Statistical Features Segmentation Technique For MR Images Of Brain’sTumors

Abstract

Medical image analysis has great significance in the field of treatment, especially
in non-invasive and clinical studies. Medical imaging techniques and it analysis and
diagnoses analysis tools enable the physicians and Radiologists to reach at a specific
diagnosis. In this study, MR images have been used for discriminating the infected
tissues from normal brain’s tissues. A semi-automatic segmentation technique based
on statistical futures has been introduced to segment the brain’s MR image tissues.
The proposed system used two stages for extracting the image texture features. The
first stage is based on utilizing the 1st order statistical futures histogram based
features such as (the mean, standard deviation, and image entropy ) which is local in
nature, while the second stage is based on utilizing the 2nd order statistical futures
(i.e Co-Occurrence matrices features).
Similar coloring and semi-equal statistical features of the tumor area and the Gray
Matter (GM) brain’s tissue was the main encountered problem in the first presented
segmentation method. To overcome this problem, an adaptive multi-stage
segmentation technique is presented, in which the mean value of each pre-segmented
classes has been used to distinguish the tumor tissue from others. The segmentation
process is followed by a 2nd order classification method to assign image pixels
accurately to their regions, using the invariant moments parameters weighted
together with the Co-Occurrence parameters. Different samples of MR images for
normal and abnormal brains (i.e. T1 and T2-weighted) have been tested, for different
patients.