Visual Analytics Methods for Modeling in Medical Imaging
Final Report Abstract
Medical imaging plays an important role in clinical practice, for example in treatment planning or computer-aided diagnosis. In this respect, segmentation of medical images is a necessary prerequisite. Frequently used segmentation algorithms are based on statistical shape models (SSMs). By modeling an organ’s shape variability, they enable segmentation of organs which can not be segmented using image intensities only. For building an SSM, models have to be selected that fit the high-dimensional training data well. Due to the lack of prior information on the data, standard models are frequently chosen. However, they do not necessarily describe the data in an optimal way. A poor choice of the model is not apparent until the segmentation algorithm is evaluated. Visual analytics methods can provide valuable tools for supporting this modeling process. The aim of this project was to develop new Visual Analytics methods for fitting SSMs in medical image segmentation. Our approach combined interactive data visualization, data analysis and model steering in all stages of the process. In this way, the user is provided with a deeper insight into the correspondence between data and model result. We created several solutions that support several stages of medical modeling: 1. As medical image modeling data extraction is very costly, only few datasets are available for testing. Therefore, we developed new tools for interactive creation of test datasets with variable data characteristics. 2. The data for medical modeling need to be pre-processed for their use in segmentation. The pre-processing requires setting of suitable parameters in order to provide good modeling results. We developed new tools that allow the users to visually inspect input data properties, their match with model assumptions. Moreover, they allow for analyzing the impact of parameter settings on output quality. 3. Many medical models work iteratively for producing segmentation result. In order to be able to improve them, it is needed to understand why and how the model works. For this purpose, we developed new methods, that allow the users to analyze the progress of segmentation algorithm with respect to the final output quality. 4. Medical experts need to evaluate model results on a test dataset of several organ instances and to compare several algorithms for determining best algorithms. This is difficult, as several results need to be compared simultaneously. We developed a new way of analyzing segmentation quality across a dataset. Moreover, we presented new ways of interactive exploration of the output quality. This allows to determine best algorithms on regional level, not only on whole organs. 5. The exploration of datasets and models with visual analytics system is a long process consisting of many steps. In order to be able to reproduce the results, one needs to track user’s activity. The tracking is based on a special taxonomy. We developed a new interaction taxonomy that unifies activities in data exploration and analysis.
Publications
- PCDC - On the Highway to Data - A Tool for the Fast Generation of Large Synthetic Data Sets. International Workshop on Visual Analytics (EuroVA), pp. 7-11, 2012
Bremm S., von Landesberger T., Heß M., Fellner D.
- Visual Analytics for model-based medical image segmentation: Opportunities and challenges. Expert Syst. Appl.. 40 (12) 2012
von Landesberger T, Bremm S, Kirschner M, Wesarg S, Kuijper A.
(See online at https://doi.org/10.1016/j.eswa.2013.03.006) - Visual analytics methods for categoric spatio-temporal data, 2012 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 183,192, 14-19 Oct. 2012
von Landesberger, T.; Bremm, S.; Andrienko, N.; Andrienko, G.; Tekusova, M.
- Opening up the “black box” of Medical Image Segmentation with Statistical Shape Models. The Visual Computer. v. 29(9): pp. 893-905, 2013
von Landesberger T., Andrienko G., Andrienko N., Bremm S., Kirschner M., Wesarg S., Kujper A.
(See online at https://doi.org/10.1007/s00371-013-0852-y) - Interaction taxonomy for tracking of user actions in visual analytics applications. In Handbook of Human Centric Visualization (pp. 653-670). Springer New York, 2014
von Landesberger, T., Fiebig, S., Bremm, S., Kuijper, A., & Fellner, D. W.
