Xu Jianfeng, Gu Lixu, Qi Wenyuan
Department of Computer Science, Shanghai Jiaotong University, Shanghai, P.R. China.
Conf Proc IEEE Eng Med Biol Soc. 2005;2005:6460-3. doi: 10.1109/IEMBS.2005.1615978.
An accurate and effective segmentation technique is the basis of the ideal dynamic heart modeling. In this paper, a novel multistage approach is proposed to perform the segmentation for the heart modeling and its multistage segmentation procedure orderly consists of four stages: Morphological Recursive Erosion (MRE); Fast Marching (FM); 3D Morphological Reconstruction (MR) and Morphological Recursive Dilation (MRD). To prove its accuracy and effectiveness, the approach is tested on 3 CT datasets of beating heats with each set containing ten individual volumes throughout a cardiac cycle. In order to validate the segmentation results, a novel Ra0dial Distance Based Validation (RDBV) method is also presented in this paper that provide the Global Accuracy (GA) measure to evaluate the segmentation accuracy. GA is calculated based on a Local Radial Distance Error (LRDE), which is along the radii emitted from the points along the skeleton of the object, to accommodate the complicated cardiac structure. The RDBV is improved in the universality and ability to reflect significant local errors in global accuracy function. The average accuracy of the proposed segmentation approach using the RDBV is 0.783.
