A measure of confidence for Phase-Based Motion Estimator applied to 2D US-TO images

Recent studies showed the potential of the Phase Based Motion Estimator when applied to ultrasound images containing transverse oscillations. Even though the emergence of ultrafast or high frame rates imaging overcame some of the prior limitations of the algorithm regarding its capability to detect large displacement values, the method still presents an important lack of robustness. In this paper, we propose an improvement of the Phase Based Motion Estimator, by introducing a measure of confidence. The robustness of the method is indeed highly influenced by the orientations of the patterns present in radio frequency images containing transverse oscillations. The current phase based method relies on a specific pattern, containing pixels oriented in the two perpendicular directions of the oscillations. Any deviation from the assumed image model results in a underperforming estimate. The detection of the orthogonal patterns, which are in concordance with the theoretical image model, could guarantee the favorable outcome of the algorithm. A promising tool for identifying the desired patterns is given by Gabor filters oriented in 0 and 90 degrees direction. The magnitude response of the filter to the image orientations can be then grouped in two classes by applying the k-Means method. The class described by high intensity values is expected to guarantee the accuracy of the motion estimator. By keeping the regions corresponding with the first class and discarding the ones identified as belonging to the second class, we eliminate the underperforming estimates and increase the robustness of the method. The validation on simulated ultrafast images improved estimates with error reduced by 73% in the axial direction and 53% in the lateral direction, compared to the case when no measure of confidence was used.