Abstract
In this research, we propose a model-based 3-D myocardial segmentation method that extracts a left ventricle region from cardiac CT images. The proposed method employs a shape model, which has region label, pixel value, and edge information. A target image is segmented by region labels from the registered shape model; the registration process refers objective functions based on the pixel correlation and the edge distance. To evaluate properties of the proposed method, we applied our method to generated model data that imitate a left ventricle of heart. We compared segmentation performances between three methods; registering shape model to target image with edge, with pixel, and with both (the proposed method).The proposed method achieved the accurate, fast-converged, and robust segmentation compared with the other methods.
