A Brief Survey of Image Registration Techniques in Dynamic Myocardial Perfusion CT Imaging

This survey provides an overview of image registration techniques in dynamic myocardial perfusion Computed Tomography (CT) imaging by reviewing 25 selected papers. The analysis focuses on three key aspects: (i) the methods and techniques employed for image registration including traditional and deep learning-based approaches; (ii) the performance achieved by each method with an emphasis on identifying the best-performing techniques based on metrics such as Dice coefficient, Target Registration Error (TRE) and Structural Similarity Index (SSIM); and (iii) the datasets utilized in these studies including both benchmark and real-time clinical datasets. The survey highlights current trends, comparative performance and key challenges in dynamic myocardial perfusion CT registration for providing insights for future research directions in improving accuracy, robustness and clinical applicability. The best performances were achieved by Recursive cascade network (RCN) with the highest Dice coefficient (0.9962) and SSIM (0.90) and Lung respiratory motion estimation with Kalman filtering and 4D CT (LRME-4DCT) with the lowest TRE$\boldsymbol{(} \mathbf{0. 8 5 ~ m m} \boldsymbol{)}$.