A novel model-based approach to left ventricle segmentation

In this paper a parametric model of the left ventricle is presented, whose task is to improve the segmentation results obtained by the use of standard algorithms. An individual model is built on the basis of properly designated sections. Incorrectly designated sections should be replaced with ellipses evaluated using the presented model. While elaborating the model a database has been used consisting of cardiac images delineated by experts. The model is based on parametric curves and regression analysis. A segmentation algorithm based on the Kernelized Weighted C-Means clustering and automatic segmentation correctness coefficients has been proposed. Eventually the model should also work with other segmentation algorithm. By improving the segmentation results with the model, the error has been reduced from a clinically unacceptable to the interobserver variability. The model is most useful while assessing the epicardium at end-systole and the heart weight.