RETINAL MICROANEURYSMS DETECTION USING LOCAL CONVERGENCE INDEX FEATURES

Abstract

Retinal microaneurysms (MAs) are the earliest clinical sign of diabetic retinopathy disease. Detection of microaneurysms is crucial for the early diagnosis of diabetic retinopathy and prevention of blindness. In this paper, a novel and reliable method for automatic detection of microaneurysms in retinal images is proposed. In the first stage of the proposed method, several preliminary microaneurysm candidates are extracted using a gradient weighting technique and an iterative thresholding approach. In the next stage, in addition to intensity and shape descriptors, a new set of features based on local convergence index filters is extracted for each candidate. Finally, the collective set of features is fed to a hybrid sampling/boosting classifier to discriminate the MAs from non-MAs candidates. The method is evaluated on images with different resolutions and modalities (RGB and SLO) using six publicly available datasets including the Retinopathy Online Challenges dataset (ROC). The proposed method achieves an average sensitivity score of 0.471 on the ROC dataset outperforming state-of-the-art approaches in an extensive comparison. The experimental results on the other five datasets demonstrate the effectiveness and robustness of the proposed microaneurysms detection method regardless of different image resolutions and modalities.

Existing System

Morphology-based approach and support vector machine (SVM) classifier

Proposed System

In the first stage, the MA candidates are extracted using multi-scale multi-orientation gradient weighting and iterative thresholding. The gradient weighting technique is particularly useful for the detection of small objects with weak boundaries and in low contrast regions. Compared to other techniques, the multi-orientation and multi-scale gradient weighting technique enhances the boundary of gradient structures that is very characteristic of local shape, and it provides local representation which have an easily controllable degree of invariance to local geometric transformations such as translations and rotations. Afterwards, the method generates a set of features for each candidate depending upon their intensity, shape and LCF responses. The LCF filters are based on gradient convergence but not intensity and as such can represent low contrast MAs which otherwise would be easily lost in the background noise. The true MA candidates are then selected using a hybrid sampling/boosting classifier to avoid the drawback of imbalanced data learning and to improve the performance of MA detection. The RUSBoost (with decision trees as the weak learners) is a suitable classifier since we deal with a skewed set with the minority of MA candidates and the majority of non-MA candidates.

Architecture

BLOCK DIAGRAM