Home
According to recent reports from the World Health Organization (WHO), myopia now affects 30% of the global population and continues to rise steadily. Given that early detection is critical for timely and effective treatment, this study evaluates and compares the performance of two prominent convolutional neural network (CNN) architectures—ResNet50 and InceptionV3—for the automated classification of myopia using blue-light fundus imagery. Employing a cross-sectional, non-experimental design, the research utilized a large-scale dataset of 124,794 images (63,294 Myopia; 61,500 Normal). The data were partitioned into training (70%), validation (20%), and testing (10%) sets. Models were implemented in Python using TensorFlow and Keras, leveraging the Google Colab Pro environment with A100 GPU acceleration. To mitigate overfitting and enhance generalization, rigorous preprocessing and data augmentation techniques were applied. Experimental results indicate that both architectures achieved exceptional performance; notably, InceptionV3 outperformed ResNet50, achieving a validation accuracy of 99.97% and a significantly lower loss of 0.0075. These results confirm the robustness of deep learning models for clinical myopia screening in response to the growing global prevalence.