Automatic microaneurysm detection using Multi-level Threshold based on ISODATA

Abstract

Diabetic Retinopathy is one of the most serious diseases that can lead to blindness. The small swelling blood regions or microaneurysms are the early sign of Diabetic Retinopathy. Detecting that a patient has got the Diabetic Retinopathy at the earliest stage as possible can help to prevent him/her from the vision lost. However, automatic microaneurysm detection is still a challenging topic for medical image processing researchers. This is because of the varieties of microaneurysm characteristic such as size, contrast, shape, and data distribution. In this paper, we propose an approach to automatically detect microaneurysms using Multi-level Threshold based on ISODATA. The proposed method consists of two main steps: 1) preprocessing and 2) feature extraction. In the preprocessing step, Contrast Limited Adaptive Histogram Equalization, Gaussian Filter and Median Filter are applied to enhance the image quality. Next, in the feature extraction step, Multi-level Threshold based on ISODATA and Noise Removing Techniques are adopted to remove non-microaneurysm objects. The 89 retinal fundus images from a public database DIARETDB1 are used as a dataset. By comparing with the ground truth, the proposed approach provides the reasonable results with sensitivity of 62.82%, specificity of 93.60% and accuracy of 93.43%.