Superpixels for the Visible Human Project and human anatomical voxel model construction

We have previously reported a Full Body Image segmentation tool, the FBIseg tool [1], for the Visible Human Project and similar projects around the world. The purpose of this paper is to present the novel superpixel approach for multi-organ multi-tissue segmentation in medical images and to demonstrate the efficacy and usefulness of the superpixel-FBIseg framework by segmenting the torso CT of a 14-year-old female and producing a human anatomical voxel model.
Full-body image segmentation can be used in delineating the organs and tissues in the Visible Human Project and other similar projects. It is also the approach in constructing human anatomical voxel models from clinical volumetric imaging data such as CT or MRI for dosimetry calculation [2]. Though a key step in many
research areas, literature shows that full-body segmentations were mostly manually performed with limited computer assistance. As such, it is notoriously laborious and time-consuming. An automatic system for full body segmentation would be desirable but has not yet been achieved. Machine segmentation has many advantages such as speed, reproducibility, pattern recognition and is not subject to human errors. However, it lacks expert knowledge. Our novel superpixel approach to full body segmentation reduces the large number of pixels in an image to a small number of superpixels. Expert knowledge is then incorporated with the use of the FBIseg tool.