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

Abstract

Background Globally, glaucoma is the second leading cause of blindness. Detecting glaucoma in the early stages is essential to avoid disease complications, which lead to blindness. Thus, computer-aided diagnosis systems are powerful tools to overcome the shortage of glaucoma screening programs. Methods A systematic search of public databases, including PubMed, Google Scholar, and other sources, was performed to identify relevant studies to overview the publicly available fundus image datasets used to train, validate, and test machine learning and deep learning methods. Additionally, existing machine learning and deep learning methods for optic cup and disc segmentation were surveyed and critically reviewed. Results Eight fundus images datasets were publicly available with 15,445 images labeled with glaucoma or non-glaucoma, and manually annotated optic disc and cup boundaries were found. Five metrics were identified for evaluating the developed models. Finally, three main deep learning architectural designs were commonly used for optic disc and optic cup segmentation. Conclusion We provided future research directions to formulate robust optic cup and disc segmentation systems. Deep learning can be utilized in clinical settings for this task. However, many challenges need to be addressed before using this strategy in clinical trials. Finally, two deep learning architectural designs have been widely adopted, such as U-net and its variants.

Assessing coarse-to-fine deep learning models for optic disc and cup segmentation in fundus images

“…We used the same U-Net architecture as backbone for all models due to its popularity for OD/OC segmentation. [7][8][9]18 Its encoder consisted of 4 convolutional blocks with increasing output depths of 64, 128, 256 and 512, each followed by 2 × 2 max-pooling layers for downsampling. These blocks comprised two consecutive convolutional layers with 3 × 3 filters, each followed by a batch normalization layer and a ReLU activation.…”

Artificial intelligence for glaucoma: state of the art and future perspectives

“…Research on machine-learning (ML) and deep-learning (DL) models/algorithms (both terms are used henceforth interchangeably) for glaucoma detection has primarily focused on specific imaging techniques individually or multimodal imaging, using state-of-the-art convolutional neural network (CNN) models as the most common DL architecture [2 ▪ ,3]. Most studies on glaucoma detection have relied on color fundus photography (CFP) due to its accessibility and low cost [2 ▪ ,3–6]. Optical Coherence Tomography (OCT)-based DL models using conventional 2D B scans, 3D volumetric scans or OCT-angiography have yielded excellent accuracy [3–5,7 ▪ ].…”

“…Data quality: check for its consistency, completeness, accuracy and timeliness; Standardizing image quality criteria (such as with OCT quality index thresholds) or creating image processing pipelines may be required; in an effort to improve image segmentation, recent research has focused on strategies with dedicated DL networks, in an effort to bypass the need for manual segmentation by experts (‘ground truth’) [6,43–45];…”

Report of the International Conference on Generalized Functions (Debrecen, Hungary, 1984)

“…Numerous further lesions and anatomical structures in retinal images, such as exudates [50] and the optic disk [51], are targeted by segmentation and detection algorithms. The possibility of evaluating their performance in the FoV region or using all pixels in the images is present in all related problems.…”