| Popis: |
Cardiac segmentation of CT images is important to diagnose CVDS quantitatively in clinic. In this work, a cardiac segmentation method is proposed which is multi-atlas coarse-refine registration based. In the proposed method, the cardiac rough region and it's accurate segmentation are recognized with two rounds of registration. The proposed method is evaluated quantitatively and qualitatively on 20 2D cardiac CT images. In addition, we also applied the proposed method to multiple situations where cardinalities of the atlases are equal whereas the atlases are different to evaluate its robustness to the atlas. Experimental results demonstrate the effectiveness and robustness of the proposed method. |
