Earlier Detection of Microaneurysms in Fundus Image using VGG-19 with Dual Attention Mechanism

Diabetic retinopathy (DR) is characterized by the presence of retinal microaneurysms (MAs) as its primary symptom. The automated identification of MA is beneficial in aiding physicians with the process of diagnosing and treating the condition. Previous algorithms mostly emphasized the characteristics of the target, but it is also valuable to investigate the local structural properties of both the target and background. This work aims to offer an automated method for detecting microaneurysms (MAs) in fundus retina pictures. The proposed technique was tested using the Kaggle Diabetic Retinopathy database, which consists of 35,126 color fundus images. The methodology encompasses the stages of pre-processing, feature extraction, and classification. The images underwent pre-processing, which involved filtering and enhancing the contrast. Then, 3D-CNN based feature extractor was adopted along with VGG-19 Dual Attention Mechanism (VDD-19_DAM) classifier helps to classify three categories such as saccular, fusiform and focal bulges. The performance of the classifier was evaluated in terms of specificity, sensitivity, and accuracy which attained 99.8%, 94.6% and 95%, respectively.