CNN with Multiple Inputs for Automatic Glaucoma Assessment Using Fundus Images

In the area of ophthalmology, glaucoma affects an increasing number people. It is a major cause blindness. Early detection avoids severe ocular complications such as glaucoma, cystoid macular edema, or diabetic proliferative retinopathy. Intelligent artificial intelligence has been confirmed beneficial for assessment. this paper, we describe approach to automate diagnosis using funds images. The setup proposed framework in order: Bi-dimensional Empirical Mode Decomposition (BEMD) algorithm applied decompose Regions Interest (ROI) components (BIMFs+residue). CNN architecture VGG19 implemented extract features from decomposed BEMD components. Then, fuse same ROI bag features. These last very long; therefore, Principal Component Analysis (PCA) are used reduce dimensions. bags obtained input parameters classifier based on Support Vector Machine (SVM). To train built models, have two public datasets, which ACRIMA and REFUGE. For testing our part REFUGE plus four other RIM-ONE, ORIGA-light, Drishti-GS1, sjchoi86-HRF. overall precision 98.31%, 98.61%, 96.43%, 96.67%, 95.24%, 98.60% ACRIMA, REFUGE, sjchoi86-HRF respectively, by model trained Again accuracy 98.92%, 99.06%, 98.27%, 97.10%, 96.97%, 96.36% training ACRIMA. experimental results different datasets demonstrate efficiency robustness approach. A comparison with some recent previous work literature shown significant advancement proposal.