Optimizing Glaucoma Diagnosis with Ensemble Deep Learning Models Integrating CNN, ResNet50, and MLP Architectures

Glaucoma a pervasive, incurable eye condition which diminishes eyesight and worsens the standard of life. It damages the optic nerve, which is in charge of maintaining vision in the rear of the eye and causes long-term vision loss. There are three categories in the dataset, A abnormal indicating angle-closure glaucoma and early-glaucoma indicating open-closure glaucoma and a normal eye for prediction. In this article, we formulate various deep learning models. The Convolutional Neural Network (CNN) incorporates fully-connected layers at its conclusion, leveraging the extracted features from the image for the ultimate prediction. The ResNet50 model, where the numeral “50” denotes the total count of layers, encompassing both convolutional and fully connected layers. Additionally, the Multilayer Perceptron encompasses an input layer, one or more hidden layers, and an output layer. and ensembling these models to make a single model to improve the accuracy. In this ensembling method, different models can be separately trained using the same dataset (glaucoma dataset from ORIGA). Ensembling can lessen overfitting, increase robustness, and boost performance as a whole. The accuracy of the ensembling model is 89% which is more when compared to other individual models.