Extended statistical shape models based on probabilistic correspondences for 3D segmentation of medical images

The objective of this project is the development of model- and knowledge-based methods for shape analysis as well as automatic 3D-segmentation of diagnostically and therapeutically relevant objects in medical image volumes. The incorporation of a priori knowledge about the shape and the context of image structures allows a more robust segmentation of structures that feature weak edges or inhomogeneous intensities.

The probabilistic approach developed here is aimed to augment the possibilities of representing the natural 3D shape variability of anatomical structures in statistical shape models and to allow an optimal exploitation of shape information even in training data containing few observations. The idea is to integrate the probabilistic model into a flexible segmentation algorithm that should be able to deal with complex segmentation problems as non-spherical topologies or multi-object segmentation.

The new methods are evaluated on clinically relevant segmentation problems in the fields of radiation therapy and computer-aided intervention planning.

Probabilistic statistical shape model of the kidney. (a) shows the mean shape. (b-e) show the shape variations according to the first (b,c) and second (d,e) variation mode.

The project is funded by Deutsche Forschungsgemeinschaft (DFG: HA 2355/7-1).

Selected Publications

  1.  Hufnagel, H., Pennec, X., Ehrhardt, J., Handels, H. and Ayache, N. (2007). Shape Analysis Using a Point-Based Statistical Shape Model Built on Correspondence Probabilities. Proceedings of the MICCAI'07: 959-967.
  2. Hufnagel, H., Pennec, X., Ehrhardt, J., Ayache, N. and Handels, H. (2008). "Generation of a Statistical Shape Model with Probabilistic Point Correspondences and EM-ICP." International Journal for Computer Assisted Radiology and Surgery (IJCARS) 2(5): 265-273.
  3. Hufnagel, H., Ehrhardt, J., Pennec, X., Ayache, N. and Handels, H. (2009a). "Computing of Probabilistic Statistical Shape Models of Organs Optimizing a Global Criterion." Methods of Information in Medicine 48(4): 314-319.
  4. Hufnagel, H., Ehrhardt, J., Pennec, X., Schmidt-Richberg, A. and Handels, H. (2009b). Level Set Segmentation Using a Point-Based Statistical Shape Model Relying on Correspondence Probabilities. Proc. of MICCAI Workshop Probabilistic Models for Medical Image Analysis (PMMIA'09): 34-44.
  5. Hufnagel, H., Ehrhardt, J., Pennec, X. and Handels, H. (2009c). Application of a Probabilistic Statistical Shape Model to Automatic Segmentation. World Congress on Medical Physics and Biomedical Engineering, WC 2009, München: 2181-2184.
  6. Hufnagel, H., Ehrhardt, J., Pennec, X., Schmidt-Richberg, A. and Handels, H. (2010a). Coupled Level Set Segmentation Using a Point-Based Statistical Shape Model Relying on Correspondence Probabilities. Proc. SPIE Symposium on Medical Imaging 2010: 6914 6914T6911-6914T6918.

Project Team

Dipl.- Inf. Heike Hufnagel
Dr. Jan Ehrhardt
Prof. Dr. Heinz Handels

Cooperation Partners

Prof. Dr. Nicholas Ayache
Dr. Xavier Pennec
INRIA, Institut National de Recherche en Informatique et en Automatique, Epidaure Group, Sophia Antipolis Cedex, Frankreich