An Attention-oriented U-Net Model and Global Feature for Medical Image Segmentation

Yandong Han ； Jiangjiang Li

《淡江理工學刊》 23卷4期 (2020/12) Pp. 731-738

https://doi.org/10.6180/jase.202012_23(4).0018

摘要

Aiming at the problem that medical images are easily disturbed by various factors, resulting in uneven gray scale and blurred boundary, which increases the difficulty of segmentation. In this paper, we propose an attention-oriented U-Net model for medical image segmentation. The original convolutional layer is replaced by reside-density module, and the transpose convolution and scaling convolution modules are used for up-sampling. Meanwhile, the feature layers at different levels are processed by attention mechanism, which can extract more features and promote network convergence. Experimental results show that the proposed U-Net structure can improve the precision and efficiency of medical image segmentation.

關鍵字

U-Net model ； medical image segmentation ； attention ； reside-density module