Automated Screening for Ocular Abnormalities: Leveraging Data Augmentation for Improved Diagnostic Accuracy

Early and accurate diagnosis of ocular diseases is crucial for timely intervention and improved patient outcomes. This study explores the potential of deep learning for automated classification of four common eye conditions: cataract, diabetic retinopathy, glaucoma, and normal eyes. We propose a meticulously fine-tuned EfficientNetV2L architecture, specifically designed for this task. Our methodology incorporates a comprehensive preprocessing pipeline to enhance image quality and extract relevant features. We address the challenge of limited data availability by employing a comprehensive data augmentation strategy, effectively expanding the training dataset. Our research utilizes the Ocular Image Database (OIH), comprising 4217 images, and evaluates the model’s performance through extensive testing on unseen data. The proposed approach achieves a remarkable accuracy of 98.88% in classifying ocular diseases. This research contributes to the growing body of evidence supporting the transformative potential of AI in healthcare, paving the way for more advanced and personalized patient care in the future.