Key points are not available for this paper at this time.
Glaucoma in the retina is considered a significant cause of irreparable vision loss, and automated glaucoma detection on fundus images has seen the development of several approaches recently. Although early diagnosis and treatment can prevent possible blindness, manually segmenting the optic disc depends on the availability of a large number of labeled samples for the training phase. The proposed method utilizes U-Net for segmentation based on labeled and unlabeled data for prediction and data enhancement with a large number of feature channels in up sampling. This approach can improve overall performance by considering pixel-level and image-level labels during network training and encoding the data train, differentiating U-Net from Convolutional Neural Network (CNN). Classification is performed using CNN for diagnosing glaucoma, evaluated on a large training dataset. The obtained results demonstrate that the proposed U-Net model achieves better accuracy, reaching 0.99 on the Drishti dataset. These results ensure accurate segmentation and classification performance compared to other existing methods such as CNN and Multi-layer Neural Network.
Neelagar et al. (Fri,) studied this question.