Quantitative Evaluation and Comparison of Lung Contour Extraction Methods

Abstract

Pulmonary contours in chest CT images are closed curves that may, in some cases, exhibit multiple concave and convex irregularities. In this paper we assess the quality of pulmonary contours obtained by several fully automated methods. A comparison among these contours and corresponding manually drawn reference contours was performed and several figures of merit were assigned to each one of the automated contours extraction algorithms. Likeness between a computed contour and reference contour was measured through the Pratt figure of merit, similarity index, mean error and fraction of errors greater than 5 pixels. For each of these figures we present an exploratory statistical analysis and discuss its sensibility to the longitudinal location of the CT slice