Publication

Abstract : Diabetic retinopathy (DR) is an eye ailment affecting retinal blood vessels. Many deep learning methods for detecting DR have been presented as manual diagnosis is time-consuming and inconvenient. InceptionV3 architecture was modified using a soft attention module in this proposed framework. The attention technique’s basic concept is to concentrate on specific relevant parts by assigning the weights accordingly. Contrast Limited Adaptive Histogram Equalization (CLAHE) is the pre-processing method applied initially to the fundus images to improve the contrast level. Along with this, augmentation has been done to increase the number of images, which are then trained and validated using a modified InceptionV3 model. The experimental results show that the suggested model better diagnoses all stages of DR than existing techniques and outperforms the existing model on the IDRiD and DDR datasets.