Prediction Of Response To Neoadjuvant Chemotherapy For Breast Cancer Based On DCE-MRI Mapping Pattern Analysis

Breast cancer,as the most common malignant tumor in women,poses a great threat to women’s physical health.Neoadjuvant chemotherapy,which uses chemotherapy drugs to shrink the size of tumors or control their growth,is widely used in the treatment of breast cancer.However,neoadjuvant chemotherapy is not suitable for all breast cancer patients,and there are still some patients with unsatisfactory chemotherapy effect,so as to delay the best treatment opportunity.Therefore,if accurate prediction of the efficacy of neoadjuvant chemotherapy can be achieved,it will provide physicians with more effective clinical recommendations and help improve the postoperative survival quality of patients.Dynamic Contrast Enhanced Magnetic Resonance Imaging(DCE-MRI)is widely used in predicting the efficacy of neoadjuvant chemotherapy because it contains rich morphological and hemodynamic information of the lesion tissue.However,due to tumor heterogeneity,the accuracy of traditional prediction of chemotherapy efficacy based on original image features needs to be improved.In this study,by mining the hemodynamic features and spatial and temporal information on images,a new method for predicting efficacy of adjuvant chemotherapy based on signal decomposition analysis,spatial projection and temporal projection mapping models was proposed.Specific research contents are as follows:(1)Prediction of efficacy of neoadjuvant chemotherapy based on signal decomposition mapping model.DCE-MRI described the hemodynamic characteristics of the tumor by measuring signal intensity changes at different time points.The Convex Analysis of <e:4>(CAM)decomposition algorithm could extract the image signal intensity modes,and the probability matrix of each patient was obtained using the extracted unified signal modes.The number of signal decomposition K under CAM decomposition was specified to obtain different signal patterns.In this experiment,the image omics features were extracted from the probability matrix graph obtained by signal pattern decomposition with K=3-5.Recursive Feature Elimination(RFE)and Lasso were used for feature selection.Three classifiers,logistic regression,support vector machine and Random Forest(RF),were designed to predict the efficacy of neoadjuvant chemotherapy.The results showed that when the number of signal modes K=5,RFE+RF model had the best classification effect(AUC=0.750).(2)Study on the efficacy prediction of neoadjuvant chemotherapy based on time projection mapping model.ResidualNetwork(ResNet)model was constructed for prediction research,and model classifiers,loss functions and training algorithms were designed.The experiment used ResNet networks with different depths and tested the prediction effect of different sequences(preenhancement sequence,mid-enhancement sequence,late enhancement sequence,subtraction sequence).First,the efficacy of each sequence was predicted,and the subtraction sequence under the ResNet50 network model had the best predictive effect(AUC=0.686).Multiple sequences were combined into multiple channels as input to predict the curative effect.Results Six-channel images as input in ResNet50 network model had the best predictive effect(AUC=0.732).The maximum value,median value,minimum value and range were projected for different time series respectively,and the constructed mapping pattern was used to predict the chemotherapy efficacy,among which the mapping pattern based on the maximum projection had the best prediction effect(AUC=0.747).(3)Study on the efficacy prediction of neoadjuvant chemotherapy based on spatial projection mapping pattern map.By selecting multiple image sections adjacent to the maximum diameter of the tumor,different statistical mappings(maximum,median,minimum,range)were performed to construct the spatial projection mapping model map for the prediction of the efficacy of neoadjuvant chemotherapy.Firstly,the mapping pattern map of each sequence was used to predict the chemotherapy efficacy,among which the range feature mapping pattern map of the late enhanced sequence as the input predicted the best effect(AUC=0.755),which was significantly improved compared with the single sequence image without the mapping pattern map.The mapping pattern map with the best predictive effect of each sequence was selected to construct the multisequence pattern map and predict the efficacy of neoadjuvant chemotherapy(AUC=0.781).The spatial and channel attention modules were introduced to improve the network’s ability to search for important information of images,and the results showed that the prediction result of efficacy was improved.Finally,the optimal mapping pattern map on time and space projection 1was used for model fusion of the feature layer and the decision layer,among which the prediction effect was the best under the parallel connection of the feature layer(AUC=0.803).According to the features of DCE-MRI imaging omics,a study was carried out to predict the efficacy of neoadjuvant chemotherapy using signal pattern decomposition,spatial pattern projection and time pattern projection.Different image sequences were used to predict the efficacy,and attention mechanism and model fusion methods were introduced to further improve the prediction effect.The results showed that the characteristic parallel fusion method combined with temporal and spatial model projection had the best predictive effect(AUC=0.803),indicating that these three mapping model methods had correlation with the efficacy prediction of neoadjuvant chemotherapy.