U-Net-based gannet sine cosine algorithm enabled lesion segmentation and deep CNN for diabetic retinopathy classification

ABSTRACT

Diabetic retinopathy (DR) is a reason for blindness. This disease raises a threat of developing deteriorations in blood vessels supplied by the retina. In this research, U-Net_Gannet Sine Cosine Algorithm (U-Net_GSCA) is used for DR segmentation, and DR classification is made by the GSCA_Deep Convolutional Neural Network (GSCA_DeepCNN). In the pre-processing process the Region of Interest (ROI) is extracted. For pre-processing process, it uses Laplacian filter. Color spacing, rotation, and brightness are the augmentation techniques used for image augmentation. U-Net which is tuned by GSCA is used for lesion segmentation. Furthermore, GSCA is the integration of GOA with SCA. The segmentation of the artery and vein is performed utilising the sparking process. In the feature extraction phase, they uses augmented and segmented images. From this images, the feature extraction is made. Finally, the classification of DR is made by the deep CNN that is also trained by GSCA. Furthermore, U-Net_GSCA achieved 92.6% segmentation accuracy, GSCA_DeepCNN achieved 92.2% classification accuracy, 92.5% sensitivity, and 91.9% specificity.

KEYWORDS:

• Gannet optimisation algorithm (GOA)
• Sine cosine algorithm (SCA)
• laplacian filter
• U-Net
• deep convolutional neural network (deep CNN)

Disclosure statement

No potential conflict of interest was reported by the author(s).