Automatic Microaneurysyms Detection using Deep Learning from Fundus Images

Microaneurysms are regarded as one of the first warning indicators of diabetic retinopathy (DR), which may cause gradual vision loss and eventually blindness (MAs). These MAs are practically round in form, black in colour, and very small in size, all of which make them difficult for human ophthalmologists to detect. Here, precise early diagnosis of microaneurysms aids in curing DR before irreversible blindness sets in. The suggested technique uses a combination of the VGG-19 and Inception-v3 pre-trained convolutional neural network (CNN) models to conduct early identification of MAs. Using public datasets like "E-Ophtha" and "DIARETDB1," the suggested method was able to obtain 96% and 94% classification accuracy, respectively, in performance evaluations. The new method was also more sensitive and specific than existing methods for detecting microaneurysms.