of tumors.However,breast cancer CAD system still faces many challenges at the present stage.When the number of benign and malignant samples collected differs greatly from that of malignant samples,or the number of malignant samples with different degrees of malignancy is unbalanced,the diagnostic performance of the CAD system will decline.To this end,this article mainly studies the construction of a breast cancer CAD system with unbalanced data.The research content of this article mainly includes the following three aspects:(1)Research on deep convolutional generative adversarial networks based on Wasserstein distance: High-quality data sets are very important to the training process of machine learning.The data must not only cover a wide range,but also be as balanced as possible.However,there are often underrepresented classes in real databases,which affects classification performance.Based on this,this research proposes a Wasserstein distance-based deep convolutional generative adversarial network to generate artificial synthetic data to balance the unbalanced data set,and measure the similarity between the original data and the artificial synthetic data through different distance measurement parameters.This algorithm is compared with other popular data generation algorithms.The results show that the artificially synthesized data generated by the algorithm in this paper is more similar to the original data.(2)Research on breast cancer benign and malignant CAD system based on unbalanced data: Existing breast cancer CAD systems often result in a low recognition rate of minority samples due to the imbalance of benign and malignant sample data.In response to the above problems,this study proposed a breast cancer benign and malignant CAD system based on unbalanced data,and conducted comparative experiments on a variety of classifiers.The results are compared with traditional CAD systems and existing more popular data equalization algorithms.The results show that the algorithm proposed in this paper has higher classification accuracy.(3)Research on the CAD system for molecular classification of breast cancer based on unbalanced data: When traditional methods deal with multi-classification of unbalanced data,the two-classification algorithm is usually directly extended to multi-classification,which leads to the performance degradation of the algorithm.In order to achieve precise treatment of breast cancer,this paper constructs a CAD system for molecular classification of breast cancer based on unbalanced data.First,a decomposition strategy is used to decompose the sample into multiple binary sub-categories,and then data equalization is performed on each binary sub-classifier,and the competitiveness between the two classifiers is defined by a distance-weighted method,and finally output through voting final result.The experiment is compared with the existing more popular data equalization algorithms,and the results show that the method proposed in this study effectively improves the diagnostic accuracy of the Her-2 overexpression type and Luminal A breast cancer.