Interpreting Deep Learning Studies in Glaucoma: Unresolved Challenges

Advancements in Glaucoma Diagnosis: The Role of AI in Medical Imaging

The progress of artificial intelligence algorithms in digital image processing and automatic diagnosis studies of the eye disease glaucoma has been growing and presenting essential advances to guarantee better clinical care for the population. Given the context, this article describes the main types of glaucoma, traditional forms of diagnosis, and presents the global epidemiology of the disease. Furthermore, it explores how studies using artificial intelligence algorithms have been investigated as possible tools to aid in the early diagnosis of this pathology through population screening. Therefore, the related work section presents the main studies and methodologies used in the automatic classification of glaucoma from digital fundus images and artificial intelligence algorithms, as well as the main databases containing images labeled for glaucoma and publicly available for the training of machine learning algorithms.

Optic Nerve Head Abnormalities in Nonpathologic High Myopia and the Relationship With Visual Field

PURPOSE

To describe the optic nerve head (ONH) abnormalities in nonpathologic highly myopic eyes based on swept-source optical coherence tomography (OCT) and the relationship with visual field (VF).

DESIGN

Secondary analysis from a longitudinal cohort study.

METHODS

Highly myopic patients without myopic maculopathy of category 2 or higher were enrolled. All participants underwent a swept-source OCT examination focused on ONH. We differentiated between 3 major types (optic disc morphologic abnormality, papillary/peripapillary tissue defect, and papillary/peripapillary schisis) and 12 subtypes of ONH abnormalities. The prevalence and characteristics of ONH abnormalities and the relationship with VF were analyzed.

RESULTS

A total of 857 participants (1389 eyes) were included. Among the 1389 eyes, 91.86%, 68.61%, and 34.92% of them had at least 1, 2, or 3 ONH abnormalities, respectively, which corresponded to 29.55%, 31.79%, and 35.67% of VF defects, respectively. Among the 12 subtypes of the 3 major types, peripapillary hyperreflective ovoid mass-like structure, visible retrobulbar subarachnoid space, and prelaminar schisis were the most common, respectively. Perimetric defects corresponding to OCT abnormalities were more commonly found in eyes with peripapillary retinal detachment, peripapillary retinoschisis, and peripapillary hyperreflective ovoid mass-like structure. Glaucoma-like VF defects were more common in eyes with deep optic cups (28.17%) and with optic disc pit/pit-like change (18.92%).

CONCLUSIONS

We observed and clarified the ONH structural abnormalities in eyes with nonpathologic high myopia. These descriptions may be helpful to differentiate changes in pathologic high myopia or glaucoma.

Review of Visualization Approaches in Deep Learning Models of Glaucoma

Glaucoma is a major cause of irreversible blindness worldwide. As glaucoma often presents without symptoms, early detection and intervention are important in delaying progression. Deep learning (DL) has emerged as a rapidly advancing tool to help achieve these objectives. In this narrative review, data types and visualization approaches for presenting model predictions, including models based on tabular data, functional data, and/or structural data, are summarized, and the importance of data source diversity for improving the utility and generalizability of DL models is explored. Examples of innovative approaches to understanding predictions of artificial intelligence (AI) models and alignment with clinicians are provided. In addition, methods to enhance the interpretability of clinical features from tabular data used to train AI models are investigated. Examples of published DL models that include interfaces to facilitate end-user engagement and minimize cognitive and time burdens are highlighted. The stages of integrating AI models into existing clinical workflows are reviewed, and challenges are discussed. Reviewing these approaches may help inform the generation of user-friendly interfaces that are successfully integrated into clinical information systems. This review details key principles regarding visualization approaches in DL models of glaucoma. The articles reviewed here focused on usability, explainability, and promotion of clinician trust to encourage wider adoption for clinical use. These studies demonstrate important progress in addressing visualization and explainability issues required for successful real-world implementation of DL models in glaucoma.

Applications of Artificial Intelligence and Deep Learning in Glaucoma

Diagnosis and detection of progression of glaucoma remains challenging. Artificial intelligence-based tools have the potential to improve and standardize the assessment of glaucoma but development of these algorithms is difficult given the multimodal and variable nature of the diagnosis. Currently, most algorithms are focused on a single imaging modality, specifically screening and diagnosis based on fundus photos or optical coherence tomography images. Use of anterior segment optical coherence tomography and goniophotographs is limited. The majority of algorithms designed for disease progression prediction are based on visual fields. No studies in our literature search assessed the use of artificial intelligence for treatment response prediction and no studies conducted prospective testing of their algorithms. Additional challenges to the development of artificial intelligence-based tools include scarcity of data and a lack of consensus in diagnostic criteria. Although research in the use of artificial intelligence for glaucoma is promising, additional work is needed to develop clinically usable tools.

Citation trends in ophthalmology articles and keywords in mainland China, Hong Kong, and Taiwan since 2013 using temporal bar graphs (TBGs): Bibliometric analysis

BACKGROUND

We selected authors from mainland China, Hong Kong, and Taiwan (CHT) to examine citation trends on articles and keywords. The existence of suitable temporal bar graphs (TBGs) for displaying citation trends is unknown. It is necessary to enhance the traditional TBGs to provide readers with more information about the citation trend. The purpose of this study was to propose an advanced TBG that can be applied to understand the most worth-reading articles by ophthalmology authors in the CHT.

METHODS

Using the search engine of the Web of Science core collection, we conducted bibliometric analyses to examine the article citation trends of ophthalmology authors in CHT since 2013. A total of 6695 metadata was collected from articles and review articles. Using radar plots, the Y-index, and the combining the Y-index with the CJAL scores (CJAL) scores, we could determine the dominance of publications by year, region, institute, journal, department, and author. A choropleth map, a dot plot, and a 4-quadrant radar plot were used to visualize the results. A TBG was designed and provided for readers to display citation trends on articles and keywords.

RESULTS

We found that the majority of publications were published in 2017 (2275), Shanghai city (935), Sun Yat-Sen University (China) (689), the international journal Ophthalmology (1399), the Department of Ophthalmology (3035), and the author Peizeng Yang (Chongqing) (65); the highest CAJL scores were also from Guangdong (2767.22), Sun Yat-Sen University (China) (2147.35), and the Ophthalmology Department (7130.96); the author Peizeng Yang (Chongqing) (170.16) had the highest CAJL; and the enhanced TBG features maximum counts and recent growth trends that are not included in traditional TBGs.

CONCLUSION

Using the Y-index and the CJAL score compared with research achievements of ophthalmology authors in CHT, a 4-quadrant radar plot was provided. The enhanced TBGs and the CJAL scores are recommended for future bibliographical studies.

Daily Patterns of Accelerometer-Measured Movement Behaviors in Glaucoma Patients: Insights From UK Biobank Participants

PURPOSE

The purpose of this study is to compare daily patterns of accelerometer-measured movement behaviors between glaucoma patients and those without glaucoma.

METHODS

From 2013 to 2015, 106,053 UK Biobank participants took part in a 7-day accelerometer test. Based on established algorithms, continuous accelerometer data were classified into 4 movement behaviors: moderate to vigorous physical activity (MVPA), light physical activity, sedentary behaviors, and sleep. Glaucoma and other covariates were defined according to baseline assessments and inpatient diagnosis records. Negative binomial regression models were used to compare daily patterns of movement behaviors between glaucoma patients and those without glaucoma.

RESULTS

Accelerometer data from 1262 glaucoma patients and 81,551 participants without glaucoma were included. Compared with participants without glaucoma, glaucoma patients spent 4.7% less time on MVPA in multivariable models [mean=28.3 vs 31.4 min/d; incidence-rate ratio (IRR) 0.953, 95% confidence interval (CI): 0.910-0.998; P=0.044], which was mainly attributed to the decreased MVPA time during 18:00-23:59 (IRR=0.863, Bonferroni-corrected 95% CI: 0.768-0.970; P=0.002). Subgroup analyses indicated that compared with those with normal body mass index, the decreased MVPA time was more pronounced in participants with overweight and obesity (IRR=0.912, Bonferroni-corrected 95% CI: 0.851-0.978; P for interaction=0.007). No significant association was found between glaucoma and time spent on other movement behaviors including light physical activity, sedentary behaviors, and sleep.

CONCLUSIONS

Daily patterns of movement behaviors were significantly changed in glaucoma patients. Compared with those without glaucoma, glaucoma patients spent less time on MVPA, especially in the evening.

The Glaucoma Suspect Problem: Ways Forward

The diagnosis of glaucoma depends upon indentification of characteristic damage to the optic nerve and retinal fiber layer. In many cases, however, clinicians find it difficult to ascertain whether glaucomatous damage is present or absent. These patients are often labeled as "glaucoma suspects," which creates a subpopulation of individuals without clear-cut disease who nonetheless must remain under surveillance. Most will never go on to develop glaucoma, yet the need for ongoing monitoring burdens clinics and health care systems. In this perspective, we illustrate possible directions and novel approaches that can be used to remedy this situation by integrating current technologies into clinical practice. In particular, we suggest that optical coherence tomography be better utilized to methodologically classify these eyes into glaucomatous and healthy categories.

Minimally Invasive Glaucoma Surgery in Primary Angle-Closure Glaucoma

Primary angle-closure glaucoma (PACG) is responsible for half of the glaucoma-related blindness worldwide. Cataract surgery with or without trabeculectomy has been considered to be the first-line treatment in eyes with medically uncontrolled PACG. While minimally invasive glaucoma surgery has become an important surgical approach for primary open-angle glaucoma, its indications and benefits in PACG are less clear. This review summarizes the efficacy and safety profile of minimally invasive glaucoma surgery in PACG to unfold new insights into the surgical management of PACG.

Development and Clinical Validation of Semi-Supervised Generative Adversarial Networks for Detection of Retinal Disorders in Optical Coherence Tomography Images Using Small Dataset

PURPOSE

To develop and test semi-supervised generative adversarial networks (GANs) that detect retinal disorders on optical coherence tomography (OCT) images using a small-labeled dataset.

METHODS

From a public database, we randomly chose a small supervised dataset with 400 OCT images (100 choroidal neovascularization, 100 diabetic macular edema, 100 drusen, and 100 normal) and assigned all other OCT images to unsupervised dataset (107,912 images without labeling). We adopted a semi-supervised GAN and a supervised deep learning (DL) model for automatically detecting retinal disorders from OCT images. The performance of the 2 models was compared in 3 testing datasets with different OCT devices. The evaluation metrics included accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curves.

RESULTS

The local validation dataset included 1000 images with 250 from each category. The independent clinical dataset included 366 OCT images using Cirrus OCT Shanghai Shibei Hospital and 511 OCT images using RTVue OCT from Xinhua Hospital respectively. The semi-supervised GANs classifier achieved better accuracy than supervised DL model (0.91 vs 0.86 for local cell validation dataset, 0.91 vs 0.86 in the Shanghai Shibei Hospital testing dataset, and 0.93 vs 0.92 in Xinhua Hospital testing dataset). For detecting urgent referrals (choroidal neo-vascularization and diabetic macular edema) from nonurgent referrals (drusen and normal) on OCT images, the semi-supervised GANs classifier also achieved better area under the receiver operating characteristic curves than supervised DL model (0.99 vs 0.97, 0.97 vs 0.96, and 0.99 vs 0.99, respectively).

CONCLUSIONS

A semi-supervised GAN can achieve better performance than that of a supervised DL model when the labeled dataset is limited. The current study offers utility to various research and clinical studies using DL with relatively small datasets. Semi-supervised GANs can detect retinal disorders from OCT images using relatively small dataset.

Systematic bibliometric and visualized analysis of research hotspots and trends on the application of artificial intelligence in diabetic retinopathy

BACKGROUND

Artificial intelligence (AI), which has been used to diagnose diabetic retinopathy (DR), may impact future medical and ophthalmic practices. Therefore, this study explored AI's general applications and research frontiers in the detection and gradation of DR.

METHODS

Citation data were obtained from the Web of Science Core Collection database (WoSCC) to assess the application of AI in diagnosing DR in the literature published from January 1, 2012, to June 30, 2022. These data were processed by CiteSpace 6.1.R3 software.

RESULTS

Overall, 858 publications from 77 countries and regions were examined, with the United States considered the leading country in this domain. The largest cluster labeled "automated detection" was employed in the generating stage from 2007 to 2014. The burst keywords from 2020 to 2022 were artificial intelligence and transfer learning.

CONCLUSION

Initial research focused on the study of intelligent algorithms used to localize or recognize lesions on fundus images to assist in diagnosing DR. Presently, the focus of research has changed from upgrading the accuracy and efficiency of DR lesion detection and classification to research on DR diagnostic systems. However, further studies on DR and computer engineering are required.