Data Augmentation of Domain Learning in Optic Nerve Combined Cup-Disc Segmentation with a Few Labeled Data

In recent years, deep learning's powerful learning capabilities have found applications in medical image analysis, a field where manual segmentation and annotation are both time-consuming and costly. However, the limited amount of medically labeled data presents a challenge for deep learning models that require large datasets to learn effectively. This paper focuses on the segmentation of fundus images for automatic glaucoma assessment and proposes the use of a Generative Adversarial Network (GAN) with a domain-based learning method to expand the data space and identify a suitable segmentation network structure and loss function for learning with low volume data. The proposed method's effectiveness is demonstrated through experiments on several benchmark datasets of fundus images. For the best performance, only 5 percentages of training data are required for the proposed method to achieve the same level of accuracy as state-of-the-art models trained on large-volume data.