Material Decomposition of Dual-Energy CT Based on CBAM Attention U-Net

作者：Xinrui Zhang, Lei Li, Bin Yan · 发表于：Journal of Physics Conference Series · 年份：2022 · DOI：10.1088/1742-6596/2363/1/012020 · 被引用次数：3 · 研究领域：Advanced X-ray and CT Imaging、Medical Imaging Techniques and Applications、Radiation Dose and Imaging

Spectral CT has the ability of quantitative material analysis by exploring the difference in the attenuation properties under different X-ray energies, and has become an important direction for the development of CT technology. However, due to the non-linearity and inconsistency of the observation, there is noise boosting in material decomposition and serious decrease of basis material images. In this paper, we design a U-Net convolutional neural network structure with double-entry and double-out to realize the effective material decomposition of high- and low-energy images. When designing the network, we introduce the attention mechanism, and add CBAM modules combining channel attention and spatial attention to improve the overall performance of the network. We also design an ablation experiment to test the module from a vertical perspective, and compare the material decomposition performance of each network from a horizontal perspective. The results show that the proposed Attention U-Net (AT-U-Net) has advantages in reducing noise boosting and improving the quality of basis material images.