Deep Convolutional Neural Network Prediction For Glaucoma Detection Using OCT and OCT-Angiography Disc-and Macula-Centered Images and Their Combined Power

This study evaluates the effectiveness of deep convolutional neural networks (CNN) in glaucoma detection based on a multimodal approach using Optical Coherence Tomography (OCT) and OCT Angiography (OCTA) images centered on the disc and macula regions. The research focuses on binary classification for glaucoma diagnosis. The study introduces a novel intermediate fusion architecture, named GlauOCTA, which enables the modeling of interactions between features of different modalities while offering adaptability to varying numbers of input images. The results demonstrate that the combination of disc-and macula-centered images from both the OCT and OCTA technologies consistently outperforms individual region images. Specifically, when considering only the disc or macula, OCT+OCTA yields the best accuracy and area under the curve (AUC). However, when both the disc and macula images are available, OCTA alone provides the most accurate results. These findings provide valuable insights for enhancing diagnostic precision in glaucoma classification and have potential implications for improving glaucoma detection in clinical practice.