AUTOMATIC DETECTION OF HIGH MYOPIA AND PATHOLOGICAL MYOPIA USING FUNDUS IMAGE

Abstract—Myopia, particularly high myopia (HM) and pathological myopia (PM), has become a growing global concern, especially among the younger population. Early and accurate detection is essential to prevent irreversible vision loss caused by progressive retinal damage. This paper presents an automated diagnostic system for classifying fundus images into normal, high myopia, and pathological myopia categories using deep learning techniques. The proposed method involves preprocessing steps for image enhancement, feature extraction using DenseNet121, and classification using a three-class architecture. Comparative evaluation with models such as VGG16, GoogleNet, and standard CNNs revealed that DenseNet121 achieved the best performance, with 94% accuracy and an F1-score of 92. The system also incorporates a user-friendly graphical interface and database integration for real-world deployment. This research demonstrates the viability of using deep learning to support ophthalmologists in early diagnosis and long-term patient monitoring, especially in remote or resource-constrained settings.