Novel methods in retinal vessel calibre feature extraction for systemic disease assessment

Aliahmad, B 2014, Novel methods in retinal vessel calibre feature extraction for systemic disease assessment, Doctor of Philosophy (PhD), Electrical and Computer Engineering, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Aliahmad.pdf Thesis application/pdf 52.26MB
Title Novel methods in retinal vessel calibre feature extraction for systemic disease assessment
Author(s) Aliahmad, B
Year 2014
Abstract Retina and its vascular network have unique branching characteristics morphology of which will change as a result of some systemic diseases, including hypertension, stroke and diabetes. Therefore, retinal image has been used as non-invasive screening window for risk assessment and prediction of such disease condition especially at the baseline. The assessment is based on a number of features among which vessel diameter (both individual and summary) and fractal dimension (FD) are the ones mostly associated with risk of diabetes and stroke. The association is linked to the higher risk of diabetes and stroke in people with narrower retinal arteriole diameter or change in overall fractal dimension independent of any risk factor (i.e. blood pressure, cardiovascular risk factors). Diameter measurement requires vessel edges to be located and tracked however; accurate edge perception is subject to image contrast, shadows, lighting condition and even presence of retinopathy legions close to vessel boundaries. This will lead to imprecision and inconsistencies between different automatic measurement techniques and may affect the significance of its association with disease condition in risk-assessment studies. As accuracy and success of diameter measurement is subject to large variations due to image artifacts it may not be suitable for fully automatic applications. In order to compensate for such error, at first two novel automatic vessel diameter measurement techniques were proposed and validated which were more robust in the presence of such image artifacts compared to similar methods. However, sometimes the exact edge location and actual diameter value is not of interest. In most case-control studies, it is of importance to comparatively evaluate the variations in retinal vessel diameter as a sign of retinopathy such as arteriolar nicking as an example of hypertensive retinopathy. Vessel diameter is often required to be compared with a reference value in many analytical assessments for diagnostic purpose. This includes monitoring the diameter variations of a specific vessel segment within single subject overtime or across multiple subjects. This helps ophthalmologists to understand whether it has undergone any significant change and perhaps associate it with a disease abnormality. A technique that can effectively quantify that change without being impaired by image artifacts is of more importance and one of the rationales of this study. This research hypothesized an edge independent solution for quantifying diameter variations when the actual diameter value is not required and proposed a new feature based on fractal analysis of vessel cross-section profile as a time series signal. This feature provides a link between FD as a global measure of the complexity and diameter variation as local property of a specific vessel segment. The clinical application of this feature has been validated on two population studies which showed promising result for assessment of mild non-proliferative diabetic retinopathy and 10-year stroke. This research work has also investigated whether the FD of retinal microvasculature would be affected by cyclic pulsations of retinal vessels and whether ECG synchronization is required prior to taking fundus images to compensate for this potential source of variations.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Electrical and Computer Engineering
Keyword(s) Retinal image analysis
Diameter measurement
stroke and diabetes risk assessment
Version Filter Type
Access Statistics: 229 Abstract Views, 387 File Downloads  -  Detailed Statistics
Created: Fri, 26 Sep 2014, 09:14:20 EST by Maria Lombardo
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us