Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.784794
Title: Automated characterisation and classification of liver lesions from CT scans
Author: Alahmer, Hussein
ISNI:       0000 0004 7970 339X
Awarding Body: University of Lincoln
Current Institution: University of Lincoln
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Cancer is a general term for a wide range of diseases that can affect any part of the body due to the rapid creation of abnormal cells that grow outside their normal boundaries. Liver cancer is one of the common diseases that cause the death of more than 600,000 each year. Early detection is important to diagnose and reduce the incidence of death. Examination of liver lesions is performed with various medical imaging modalities such as Ultrasound (US), Computer tomography (CT), and Magnetic resonance imaging (MRI). The improvements in medical imaging and image processing techniques have significantly enhanced the interpretation of medical images. Computer-Aided Diagnosis (CAD) systems based on these techniques play a vital role in the early detection of liver disease and hence reduce liver cancer death rate. Moreover, CAD systems can help physician, as a second opinion, in characterising lesions and making the diagnostic decision. Thus, CAD systems have become an important research area. Particularly, these systems can provide diagnostic assistance to doctors to improve overall diagnostic accuracy. The traditional methods to characterise liver lesions and differentiate normal liver tissues from abnormal ones are largely dependent on the radiologists experience. Thus, CAD systems based on the image processing and artificial intelligence techniques gained a lot of attention, since they could provide constructive diagnosis suggestions to clinicians for decision making. The liver lesions are characterised through two ways: (1) Using a content-based image retrieval (CBIR) approach to assist the radiologist in liver lesions characterisation. (2) Calculating the high-level features that describe/ characterise the liver lesion in a way that is interpreted by humans, particularly Radiologists/Clinicians, based on the hand-crafted/engineered computational features (low-level features) and learning process. However, the research gap is related to the high-level understanding and interpretation of the medical image contents from the low-level pixel analysis, based on mathematical processing and artificial intelligence methods. In our work, the research gap is bridged if a relation of image contents to medical meaning in analogy to radiologist understanding is established. This thesis explores an automated system for the classification and characterisation of liver lesions in CT scans. Firstly, the liver is segmented automatically by using anatomic medical knowledge, histogram-based adaptive threshold and morphological operations. The lesions and vessels are then extracted from the segmented liver by applying AFCM and Gaussian mixture model through a region growing process respectively. Secondly, the proposed framework categorises the high-level features into two groups; the first group is the high-level features that are extracted from the image contents such as (Lesion location, Lesion focality, Calcified, Scar, ...); the second group is the high-level features that are inferred from the low-level features through machine learning process to characterise the lesion such as (Lesion density, Lesion rim, Lesion composition, Lesion shape,...). The novel Multiple ROIs selection approach is proposed, in which regions are derived from generating abnormality level map based on intensity difference and the proximity distance for each voxel with respect to the normal liver tissue. Then, the association between low-level, high-level features and the appropriate ROI are derived by assigning the ability of each ROI to represents a set of lesion characteristics. Finally, a novel feature vector is built, based on high-level features, and fed into SVM for lesion classification. In contrast with most existing research, which uses low-level features only, the use of high-level features and characterisation helps in interpreting and explaining the diagnostic decision. The methods are evaluated on a dataset containing 174 CT scans. The experimental results demonstrated that the efficacy of the proposed framework in the successful characterisation and classification of the liver lesions in CT scans. The achieved average accuracy was 95:56% for liver lesion characterisation. While the lesion's classification accuracy was 97:1% for the entire dataset. The proposed framework is developed to provide a more robust and efficient lesion characterisation framework through comprehensions of the low-level features to generate semantic features. The use of high-level features (characterisation) helps in better interpretation of CT liver images. In addition, the difference-of-features using multiple ROIs were developed for robust capturing of lesion characteristics in a reliable way. This is in contrast to the current research trend of extracting the features from the lesion only and not paying much attention to the relation between lesion and surrounding area. The design of the liver lesion characterisation framework is based on the prior knowledge of the medical background to get a better and clear understanding of the liver lesion characteristics in medical CT images.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.784794  DOI: Not available
Keywords: G400 Computer Science
Share: