Research On Prediction Of Response To Neoadjuvant Chemotherapy In Breast Cancer Based On Magnetic Resonance Images

Breast cancer is the most common cancer among women worldwide,accounting for up to 30% of all female cancer cases.In China,the incidence of breast cancer in women has been increasing year by year.Neoadjuvant Chemotherapy(NAC)is a widely used standard treatment for breast cancer,which can reduce the size of the tumor,reduce the difficulty of surgery,and increase the breast preservation rate.However,due to individual differences among breast cancer patients and the diversity of tumor components,there is significant variability in the efficacy of NAC treatment.For patients who do not benefit from NAC,adverse events such as tumor progression may occur.Therefore,the focus of this study is on the early prediction of NAC efficacy in breast cancer,aiming to address this clinical problem.The specific research content and innovative points are as follows:(1)A breast cancer lesion detection model based on an improved Faster R-CNN is proposed to address the issue of missed detection of small breast cancer lesions.The model utilizes ResNet50 and Feature PyramidNetworks(FPN)for multi-scale feature extraction,enhancing the accuracy of lesion detection.The ROI Align technique is employed in the feature mapping layer to reduce biases introduced by feature mapping.In the test set,this model is compared with nine other deep learning detection models,and the improvement strategy is evaluated.Experimental results demonstrate that the model performs excellently in the detection of small lesions in breast cancer DCE-MRI images,achieving an accuracy rate of 82.52%.(2)To address the issue of strong interpretability but poor predictive performance in traditional radiomics-based efficacy prediction methods,a RSLA-Stacking model based on the Stacking algorithm is proposed.This model combines the advantages of four different machine learning classifiers and effectively reduces the risk of network overfitting.This study also explores the role of different sequences of DCE-MRI and clinical pathological information in efficacy prediction among the four machine learning classifiers and compares their predictive performance with the RSLA-Stacking model.Experimental results demonstrate that the model performs best in terms of accuracy,achieving an accuracy rate of 83.05%.(3)In response to the issue of poor predictive performance in existing end-to-end deep learning efficacy prediction methods,a breast cancer efficacy prediction model based on wavelet transform and attention mechanism is proposed.This model incorporates discrete wavelet transform(DWT),channel attention mechanism(SER),and spatial channel attention mechanism(SCAR)modules to learn the multi-scale information and important spatial channel information of the images,thereby improving the predictive performance.Compared to other deep learning classification models,WAVGG achieves a prediction accuracy of 91.15%.Additionally,this model performs end-to-end efficacy prediction,unlike the RSLA-Stacking model,leading to better predictive performance.Furthermore,the results of ablation experiments further validate the importance of each module in efficacy prediction within this model.