Glaucoma Assessment from OCT images using Capsule Network

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis relies on analyzing retinal changes, often using Optical Coherence Tomography (OCT).
• Current methods may require complex segmentation of retinal nerve fiber layer and optic nerve head structures.
• Deep learning offers potential for automated and accurate glaucoma detection from OCT data.

Purpose of the Study:

• To propose a deep learning Capsule network for direct glaucoma classification from 3D OCT volumes.
• To evaluate the performance of the proposed Capsule network against existing methods.
• To demonstrate a segmentation-free approach for glaucoma detection.

Main Methods:

• A Capsule network was developed to directly process 3D OCT volumes for glaucoma classification.
• The network was trained exclusively on labeled OCT data, bypassing the need for structure segmentation.
• Performance was evaluated using the area under the ROC curve (AUC) on a dataset of 50 volumes.

Main Results:

• The proposed Capsule network achieved an AUC of 0.97 for glaucoma classification.
• This performance significantly surpasses traditional machine learning and segmentation-based OCT analysis methods.
• The network demonstrated superior performance compared to 3D Convolutional Neural Networks with fewer parameters and training epochs.

Conclusions:

• Deep learning Capsule networks offer a highly effective and efficient approach for glaucoma classification using 3D OCT data.
• A segmentation-free deep learning strategy can achieve state-of-the-art diagnostic accuracy in glaucoma detection.
• This method presents a promising advancement for automated glaucoma diagnosis and management.