Estimation of the Spatial Resolution of a 2D Strain Estimator Using Synthetic Cardiac Images

Although myocardial deformation imaging is often used in clinical practice, to date, the spatial resolution of any speckle tracking algorithm has not been tested. The minimum size of the resolvable dysfunctional areas with these techniques remains unknown while this is relevant for reliable clinical diagnosis. The current study therefore aimed at determining the spatial resolution of our previously validated non-rigid image registration (NRIR)algorithm in 2D ultra-realistic in-silico B-mode images with eight varying sizes of dysfunctional regions. A correlation coefficient obtained between the NRIR-end-systolic (ES)strains and the true ones was 0.97, indicating a good tracking performance. The obtained mean ES strain errors and limits of agreement were 0.19% and -3.37~3.76% respectively (p <; 0.01). Based on the pixel-wise ES strain differences between the normal heart model and the model with dysfunctional regions, the sensitivity, specificity and accuracy to detect dysfunction were computed, where a region was considered resolved if more than 60% of the pixels were correctly classified. Furthermore, a correlation coefficient of 0.92 as determined by the correlation between the NRIR-estimated sizes of the dysfunctional regions against the ground truth sizes is suggestive of good tracking resolution. In this way our study revealed that the discriminative power of our algorithm was adequate to resolve dysfunctional areas down to 1.9 cm.