Retinal Glaucoma Public Datasets: What Do We Have and What Is Missing?

Abstract

Public databases for glaucoma studies contain color images of the retina, emphasizing the optic papilla. These databases are intended for research and standardized automated methodologies such as those using deep learning techniques. These techniques are used to solve complex problems in medical imaging, particularly in the automated screening of glaucomatous disease. The development of deep learning techniques has demonstrated potential for implementing protocols for large-scale glaucoma screening in the population, eliminating possible diagnostic doubts among specialists, and benefiting early treatment to delay the onset of blindness. However, the images are obtained by different cameras, in distinct locations, and from various population groups and are centered on multiple parts of the retina. We can also cite the small number of data, the lack of segmentation of the optic papillae, and the excavation. This work is intended to offer contributions to the structure and presentation of public databases used in the automated screening of glaucomatous papillae, adding relevant information from a medical point of view. The gold standard public databases present images with segmentations of the disc and cupping made by experts and division between training and test groups, serving as a reference for use in deep learning architectures. However, the data offered are not interchangeable. The quality and presentation of images are heterogeneous. Moreover, the databases use different criteria for binary classification with and without glaucoma, do not offer simultaneous pictures of the two eyes, and do not contain elements for early diagnosis.

Clinical Perspectives on the Use of Computer Vision in Glaucoma Screening

“…The CDRs computed using the segmented masks were very close to the ground truth (GT) masks classified by experts, reinforcing that CNNs can make an assessment like that carried out by a clinician. However, in most works, the validation of DL algorithms was compared to human classification results as a reference standard, occurring in the same way with public databases [ 47 ], an approach that can present severe limitations, as they tend to exaggerate or underestimate the probability of glaucoma given the significant variability of the optic disc region, the low reproducibility and inter-examiner agreement, the different experiences of the evaluators, and the lack of references to screen for the glaucomatous papilla [ 4 ].…”

“…The diagnostic accuracy of the model using color photographs suggests that deep learning (DL)-based architectures have the potential to standardize and automate the classification of chronic open-angle glaucoma. However, further studies are needed to implement these technologies [ 36 , 37 ] and bring them closer to reality by analyzing a set of data less dependent on human interpretation, including retinal photographs and clinical markers, symptoms, ocular pressure, family and personal history [ 47 ], and complementary exam results such as VF and OCT.…”

On the Analyses of Medical Images Using Traditional Machine Learning Techniques and Convolutional Neural Networks

“…The top-performing groups to date have all employed CNNs in the 2022 and 2021 challenges such as Diabetic Retinopathy Analysis Challenge (DRAC-2022) sponsored by MICCAI 2022, RAVIR: A Dataset and Methodology for the Semantic Segmentation and Quantitative Analysis of Retinal Arteries and Veins in Infrared Reflectance Imaging 2022 [ 269 ], AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge sponsored by IEEE ISBI 2022 [ 270 , 271 ].…”

The application of artificial intelligence in glaucoma diagnosis and prediction

“…Firstly, the dataset is homogeneous in terms of features. Because the datasets being used are publicly available ( Camara et al, 2022 ), the classification of images lacks specificity that is often consistent with homogeneous diseases or populations, making it difficult to ensure the fairness of the validated results. Secondly, the utilization of clinical data is very low.…”