Predicting Glaucoma With Hyperparameter-Tuned Convolutional Neural Networks As Clinical Diagnostics

Abstract

Glaucoma is responsible for a significant percentage of irreversible blindness worldwide and is often asymptomatic until very late in its course. As with many diseases and conditions, the earlier glaucoma is detected, the less likely the patient will have permanent optic nerve damage. Traditional clinical screening techniques, sadly, suffer from the constraints of subjectivity, access, and cost. Based on clinical diagnosis in conjunction with hyperparameter-optimized Convolutional Neural Network models, the research advises a hybrid clinical and artificial intelligence approach to identify glaucomatous eyes from retinal fundus images. The suggested model outperformed ResNet and U-Net, two common deep learning techniques that are frequently applied to related issues. During testing, it achieved a 96% accuracy rate and an AUC of 0.96. Overall, the research presents a baseline for the potential of AI-informed diagnostic systems to augment the abilities of practitioners in a new way to conduct scalable glaucoma screening and early detection with an emphasis for clinical practice in areas without adequate access to such screening. Future directions involve building generalization of our model using different datasets, developing real-time, multimodal data analysis through clinical decision support systems, and integration on an edge device with low-latency computational capacity.