Efficient Diagnosis of Retinal Diseases Using Convolutional Neural Networks

The eye is a vital sensory organ that enables us to fulfill all our life’s needs. Diseases affecting such a vital organ can have a detrimental impact on our lives. Although certain eye conditions are easily managed, others may result in lasting damage or loss of sight if not identified promptly. Problems within the retina or improper image focus on the retina may result in loss of eyesight. Optical Coherence Tomography (OCT) can identify diseases using retinal images taken from a side-angle view. Medical images are analyzed using Convolutional Neural Networks (CNNs) to automatically diagnose diseases. Doctors may reach varying conclusions when diagnosing diseases based on medical images. These conclusions may even contain human error. These challenges can be overcome with the use of CNNs. When creating a CNN architecture, many hyperparameter values need to be determined at the beginning before the training phase. A well-structured design is crucial for the successful performance of CNNs. The lengthy training time of CNNs makes testing every hyperparameter combination a very time-intensive process. This research determined the best hyperparameters for CNNs by means of Bayesian optimization. The study employed a dataset comprising four categories: DME, CNV, DRUSEN, and NORMAL. With Bayesian optimization, this proposed model reached an accuracy and F1 score of 99.69%, outperforming existing research findings. The proposed model will also help doctors to make decisions and speed up the decision-making process.