To search, Click below search items.


All Published Papers Search Service


MADeep-Automatic Microaneurysms Detection on Retinal Fundus by using region growing and deep neural networks


Qaisar Abbas


Vol. 19  No. 1  pp. 161-166


Detection of microaneurysms (MAs) is required by ophthalmologists as a part of eye screening process. It is difficult for them to detect and count MAs through the existing system due to its complex structure. If MAs lesions are automatically detected then it can definitely release the workload on experts. Therefore in this paper, the competition region growing algorithm (CRGA) and variants of deep neural networks (DNNs) techniques are used to automatically detect microaneurysms (MAs) lesions. This new proposed system is known as MADeep that contains three main phases such as localization of MAs regions through a CRGA, segmentation of candidate regions of red lesions by the region-based convolutional neural network (R-CNN) and detection of MAs regions by stack-based autoencoders (SAEs). The MADeep system is tested and evaluated on a set of 600 retinal images that contains 1024 MAs, which are obtained from three different online sources. On average, the MADeep achieved sensitivity (SE) of 92%, specificity (SP) of 95%, true positive rate (TPR) of 93%, false positive rate (FPR) of 65% and area under ROC curve (AUC) of 0.94. The experimental results indicate that the proposed MADeep system is better than other state-of-the-art systems.


Microaneurysms, Diabetic retinopathy, Retinal fundus images, Deep neural network, Region-based convolutional neural network, Autoencoders, region growing.