An Ocular Feature-Based Novel Biomarker Determination for Glaucoma Diagnosis Using Supervised Machine Learning and Fundus Imaging

Abstract

Glaucoma, an incurable eye disease, is a significant global health concern recognized by the World Health Organization. It progresses by increasing eye pressure, damaging the optic nerve, and leading to blindness. Regular eye exams are crucial for early detection and preventing vision loss, as early stages are often asymptomatic. Advanced feature engineering and machine learning are key to uncovering new glaucoma biomarkers, aiding early diagnosis and improving automated systems for ophthalmologists. The glaucoma biomarkers in the realm of machine learning are the features that act as a metaphor of biological biomarkers. Most research has concentrated on either structural or nonstructural feature selection strategies, with limited analysis on combined feature sets. In this letter, a new reduced feature set is investigated by combining both handcrafted structural and nonstructural features that act as a novel glaucoma biomarker for efficient and robust glaucoma diagnosis system. The proposed biomarker is a combination of structural and nonstructural (statistical, spectral, and geometric) features, which has been rigorously evaluated on the publicly available, large, and diverse standardized multi-channel dataset (SMDG)-19 glaucoma dataset. The classification accuracy achieved from Extra Tree Classifier is 85.42 % using tenfold cross-validation approach. In light of this, the suggested method's outcome set it apart from other State-of-the-Art models in biomarker determination and makes it a unique choice for ophthalmologists seeking a glaucoma biomarker for diagnosis systems.