基于U型网络的X线头影图像特征点自动定位

为了提高口腔临床诊断效率，提出一种基于U型生成对抗网络的X线头影测量图像结构特征点自动定位方法。该方法在生成对抗网络框架下构建U型生成器网络，用于学习从当前图像到目标特征点偏移距离图的映射；再构建鉴别器网络以判断预测的偏移距离图是否与真实数据一致。将新采集的X线头影测量图像作为已训练的U型生成器网络输入，得到新图像针对目标特征点的偏移距离图，然后通过回归投票方法从预测出的偏移距离图中获得检测目标特征点坐标。实验结果表明，该方法相较于其他方法具有更高的检测成功率，能较准确地获得X线头影测量图像中结构特征点的位置。 

Automatical Localization of X-ray Cephalometric Images Landmarks via U-type Network

In order to improve the efficiency of oral clinical diagnosis, this paper proposes an automatic location method for structural feature points of X-ray cephalometric images based on U-type generative network. This method constructs a U-type generator network under the framework of a generative adversarial network to learn the mapping from the current image to the target landmark offset distance map; then builds a discriminator network to determine whether the predicted offset distance map is consistent with the real data. That is to say, in this paper, image is the output of the U-type generating adversarial network, rather than the common output displacement or coordinate value. The newly acquired X-ray cephalometric image is used as the input of the trained U-type generator network to obtain the offset distance map of the new image to the target feature point, and then the detection is obtained from the predicted offset distance map through the regression voting method Target feature point coordinates. The experimental results show that the method in this paper gets higher detection rate than others,and it can accurately obtain the position of the structural feature points in the X-ray cephalometric image.