Evaluation of a cardiac ultrasound segmentation algorithm using a phantom

This paper evaluates the performance of a level set algorithm for segmenting the endocardium in short-axis ultrasound images. The evaluation is carried out using an anthropomorphic ultrasound phantom. Details of the phantom design, including comparison of the ultrasound parameters with in-vitro measurements, are included. In addition to measuring segmentation accuracy, the effectiveness of the energy minimization scheme is also determined. It is argued that using the phantom along with global minimization algorithms (simulated annealing and random search) makes is possible to assess the minimization strategy.