Predicting and Diagnosing Disorders
There is some strong research evidence about the relationships between vascular
system disorders and vascular network measurements. These relationships are currently
somewhat fuzzy due to the imprecision of current measurement techniques, lack of
software and insufficient cooperation between medical and non-medical researchers.
We have built accurate and precise measurement algorithms, and have established
a strong cooperation relationship with medical researchers by introducing an automated
system without human intervention. This greatly reduces the workload involved in
extracting manual or semi manual measurements, potentially allowing us to greatly
increase the number of samples in medical studies. This allows for the study of
correlations and differentiation between retinal graphs, both between subjects and
(for progression) within subjects. This will support the quantification of pathological
conditions including venous beading, arterial nicking, tortuosity and neovascularization.
The accurate segmentation of retinal vessels provide an opportunity for researchers
who are working on non vessel features such as haemorrhages and microaneurysms (HMAs),
hard exudates, drusen and cotton wool spots, to develop their methods without any
negative side effects caused by the vessel network. During processing the ESP algorithm
detects some non-vessel objects and subsequently classifies them as non vessel segments.
These may be useful for subsequently classification as potential lesions.
For many retinal diseases the primary indicators are lesions – either dark lesions
(HMAs) or white lesions (exudates, drusen and cotton wool spots). These lesions
are often faint and difficult to accurately detect and classify. We have developed
methods based on level set methods (contour finding) that very accurately segment
potential lesions, leading to improved classification accuracy. We are studying
the use of this technology in diagnosis of diabetic retinopathy.