Multi-Modal Imaging-Based Feature Fusion for Accurate Glaucoma Diagnosis with Deep Learning

Abstract

Glaucoma, a leading cause of irreversible blindness worldwide, affects over 76 million people currently and is projected to increase to 111.8 million by 2040 according to the World Health Organization. Glaucoma is often asymptomatic and progresses gradually, making early detection and treatment crucial to prevent irreversible vision loss. Ophthalmologists commonly use eye analysis machines such as ophthalmoscopy and fundus photography to obtain scans for accurate diagnosis and treatment of glaucoma. However, clinical-decision support techniques in medical imaging face challenges in interpretability and reliability. To overcome these challenges and accurately detect glaucoma, this study presents a novel Fire Hawk Optimized Deep Learning Based Retinal Disease Grading and Classification (FHODLB-RDGC) utilizing multi-modal feature fusion from Optical Coherence Tomography (OCT) and Fundus images. Radiomics features for OCT images using DL Technique and cup-to-disc ratio for fundus images using NestNet model are extracted using Deep Learning (DL) techniques and merged. The proposed FHODLB-RDGC technique is of great interest in the field of ophthalmology and has demonstrated enhanced retinal disease classification results in extensive simulations on a benchmark database, particularly in accurately diagnosing glaucoma. Therefore, glaucoma, a severe and potentially blinding retinal disease, has the potential of being more accurately recognised and treated early on owing to the FHODLB-RDGC technique.