Abstract: Due to the complex structure and small size of the hippocampus, it is difficult to segment the hippocampus accurately. To address this issue, a method of the generative adversarial network (GAN) segmentation hippocampus method based on global and local feature information (GLGAN) is proposed. Firstly, in order to improve the stability of the network and the accuracy of hippocampus segmentation, and reduce the problems of information loss and gradient explosion, the global generative adversarial network method (GGAN) is proposed by improving the generator and loss function of GAN. Secondly, since the discriminator is essentially a binary classifier and is not sensitive to small local changes, a GAN method of dual discriminator network structure with global and local feature information is proposed. Finally, a total loss function is designed to balance GAN adversarial loss and 3D u-net segmentation loss. The experimental results show that the GLGAN is conducive to the intensive evaluation of the hippocampus, promotes the discriminator to push the mask value generated by the generator to a more realistic distribution, and improves the segmentation accuracy of the hippocampus. The Dice coefficient and IOU of GLGAN segmentation of hippocampus are 0.804 and 0.672 respectively.

Key words: Generative adversarial network (GAN), 3D CNN, Segmentation, Hippocampus, 3D u-net