Longitudinal Analysis Of DCE-MRI For Prediction Of Neoadjuvant Chemotherapy Response In Breast Cancer

Breast cancer is the most common disease among female cancer patients worldwide and the corresponding patient population is also gradually becoming younger,so precise treatment of breast cancer is crucial.In clinical treatment,neoadjuvant chemotherapy has been widely used.Studies have shown that patients who have achieved better response to neoadjuvant chemotherapy,especially those who have achieved complete pathological remission,will have a much higher survival rate.However,due to the individual differences among patients,some breast cancer patients cannot achieve the desired chemotherapy effect after neoadjuvant chemotherapy,and their disease even progresses during chemotherapy.To maximize the timing of treatment and to reduce the suffering of patients with ineffective chemotherapy,it is important to predict the response to neoadjuvant chemotherapy early.Currently,there are many studies showing that the response to neoadjuvant chemotherapy can be evaluated by DCE-MRI of the breast,which is a non-invasive test and has been widely used in the clinical management of breast cancer.However,in clinical diagnosis,the image analysis based on the subjective experience of the physician cannot provide a comprehensive and accurate assessment of the pathological information of breast cancer.The Radiomics generated by the fusion of big data and medical imaging can effectively improve the accuracy of diagnosis and treatment.Pre-chemotherapy and early chemotherapy DCE-MRI images were used in predicting the response to neoadjuvant chemotherapy.The images of 114 patients who underwent neoadjuvant chemotherapy for breast cancer were collected,and the data set was divided into two groups according to the Miller & Payne grading system: the chemotherapy non-responsive group and the chemotherapy responsive group.The lesion areas in the DCE-MRI images of patients before and early in chemotherapy were segmented,and the corresponding shape,statistical and textural features were extracted.Although there have been studies to predict the response to neoadjuvant chemotherapy by imaging histological methods and DCE-MRI images,such studies have not discussed and studied pre-chemotherapy and early chemotherapy images separately when making predictions.In this paper,we not only predicted the response based on the pre-chemotherapy and early chemotherapy images,but also combined the two images for longitudinal analysis to further improve the accuracy of response prediction.In the prediction of chemotherapy response using pre-chemotherapy and early chemotherapy images,we performed univariate and multivariate prediction analyses of shape features,statistical features,textural features and all features.The prediction performance of pre-chemotherapy and early chemotherapy images in response prediction was comprehensively evaluated and compared based on the prediction results of various types of features.In the experimental process,the prediction analysis was performed under the LOOCV,and the SVM-RFE was used to find the optimal subset of features.The prediction model was comprehensively evaluated by plotting the receiver operating characteristic curve of the prediction results and calculating the area under the curve,sensitivity,specificity,F1 score and other indicators.The experimental results showed that the early chemotherapy images had better predictive performance than the pre-chemotherapy images.In the study of chemotherapy response prediction based on longitudinal analysis,longitudinal analysis was performed not only on the two DCE-MRI images before and after chemotherapy,but also on the extracted features.The key to the longitudinal analysis is the alignment of the images before and after chemotherapy,but the interval between the two images is long and there are many factors affecting the images when they are taken,so the longitudinal analysis of the images needs to be performed by calculating the deformation rate using the deformation alignment method.In contrast,the feature longitudinal time analysis process is relatively simple,directly using the image features already extracted before and early in chemotherapy,which can be achieved by the corresponding calculation.The experimental results showed that the results of longitudinal analysis correlated with chemotherapy response and achieved good prediction results in predicting chemotherapy response.The previous studies were fused from two aspects,feature-based level fusion and decision-based level fusion.The prediction models before and after fusion were compared using the same model evaluation method.The experimental results showed that the fused prediction models achieved better prediction results for the test set,with better generalization ability compared to the individual prediction models before fusion.The study showed that there is a correlation between pre-chemotherapy and early chemotherapy DCE-MRI images and neoadjuvant chemotherapy response,and the prediction models constructed by Radiomics and longitudinal analysis can predict chemotherapy response more accurately,which can provide some guidance value for the treatment of breast cancer patients.