Few-shot Learning Using a Small-Sized Dataset of High-Resolution FUNDUS Images for Glaucoma Diagnosis

Deep learning has recently attracted a lot of attention, mainly thanks to substantial gains in terms of effectiveness. However, there is still room for significant improvement, especially when dealing with use cases that come with a limited availability of data, as is often the case in the area of medical image analysis. In this paper, we introduce a novel approach for early diagnosis of glaucoma in high-resolution FUNDUS images, only requiring a small number of training samples. In particular, we developed a predictive model based on a matching neural network architecture, integrating a high-resolution deep convolutional network that allows preserving the high-fidelity nature of the medical images. Our experimental results show that our predictive model is able to obtain higher levels of effectiveness than vanilla deep convolutional neural networks.