Soft Attention based Densenet for Severity Grading of Diabetic Macular Edema

Diabetic Macular Edema is a diabetic complication that leads to the blindness in working age-population. DME causes severe damage to the blood vessels in the retinal tissue and leads to the leakage of fluid and distorted vision. Diabetic retinopathy (DR) is considered to be a primary cause of DME. The presence of DME is identified by the hard exudates (bright lesions) in the retinal images. We can prevent the blindness in diabetic patient by the early detection of diabetic macular edema. There are various methodologies for diagnosing and predicting the severity of the DME. The proposed work uses Soft Attention based Densenet (SA -Densenet) for predicting the severity levels (focal DME & Diffuse DME) of the diabetic Macular Edema. The proposed framework uses the kaggle fundus dataset for predicting the severity level of DME. Throughout the experiment, measures including accuracy, precision, and recall are assessed to check the performance of the proposed work. Comparatively the proposed strategy is the most accurate in predicting diabetic macular edema from fundus images.