A new cardiac motion vector field estimation method based on the optical-flow method with additional constraint from motion of an anatomical feature in 4D cardiac PET

Objectives The goal is to develop and evaluate a new cardiac motion vector field (MVF) estimation method based on the conventional optical-flow algorithm with an additional constraint from the motion of an anatomical feature from 4D cardiac gated PET images to improve cardiac motion estimation accuracy.

Methods Noisy cardiac-gated PET projection data were generated from the 4D XCAT phantom with realistic cardiac MVF model with 8 gated frames and reconstructed using the STIR package. A customized high-resolution PET scanner with a system resolution of 0.78 mm was used in the analytical simulation for accurate feature extraction and the iterative OS-EM algorithm was used in image reconstruction. The interventricular sulcus (IS) was extracted from 3D cardiac PET image at each gated frame, and the motion along its length was estimated and decomposed into the radial, circumferential, and longitudinal components. Similar to our previous implementation, a sulcus-based initial cardiac MVF estimate was extrapolated from the motion of extracted IS based on the a priori knowledge about the spatial variation of three cardiac motion components in myocardium. It was also used as an additional constraint in the new optical-flow algorithm to enforce sulcus matching between two cardiac frames. The constraint was assigned larger weight for voxels closer to the IS. The new motion estimation algorithm was combined with the S-initial, which was based on the extracted IS motion and cardiac motion model, for feature-based cardiac motion estimation. In comparison, the conventional optical flow algorithm was applied with initial estimates: 0-initial, where the initial cardiac MVF has zero values; T-initial, where the ‘true’ cardiac MVF of the 4D XCAT phantom was used; as well as the S-initial. The motion estimation results with four methods were compared to the true cardiac MVF of the 4D XCAT phantom for quantitative evaluation.

Results In frames with significant cardiac motion, the cardiac MVF estimates using the conventional optical-flow algorithm and the S-initial were more accurate than that using the 0-initial, and were comparable to that using the T-initial. When the new motion estimation method with the additional constraint term and the S-initial were used, the accuracy of the cardiac MVF estimate is further improved, surpassing that of the conventional motion estimation method using the T-initial.

Conclusions A cardiac MVF estimation method based on the conventional optical flow algorithm and an additional constraint term that incorporate the motion of the IS was developed and evaluated using analytically simulated 4D cardiac-gated MP PET data. The new motion estimation method provides improved estimate of the cardiac MVF, providing important additional diagnostic information from 4D MP PET studies.