In recent decades, Diabetic Retinopathy (DR) is a progressive eye disease that causes severe eye injuries if it is not detected and treated on time. Accurate microaneurysms detection is a vital step for early detection of DR, because it is the primary sign of disease. In this paper, a six-phase model is introduced for detecting microaneurysms from the fundus retinal images for early diagnosis of DR. Initially, lower light retinal image enhancement, image normalization, gradient weighting and shade correction are applied for improving the visibility level of fundus retinal images, which are acquired from the e-ophtha, and DiaRetDB1 datasets. Further, the hessian-based filter, and Otsu thresholding with the morphological operator are employed to eliminate blood vessel regions from the microaneurysms regions. Next, a grey wolf optimizer is used for predicting the correctness of the segmented microaneurysms regions. After segmentation, feature extraction: shape and Gray Level Co-occurrence Matrix (GLCM) features and classification: Modified K Nearest Neighbor (MKNN) are used to extract features from microaneurysms regions and to classify microaneurysms and non-microaneurysms regions. The simulation result showed that the proposed model achieved effective performance in microaneurysms detection compared to the existing models such as H-maximamultilevel thresholding-multilayer perceptron and statistical geometrical features. The proposed model achieved 99.10% and 99.90% of accuracy on e-ophtha and DiaRetDB1 datasets, which are effective related to the existing models in microaneurysms detection.