Adaptive techniques for brain tumor detection

العناوين الأخرى

تقنيات مطورة للكشف عن أورام الدماغ

مقدم أطروحة جامعية

Abdun, Rabab Sadun

مشرف أطروحة جامعية

Ali, Salih Mahdi
Abbud, Luayy Kazim

الجامعة

جامعة بغداد

الكلية

كلية العلوم

القسم الأكاديمي

قسم الفيزياء

دولة الجامعة

العراق

الدرجة العلمية

دكتوراه

تاريخ الدرجة العلمية

2014

الملخص الإنجليزي

Brain tumors are one of the major causes of death among people.

The chances of survival can be increased if the tumor is detected correctly at its early stage.

Detection of tumors in MRI of brain is not an easy task when the tumor is overlapped with dense brain tissues.

Despite numerous efforts and acceptable results in brain tumors segmentation, accurate and reproducible segmentation and characterization of abnormalities still a challenging task due to the varieties in tumor shapes, locations and the change in intensities of various types of tumors.

The aim of the current study is to introduce an adaptive global automatic technique that can adequately process all slices of Magnetic Resonance Imaging (MRI) of brain; i.e.

detecting, isolating and extracting the tumor regions especially when only a single source of information is available (e.g.

gray tone MR Images).

Many segmentation methods including clustering algorithms, region growing and artificial neural networks of both schemes of training supervised and unsupervised were suggested and used to achieve optimal separation of brain's tissues, which leads to better isolation of the abnormalities (tumors).

To extract the brain's matter; improved skull stripped methods depending on traditional and non-traditional methods were proposed; i.e.

skull stripping utilizing contour filling method, deformable contour algorithm based on traditional and active contour, modified contour technique based on KMeans clustered image and skull stripping utilizing morphological operations and active contour.

The final results were promising compared with the common known methods.

The most important factor that affects the results of skull stripping methods was the adopted dataset.

In tumor isolation, the proposed methods were: contour based segmentation utilizing contour and contour filling methods, K-Means based on Fuzzy C-Mean (FCM), K-Means and FCM based on intensity and location, K-Means and FCM based on Gray Level Cooccurrence Matrix (GLCM) features, Possibilistic Fuzzy C-Mean based on image histogram.

Untraditional methods like hierarchical self-organization feature map, and hierarchical self-organization feature map based on FCM.

Moreover, supervised neural networks based on image intensity and standard deviation once, and based on GLCM features another are also used to modify the overall efficiency suitable for the work configuration, which is a very hard case of segmentation due to the overlapping spectrum fingerprint of tissues.

The results of these methods were acceptable.

Increasing the used extracted features (information) improves the quality to a certain limit (especially when the features gained from single source of data), but when implementing several sources the results showed better quality.

According to the radiologist consultations the K-Means and K-Means based on intensity and location techniques have yield more accurate than fuzzy methods.

The general conclusion is that, there is no one global automatic technique suitable for tumor detection in all slices due to the differences of these slices.

التخصصات الرئيسية

الفيزياء

عدد الصفحات

298

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-741082