Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images

Over the last few decades, medical image retrieval has become one of the most exciting and fastest growing research areas in the application of image retrieval due to the need for computer-assisted classification, query, and retrieval methods for large medical image archives. These automation proces...

Full description

Saved in:

Bibliographic Details
Main Author:	Wan Zaki, Wan Mimi Diyana
Format:	Thesis
Published:	2012
Subjects:	R Medicine (General)
Tags:	Add Tag No Tags, Be the first to tag this record!

id	my-mmu-ep.5440
record_format	uketd_dc
spelling	my-mmu-ep.54402014-04-21T02:57:34Z Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images 2012-02 Wan Zaki, Wan Mimi Diyana R Medicine (General) Over the last few decades, medical image retrieval has become one of the most exciting and fastest growing research areas in the application of image retrieval due to the need for computer-assisted classification, query, and retrieval methods for large medical image archives. These automation processes offset the high cost of manual annotation by medical experts, which is cumbersome, prone to errors, and prohibitively expensive as well as dependent on human subjectivity. One solution to such problems is to fully automate the segmentation and feature extraction processes for the development of content-based medical image retrieval (CBMIR). The existing approaches strongly focus on a particular imaging modality with the queries restricted to a well-defined diagnostic background. Hence, the main motivation of this thesis is to develop a fully automated segmentation approach together with a reliable feature extraction method to be used in a CBMIR system for intracranial haemorrhage (ICH) in Computed Tomography (CT) brain images. To overcome the volume partial effect and inconsistency of grey-level values of the CT brain images, multi-level thresholding methods are proposed. These level by-level segmentation approaches are fully automated and able to extract the intracranial and skull information from the CT brain images. The intracranial is subsequently further segmented into cerebrospinal fluid, brain tissues, and other homogenous regions useful for detecting any abnormalities (i.e., bleeding, calcification, misaligned ventricles) that may be present in the brain. In addition, the extracted skull is useful to represent a skull feature vector for skull fracture detection. The proposed approach promotes more effective segmentation compared to other fully automated methods discussed in previous literature. To develop a reliable and efficient CBMIR, namely for ICH, a Binary Coherent Vector (BCV) approach, which is part of the feature extraction process, is proposed. This work demonstrates that the combination of geometric shapes and Hu moment invariant features provides the best feature vectors for distinguishing haemorrhage shape, resulting in an average precision rate of up to 80% during the first 60% of the average recall using both normalized Manhattan and normalized Euclidean as well as Mahalanobis distance metrics. These results are promising and provide a strong basis for the application of CBMIR, specifically for CT brain images. 2012-02 Thesis http://shdl.mmu.edu.my/5440/ http://library.mmu.edu.my/diglib/onlinedb/dig_lib.php phd doctoral Multimedia University Faculty of Engineering
institution	Multimedia University
collection	MMU Institutional Repository
topic	R Medicine (General)
spellingShingle	R Medicine (General) Wan Zaki, Wan Mimi Diyana Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images
description	Over the last few decades, medical image retrieval has become one of the most exciting and fastest growing research areas in the application of image retrieval due to the need for computer-assisted classification, query, and retrieval methods for large medical image archives. These automation processes offset the high cost of manual annotation by medical experts, which is cumbersome, prone to errors, and prohibitively expensive as well as dependent on human subjectivity. One solution to such problems is to fully automate the segmentation and feature extraction processes for the development of content-based medical image retrieval (CBMIR). The existing approaches strongly focus on a particular imaging modality with the queries restricted to a well-defined diagnostic background. Hence, the main motivation of this thesis is to develop a fully automated segmentation approach together with a reliable feature extraction method to be used in a CBMIR system for intracranial haemorrhage (ICH) in Computed Tomography (CT) brain images. To overcome the volume partial effect and inconsistency of grey-level values of the CT brain images, multi-level thresholding methods are proposed. These level by-level segmentation approaches are fully automated and able to extract the intracranial and skull information from the CT brain images. The intracranial is subsequently further segmented into cerebrospinal fluid, brain tissues, and other homogenous regions useful for detecting any abnormalities (i.e., bleeding, calcification, misaligned ventricles) that may be present in the brain. In addition, the extracted skull is useful to represent a skull feature vector for skull fracture detection. The proposed approach promotes more effective segmentation compared to other fully automated methods discussed in previous literature. To develop a reliable and efficient CBMIR, namely for ICH, a Binary Coherent Vector (BCV) approach, which is part of the feature extraction process, is proposed. This work demonstrates that the combination of geometric shapes and Hu moment invariant features provides the best feature vectors for distinguishing haemorrhage shape, resulting in an average precision rate of up to 80% during the first 60% of the average recall using both normalized Manhattan and normalized Euclidean as well as Mahalanobis distance metrics. These results are promising and provide a strong basis for the application of CBMIR, specifically for CT brain images.
format	Thesis
qualification_name	Doctor of Philosophy (PhD.)
qualification_level	Doctorate
author	Wan Zaki, Wan Mimi Diyana
author_facet	Wan Zaki, Wan Mimi Diyana
author_sort	Wan Zaki, Wan Mimi Diyana
title	Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images
title_short	Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images
title_full	Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images
title_fullStr	Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images
title_full_unstemmed	Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images
title_sort	content-based medical image analysis and retrieval of intracranial haemorrhages ct brain images
granting_institution	Multimedia University
granting_department	Faculty of Engineering
publishDate	2012
_version_	1747829575312736256

Content-based medical image analysis and retrieval of intracranial haemorrhages CT brain images

Similar Items