DAU‐Net: An unsupervised 3D brain MRI registration model with dual‐attention mechanism

Abstract

AbstractMedical image registration plays an important role in clinical treatment. However, the convolution‐based registration frameworks fail to address the local excessive deformation between images. Furthermore, the folding point in the displacement vector field (DVF) reduces the reliability of registration results. In this study, we propose a dual‐attention mechanism‐based U‐shaped registration framework (dubbed DAU‐Net). Firstly, the multi‐scale attention mechanism is introduced to extract the long‐range dependence to deal with the local excessive deformation. Then, the channel attention mechanism is proposed to enhance the information fusion between channels, which not only fuses the features between different layers in the dual‐attention network but also improves the non‐linear mapping ability of the registration network. In the end, the objective function with the folding penalty regularization term is designed to improve the smoothness of the DVF. The model is evaluated on LPBA40 and Mindboggle101 open datasets. The registration accuracy in LPBA40 and Mindboggle101 datasets has been increased by 2.9% and 3.1%, respectively, while the folding rate is reduced by nearly 40 times compared with VoxelMorph. Combined multi‐scale attention mechanism with channel attention mechanism, the registration accuracy of DAU‐Net is improved. By utilizing the folding penalty regularization term, the folding rate is decreased significantly.