Unsupervised Structure-Aware Cascaded Recurrent Network Based Nonrigid Motion Correction for Ultrafast Power Doppler Imaging

The visualization and quantitation of microvessels attract increasing attention because microvessel density and hemodynamics are closely related to disease and neural development. Ultrafast power Doppler imaging (uPDI) is a non-invasive technology to image and quantify microvessels with high sensitivity. However, it is sensitive to tissue motion, which is unavoidable during in-vivo imaging. Current motion correction methods used in uPDI either cannot estimate nonrigid motion or cannot be implemented quickly. Therefore, in this study, we propose an unsupervised structure-aware recurrent cascaded coarse-to-fine motion correction network for uPDI. Dealing with complex and large physiological motions, we adapt a coarse-to-fine strategy that first performs affine transformation correction and then nonrigid motion correction. Considering that affine transformation has less impact on image correlation, recurrent cascaded architecture is introduced during the inference stage of the affine network to improve the accuracy of affine transformation correction. Edge structures are automatically extracted to guide the proposed method to register well globally. The uPDI images generated after motion correction are visualized and the profiles of local microvessels, full width at half maximum (FWHM), and blood-to-tissue ratio (BTR) are calculated to evaluate the motion correction performance of the proposed method. Normalized cross-correlation based nonrigid motion correction and a phase correlation-based motion correction method are selected as the comparison methods. The results show that the proposed method can generate better uPDI images with lower FWHM, higher BTR, and lower computational time, demonstrating that it can be a competitive motion correction tool for uPDI.