Glaucoma Grading VIA Mean Defect Back Propagation From OCT Images

Glaucoma is one of the most prevalent causes of irreversible blindness worldwide. In clinical practice, the most commonly used methods for diagnosing glaucoma are based on fundus and Optical Coherence Tomography (OCT) imaging, tonometry, perimetry and campimetry. OCT image is better suited to detect changes in disease progression than other existing imaging techniques. In addition, using the mean defect index (MD), the campimetry test allows evaluating the evolution of visual field loss. Convolutional Neural Networks (CNNs) can facil-itate automatic diagnosis and grading of glaucoma. However, few studies have addressed this issue using OCT images. In this paper, we aim to improve the grading of glaucoma patients through a neural network that combines the information provided by both tests, OCT images and MD index. Using a two-term loss function (
$\mathcal{L}_{c}$
and
$\mathcal{L}_{R}$
), the proposed framework obtains from an OCT image the glaucoma grade as well as its corresponding MD value. This methodological core outperforms the classification baseline framework by 3.21 %, getting a test accuracy of 0.9259. This approach would not only improve the classification of glaucoma but also extracts the MD index from OCT images without the need for a visual field (VF) test.