A Serial Image Segmentation and Registration Network (SISRN) for Dynamic Enhanced Magnetic Resonance Image Sequences

Chronic kidney disease (CKD) significantly affects patients' quality of life, and early detection and intervention are necessary. Dynamic renal glomerular filtration rate (GFR) analysis from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) often serves as a non-invasive diagnostic approach for early detection of CKD. However, the segmentation and registration of multi-phase DCE-MRI remains time-consuming and challenging. In this paper, we propose a serial image segmentation and registration network (SISRN) to efficiently align DCE-MRI sequences. Our approach uniquely incorporates segmentation into the registration network, ensuring that the image similarity loss is influenced by both the accuracy of image matching and the segmentation results. Additionally, a temporal consistency loss is applied to ensure the longitudinal smoothness and continuity of segmentation. The feature fusion layer is also used to integrate the segmentation and registration paths of joint training. Experimental results demonstrate the effectiveness of the proposed method, highlighting its potential for clinical application.