A VMamba Attention Network for Efficient Deformable Image Registration in Lung 4DCT

Image registration is a foundational technique in medical image analysis; however, it is often hindered by issues such as slow processing speed and limited accuracy. In this work, we propose the Visual Mamba Attention Network (VMANet) for lung image registration. VMANet incorporates a dual-stream Mamba Attention module, enabling the effective capture of global spatial and channel information. This enhances its ability to perform precise feature extraction and deformation field prediction, making it well-suited for image registration tasks. Finally, a spatial transformation layer leverages the deformation field to warp the moving image to the fixed image. Comprehensive experiments demonstrate that our approach outperforms existing convolutional neural network-based registration methods in terms of both registration accuracy and real-time performance. The proposed VMANet model thus shows significant promise for applications in pulmonary studies.