| Breast cancer is a serious hazard to women’s health.Neoadjuvant chemotherapy is widely used as an important way to treat breast cancer.Clinically,the pathological complete remission rate and late survival rate of patients who received chemotherapy and achieved a good prognosis response were effectively improved.Neoadjuvant chemotherapy usually lasts for 6 to 8 treatment cycles.Due to individual differences among cases,some patients may not be effective after chemotherapy and may even have side effects.By introducing breast imaging in the early stage of chemotherapy(the first two courses of treatment),it will be of great significance to predict the final curative effect after chemotherapy according to the longitudinal time changes of tumor and microenvironment,which will help to accurately grasp the timing of treatment and adjust the diagnosis and treatment plan in time.Dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)has been widely used in the diagnosis of breast cancer.However,due to the high heterogeneity of the tumor and the microenvironment and the partial volume effect of the image,it is difficult to further improve the accuracy of the efficacy prediction based on the traditional original image features.In this study,based on the images of various stages of tumor(before chemotherapy and early chemotherapy),the deep matrix decomposition method of images was introduced,and the prediction model of curative effect was constructed and comprehensively analyzed based on longitudinal time radiomics characteristics and dynamic pattern changes.The specific research content includes the following aspects:(1)Efficacy prediction analysis based on longitudinal temporal multi-regional imaging features:Based on the longitudinal temporal features of pre-chemotherapy and early chemotherapy images,a single-feature prediction model was established to analyze the correlation between different features and curative effect.The results show that among the salient features of multi-region images,morphological features have better predictive performance(AUC=0.723).A multi-feature model was established,and predictive analysis was conducted through the model fusion of the feature layer and the decision-making layer.The results showed that there was a close relationship between the longitudinal temporal characteristics of the multi-region images and the curative effect.At the same time,the complementary information between the multi-region images was used to further improve the prediction performance of the model(AUC =0.788).(2)Study on curative effect prediction based on depth matrix decomposition images: the depth matrix decomposition model was introduced for image decomposition to obtain potential dynamic enhancement modes.After decomposition,the unified signal mode matrix and mode probability matrix are obtained,and the image features of different sub-signal modes recovered according to the mode probability are analyzed.In the prediction tasks of pre-chemotherapy and early chemotherapy imaging features,early chemotherapy imaging showed better predictive performance(AUC=0.834).In the task of image longitudinal temporal features and model fusion,both the longitudinal temporal features of the decomposed images and the fusion model showed better prediction results than before decomposition(AUC=0.888).It further proves the application value of image depth decomposition in curative effect prediction,combining heterogeneity information reflected by multimodal images and longitudinal temporal features to further enhance the predictive ability of curative effect based on feature level.(3)Research on curative effect prediction based on deep pattern change characteristics: Using the change information of different dynamic patterns after decomposition,statistical analysis was performed on the probability distribution of pre-chemotherapy and early chemotherapy patterns in different curative effect groups,which showed that tumor imaging was more effective in chemotherapy-effective groups.The middle distribution is more significantly different(P<0.01).In the analysis of the correlation between the pattern change characteristics and curative effect after coding,it was found that there was a significant difference between the pattern change characteristics and the curative effect from before chemotherapy to the early stage of chemotherapy(P<0.05).To verify the curative effect prediction value of pattern change-related features,pattern probability features and pattern change features were used for predictive analysis.The results showed that the pattern probability matrix after deep decomposition could be used as a potential marker for therapeutic effect prediction.Through the combination of multi-region images,further the predictive ability of curative effect was improved(AUC=0.892).In this study,the potential dynamic patterns of tumor and gland regions were identified through deep decomposition.Based on the radiomics characteristics of different sub-signal patterns and the change characteristics of depth patterns after decomposition,a curative effect prediction model was established,evaluated,and analyzed.The results show that compared with the traditional curative effect prediction method based on the original image features in a single time dimension,the depthdecomposed image features in longitudinal time show better predictive ability.In the analysis of deep mode changes,the dynamic mode changes of different treatment groups showed significant differences,indicating that there is a potential correlation between this deep decomposition mode and the treatment effect,and it is expected to be used as a new potential marker for the prediction of treatment effect.Breast cancer treatment provides valuable reference. |
PDF Full Text Request