Identification and Classification of Ocular Surface Squamous Neoplasia Using YOLO and Traditional Feature Extraction and Fusion Methods

Ocular Surface Squamous Neoplasia (OSSN) is a potentially life-threatening eye condition that encompasses a range of abnormalities, from dysplasia to invasive carcinoma. Early detection and intervention are crucial for improving treatment outcomes and minimising complications. This study suggests a novel hybrid method for detecting and categorising OSSN that combines the advantages of conventional feature extraction techniques with deep learning-based object detection. The approach integrates the deep learning-based object detection framework, YOLO (You Only Look Once), with traditional feature extraction techniques, including texture descriptors. The YOLO algorithm detects OSSN lesions in ocular images, while conventional feature extraction methods derive relevant features from these identified lesions. A feature fusion strategy integrates features from the deep learning algorithm with handcrafted features, thus enhancing overall classification performance. Various classification algorithms, including K-Nearest Neighbours (KNN), Support Vector Machine (SVM), Random Forest, and XGBoost, are trained using the fused feature set. The experiment results show that the proposed hybrid strategy can identify and categorize OSSN in ocular images. Performance metrics for the feature fusion model indicate accuracies of 99.04% for KNN, 98.1% for SVM, 97.14% for Random Forest, and 92.38% for XGBoost classifiers. These findings imply that the proposed hybrid method, which identifies and classifies OSSN in ocular images with excellent accuracy and resilience, is a viable tool for clinical applications.