MESA: Complete approach for design and evaluation of segmentation methods using real and simulated tomographic images

In this paper we present MESA: a platform for design and evaluation of medical image segmentation methods. The platform offers a complete approach for the method creation and validation using simulated and real tomographic images. The system consists of several modules that provide a comprehensive workflow for generation of test data, segmentation method development as well as experiment planning and execution. The test data can be created as a virtual scene that provides an ideal reference segmentation and is also used to simulate the input images by a virtual magnetic resonance imaging (MRI) scanner. Both ideal reference segmentation and simulated images could be utilized during the evaluation of the segmentation methods. The platform offers various experimental capabilities to measure and compare the performance of the methods on various data sets, parameters and initializations. The segmentation framework, currently based on deformable models, uses a template solution for dynamical composition and creation of two- and three-dimensional methods. The platform is based on a client–server architecture, with computational and data storage modules deployed on the server and with browser-based client applications. We demonstrate the platform capabilities during the design of segmentation methods with the use of simulated and actual tomographic images.