显著性引导及不确定性监督的深度编解码网络

基于U-Net的编码-解码网络及其变体网络在医学图像语义分割任务中取得了卓越的分割性能.然而，网络在特征提取过程中丢失了部分空间细节信息，影响了分割精度.另一方面，在多模态的医学图像语义分割任务中，这些模型的泛化能力和鲁棒性不理想.针对以上问题，本文提出一种显著性引导及不确定性监督的深度卷积编解码网络，以解决多模态医学图像语义分割问题.该算法将初始生成的显著图和不确定概率图作为监督信息来优化语义分割网络的参数.首先，通过显著性检测网络生成显著图，初步定位图像中的目标区域；然后，根据显著图计算不确定分类的像素点集合，生成不确定概率图；最后，将显著图和不确定概率图与原图像一同送入多尺度特征融合网络，引导网络关注目标区域特征的学习，同时增强网络对不确定分类区域和复杂边界的表征能力，以提升网络的分割性能.实验结果表明，本文算法能够捕获更多的语义信息，在多模态医学图像语义分割任务中优于其他的语义分割算法，并具有较好的泛化能力和鲁棒性.

Deep Encoder-decoder Network with Saliency Guidance and Uncertainty Supervision

The encoder-decoder network based on U-Net and its variants have achieved excellent performance in semantic segmentation of medical images. However, part of spatial detailed information is lost during feature extraction, which affects its accuracy in the segmentation task. For another, the ability of generalization and robustness of these models are unsatisfactory in multiple modalities of medical image semantic segmentation tasks. In view of the above problems, we propose a deep convolutional encoder-decoder network with saliency guidance and uncertainty supervision, to solve the semantic segmentation problem in multimodal medical images. In this method, the initial generated saliency map and uncertainty probability map are used as the supervised information to optimize the parameters of semantic segmentation network. Firstly, saliency map is generated by saliency detection network to locate the target region of image preliminarily. Then, the uncertainty probability map is generated by calculating the set of pixel points with uncertain classification, which based on saliency map. Finally, these two maps are sent into the multi-scale feature fusion network together with the original image, guiding the network to focus on learning the feature of target region, and enhancing the representational capacity of the uncertain classification regions and complex boundaries, so as to improve the segmentation performance of the network. Experimental results show that the proposed method can capture more semantic information which outperforms existing semantic segmentation methods in multimodal medical image semantic segmentation tasks, and has strong generalization capability and robustness.