Visual Analytics for the Exploration and Assessment of Segmentation Errors

Abstract
Several diagnostic and treatment procedures require the segmentation of anatomical structures from medical images. However, the automatic model-based methods that are often employed, may produce inaccurate segmentations. These, if used as input for diagnosis or treatment, can have detrimental effects for the patients. Currently, an analysis to predict which anatomic regions are more prone to inaccuracies, and to determine how to improve segmentation algorithms, cannot be performed. We propose a visual tool to enable experts, working on model-based segmentation algorithms, to explore and analyze the outcomes and errors of their methods. Our approach supports the exploration of errors in a cohort of pelvic organ segmentations, where the performance of an algorithm can be assessed. Also, it enables the detailed exploration and assessment of segmentation errors, in individual subjects. To the best of our knowledge, there is no other tool with comparable functionality. A usage scenario is employed to explore and illustrate the capabilities of our visual tool. To further assess the value of the proposed tool, we performed an evaluation with five segmentation experts. The evaluation participants confirmed the potential of the tool in providing new insight into their data and employed algorithms. They also gave feedback for future improvements.
Description

        
@inproceedings{
10.2312:vcbm.20161287
, booktitle = {
Eurographics Workshop on Visual Computing for Biology and Medicine
}, editor = {
Stefan Bruckner and Bernhard Preim and Anna Vilanova and Helwig Hauser and Anja Hennemuth and Arvid Lundervold
}, title = {{
Visual Analytics for the Exploration and Assessment of Segmentation Errors
}}, author = {
Raidou, Renata G.
 and
Marcelis, Freek J. J.
 and
Breeuwer, Marcel
 and
Gröller, Eduard
 and
Vilanova, Anna
 and
Wetering, Huub M. M. van de
}, year = {
2016
}, publisher = {
The Eurographics Association
}, ISSN = {
2070-5786
}, ISBN = {
978-3-03868-010-9
}, DOI = {
10.2312/vcbm.20161287
} }
Citation