LightMHS: A Smaller Hippocampus Segmentation Network Based on MobileViT

Abstract

The morphological analysis and volume measurement of the hippocampus is critical for the study of many brain diseases. Therefore, to assist doctors in diagnosis and treatment, a light and accurate hippocampus segmentation method is urgently needed in clinical practice. U-net and its latest variants have become a hot network for medical image segmentation in recent years, and the architecture based on Transformer has also received extensive attention. However, many networks are not ideal in practical application due to their large number of parameters and high computational complexity. We propose a lightweight 3D hippocampus segmentation model which combines the advantages of CNN and Vision Transformer (ViT). In order to obtain local context information, the encoder first utilizes 3D CNN to extract spatial feature maps, and we propose an attention module to learn spatial and channel relationships. Considering the importance of local features and global semantics for 3D segmentation, we introduce a lightweight ViT to further model local and global information. To evaluate the effectiveness of encoder feature representation, we design three decoders of different complexity to generate segmentation maps. We validated our model on three public hippocampus datasets, and the experimental results show that compared with other models, we achieved more beneficial performance with fewer parameters and lower computational complexity.

Note:
Funding Information: The authors are grateful to the reviewers for their valuable comments, which have greatly improved the paper. This work was supported by the Natural Science Foundation of Jiangsu Province under Grant BK20190079 and the National Key Research and Development Program of China under Grant SQ2017YFGH001005.
Declaration of Interests: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Keywords: Vision Transformer, 3D CNN, Lightweight, Hippocampus Segmentation