Visualization for Validation and Improvement of Three-dimensional Segmentation Algorithms

title="Visualization for Validation and Improvement of Three-dimensional Segmentation Algorithms",
booktitle="Data Visualization 2005 (Proceedings of the EUROGRAPHICS - IEEE VGTC Symposium on Visualization 2005)",
author="Gunther H. Weber AND Cristian Luis Hendriks Luengo AND Soile V. E. Keränen AND Scott E. Dillard AND Derek Ju AND Damir Sudar AND Bernd Hamann ",
keywords="image processing, volume visualization, gene expression",
editor=" Ken Brodlie AND David Duke AND Ken Joy ",
organization="EUROGRAPHICS - IEEE VGTC",
publisher="Eurographics Association",
address="PO Box 16, CH-1288 Aire-la-Ville, Switzerland",
location="Leeds, United Kingdom",
eventtime="June 1--3, 2005",
abstract="The Berkeley Drosophila Transcription Network Project (BDTNP) is developing a suite of methods that will allow a quantitative description and analysis of three dimensional (3D) gene expression patterns in an animal with cellular resolution. An important component of this approach are algorithms that segment 3D images of an organism into individual nuclei and cells and measure relative levels of gene expression. As part of the BDTNP, we are developing tools for interactive visualization, control, and verification of these algorithms. Here we present a volume visualization prototype system that, combined with user interaction tools, supports validation and quantitative determination of the accuracy of nuclear segmentation. Visualizations of nuclei are combined with information obtained from a nuclear segmentation mask, supporting the comparison of raw data and its segmentation. It is possible to select individual nuclei interactively in a volume rendered image and identify incorrectly segmented objects. Integration with segmentation algorithms, implemented in MATLAB, makes it possible to modify a segmentation based on visual examination and obtain additional information about incorrectly segmented objects. This work has already led to significant improvements in segmentation accuracy and opens the way to enhanced analysis of images of complex animal morphologies.",
back to publication