3D–3D Rigid Registration of Echocardiographic Images With Significant Overlap Using Particle Filter

Abstract

The precise alignment of 3D echocardiographic images taken from different views has been shown to enhance image quality and increase the field of view. This study proposes a novel sequential Monte Carlo (SMC) algorithm for the 3D-3D rigid registration of echocardiographic images with significant overlap that is robust to the noise present in ultrasound images. The algorithm estimates the translational and rotational components of the rigid transform through an iterative process and requires an initial approximation of the rotation and translation limits that depend on the dimension of the image and the initial overlap between images. The registration is performed in two ways: the same transform approach applies the transform computed for the end diastolic frame to all frames of the cardiac cycle, whereas the unique transform approach registers each frame independently. The proposed SMC and exhaustive search algorithms were evaluated for 3D transthoracic echocardiographic volumes recorded from 3 patients and 3 volunteers who participated in two different research studies conducted at the Mazankowski Alberta Heart Institute. The evaluations demonstrate that the same transform approach yielded a Dice score value of $0.716~\pm ~0.041$ for the left ventricle and required less computational time than the unique transform approach or exhaustive search. It was found that the SMC algorithm performs better than the exhaustive search at the 0.05 significance level using the paired t test. The accuracy was improved further using the Simple Elastix non-rigid registration algorithm to fix misalignments due to movement and breathing, with an overall Dice score value of $0.787~\pm ~0.038$ .

Keywords

• Sequential Monte Carlo
• particle filter
• 3D-3D rigid registration
• echocardiographic images