Design and development of pervasive classifier for diabetic retinopathy

Diabetic retinopathy is a serious complication of diabetes mellitus which can eventually lead to blindness around 10% of patients with diabetes develop sight threatening retinopathy. Diabetic retinopathy is a condition for which treatment is available. The treatment is effective in preventing sight loss. Hence there has been a major impact on automatic screening for diabetic retinopathy. Detection of sight threatening retinopathy is an important aim of the screening programme. Four types of lesions namely microaneurysm, haemorrhages, soft exudates and hard exudates are predominant stages of diabetic retinopathy. For detection of these lesions feature extraction is an important attribute. Various types of feature extraction methods are already available. But these methods are limited in their application. Gabor filter is used for detection and differentiation of bright lesions, wavelet transform is used for detecting microaneurysm alone. As a consequence we have to rely on various feature extraction techniques to discriminate the pathologies. This paper uses AM-FM feature extraction method which is more advantageous than available methods. Contrary to other methods the same system is applied to detect red lesions (microaneurysm and haemorrhages) and hard exudates. After the features are extracted an automatic classification system based on partial least square is used to discriminate the pathologies.