An empirical parameter selection method for endocardial border identification algorithms

Identification of the cardiac borders is a basic goal in the analysis of echocardiographic images. For this reason there has been a great deal of interest in automatic border identification methods. Numerous studies have been published, describing a wide variety of proposed methods. In spite of the large number of published studies, however, the relative merits of the various methods remain unclear. Furthermore, the optimal configuration of each method is in most cases unknown. One of the reasons for this is the lack of a well-defined, rigorous method for evaluating and comparing alternative algorithms and/or configurations. In this article we describe a systematic, empirical procedure for evaluating border identification algorithms. The method is based on a strict metrical measure of border error, and a set of test images which reflect the heterogeneity of typical clinical images. The methodology can be used to compare alternative analysis methods, or to determine the optimal configuration for any particular analysis method. We then apply this methodology to the problem of determining parameter values for several of the most commonly used analysis methods.