期刊
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
卷 68, 期 -, 页码 -出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.bspc.2021.102697
关键词
Medical image fusion; CNNs; Spiking cortical model; Local extrema scheme; Siamese network
资金
- Famous teacher of teaching of Yunnan 10000 Talents Program, major science and technology project of Yunnan Province [2018ZF017]
- National Natural Science Foundation of China [62066047, 61861045, 21461029, 21561033]
- Yunnan University's Research Innovation Fund for Graduate Students [2020298]
This paper proposes a novel framework combining the local extrema scheme and Siamese network to improve the decomposition efficiency of the scheme and customize the fusion strategy. By enhancing the decomposition efficiency and fusion strategy, the framework achieves better performance compared to other state-of-the-art methods according to extensive experimental results and metrics.
Multimodal medical image fusion is an auxiliary approach to help doctors diagnose diseases accurately leveraging information enhancement technology. Up to now, none of the fusion strategies is authoritative. Exploring methods with excellent performance is still the theme of image fusion works. The local extrema scheme (LES) and convolutional neural networks (CNNs) perform remarkable in medical image fusion tasks. However, the low decomposition efficiency of the LES and the limitations of CNNs should be addressed. Therefore, a novel framework proposed by combining the local extrema scheme and a Siamese network. This paper tried to solve the mentioned issues by improving the decomposition efficiency of LES and customizing the fusion strategy. Initially, the multi-scale local extrema scheme (MSLES) is introduced to decompose the source image into a series of detailed layers and a smoothed layer. Simultaneously, an adaptive dual-channel spiking cortical model (ADCSCM) based on the image information entropy (EN) is constructed to fuse the smoothed layer, and subsequently a feasible weight allocation strategy is designed by combining the Siamese network and EN to fuse the detailed layers. Ultimately, the informative image is reconstructed with the fused smoothed layer and detailed layers. By analyzing the extensive experimental results and metrics, the proposed framework achieves better performance against other state-of-art methods.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据