Use this URL to cite or link to this record in EThOS:
Title: An automated classification system to determine malignant grades of brain tumour (glioma) in magnetic resonance images based on meta-trainable multiple classifier schemes
Author: Al-Zurfi, Ahmed
ISNI:       0000 0004 7966 9208
Awarding Body: University of Salford
Current Institution: University of Salford
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
The accurate classification of malignant grades of brain tumours is crucial for therapeutic planning as it impacts on the tumour's prognosis, where the higher the malignancy levels of the brain tumour are, the higher the mortality rate is. It is also essential to provide patients with appropriate clinical management that may prolong survival and improve their quality of life. Determining the malignant grade of a brain tumour is a critical challenge because different malignant grades of brain tumours, in some cases, have inconsistent and mixed morphological characteristics. Consequently, the visual diagnosis using only the naked eye is a very complex and challenging task. The most common type of brain tumour is glioma. According to the World Health Organisation, low-grade glioma, which includes grade I and grade II are the least malignant, slow growing, and respond well to treatment. While, high-grade gliomas, which include grade III and grade IV are extremely malignant, have a poor prognosis and may lead to a high mortality rate. Hence, the motivation to develop an automated classification system to predict the malignant grade of glioma is the aim of this research. To achieve this aim, several novel methods were developed and this includes new methods for the extraction of statistical measures, selection of the dominant predictors, and the fusion of multi-classification models. The integration of these stages generates an accurate and automated decision system to determine the malignant grade of glioma. The feature extraction starts from the viewpoint that the objective measure of the brain tumour descriptors in MR images lead to an accurate classification of malignant brain tumours. This work starts from the standpoint that meta-trainable fusion of multiple classifier models can offer a better classification accuracy to recognise the malignant grade of glioma in MR images. This study developed a novel strategy based on two stages of multiple classifier systems for glioma grades. In the first stage, different machine learning algorithms were used. In the second stage, a systematic trainable combiner was designed based on deep neural networks. This research was validated using four benchmark datasets of MR images, which are publicly available and confirmed with the histopathological diagnosis. The proposed system was also evaluated and compared against different traditional algorithms; the experimental results showed that the proposed system has successfully achieved better and optimal discrimination in glioma grades on all dataset.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available