Improved U-Net Performance with Augmentation for Retinal Optic Segmentation

Glaucoma is a complex eye condition that involves damage to the optic nerve and causes permanent loss of vision. Early detection of glaucoma is an important part of the eye that plays a role in vision, namely the retinal optic. The retinal optic consists of two parts, namely the optic disc and the optic cup. This study aims to build an automatic segmentation model by combining the augmentation and automatic segmentation stages. The augmentation techniques used in this research are rotating, flipping, and sharpening. The augmentation techniques aim to increase the amount and variety of data to improve the quality of segmentation results. The U-Net architecture is used as a segmentation model. The combination of augmentation and U-Net architecture is expected to meet the needs of the U-Net architecture. The U-Net architecture requires large amounts of data. The combination of augmentation and U-Net architecture can improve better performance. Measurement parameters to evaluate the performance of the proposed method are accuracy, sensitivity, specificity, and Intersection over Union (IoU). The results of the proposed method obtained an accuracy of 96%, sensitivity of 92%, specificity of 90%, and IoU of 96%. These results demonstrate that the proposed method for optic disc and optic cup segmentation has excellent and robust capabilities. The proposed method can be used as a model for developing automatic glaucoma detection applications in the medical field.