Performance analysis of automated lesion detection of diabetic retinopathy using morphological operation

Diabetic retinopathy is one of the most common complications of diabetes. Hemorrhages, micro-aneurysms, exudates are one of the earlier signs of diabetic retinopathy. This paper proposes an algorithm of matched filter with morphological operation for the detection of lesions in the fundus retinal image. Contrast limited adaptive histogram equalization method is used for the extraction of vessels. The noise is removed from the images using matched filter. After enhancement, the thresholding is applied for vessel extraction. The threshold of the image is done by the iterative self-organizing data analysis technique algorithm method. After removal of optic disk and the blood vessels from the retinal image, the morphology method is used to easily identify different types of lesions. The diabetic retinopathy images were collected from DIARETDB1; the morphology operation method is analyzed using the metrics of sensitivity, specificity and accuracy. The proposed method's detection accuracy value for the recognition of micro-aneurysms, exudates and hemorrhages was 98.43%, 98.06% and 98.68% compared with the results of the differential evolution algorithm. Detection of lesion such as micro-aneurysms, hemorrhages and exudates was possible. When compared with the differential evolution algorithm, morphological method achieved good accuracy for the detection of diabetic retinopathy.

Automated summary

This paper presents a new algorithm for detecting diabetic retinopathy, which combines matched filter with morphological operation to achieve easy identification of different lesion types and vessel extraction. Experimental results demonstrate the high accuracy of this method in lesion recognition.