Research On Segmentation Algorithm Of Breast Tumor In DCE-MRI Based On Convolutional Neural Network

Breast cancer is the most common malignant tumor among women all over the world.Clinical studies have shown that dynamic contrast-enhanced magnetic resonance imaging（DCE-MRI）can simultaneously display the physiological characteristics and anatomical structures of breast tumors,and is an effective tool for breast tumor diagnosis.Accurate segmentation of breast tumor in DCE-MRI can provide a reliable basis for clinical diagnosis,treatment and prognosis monitoring of tumors.However,due to the different positions,sizes and shapes of breast tumors,and the low contrast and noise distribution of MRI images,the tumor borders are blurred and difficult to identify.In addition,the size of the tumor is too small,causing a serious imbalance of categories.All of these have brought great challenges to the precise segmentation of breast tumor in DCE-MRI.To solve the above problems,this dissertation proposes a fusion network with multi-scale residuals and dual-domain attention and a hybrid loss function with adaptive weights to segment tumors in breast DCE-MRI.First of all,preprocessing operations such as cropping,enhancement,denoising,amplification and standardization of breast DCE-MRI data are required.Then,a segmentation model of a fusion network with multi-scale residuals and dual-domain attention is designed.The model uses a multi-scale residual block composed of multi-scale convolution as the basic building block.Multi-scale residual block improves the network’s ability to recognize targets of different sizes and the model’s robustness by extracting multi-scale features and optimizing gradient propagation.At the same time,a dual-domain attention unit is integrated into the network to guide the network to extract favorable features more effectively,so as to improve the network’s ability of edge contour recognition and boundary preservation.Next,a hybrid loss function with adaptive weights is proposed to improve the direction of network optimization.This function mitigates the effects of extreme imbalance between positive and negative samples and guides the network to pay more attention to the difficult classification and segmentation samples by adjusting the attention degree of difficult samples and simple samples.In addition,the function also improves the generalization ability of the model.Finally,the proposed network and hybrid loss function are applied to the task of breast tumor segmentation in DCE-MRI.The effectiveness of the segmentation network and its modules and loss function is verified through comparative experiments and result analysis.Comparative experiments on the data obtained by Ruijin Hospital Affiliated to Shanghai Jiaotong University show that the method proposed in this paper achieves a Dice similarity coefficient of 0.8063 on the task of breast tumor segmentation in DCE-MRI,and the method is superior to other methods in this experiment on all segmentation metrics.In addition,the proposed modules and loss function have effectively improved the segmentation accuracy of the network.The experimental results show that the method proposed in this paper has better segmentation performance and smaller model capacity,which has certain reference significance and clinical application value.