Hybrid CNN-GRU based Machine Vision Multimodal Deep Learning Model for Glaucoma Detection in Retinal Fundus Image

One of the most common and well-known conditions is glaucoma, but patients, especially those over 60, may lose their vision if the condition is left untreated or misdiagnosed. In the proposed work, we classify a picture as either healthy or glaucomatous using convolutional neural networks. The assessment of glaucomatous optic nerve and retinal damage may be achieved by structural evaluation using optical coherence tomography (OCT), and the degree of vision loss can be ascertained through functional visual field (VF) testing. In this work, we used a convolutional neural network (CNN) and a long short-term memory (LSTM) network to develop a multimodal deep learning model for glaucoma progression prediction. We used OCT scans, VF values, clinical, and demographic information from 89 glaucoma patients who saw five different physicians in a 12-month period. We showed that our novel generative model-based method can predict progression seven months ahead of time with the greatest AUC of 0.85. Moreover, the incorporation of artificial future images enabled the model to accurately forecast vision loss up to 8 months ahead of time with an AUC of 0.83, in contrast to the use of just structural (AUC = 0.69) or functional metrics (AUC = 0.71). The viability of using artificial follow-up OCT pictures to detect glaucoma development early is clarified by this study. As part of transfer learning, hyperparameters including the learning rate, optimizers, and activation function are changed throughout experimentation. The work offers practicing ophthalmologists an alternative point of view.