An automated system for the classification and segmentation of brain tumours in MRI images based on the modified grey level co-occurrence matrix

Hasan, A ORCID: https://orcid.org/0000-0001-9151-6958 2017, An automated system for the classification and segmentation of brain tumours in MRI images based on the modified grey level co-occurrence matrix , PhD thesis, University of Salford.

[img]
Preview
PDF
Download (5MB) | Preview

Abstract

The development of an automated system for the classification and segmentation of brain tumours in MRI scans remains challenging due to high variability and complexity of the brain tumours. Visual examination of MRI scans to diagnose brain tumours is the accepted standard. However due to the large number of MRI slices that are produced for each patient this is becoming a time consuming and slow process that is also prone to errors. This study explores an automated system for the classification and segmentation of brain tumours in MRI scans based on texture feature extraction. The research investigates an appropriate technique for feature extraction and development of a three-dimensional segmentation method. This was achieved by the investigation and integration of several image processing methods that are related to texture features and segmentation of MRI brain scans. First, the MRI brain scans were pre-processed by image enhancement, intensity normalization, background segmentation and correcting the mid-sagittal plane (MSP) of the brain for any possible skewness in the patient’s head. Second, the texture features were extracted using modified grey level co-occurrence matrix (MGLCM) from T2-weighted (T2-w) MRI slices and classified into normal and abnormal using multi-layer perceptron neural network (MLP). The texture feature extraction method starts from the standpoint that the human brain structure is approximately symmetric around the MSP of the brain. The extracted features measure the degree of symmetry between the left and right hemispheres of the brain, which are used to detect the abnormalities in the brain. This will enable clinicians to reject the MRI brain scans of the patients who have normal brain quickly and focusing on those who have pathological brain features. Finally, the bounding 3D-boxes based genetic algorithm (BBBGA) was used to identify the location of the brain tumour and segments it automatically by using three-dimensional active contour without edge (3DACWE) method. The research was validated using two datasets; a real dataset that was collected from the MRI Unit in Al-Kadhimiya Teaching Hospital in Iraq in 2014 and the standard benchmark multimodal brain tumour segmentation (BRATS 2013) dataset. The experimental results on both datasets proved that the efficacy of the proposed system in the successful classification and segmentation of the brain tumours in MRI scans. The achieved classification accuracies were 97.8% for the collected dataset and 98.6% for the standard dataset. While the segmentation’s Dice scores were 89% for the collected dataset and 89.3% for the standard dataset.

Item Type: Thesis (PhD)
Contributors: Meziane, F (Supervisor) and Aspin, R (Supervisor)
Schools: Schools > School of Computing, Science and Engineering
Depositing User: A Hasan
Date Deposited: 18 Oct 2021 07:48
Last Modified: 18 Oct 2021 07:48
URI: http://usir.salford.ac.uk/id/eprint/42565

Actions (login required)

Edit record (repository staff only) Edit record (repository staff only)

Downloads

Downloads per month over past year