基于多尺度残差双域注意力网络的乳腺动态对比度增强磁共振成像肿瘤分割方法

针对乳腺肿瘤大小形态多变、边界模糊以及前景与背景间严重类不平衡的问题，该文提出一种多尺度残差双域注意力融合网络。该网络以多尺度卷积构成的多尺度残差块作为基本搭建模块，通过提取多尺度特征和优化梯度传播通道提高其识别不同尺寸目标的能力，同时融入双域注意力单元，提高网络的边缘识别和边界保持能力。另外该文提出一种混合自适应权重损失函数改善网络优化方向，缓解正负样本极度不均衡的影响。实验结果表明，该文所提方法的平均骰子相似系数(Dice)值达到0.806 3，较U形网络(UNet)提高5.3%，参数量下降73.36%，具有更优的分割性能。

Segmentation Algorithm of Breast Tumor in Dynamic Contrast-Enhanced Magnetic Resonance Imaging Based on Network with Multi-scale Residuals and Dual-domain Attention

Considering the problems of breast tumor size and shape change, blurred boundary and severe class imbalance between foreground and background, a multi-scale residual dual-domain attention fusion network is proposed. In this network, multi-scale residual blocks composed of multi-scale convolution are used as the basic building modules. 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. Meanwhile, the dual-domain attention units are integrated into the network to improve the ability of edge recognition and boundary preservation. The hybrid loss function with adaptive weight is proposed, it can improve the optimization direction of the network, alleviate the influence of the extreme imbalance of positive and negative samples. The experimental results show that the average Dice value of the method proposed in this paper reaches 0.8063, which is 5.3% higher than that of U-shaped Network (UNet), and the number of parameters is reduced by 73.36%, which has better segmentation performance.