基于Dense-UNet++的关节滑膜磁共振图像分割

为解决以往基于深度学习的滑膜磁共振图像分割模型存在的分割精度较低、鲁棒性较差、训练耗时等问题，本文提出了一种基于Dense-UNet++网络的新模型，将DenseNet模块插入UNet++网络中，并使用Swish激活函数进行训练．利用1 036张滑膜磁共振图像数据增广后的14 512张滑膜图像对模型进行训练，并利用68张图像进行测试．结果显示，模型的平均DSC系数为0.819 9，交叉联合度量（IOU）为0.927 9．相较于UNet、ResUNet和VGG-UNet++网络结构，DSC系数和IOU均有提升，DSC振荡系数降低．另外在应用于相同滑膜图像数据集和使用相同的网络结构时，Swish函数相比ReLu函数有助于提升分割精度．实验结果表明，本文提出的算法对于滑膜磁共振图像的病灶区域的分割有较好的效果，能够辅助医生对病情做出判断．

Magnetic Resonance Images Segmentation of Synovium Based on Dense-UNet++

To further improve the segmentation accuracy, robustness, and training efficiency of existing articular synovium segmentation algorithms, a new deep learning network based on Dense-UNet++ was proposed. First, we inserted the DenseNet module into the UNet++ network, then applied the Swish activation function to train the model. The network was trained through 14 512 synovial images augmented from 1 036 synovial images, and tested through 68 images. The average accuracy of the model reached 0.819 9 for dice similarity coefficient (DSC), and 0.927 9 for intersection over union (IOU) index. Compared with UNet, ResUNet and VGG-UNet++, DSC coefficient and IOU index were improved, and DSC oscillation coefficient reduced. In addition, when applied in the same synovial image set and using the same network structure, the Swish function can help improve the accuracy of segmentation compared with the ReLu function. The experimental results show that the proposed algorithm performs better in segmenting articular synovium and may assist doctors in disease diagnosis.