[Back]

P. Dvorak, K. Bartusek, W. Kropatsch:
"Automated Segmentation of Brain Tumor Edema in FLAIR MRI Using Symmetry and Thresholding";
Talk: 34th PIERS 2013, Stockholm; 2013-08-12 - 2013-08-15; in: "Proceedings of PIERS 2013", Stockholm (2013), ISBN: 978-1-934142-26-4; 936 - 939.

Nowadays, the brain tumor detection and segmentation in MR images is a de-
veloping issue. There are many research teams producing di®erent and interesting methods and
algorithms for this particular task of medical image processing. Many of them are semi-automatic,
but the aim of current research, and of this work, is to ¯nd a fully automatic method.
This paper focuses on the automatic edema segmentation in FLAIR images. This type of contrast
images was selected because of the visibility and manifestation of edema in this image type. Since
in axial plane of healthy brain, the approximate left-right symmetry exists, it is used as the prior
knowledge for searching the approximate edema location. It is assumed that the edema is not
located symmetrically in both hemispheres, which is met in most cases. For the detection, the
multi-resolution approach is used. Since the edemas manifest as a hyperintense area in FLAIR
images, it is extracted using thresholding. For the automatic determination of the threshold, the
Otsu's algorithm is used. This work does not deal with the tumor presence detection. One of our
previous work focuses on this topic. The main reason for the edema segmentation is for the tumor
classi¯cation. This will be carried out by applying the resulting mask of the proposed method to
perfusion MR images. Since perfusion images are of very low contrast, the pathological area, it
means the tumor and a potential edema around it, has to be detected and segmented in another
type of MR images.