Image segmentation for MRI brain tumor

Abstract

The task of detecting the position of tumor (the so called gross tu- mor volume-GT)in the body of the patient is the starting point for

a medical treatment. In the conformal radiotherapy the tumor cells

are irradiated and killed with high precision without damaging the

neighboring healthy tissues. Usually, a medical doctor recognizes

the GTV and designs its border lines manually on the computer

tomography slices (CT-SCAN).The CT-SCAN procedure affects

the consequence planning target volume (PTV) of irradiation, ei- ther if it is decided by the same medical doctor or by the automatic

supporting system. The main objective of this study is to design

a computer system able to detect the presence of a tumor in the

digital images of the brain and to accurately define is border line.

The basic assumption is that different local textures in images can

describe different physical characteristics corresponding to differ- ent objects. We will be using this in wide range of field such as Bio

medical informatics, Oceanography, Computer vision. We assume

that the local texture of tumor cells is highly different from local

texture of other biological tissues. Thus texture measurement in

the image could be part of an effective discrimination technique

between healthy tissues and possible tumor areas.

A computer system has been design to recognize the typical fea-

1

tures of the tumor from the digital images of the brain. The basic

concept is that local texture in the images can reveal the typical

regularities of the biological structures. Thus, the textual features

have been extracted using a co-occurrence matrix approach. The

level of recognition, among three possible types of image areas

are: tumor, non tumor, non tumor and back ground. We are focus- ing on tumor image segmentation .