3D translation estimation using the monogenic orientation vector

Three dimensional motion estimation is an active research field in ultrasound imaging, motivated by the recent progress in 3D acquisition. Out-of-plane motion and pattern decorrelation induced by azimuthal displacement yield bias in 2D estimation methods. However, 3D estimation may easily become time-consuming because of the large amount of data. Authors proposed different ways to estimate 3D displacement, for the most part using Normalized Cross-Correlation (NCC) combined with original refinement methods. In this paper, we propose an alternative which uses the 3D local orientation, obtained using the monogenic signal, in order to estimate 3D local translations. We show that the use of local orientation provides better results than the NCC and the classical optical flow approach. Results on a 3D simulated ultrasound volume show that the proposed estimation is more robust to noise than classical methods. Giving a signal to noise ratio of 25 dB, results show that the mean absolute error of our orientation-based optical flow estimator is respectively (47.8%, 62.1%, 84.9%) lower than the one generated when using intensity-based optical flow in lateral, azimuthal and axial direction.