Rapid and accurate acquisition of hippocampal volume changes in brain M RI images is of great significance for the diagnosis of diseases such as Alzheim er′s disease.The proportion of hippocampus in the brain is small and its bounda ry with the surrounding structure is not obvious,which makes the hippocampus se gmentation in nuclear magnetic resonance imaging based on deep learning difficul t.Aiming at these two problems,a hippocampal segmentation algorithm based on b oundary-enhanced loss training mechanism with attention mechanism is proposed. T he main contributions are as follows:1) A U-shaped three-dimensional convoluti on al neural network with attention mechanism is designed; 2) A boundary-enhanced l oss function is proposed to solve the problem of imbalance between the hippocamp us and the background because of the large difference in size.At the same time, the network pays more attention to the learning of the hippocampus boundary dur ing training.On the European Alzheimer′s Disease Association and the Alzheimer ′ s Neuroimaging Initiative dataset,the performance of the loss function and netw ork structure proposed in this paper are analyzed and discussed.Compared with s everal advanced deep learning-based 3D segmentation algorithms,the experimenta l results show that the proposed algorithm performs better,and the Dice accuracy of 89.41% is similar to the manual segmentation accuracy.