A Hybrid CLAHE-GAMMA Adjustment and Densely Connected U-NET for Retinal Blood Vessel Segmentation using Augmentation Data

The retina is a thin layer on the back of the eyeball that is sensitive to light. Retinal blood vessels function to supply blood and oxygen to the retinal tissue. If there is a disturbance in these blood vessels, we need to detect a disease or other disorder in the retina. One of the stages in image recognition is segmentation. This study used the Convolutional Neural Network method with U-Net architecture for retinal image segmentation. Before the segmentation stage, preprocessing processes such as grayscale conversion, standardization, Contrast Limited Adaptive Histogram Equalization (CLAHE), and gamma adjustment are carried out. Furthermore, the retinal image that has been improved in quality will be segmented. This segmentation separates the background and foreground to obtain retinal blood vessels. The last stage measures the performance of the segmentation results using the parameters, namely accuracy, specificity, sensitivity, F1 score, and Jaccard similarity score. The dataset used in this study is the DRIVE. Data augmentation is done to increase the amount of data. It tested the results using the activation function Sigmoid and rectified linear unit (ReLU) with different kernel sizes, namely 3 x 3, 5 x 5, and 7 x 7. The best research results with this method get an accuracy of 95.48%, sensitivity 74.91%, specificity 98.48%, F1 score 80.84%, and Jaccard similarity score 67.84% using ReLU activation function with kernel size 5 x 5. The method used is better and more efficient for image segmentation.