Machine Learning and Deep Learning Techniques for Optic Disc and Cup Segmentation - A Review.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is the second leading cause of global blindness.
• Early detection is vital to prevent severe complications and vision loss.
• Computer-aided diagnosis systems offer a solution to glaucoma screening shortages.

Purpose of the Study:

• To overview publicly available fundus image datasets for training and testing machine learning models.
• To critically review existing machine learning and deep learning methods for optic cup and disc segmentation.

Main Methods:

• Systematic search of public databases (PubMed, Google Scholar, etc.).
• Identification and review of studies on fundus image datasets and segmentation methods.
• Analysis of deep learning architectures for optic disc and optic cup segmentation.

Main Results:

• Eight public fundus image datasets (15,445 images) labeled for glaucoma and annotated for optic disc/cup boundaries were identified.
• Five evaluation metrics for model performance were established.
• Three common deep learning architectures for optic disc and optic cup segmentation were identified.

Conclusions:

• Deep learning shows promise for clinical application in optic nerve head segmentation.
• Further research is needed to address challenges before clinical trials.
• U-net and its variants are widely adopted deep learning architectures for this task.