Joint multi-object registration and segmentation of left and right cardiac ventricles in 4D cine MRI

Abstract

The diagnosis of cardiac function based on cine MRI requires the segmentation of cardiac structures in the images, but the problem of automatic cardiac segmentation is still open, due to the imaging characteristics of cardiac MR images and the anatomical variability of the heart.In this paper, we present a variational framework for joint segmentation and registration of multiple structures of the heart. To enable the simultaneous segmentation and registration of multiple objects, a shape prior term is introduced into a region competition approach for multi-object level set segmentation. The proposed algorithm is applied for simultaneous segmentation of the myocardium as well as the left and right ventricular blood pool in short axis cine MRI images. Two experiments are performed: first, intra-patient 4D segmentation with a given initial segmentation for one time-point in a 4D sequence, and second, a multi-atlas segmentation strategy is applied to unseen patient data. Evaluation of segmentation accuracy is done by overlap coefficients and surface distances. An evaluation based on clinical 4D cine MRI images of 25 patients shows the benefit of the combined approach compared to sole registration and sole segmentation. PURPOSECardiovascular diseases are reported to be the leading cause of death in the western countries. 1 Cine MRI has been proven to be an accurate and reproducible imaging modality for the diagnosis of cardiac function, 2 however, the computation of quantitative measures as ventricle volumes, masses and ejection fraction requires the segmentation of the left (LV) and right (RV) ventricles from cine MR images. Despite of numerous automatic cardiac segmentation approaches in the literature, the problem of LV+RV segmentation in MRI is still open, due to the characteristics of cardiac MR images and to the great variability of the images among patients. 3 Cardiac segmentation algorithms for 4D cine MRI images include intensity-based methods, deformable models, statistical shape and appearance models and atlas-patient registration methods (see Petitjean et al. 3 for an overview). Different 3D and 4D registration-based algorithms have been reported to show good correlations to manual ground-truth segmentations. 4-7 However, the variability of shape and appearance of the heart in MRI is a challenge for registration-based heart segmentation approaches.In this paper, we propose an algorithm for the simultaneous level set-based segmentation and non-linear registration of multiple objects in two images. Some approaches for joint registration and segmentation can be found in the literature, but most of these methods are designed for the segmentation of a single object 8-10 or are constrained to linear registration. 11 In our approach, each object to segment is represented by a level set function and region competition 12 is used as segmentation approach. The joint segmentation and registration problem is solved by minimizing a variational functional that incorporates prior shape knowledge, prior intensity informat...

A Novel 3D-to-3D Diffeomorphic Registration Algorithm With Applications to Left Ventricle Segmentation in MR and Ultrasound Sequences

“…Others have developed methods to jointly performed segmentation and registration for delineation of the blood pools of the left and right ventricles and the myocardium [9]. The authors combine a level-set segmentation along with registration of two volumes, where multiple anatomical regions are able to be segmented.…”

“…The derivatives for each of the f 1 , f 2 , f 3 , and f 4 terms are computed with respect to x, y, and z. Three Poisson equations of the form are given below, where the relevant derivative terms have been combined from (9), where f i k = d f i dk , with i=1,2,. .…”

Left Ventricle Motion Estimation in Cine MRI With Multilayer Iterative Deformable Graph Matching

“…Their algorithm's performance was evaluated by computing the average SCDs for the right ventricle (RV) and LV endocardium and LV epicardium contours. Ehrhardt et al [58] conducted experiments showing that the mean SCD of the RV is approximately 3 mm and the mean SCD of LV is approximately 2 mm for the HSC dataset. Based on this conclusion, we determined the mean SCD of the LV to be approximately 2.29 mm in [57].…”

An Overview on Image Registration Techniques for Cardiac Diagnosis and Treatment

“…Furthermore, the LV has a relatively simple geometry with thick myocardial walls, making its automated segmentation more feasible. Compared with LV, studies on image registration for the segmentation of the right ventricle (RV) [ 107 , 111 ] and LA [ 65 , 112 , 113 ] are few due to the more complex geometry of these chambers and their much thinner walls. However, these chambers are associated with many critical diseases, such as modeling in patients with pulmonary hypertension [ 111 , 114 ] or LA enlargement [ 65 , 112 , 113 , 115 , 116 ].…”