A Neural Network Learning Approach Using ConvNet Models to Consolidate Corneal Ulcer Detection in Opthalmology

Corneal Ulcer is an infection induced medical condition in which the patients get open sores in their eyes. In this paper, we investigate the use of deep learning in the automated classification of such ulcers using the SUSTech-SYSU dataset, released by the Zhongshan Ophthalmic Center at Sun Yat-sen University. The dataset consists of 712 pictures of patients with different kinds, degrees, and classifications of corneal ulcers. After fluorescein staining, the dataset is collected and utilized to improve deep learning models. Pre-trained models for Eye Corneal Ulcer (ECU) image categorization are trained using the deep learning Convolutional Neural Networks (CNN) architecture, and a customized model is created to improve test accuracy and validation. 7,200 training photos, 1,800 validation images, and 3,000 testing images are included in the dataset for assessment. The modified model employs an Adam optimizer for multi-class classification and categorical cross-entropy as the loss function in a sequential architecture for feature extraction and classification. The validation set is used to perform hyperpa-rameter tuning, which maximizes model performance. Training accuracy is 99% and validation accuracy for the customized model is 90%. The goal of this research is to create an automated system for classifying corneal ulcers, which could increase the efficiency of ophthalmologists in the diagnosis and treatment of corneal infections.