Segmentation and grading of diabetic retinopathic exudates using error-boost feature selection method

This paper proposes a method to segment the exudates and lesions in retinal fundus images and classify using selective brightness feature. The exudates are segmented from background and their size is also measured. The segmentation is done by extraction of pixels which fall in the color range of the spots. The essential features inferred from the segmented image include the count of the exudates, maximum size, percentage affected, color intensity of the spot, average size and the area affected by haemorrhages. The diagnosis is supported by error-boost feature selection technique. This technique classifies the retinal images as normal or abnormal based on the features obtained from the segmented image. The abnormal images are further classified as mild, moderate or severe and there is an additional classification based on non-proliferative and severe proliferative diabetic retinopathy. The diagnosis parameter ranges for each feature are set prior to the severity classification. The error boost feature selection algorithm selects the key features which classifies the retinopathy more accurately. The obtained results seem to be clinically relevant.