Automatic left ventricle segmentation based on multiatlas registration in 4D CT images

Abstract

Cardiac CT angiography (CCTA) is widely used in the diagnosis of coronary heart disease. It can provide 4D (3D + t) sequence with high spatial and temporal resolution. Segmentation of left ventricle (LV) in 4D CCTA sequence can provide useful information for clinical practice. In this paper, we present an automatic method for LV segmentation in 4D CCTA sequence in this paper. This method mainly relies on an accurate multi-atlas registration method. Thus, we first improve the multi-atlas registration method presented by Kirişli et al. by adding an extra registration step with an estimated heart mask. Then, we use a two-stage framework based on multi-atlas registration to segment the LV in the 4D sequence. Quantitative evaluation results show that our proposed multi-atlas registration method outperforms the Kirişli's method. Finally, experimental results using two 4D CCTA sequences indicate that our method can segment LV accurately.