Deep Matrix Factorization Method Based On DCE-MRI Radiomics For Prediction Of Histological Information Of Breast Cancer

Breast cancer is the most common cancer among women.Its high incidence and mortality among women poses a serious threat to public health.At present,the early detection of breast cancer and the provision of precise treatment plans for breast cancer patients can effectively improve the survival rate of breast cancer patients.Complete histological information of breast cancer can provide patients with prognostic information,which is the key to help formulate treatment plans and predict the efficacy of diagnosis and treatment of patients.Among them,ER,PR,HER-2,Ki-67 proliferation index and histological grade information are particularly important in the formulation of personalized breast cancer treatment.DCE-MRI is one of the vital breast examination methods,which can provide breast cancer morphological and functional information.This study aims at the problem of possible missing pathology reports during the diagnosis and treatment of breast cancer.The imaging features are extracted from DCE-MRI lesion areas and quantitatively analyzed,combined with breast cancer pathology information to construct a matrix to be filled,and three types of missing information imputation model are established.Prediction model to predict missing ER,PR,HER-2,Ki-67 and histological grade information.The main research contents include:(1)Extraction of histological information and image information: retrospectively analyze the histological information of 210 cases of invasive breast cancer before surgery or chemotherapy,segment the three-dimensional breast cancer area from DCE-MRI images and extract statistical features,morphological features,and texture features.The verified random forest recursive feature elimination method screens the extracted image features,uses the union-based method to further screen the features,and combines the histological information of each patient to construct a imputation matrix.(2)Prediction of histological information of breast cancer based on matrix factorization method: According to the two ways of histological information missing random and complete missing,the NMF(non-negative matrix factorization)imputation model and the CF(user-based collaborative filtering)model are different in different histological information missing rates and usage of radiomics features.The ROC curve and AUC value of the prediction model are used to evaluate the prediction effect,and the bootstrap method is used to perform a paired T test on the same group of AUC values.Experimental results show that both the non-negative matrix factorization method and the user-based collaborative filtering method can effectively fill the missing,and the filling performance of the NMF method is better than the CF method.(3)Research on breast cancer histological information prediction based on deep matrix factorization(DMF): Introduce DMF into histological information prediction.This paper proposes a DMF model with a fully connected structure and a DMF model with a residual structure,compared the histological information prediction effect of two different DMF models with different histological information missing rates and different radiomics features.The experimental results show that the depth matrix factorization model with the fully connected structure performs best under the same missing rate and the same image features.The optimal AUC value in the test set is0.806,which is significantly better than NMF and CF prediction method.This paper uses the radiomics method to construct a joint matrix of image features and histological information,establishes a NMF-based information prediction model,a CF-based prediction model,and a DMF-based prediction model to achieve a preliminary missing pathology information prediction research.