20170316-Ward-Thesis.pdf (55.62 MB)
Download file

Automated characterization of cerebral veins

Download (55.62 MB)
posted on 21.03.2017, 03:14 authored by Phillip George Dayan Ward
The cerebral vasculature provides the oxygen and glucose the brain requires to function. Severe disruption of the supply of blood to the brain, such as during ischaemic and haemorrhagic stroke, can have catastrophic consequences. Subtle changes in normal cerebral vasculature may be a potential biomarker of early stage neurodegenerative changes, and larger changes with disease progression may provide valuable prognostic information. Quantifying the risk associated with subtle cerebrovascular changes requires the reliable measurement of vascular anatomy and oxygenation.
   The magnetic field of a MRI and the magnetic properties of iron in the blood interact. This interaction makes imaging the cerebral veins and measuring venous oxygenation possible. But, the many non-vein sources of magnetism in the brain complicate this process. The aim of the thesis was to improve automated characterisation of the cerebral veins by (i) the development of a cerebral vein segmentation method, (ii) the application of this technique to compute venous density and oxygen extraction fraction measures, (iii) the construction of a vein frequency atlas to improve anatomical-heterogeneity in vein segmentation, and (iv) development of a partial volume correction method for analysing smaller veins.
   The first experimental chapter introduces a technique for mapping the location of veins using a number of post-processing techniques that were unified into a single framework. The performance of the framework was compared with alternatives from the literature and demonstrated to provide vein maps with higher accuracy.
   The second experimental chapter constructed an analysis pipeline for mapping veins and measuring venous oxygenation. The pipeline was used to compute venous oxygen extraction fraction measures in a large brain-imaging database acquired from an elderly population of healthy individuals (n>500). Reproducible patterns of venous oxygenation and venous density were mapped throughout the brain, and anatomically-specific correlations between anatomy and oxygenation were identified.
   The third experimental chapter developed a technique for maximising vein image contrast. The technique combined three image sources of cerebral vein information with differing anatomical precision into a single image. A vein atlas (and anatomical precision maps) were calculated using a database of manually traced vein maps. Three vein segmentation techniques were used together with a number of accuracy metrics to determine that the combination of the three image sources yielded the optimal vein maps.
   The final experimental chapter addressed an error in measuring vein oxygenation with binary vein maps. Standard vein maps depict veins as a collection of cubic voxels. The effect of incorporating cylindrical geometry, rather than cubic geometry, when analysing venous maps was investigated. A partial volume technique was developed to more accurately model the cylindrical geometry of veins and improve the accuracy of venous oxygen extraction fraction measurements.
   In summary, the thesis includes contributions in vein imaging, vein identification and mapping, and vein characterization and measurement. Cerebrovascular anatomy and venous oxygenation were characterised in a healthy elderly population. Application of the methods and technique developments described in the thesis may be of benefit for future studies of cerebrovascular disease and neurodegeneration.


Campus location


Principal supervisor

David L. Dowe

Additional supervisor 1

David G. Barnes

Additional supervisor 2

Parnesh Raniga

Additional supervisor 3

Gary F. Egan

Year of Award


Department, School or Centre

Clayton School of Information Technology


Doctor of Philosophy

Degree Type



Faculty of Information Technology