A Two-step Registration Approach: Application in MRI-based Strain Calculation of the Left Ventricle

摘要

Registration and finding anatomical correspondent landmarks amongst the volumetric images in both inter and intra-subject studies are prevalent challenges, especially in the field of cardiac image analysis. On the other hand, among other functional criteria, the strain is promising to identify differences in the early stages of cardiac diseases. This study aims to develop a transformation model on the existing FFD non-rigid registration algorithm using combining the shape-based and the intensity-based approaches. A novel two-step multi-resolution non-rigid FFD-based registration algorithm was proposed to measure the radial strain of the left ventricle during the cardiac cycle using cine-MRI images. The endocardial wall shape information was introduced to improve the registration accuracy by combining the original intensity-based registration with the shape-based registration approach. The proposed algorithm was evaluated on ten sequences of cine-MR images: First, the proposed algorithm was validated by two identified 3D-landmarks and then applied to a healthy subject and an LVH patient to calculate the left ventricle regional radial strain during a cardiac cycle. Regarding two anatomical landmarks, their estimated displacements from our proposed two-step registration algorithm showed a better match with the reference values compared to the classical registration method. Moreover, a comparison of normal and abnormal radial segmental strain values calculated by our proposed algorithm showed a clear difference in their functional properties. The promising results showed that the proposed registration algorithm outperformed the conventional one in terms of the accuracies of the point-tracking.