Machine learning applied to retinal image processing for glaucoma detection: review and perspective.

Area of Science:

• Ophthalmology
• Computer Science
• Medical Imaging

Background:

• Machine learning (ML) is crucial for developing computer-aided diagnostic tools.
• Retinal image processing for glaucoma diagnosis is an active area of research.
• ML algorithms are increasingly important in ophthalmology for disease detection.

Purpose of the Study:

• To systematically review machine learning algorithms for glaucoma diagnosis and detection in retinal images.
• To analyze the different ML architectures used in retinal image processing.
• To identify the challenges and advancements in automated glaucoma detection systems.

Main Methods:

• Systematic review of publications from Scopus, PubMed, IEEEXplore, and Science Direct (2014-2019).
• Inclusion of studies employing classification processes for glaucoma detection.
• Exclusion of research utilizing the segmented optic disc method.

Main Results:

• Studies explored feature extraction, dimensionality reduction, and deep convolutional networks for retinal image analysis.
• Key differences in ML architectures include data requirements and computational costs.
• Deep learning techniques show promise but necessitate large datasets and high computational resources.

Conclusions:

• Automated glaucoma diagnosis systems are feasible using ML algorithms.
• Deep learning, despite its demands, is a promising technology for fundus imaging.
• Techniques like data augmentation and transfer learning help optimize deep learning models for glaucoma detection.