Image-Based Verification of Parametric Models in Heart-Ventricle Volumetry

Measuring the volume of heart ventricles is an important issue in heart disease diagnosis. Therefore radiologists investigate computer tomographic images, searching for representative projections of the left ventricle. In the selected image showing the left ventricle with maximum circumference, a parametric model is established by measuring width and height of the left ventricle shape. These parameters are used to approximate the contour of the ventricle with an ellipse and afterwards to calculate the volume of the ventricle assuming an ellipsoid model. In this work we develop a tool for comparing estimated volumes calculated by the parametric model with volumes based on techniques from digital image processing. Herein we treat computer tomographic image slices as volumetric data sets and define a volume by a set of segmented images and a given voxel size. We implement three different segmentation techniques and apply them on 31 medical image data sets. These segmentation techniques are Thresholding, Active Contours (Snakes) and Live Wire, a graph-theoretic approach. The algorithms are explained and the results from the volume estimations are compared to each other as well as with the parametric model. Furthermore we investigate an automation technique, a Hough transform for ellipse detection to find the ellipse contour in the image showing the maximum ventricle projection. The diversity of the images makes it necessary to utilize interactive segmentation techniques instead of fully automatic ones. The Live Wire algorithm shows the best segmentation performance of the tested algorithms.