Brain tumor segmentation



Dana Cobzas, Karteek Popuri, Neil Birkbeck, Mark Schmidt, Martin Jagersand Albert Murtha


References

Popuri, K., Cobzas, D., Mutrtha A., and Jagersand 3D Variational Brain Tumor Segmentation using Dirichlet Priors on a Clustered Feature Set , International Journal of Computer Assisted Radiology and Surgery,2011

Popuri K., Cobzas D., Jagersand M.,Shah S.L., Murtha A. 3D variational brain tumor segmentation on a clustered feature set, SPIE Medical Imaging 2009 to appear

Cobzas, D., Birkbeck, N., Schmidt, M., Jagersand, M., Murtha A. 3D Variational Brain Tumor Segmentation using a High Dimensional Feature Set, Mathematical Methods in Biomedical Image Analysis (MMBIA 2007), in conjunction with ICCV

Demos


Brain tumor segmentation


Description

Tumor segmentation from MRI data is an important but time consuming manual task performed by medical experts. Automating this process is challenging due to high diversity in appearance of tumor tissue among different patients and, in many cases, similarity with normal tissue. One other challenge is how to make use of prior information about the appearance of normal brain. In this paper we show how to incorporate prior information into a multi-dimensional volumetric features set. Using manually segmented data we learn a statistical model for tumor and normal tissue using the same feature set.


Three stages of surface evolution

We propose a variational segmentation method that uses region statistics on the multi-dimensional feature set to evolve a level set. The formulation extends the Chan-Vese region-based segmentation model in a similar way to texture-based approaches. But instead of using an unsupervised approach we learn a statistical model from a set of features specifically engineered for the MRI brain tumor segmentation task.

We experimented with three types of statistics - a generative Gaussian model, a discriminative Logistic Regression Model and a Parzen histogram calculated on each feature. Finally we used histogram of clustered featured and impose a Dirichlet prior to disambiguate the tumor from ventricle. The prior penalizes the clusters predominant in the ventricles from having a high probability in the tumor.


Segmentation results using the Gaussian and Logistic Regression statistics on the multidimensional feature set.




Segmentation results using a Parzen histogram on the clustered feature set. Comparsion of supervised (2) vs. unsupervised (1), clustered (3) vs. non-clustered (2) and the effect of the Dirichlet prior to disambiguate the tumor from the ventricles.