الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 187 |  |  |
| | | from | 187 | | |
Abstract
Image segmentation is a fundamental task in many image processing applications. It can be used in medical diagnosis. Medical images have different problems. There are multiple image segmentation techniques were developed to overcome the problems of the medical images, but resulted over-segmentation and under segmentation. There is no universal image segmentation technique used for all purposes, images and applications types. The brain tumor is a serious disease. It causes death rapidly if is not detected early.
The medical scans are used to help physicians to diagnose the tissues and put the treatment plan. These images or scans are geometrically aligned to provide the physicians with the better observations to achieve the true diagnose. The 3D image can be obtained from many consecutive two dimensions 2D slices. It may not be accurate and takes extreme time, especially if the 3D image consists of multiple 2D slices. Segmenting each 2D slice in 3D image by using the 2D segmentation techniques provides higher reliability for image quality rather than segmenting the whole 3D image once by other techniques such as atlas.
The proposed medical system in thesis provides an accurate detection of brain tumor with minimal execution time in 2D and 3D images automatically without user interaction. The 2D framework contains five stages: the preprocessing, clustering, segmentation, post processing and validation.We developed the proposed system based on two different proposed clustering techniques. The first is integration between k -means (KM) with fuzzy c-means (FCM) which is called KIFCM. The second is integration between KM and particle swarm optimization (PSO) which is called KIPSO. We applied the two clustering techniques on three different benchmark brain data sets to detect the brain tumor. from experiments, we proved the effectiveness of our techniques in segmenting the brain tumor by comparing it with five used techniques: KM, FCM, expectation maximization (EM), mean shift (MS) and PSO. KIPSO is very near in accuracy to KIFCM, but it takes less time.We used it in 3D image segmentation to reduce the whole time and increase the accuracy.
KIFCM accelerates the convergence speed of FCM and eliminates the pixels overlapping by using KM. It retains more information from the image and detects the infected spread cells in the MRI image by using FCM. KIPSO results in global optimization with less processing time. It helps KM to escape from local optima by using PSO. It helps PSO to reduce the computation time by using KM.
A 3D framework uses the stages of a 2D framework besides adding two extra sub steps in pre and post processing to deal with 3D images. We used the fourth 3D image data set which contains 122- 2D MRI slices. The system visualizes all of the 2D slices in 3D reconstruction. The system makes the volumetric vision of the 3D image to visualize the brain and the tumor from the 2D clustering slices in the 3D modeling phase.