A Comparative Analysis of Image-Based Classification and Object Detection for the Ocular Redness Grading

This study presents a comprehensive exploration into ocular red-eye grading, employing ResNet-based image classification and YOLOv8-based object detection methodologies. The comparative analysis reveals distinct strengths and weaknesses inherent in each approach. While ResNet exhibits commendable accuracy and Specificity, its susceptibility to potential overfitting suggests suitability for tasks requiring detailed feature analysis. Conversely, YOLOv8 demonstrates superior performance across accuracy, Specificity, recall, and Fl-score metrics, positioning it as a promising solution for efficient ocular red-eye grading, particularly in real-time processing scenarios. Acknowledging study limitations, including potential dataset biases and challenges associated with diverse ocular conditions, is imperative for ensuring robustness and generalizability. Future research directions emphasize refining models, incorporating additional metrics, and diversifying datasets. Consideration of AI-based methodologies from natural images and videos, such as those captured by smartphones, is suggested for further investigation. Collaborative efforts between computer vision researchers and healthcare professionals hold promise for enhancing clinical relevance. The adaptability of ResN et and YOLOv8 underscores their potential utility in clinical settings, signaling advancements in diagnostic tools and personalized patient care within ophthalmology.