Marius-Cristian Ureche, Adrian Basarab, Mircea-Florin Vaida, and Denis Kouame
affine and non-rigid image registration, local phase-based image, free-form deformation, the monogenic signal
Medical image registration is of great interest nowadays and is considered a reliable technique to improve diagnosis, therapy and analysis. The majority of pixel-based registration methods use the native intensity during the registration methods. However, in the last years newer methods based on local-phase were proposed. So far, there is no objective validation method for evaluating and comparing these two registration strategies. Hence, the objective of this paper is to comparatively analyze the registration using both approaches on synthetically images which enable us to have a ground truth. We deeply analyze the use of local phase images, obtained using the phase of the monogenic signal, in automatic image registration algorithms. In order to have a ground truth we simulate ultrasound (US) images starting from magnetic resonance (MR) slices. We use a realistic approach, constructing a map of randomly distributed ultrasound scatterers and the simulation tool Field II. Moreover, we compare the standard intensity-based approach with the local phase-based approach and study the influence of input images on the success of registration algorithms using both affine transformation and Free-Form Deformation with B-splines. We also show how the use of local phase images affects the successful use of certain simple similarity metrics like sum of squared differences (SSD), sum of absolute differences (SAD), correlation coefficient (CC) and also more complicated metrics like mutual information (MI) and residual complexity (RC).
Important Links:
Go Back
