Simulation of User Interaction for Performance Evaluation of Interactive Image Segmentation Methods

Interactive image segmentation is often employed in the context of medical image analysis, as an alternative to automatic and manual image segmentation. In the last decade it has attracted a lot of attention due to the advent of efficient algorithms able to perform in interactive speed even for large three dimensional (3D) images. However, the human integration in the segmentation process restricts its repeatability and impedes its objective evaluation. Also, it inhibits the identification of the intrinsic properties of the algorithms. In this paper we report on a framework for performance evaluation of interactive image segmentation techniques, which is based on simulated user interaction. This allows for the construction of reproducible and tractable experiments, which can form the basis of a systematic and objective performance evaluation framework for interactive segmentation methods. We demonstrate quantitative results using three interactive segmentation algorithms from the literature.