Abstract:

In order to segment spine accurately from X-ray images, a method of spine X-ray image segmentation based on deep learning is proposed, which uses U-Net network based on multi-scale feature fusion. The convolutional layer in the U-Net model is replaced with an Inception network to extract features of different scales and perform multi-scale fusion. At the same time, the residual connection layer is added in front of the skip connection, and the convolution block attention module is added in front of the first up-sampling layer. In this paper, 20 spinal X-ray images are verified by the model, and the Dice coefficient is 0.8457, which is 0.1351 higher than the recent method of spinal X-ray image segmentation.

Key words: spinal X-ray image, U-Net, image segmentation, convolution block attention module

摘要：

为了精确地从X线图像中分割脊柱，提出了一种基于深度学习的脊柱X线图像分割方法，使用基于多尺度特征融合的U-Net网络进行分割。将U-Net模型中的卷积层替换成类Inception网络来提取不同尺度的特征，并进行多尺度融合。同时在跳跃连接前增加残差连接层，并在首次上采样前添加卷积块注意力模块。该模型对20幅脊柱X线图像进行验证，Dice系数为0.845 7，与近期X线脊柱图像分割方法相比，提高了0.135 1。

关键词: 脊柱X线图像, U-Net, 图像分割, 卷积块注意力模块