Detection of micro aneurysms automatically for retinopathy screening

Abstract

Diabetic retinopathy is one among the severe diseases of eye leading to irreversible damage when left undiagnosed. Hence, robust automated system for diagnostic medical imaging is becoming a vital necessity in the healthcare industry. Microaneurysms are the first confirmed visual signs of onset of diabetic retinopathy. This work exploits the benefits of tristate median filtering and power law transform for improved candidate extraction work. It includes processing the image and normalization of intensities as microaneurysms are seen as low contrast and tiny.Tristate median filtering removes unwanted noise that would have added up while acquiring images. Power law transformation is applied to increase brightness. And makes the microaneurysms clearer in the image. The second approach involves extraction of retinal vessels from the image as miniature vessels can be falsely detected as candidate lesions. Weapply simple and effective morphological operations followed by detection of lesions by extended minima transform. Candidate features are extracted and then classified by K nearest Neighbour classifier.The performance of the proposed work is analysed giving accuracy specificity, and sensitivity values 91.5%, 82 %, 93% respectively.