Jornal article

Tesař L., Shimizu A., Smutek D., Kobatake H., Nawano S.

Computerized Medical Imaging and Graphics

2008

6 (2008)

513-520

0895-6111

Ing. Ludvík Tesař, Ph.D. (ÚTIA - Oddělení adaptivních systémů)

image segmentation, Gaussian mixture model, 3D image analysi

Texture-based segmentation of 3D CT images is adressed. The extension of Haralick 2D texture features to the 3D domain was studied. The co-occurrence matrix was calculated separately for each voxel in the image, using the co-occurrences of all voxels in a small cubic region around the voxel. The segmentation method used was model-based with a Gaussian Mixture Model. Evaluation of the proposed approach was performed using a set of 3D abdominal CT images. Statistical improvement of segmentation with 3D texture features was observed as opposed to the case without those features.