Improving Glaucoma Diagnosis Assembling Deep Networks and Voting Schemes.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma damages the optic nerve, potentially causing vision loss, often due to elevated eye pressure.
• Artificial intelligence (AI) and data science, particularly deep learning, are increasingly used in medical imaging.
• Deep learning models have shown human-level or superior performance in various medical applications.

Purpose of the Study:

• To develop and evaluate a novel soft ensemble model for improved glaucoma diagnosis.
• To combine diverse deep learning models for enhanced diagnostic accuracy.
• To leverage K-Nearest Neighbors (KNN) algorithm with deep learning for robust classification.

Main Methods:

• An ensemble model was created using three diverse deep learning architectures: Convolutional Neural Networks (CNN), CapsNets, and Convolutional Autoencoders.
• The latent spaces from these models were combined.
• The K-Nearest Neighbors (KNN) algorithm was applied to the combined latent space, using true sample labels for classification.

Main Results:

• The proposed ensemble model demonstrated improved diagnostic capabilities compared to individual deep learning models.
• The AI model achieved superior results on two distinct retinal image datasets.
• The ensemble approach enhanced the state-of-the-art performance in glaucoma diagnosis.

Conclusions:

• The novel soft ensemble AI model offers a promising approach for accurate glaucoma diagnosis.
• Combining diverse deep learning models with KNN can enhance diagnostic performance in medical imaging.
• This AI strategy shows potential for improving patient outcomes through earlier and more accurate detection of eye conditions.