Grading Fundus Images for Diabetic Retinopathy

Diabetic Retinopathy (DR) is an anomaly, found commonly among diabetic patients, which can lead to blindness if left untreated for a long period of time. Early detection of the complication, leads to better probability of cure. Lesions and cotton wool spots in retinal images, are the symptoms that help in the early detection of DR. Due to these lesions the image features will change throughout the retinal image, according to the level of the disease progression. Hence, in the proposed method, these changes in features are used to grade the retinopathy. There are three levels of disease progression and the classification method was able to classify all the three levels from the given retinal images. Also, Hough transform is used to detect the cotton wool spots since these spots have approximately the same intensity as that of the optic disc. The changes in features caused by the lesions and the cotton wool spots and exudates are the key to the detection. The feature extraction technique is Grey Level Cooccurrence Matrix (GLCM), which extracts about 44 features from the image data for further classification. Using an unsupervised classification method we have discovered the similar instances between data. The clusters are formed using K means classification technique. The retinal images, both normal and abnormal, were obtained from the Messidor database and the pixel size used was 2240 � 1488.