Multi-scale Fusion Medical Image Registration with Segmentation Mask
Fuyang Xing, Zhiyue Yan, Wenming Cao, Yubo Huang · 2023
Image registration plays an indispensable role in medical imaging analysis. CNN-based approaches have been a popular trend in medical image registration. Although it can achieve promising registration performance, they usually capture only partial information about the image. In this paper, we propose a multi-scale fusion medical image registration network with segmentation mask (MSFNet). The network utilizes multi-scale strategy optimization, which can improve the registration performance by avoiding the similarity gradients of most recursive cascade approaches to optimize at coarse resolutions. Meanwhile, we design a Self-Attention-based registration network, which can establish long-distance connections for global images. Extensive evaluations on three public lung X-rays datasets (JRST, Montgomery and Shenzhen) show that MSFNet outperformed other baseline methods on various metrics. Meanwhile, the time and GPU memory consumption are also lower.