Alpha divergence based registration of dynamic scans for MR cystography

In this paper we introduce a novel non-rigid registration algorithm using an information-theoretic measure, i.e., alpha information, which is an informatics merit for two random series and has shown its potential for image registration application. The alpha parameter in the formulation of alpha-information is tunable. This nature helps us to adaptively choose the right metrics for each situation and improve registration accuracy. In this paper, we further adapt the maximum a posteriori (MAP) principle to search an optimal solution for organ segmentation in the presence of random noise. Thus, an integrated approach is proposed to register dynamic scans for compensation of bladder wall motion and deformation via alpha registration and to segment the patient-specific bladder wall from the registration of dynamic scans, which were impaired by relatively low signal to noise ratio (SNR) in each time frame. The proposed approach was evaluated by patient datasets with experts' knowledge as ground truth. Experimental results demonstrated that the alpha registration method can effectively capture the bladder wall motion and deformation, and the MAP segmentation algorithm is adequate to delineate the bladder wall from the low SNR dynamic scans after motion compensation.