Intensified U-Net Architecture for Segmentation of Diabetic Retinopathy in Retinal Image Processing

Diagnosing diabetic retinopathy (DR) is essential to keeping patients' vision intact.Since the fovea and optic disc (OD) are significant retinal factors, it is important to recognize them, but because these procedures are intricate, there is a risk of overfitting, complexity, and errors.Hence introduced the Intensified UNet framework (IUNetA), which locates using retinal images' optic disc and fovea to address DR.With a Wiener filter for noise reduction and skip connections for low-level image information, the architecture comprises of an encoder, decoder, skip connections, and a special Atrous Convolution Double Residual Block (ACDRB).To solve the semantic gap issue, a Cocktail Attention Block (CAB) is incorporated into the skip connection.Channel compression is achieved via a 1x1 convolution layer.In order to alleviate the vanishing gradient issue, the decoder block uses encoded feature maps to retrieve segmented object information.The tangent function is then used to calculate the final output.Especially, the analysis is carried out by IDRID dataset, the IUNetA attains the accuracy as 99.9%, IoU attained 89.17%, Sensitivity attained 90% and Dice Similarity Coefficient (DSC) attained 99.14% when compared to Prior models.Thus, the overall architecture is accurately segmented and localized the optic disc and fovea in the DR.