Myocardial deformation imaging is a well-established echocardiographic technique for the assessment of myocardial function. Although some solutions make use of speckle tracking of the reconstructed B-mode images, others apply block matching on the underlying radio-frequency (RF) data in order to increase sensitivity to small inter-frame motion and deformation. However, for both approaches, lateral motion estimation remains a challenge due to the relatively poor lateral resolution of the ultrasound image in combination with the lack of phase information in this direction. Hereto, non-rigid image registration (NRIR) of B-mode images has previously been proposed as an attractive solution. However, hereby, the advantages of RF-based tracking were lost. The aim of this study was therefore to develop an NRIR motion estimator adopted to RF data sets. The accuracy of this estimator was quantified using synthetic data and was contrasted against a state of the art block matching solution. The results show that RF-based NRIR outperforms BM in terms of tracking accuracy particularly, as hypothesized, in the lateral direction. Finally, this RF-based NRIR algorithm was applied clinically, illustrating its ability to estimate both in-plane velocity components in-vivo.