Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.554283
Title: Computer aided detection and segmentation of intracranial aneurysms in CT angiography
Author: Nikravanshalmani, Alireza
Awarding Body: Kingston University
Current Institution: Kingston University
Date of Award: 2012
Availability of Full Text:
Access through EThOS:
Access through Institution:
Abstract:
Accurate detection and segmentation of intracranial aneurysms plays an important role in diagnosing and reducing the incidence of subarachnoid haemorrhage (SAH) which is associated with high rates of morbidity and mortality. This research proposes a computer aided detection (CAD) and segmentation (CAS) of intracranial aneurysm in computer tomography angiography (CTA). The efficiency of the CAD/CAS system is boosted by pre-processing the input image with non-linear diffusion to smooth the CTA data while preserving the edges. A 3D region growing-based approach is used to extract the cerebral arteries followed by entropy-based search space reduction to retain the volume of the circle of Willis (CoW) and the proximal cerebral arteries where nearly all intracranial aneurysms are located, whilst eliminating the extracranial and very distal intracranial circulation. Because cerebral aneurysms vary in size we regard the problem of cerebral aneurysm detection as an intrinsically multi-scale problem and employ a multi-scale approach to all detection analysis. Shape index analysis is employed to determine potential aneurysmal regions (PARs). Hessian analysis and gradient vector field analysis which reveal 3D local shape information are used to further characterise the initial PARs. False positive reduction is then performed based on the analysis of the shape characterisations of the PARs. A ranking score is defined based on the outcomes of the shape analysis to rank the likelihood of PARs. The system allows user to navigate through the ranked PARs and select a candidate aneurysm for further analysis (CAS). The boundary of the selected aneurysm and its parent artery is delineated by using a 3D conditional morphology-based region growing approach. The output is presented to the user to be assessed for the aneurysm orientation relative to the parent vessel. A semi-automatic process is applied to detach the aneurysm from its parent artery. To have a fine segmentation of aneurysm which can be used for characterization of the aneurysm, a 3D geodesic active contour implemented in a level set framework is applied. The volume of the separated aneurysm is quantified as a typical characterization ofthe aneurysm. The system has been validated on a clinical dataset of 62 CT A scans with average 274 slices per scan (involving 17,028 CT slices) containing 70 aneurysms. Sizes of aneurysms vary between 3-16mm. 42 CTA scans have been used as a training dataset for parameter selection and 20 CTA scans have been used as a test dataset. The sensitivity of the systems for the CAD component is 97% with the average false positive of 2.24 per dataset (0.008 per slice). CAS performance was evaluated by dual visual judgment of an expert neuroradiologist and neurosurgeon. The detection and segmentation performance indicate the approach has potential in clinical applications.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.554283  DOI: Not available
Keywords: Allied health professions and studies ; Computer science and informatics
Share: