Retinal Lesion Detection Project
The project aims to reliably detection retinal lesions – both white lesions (exudates,
drusen and cotton wool spots) and dark lesions (haemorrhages and microaneurysms
– HMEs) with a view to detection of retinal diseases. Many lesions are faint and/or
small, and there is a great variation in appearance and scale, making accurate segmentation
extremely difficult. Our approach is to concentrate on very accurate segmentation
of potential lesions, so that highly accurate classifiers can be developed. We therefore
first detect candidate lesions using peak detection algorithms, then segment using
specially developed level set methods. Following accurate segmentation it is possible
to extract features to perform successful classification.
The ONH dataset for comparative studies ia under construction.