Parameter-Space Exploration for Computed Tomography Image Analysis Algorithms

We present a tool which supports computed tomography specialists in exploring the parameter and result space of volume segmentation pipelines. In a preprocessing step, the user samples the parameter space of a number of pipelines to be analysed. Our sampling tool supports arbitrary pipelines that can be parametrized via the command line. The results from sampling the parameter spaces of these pipelines are clustered by image similarity. The user can then explore the space of possible results in an interactive visualization tool. This visualization tool highlights the variation resulting from different parametrizations and pipelines, and allows exploring the influence of each parameter. We evaluate our tool on the multimodal segmentation of a carbon-fiber-reinforced polymer specimen with impact damage, scanned with a Talbot-Lau grating interferometer computed tomography device.