A new hybrid technique for nosologic segmentation of primary brain tumors / Shafaf Ibrahim

Ibrahim, Shafaf (2015) A new hybrid technique for nosologic segmentation of primary brain tumors / Shafaf Ibrahim. PhD thesis, Universiti Teknologi MARA.

Abstract

The main purpose of this research is to design a comprehensive algorithm which aims to aid the medical practitioners mainly the radiographers, radiologists and neurologists in primary brain tumors diagnosis. Detection of primary brain tumors is inspired by the necessity of high accuracy as it deals with human life. Presently, various imaging modalities techniques have incarnated as a tool for the doctors and radiologists to help them in diagnosis and treatment domain. While these are highly accurate and fast, they still require experienced and competent medical practitioners for the proper interpretation. Thus, the involvement of information technology is highly demanded in introducing reliable, simple and accurate computer systems. This study presents an algorithm for nosologic segmentation of primary brain tumors on Magnetic Resonance Imaging (MRI) brain images. The MRI technique has been chosen as the digital imaging modality since it provides clearer image for the tissue area as compared to the other techniques that focusing more on bone study such as Computed Tomography (CT) Scan and X-ray. The purpose of segmentation is to highlight the tumor areas, whereas classification is used to identify the type of the primary brain tumors. For this purpose, an algorithm which hybridized the Intensity Based Analysis (IBA), Grey Level Co-occurrence Matrices (GLCM), Adaptive Network-based Fuzzy Inference System (ANFIS) and £article ~warm Optimization (PSO) Clustering Algorithm (CAPSOCA) is proposed. The combination of several computer vision techniques was presented which aims to deliver reproducible nosologic segmentation of primary brain tumors which are gliomas and meningiomas. A Rule Based Expert System (RBES) is used to preliminarily classify the various types of primary brain tumors. It is designed to incorporate with the CAPSOCA algorithm which intended to strengthen the classification outcomes. The performance of the proposed CAPSOCA algorithm is compared with the PSG and ANFIS algorithms separately. The performance of the algorithms is quantified by two measurements which are segmentation accuracy and classification accuracy. The segmentation accuracy is evaluated using two approaches which are tumor template matching and comparison with ground truth. On the other hand, the classification accuracy is quantified using truth table by comparing the classification outcomes with histopathology diagnosis collected from the patient's record. The study revealed that very high pixel detection is noticeable in PSG algorithm which leads to over-segmentation in the tumor areas, and indirectly affected the segmentation accuracy. On the other hand, the ANFIS is found to have limited pixel detection in nosologic segmentation of primary brain tumors. The CAPSOCA was proven to be the best algorithm for nosologic segmentation of primary brain tumors from the MRI images data. It appeared to return the highest percentages of accuracy as compared to the PSO and ANFIS for both segmentation and classification at 78.790/0 of accuracy for gliomas segmentation, 76.68% of accuracy for meningiomas segmentation and 76.30% of accuracy for classification accuracy.

Metadata

Item Type: Thesis (PhD)
Creators:
Creators
Email / ID Num.
Ibrahim, Shafaf
UNSPECIFIED
Divisions: Universiti Teknologi MARA, Shah Alam > Faculty of Computer and Mathematical Sciences
Programme: Doctor of Philosophy
Keywords: Hybrid technique; Algorithm; Nosologic segmentation; Primary brain tumors
Date: 2015
URI: https://ir.uitm.edu.my/id/eprint/16322
Edit Item
Edit Item

Download

[thumbnail of ABS_SHAFAF IBRAHIM TDRA VOL 7 IGS 15.pdf]
Preview
Text
ABS_SHAFAF IBRAHIM TDRA VOL 7 IGS 15.pdf

Download (1MB) | Preview

Digital Copy

Digital (fulltext) is available at:

Physical Copy

Physical status and holdings:
Item Status:

ID Number

16322

Indexing

Statistic

Statistic details