Deep Learning for the Diagnosis of Stage in Retinopathy of Prematurity: Accuracy and Generalizability across Populations and Cameras

Application of Artificial Intelligence in the Diagnosis, Follow-Up and Prediction of Treatment of Ophthalmic Diseases

PURPOSE

To describe the application of artificial intelligence (AI) in ophthalmic diseases and its possible future directions.

METHODS

A retrospective review of the literature from PubMed, Web of Science, and Embase databases (2019-2024).

RESULTS

AI assists in cataract diagnosis, classification, preoperative lens calculation, surgical risk, postoperative vision prediction, and follow-up. For glaucoma, AI enhances early diagnosis, progression prediction, and surgical risk assessment. It detects diabetic retinopathy early and predicts treatment effects for diabetic macular edema. AI analyzes fundus images for age-related macular degeneration (AMD) diagnosis and risk prediction. Additionally, AI quantifies and grades vitreous opacities in uveitis. For retinopathy of prematurity, AI facilitates disease classification, predicting disease occurrence and severity. Recently, AI also predicts systemic diseases by analyzing fundus vascular changes.

CONCLUSIONS

AI has been extensively used in diagnosing, following up, and predicting treatment outcomes for common blinding eye diseases. In addition, it also has a unique role in the prediction of systemic diseases.

Implementation of Artificial Intelligence in Retinopathy of Prematurity Care: Challenges and Opportunities

The diagnosis of retinopathy of prematurity (ROP) is primarily image-based and suitable for implementation of artificial intelligence (AI) systems. Increasing incidence of ROP, especially in low and middle-income countries, has also put tremendous stress on health care systems. Barriers to the implementation of AI include infrastructure, regulatory, legal, cost, sustainability, and scalability. This review describes currently available AI and imaging systems, how a stable telemedicine infrastructure is crucial to AI implementation, and how successful ROP programs have been run in both low and middle-income countries and high-income countries. More work is needed in terms of validating AI systems with different populations with various low-cost imaging devices that have recently been developed. A sustainable and cost-effective ROP screening program is crucial in the prevention of childhood blindness.

Unmasking biases and navigating pitfalls in the ophthalmic artificial intelligence lifecycle: A narrative review

Over the past 2 decades, exponential growth in data availability, computational power, and newly available modeling techniques has led to an expansion in interest, investment, and research in Artificial Intelligence (AI) applications. Ophthalmology is one of many fields that seek to benefit from AI given the advent of telemedicine screening programs and the use of ancillary imaging. However, before AI can be widely deployed, further work must be done to avoid the pitfalls within the AI lifecycle. This review article breaks down the AI lifecycle into seven steps-data collection; defining the model task; data preprocessing and labeling; model development; model evaluation and validation; deployment; and finally, post-deployment evaluation, monitoring, and system recalibration-and delves into the risks for harm at each step and strategies for mitigating them.

Performance of Automated Machine Learning in Predicting Outcomes of Pneumatic Retinopexy

Purpose

Automated machine learning (AutoML) has emerged as a novel tool for medical professionals lacking coding experience, enabling them to develop predictive models for treatment outcomes. This study evaluated the performance of AutoML tools in developing models predicting the success of pneumatic retinopexy (PR) in treatment of rhegmatogenous retinal detachment (RRD). These models were then compared with custom models created by machine learning (ML) experts.

Design

Retrospective multicenter study.

Participants

Five hundred and thirty nine consecutive patients with primary RRD that underwent PR by a vitreoretinal fellow at 6 training hospitals between 2002 and 2022.

Methods

We used 2 AutoML platforms: MATLAB Classification Learner and Google Cloud AutoML. Additional models were developed by computer scientists. We included patient demographics and baseline characteristics, including lens and macula status, RRD size, number and location of breaks, presence of vitreous hemorrhage and lattice degeneration, and physicians' experience. The dataset was split into a training (n = 483) and test set (n = 56). The training set, with a 2:1 success-to-failure ratio, was used to train the MATLAB models. Because Google Cloud AutoML requires a minimum of 1000 samples, the training set was tripled to create a new set with 1449 datapoints. Additionally, balanced datasets with a 1:1 success-to-failure ratio were created using Python.

Main Outcome Measures

Single-procedure anatomic success rate, as predicted by the ML models. F2 scores and area under the receiver operating curve (AUROC) were used as primary metrics to compare models.

Results

The best performing AutoML model (F2 score: 0.85; AUROC: 0.90; MATLAB), showed comparable performance to the custom model (0.92, 0.86) when trained on the balanced datasets. However, training the AutoML model with imbalanced data yielded misleadingly high AUROC (0.81) despite low F2-score (0.2) and sensitivity (0.17).

Conclusions

We demonstrated the feasibility of using AutoML as an accessible tool for medical professionals to develop models from clinical data. Such models can ultimately aid in the clinical decision-making, contributing to better patient outcomes. However, outcomes can be misleading or unreliable if used naively. Limitations exist, particularly if datasets contain missing variables or are highly imbalanced. Proper model selection and data preprocessing can improve the reliability of AutoML tools.

Financial Disclosures

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Assessing spectral effectiveness in color fundus photography for deep learning classification of retinopathy of prematurity

Significance

Retinopathy of prematurity (ROP) poses a significant global threat to childhood vision, necessitating effective screening strategies. This study addresses the impact of color channels in fundus imaging on ROP diagnosis, emphasizing the efficacy and safety of utilizing longer wavelengths, such as red or green for enhanced depth information and improved diagnostic capabilities.

Aim

This study aims to assess the spectral effectiveness in color fundus photography for the deep learning classification of ROP.

Approach

A convolutional neural network end-to-end classifier was utilized for deep learning classification of normal, stage 1, stage 2, and stage 3 ROP fundus images. The classification performances with individual-color-channel inputs, i.e., red, green, and blue, and multi-color-channel fusion architectures, including early-fusion, intermediate-fusion, and late-fusion, were quantitatively compared.

Results

For individual-color-channel inputs, similar performance was observed for green channel (88.00% accuracy, 76.00% sensitivity, and 92.00% specificity) and red channel (87.25% accuracy, 74.50% sensitivity, and 91.50% specificity), which is substantially outperforming the blue channel (78.25% accuracy, 56.50% sensitivity, and 85.50% specificity). For multi-color-channel fusion options, the early-fusion and intermediate-fusion architecture showed almost the same performance when compared to the green/red channel input, and they outperformed the late-fusion architecture.

Conclusions

This study reveals that the classification of ROP stages can be effectively achieved using either the green or red image alone. This finding enables the exclusion of blue images, acknowledged for their increased susceptibility to light toxicity.

Artificial Intelligence (AI) for Early Diagnosis of Retinal Diseases

Artificial intelligence (AI) has emerged as a transformative tool in the field of ophthalmology, revolutionizing disease diagnosis and management. This paper provides a comprehensive overview of AI applications in various retinal diseases, highlighting its potential to enhance screening efficiency, facilitate early diagnosis, and improve patient outcomes. Herein, we elucidate the fundamental concepts of AI, including machine learning (ML) and deep learning (DL), and their application in ophthalmology, underscoring the significance of AI-driven solutions in addressing the complexity and variability of retinal diseases. Furthermore, we delve into the specific applications of AI in retinal diseases such as diabetic retinopathy (DR), age-related macular degeneration (AMD), Macular Neovascularization, retinopathy of prematurity (ROP), retinal vein occlusion (RVO), hypertensive retinopathy (HR), Retinitis Pigmentosa, Stargardt disease, best vitelliform macular dystrophy, and sickle cell retinopathy. We focus on the current landscape of AI technologies, including various AI models, their performance metrics, and clinical implications. Furthermore, we aim to address challenges and pitfalls associated with the integration of AI in clinical practice, including the "black box phenomenon", biases in data representation, and limitations in comprehensive patient assessment. In conclusion, this review emphasizes the collaborative role of AI alongside healthcare professionals, advocating for a synergistic approach to healthcare delivery. It highlights the importance of leveraging AI to augment, rather than replace, human expertise, thereby maximizing its potential to revolutionize healthcare delivery, mitigate healthcare disparities, and improve patient outcomes in the evolving landscape of medicine.

Development of a deep learning model to distinguish the cause of optic disc atrophy using retinal fundus photography

The differential diagnosis for optic atrophy can be challenging and requires expensive, time-consuming ancillary testing to determine the cause. While Leber's hereditary optic neuropathy (LHON) and optic neuritis (ON) are both clinically significant causes for optic atrophy, both relatively rare in the general population, contributing to limitations in obtaining large imaging datasets. This study therefore aims to develop a deep learning (DL) model based on small datasets that could distinguish the cause of optic disc atrophy using only fundus photography. We retrospectively reviewed fundus photographs of 120 normal eyes, 30 eyes (15 patients) with genetically-confirmed LHON, and 30 eyes (26 patients) with ON. Images were split into a training dataset and a test dataset and used for model training with ResNet-18. To visualize the critical regions in retinal photographs that are highly associated with disease prediction, Gradient-Weighted Class Activation Map (Grad-CAM) was used to generate image-level attention heat maps and to enhance the interpretability of the DL system. In the 3-class classification of normal, LHON, and ON, the area under the receiver operating characteristic curve (AUROC) was 1.0 for normal, 0.988 for LHON, and 0.990 for ON, clearly differentiating each class from the others with an overall total accuracy of 0.93. Specifically, when distinguishing between normal and disease cases, the precision, recall, and F1 scores were perfect at 1.0. Furthermore, in the differentiation of LHON from other conditions, ON from others, and between LHON and ON, we consistently observed precision, recall, and F1 scores of 0.8. The model performance was maintained until only 10% of the pixel values of the image, identified as important by Grad-CAM, were preserved and the rest were masked, followed by retraining and evaluation.

Learning across diverse biomedical data modalities and cohorts: Challenges and opportunities for innovation

In healthcare, machine learning (ML) shows significant potential to augment patient care, improve population health, and streamline healthcare workflows. Realizing its full potential is, however, often hampered by concerns about data privacy, diversity in data sources, and suboptimal utilization of different data modalities. This review studies the utility of cross-cohort cross-category (C4) integration in such contexts: the process of combining information from diverse datasets distributed across distinct, secure sites. We argue that C4 approaches could pave the way for ML models that are both holistic and widely applicable. This paper provides a comprehensive overview of C4 in health care, including its present stage, potential opportunities, and associated challenges.

Artificial intelligence-based image analysis in clinical testing: lessons from cervical cancer screening

Novel screening and diagnostic tests based on artificial intelligence (AI) image recognition algorithms are proliferating. Some initial reports claim outstanding accuracy followed by disappointing lack of confirmation, including our own early work on cervical screening. This is a presentation of lessons learned, organized as a conceptual step-by-step approach to bridge the gap between the creation of an AI algorithm and clinical efficacy. The first fundamental principle is specifying rigorously what the algorithm is designed to identify and what the test is intended to measure (eg, screening, diagnostic, or prognostic). Second, designing the AI algorithm to minimize the most clinically important errors. For example, many equivocal cervical images cannot yet be labeled because the borderline between cases and controls is blurred. To avoid a misclassified case-control dichotomy, we have isolated the equivocal cases and formally included an intermediate, indeterminate class (severity order of classes: case>indeterminate>control). The third principle is evaluating AI algorithms like any other test, using clinical epidemiologic criteria. Repeatability of the algorithm at the borderline, for indeterminate images, has proven extremely informative. Distinguishing between internal and external validation is also essential. Linking the AI algorithm results to clinical risk estimation is the fourth principle. Absolute risk (not relative) is the critical metric for translating a test result into clinical use. Finally, generating risk-based guidelines for clinical use that match local resources and priorities is the last principle in our approach. We are particularly interested in applications to lower-resource settings to address health disparities. We note that similar principles apply to other domains of AI-based image analysis for medical diagnostic testing.

Applications of Deep Learning: Automated Assessment of Vascular Tortuosity in Mouse Models of Oxygen-Induced Retinopathy

Objective

To develop a generative adversarial network (GAN) to segment major blood vessels from retinal flat-mount images from oxygen-induced retinopathy (OIR) and demonstrate the utility of these GAN-generated vessel segmentations in quantifying vascular tortuosity.

Design

Development and validation of GAN.

Subjects

Three datasets containing 1084, 50, and 20 flat-mount mice retina images with various stains used and ages at sacrifice acquired from previously published manuscripts.

Methods

Four graders manually segmented major blood vessels from flat-mount images of retinas from OIR mice. Pix2Pix, a high-resolution GAN, was trained on 984 pairs of raw flat-mount images and manual vessel segmentations and then tested on 100 and 50 image pairs from a held-out and external test set, respectively. GAN-generated and manual vessel segmentations were then used as an input into a previously published algorithm (iROP-Assist) to generate a vascular cumulative tortuosity index (CTI) for 20 image pairs containing mouse eyes treated with aflibercept versus control.

Main Outcome Measures

Mean dice coefficients were used to compare segmentation accuracy between the GAN-generated and manually annotated segmentation maps. For the image pairs treated with aflibercept versus control, mean CTIs were also calculated for both GAN-generated and manual vessel maps. Statistical significance was evaluated using Wilcoxon signed-rank tests (P ≤ 0.05 threshold for significance).

Results

The dice coefficient for the GAN-generated versus manual vessel segmentations was 0.75 ± 0.27 and 0.77 ± 0.17 for the held-out test set and external test set, respectively. The mean CTI generated from the GAN-generated and manual vessel segmentations was 1.12 ± 0.07 versus 1.03 ± 0.02 (P = 0.003) and 1.06 ± 0.04 versus 1.01 ± 0.01 (P < 0.001), respectively, for eyes treated with aflibercept versus control, demonstrating that vascular tortuosity was rescued by aflibercept when quantified by GAN-generated and manual vessel segmentations.

Conclusions

GANs can be used to accurately generate vessel map segmentations from flat-mount images. These vessel maps may be used to evaluate novel metrics of vascular tortuosity in OIR, such as CTI, and have the potential to accelerate research in treatments for ischemic retinopathies.

Financial Disclosures

The author(s) have no proprietary or commercial interest in any materials discussed in this article.

Teleophthalmology and retina: a review of current tools, pathways and services

Telemedicine, the use of telecommunication and information technology to deliver healthcare remotely, has evolved beyond recognition since its inception in the 1970s. Advances in telecommunication infrastructure, the advent of the Internet, exponential growth in computing power and associated computer-aided diagnosis, and medical imaging developments have created an environment where telemedicine is more accessible and capable than ever before, particularly in the field of ophthalmology. Ever-increasing global demand for ophthalmic services due to population growth and ageing together with insufficient supply of ophthalmologists requires new models of healthcare provision integrating telemedicine to meet present day challenges, with the recent COVID-19 pandemic providing the catalyst for the widespread adoption and acceptance of teleophthalmology. In this review we discuss the history, present and future application of telemedicine within the field of ophthalmology, and specifically retinal disease. We consider the strengths and limitations of teleophthalmology, its role in screening, community and hospital management of retinal disease, patient and clinician attitudes, and barriers to its adoption.

Machine Learning Models for Predicting Long-Term Visual Acuity in Highly Myopic Eyes

Importance

High myopia is a global concern due to its escalating prevalence and the potential risk of severe visual impairment caused by pathologic myopia. Using artificial intelligence to estimate future visual acuity (VA) could help clinicians to identify and monitor patients with a high risk of vision reduction in advance.

Objective

To develop machine learning models to predict VA at 3 and 5 years in patients with high myopia.

Design, Setting, and Participants

This retrospective, single-center, cohort study was performed on patients whose best-corrected VA (BCVA) at 3 and 5 years was known. The ophthalmic examinations of these patients were performed between October 2011 and May 2021. Thirty-four variables, including general information, basic ophthalmic information, and categories of myopic maculopathy based on fundus and optical coherence tomography images, were collected from the medical records for analysis.

Main Outcomes and Measures

Regression models were developed to predict BCVA at 3 and 5 years, and a binary classification model was developed to predict the risk of developing visual impairment at 5 years. The performance of models was evaluated by discrimination metrics, calibration belts, and decision curve analysis. The importance of relative variables was assessed by explainable artificial intelligence techniques.

Results

A total of 1616 eyes from 967 patients (mean [SD] age, 58.5 [14.0] years; 678 female [70.1%]) were included in this analysis. Findings showed that support vector machines presented the best prediction of BCVA at 3 years (R2 = 0.682; 95% CI, 0.625-0.733) and random forest at 5 years (R2 = 0.660; 95% CI, 0.604-0.710). To predict the risk of visual impairment at 5 years, logistic regression presented the best performance (area under the receiver operating characteristic curve = 0.870; 95% CI, 0.816-0.912). The baseline BCVA (logMAR odds ratio [OR], 0.298; 95% CI, 0.235-0.378; P < .001), prior myopic macular neovascularization (OR, 3.290; 95% CI, 2.209-4.899; P < .001), age (OR, 1.578; 95% CI, 1.227-2.028; P < .001), and category 4 myopic maculopathy (OR, 4.899; 95% CI, 1.431-16.769; P = .01) were the 4 most important predicting variables and associated with increased risk of visual impairment at 5 years.

Conclusions and Relevance

Study results suggest that developing models for accurate prediction of the long-term VA for highly myopic eyes based on clinical and imaging information is feasible. Such models could be used for the clinical assessments of future visual acuity.

Characterization of Errors in Retinopathy of Prematurity Diagnosis by Ophthalmologists-in-Training in Middle-Income Countries

PURPOSE

To characterize common errors in the diagnosis of retinopathy of prematurity (ROP) among ophthalmologistsin-training in middle-income countries.

METHODS

In this prospective cohort study, 200 ophthalmologists-in-training from programs in Brazil, Mexico, and the Philippines participated. A secure web-based educational system was developed using a repository of more than 2,500 unique image sets of ROP, and a reference standard diagnosis was established by combining the clinical diagnosis and the image-based diagnosis by multiple experts. Twenty web-based cases of wide-field retinal images were presented, and ophthalmologists-in-training were asked to diagnose plus disease, zone, stage, and category for each eye. Trainees' responses were compared to the consensus reference standard diagnosis. Main outcome measures were frequency and types of diagnostic errors were analyzed.

RESULTS

The error rate in the diagnosis of any category of ROP was between 48% and 59% for all countries. The error rate in identifying type 2 or pre-plus disease was 77%, with a tendency for overdiagnosis (27% underdiagnosis vs 50% overdiagnosis; mean difference: 23.4; 95% CI: 12.1 to 34.7; P = .005). Misdiagnosis of treatment-requiring ROP as type 2 ROP was most commonly associated with incorrectly identifying plus disease (plus disease error rate = 18% with correct category diagnosis vs 69% when misdiagnosed; mean difference: 51.0; 95% CI: 49.3 to 52.7; P = .003).

CONCLUSIONS

Ophthalmologists-in-training from middle-income countries misdiagnosed ROP more than half of the time. Identification of plus disease was the salient factor leading to incorrect diagnosis. These findings emphasize the need for improved access to ROP education to improve competency in diagnosis among ophthalmologists-in-training in middle-income countries. [J Pediatr Ophthalmol Strabismus. 2023;60(5):344-352.].

Development and international validation of custom-engineered and code-free deep-learning models for detection of plus disease in retinopathy of prematurity: a retrospective study

BACKGROUND

Retinopathy of prematurity (ROP), a leading cause of childhood blindness, is diagnosed through interval screening by paediatric ophthalmologists. However, improved survival of premature neonates coupled with a scarcity of available experts has raised concerns about the sustainability of this approach. We aimed to develop bespoke and code-free deep learning-based classifiers for plus disease, a hallmark of ROP, in an ethnically diverse population in London, UK, and externally validate them in ethnically, geographically, and socioeconomically diverse populations in four countries and three continents. Code-free deep learning is not reliant on the availability of expertly trained data scientists, thus being of particular potential benefit for low resource health-care settings.

METHODS

This retrospective cohort study used retinal images from 1370 neonates admitted to a neonatal unit at Homerton University Hospital NHS Foundation Trust, London, UK, between 2008 and 2018. Images were acquired using a Retcam Version 2 device (Natus Medical, Pleasanton, CA, USA) on all babies who were either born at less than 32 weeks gestational age or had a birthweight of less than 1501 g. Each images was graded by two junior ophthalmologists with disagreements adjudicated by a senior paediatric ophthalmologist. Bespoke and code-free deep learning models (CFDL) were developed for the discrimination of healthy, pre-plus disease, and plus disease. Performance was assessed internally on 200 images with the majority vote of three senior paediatric ophthalmologists as the reference standard. External validation was on 338 retinal images from four separate datasets from the USA, Brazil, and Egypt with images derived from Retcam and the 3nethra neo device (Forus Health, Bengaluru, India).

FINDINGS

Of the 7414 retinal images in the original dataset, 6141 images were used in the final development dataset. For the discrimination of healthy versus pre-plus or plus disease, the bespoke model had an area under the curve (AUC) of 0·986 (95% CI 0·973-0·996) and the CFDL model had an AUC of 0·989 (0·979-0·997) on the internal test set. Both models generalised well to external validation test sets acquired using the Retcam for discriminating healthy from pre-plus or plus disease (bespoke range was 0·975-1·000 and CFDL range was 0·969-0·995). The CFDL model was inferior to the bespoke model on discriminating pre-plus disease from healthy or plus disease in the USA dataset (CFDL 0·808 [95% CI 0·671-0·909, bespoke 0·942 [0·892-0·982]], p=0·0070). Performance also reduced when tested on the 3nethra neo imaging device (CFDL 0·865 [0·742-0·965] and bespoke 0·891 [0·783-0·977]).

INTERPRETATION

Both bespoke and CFDL models conferred similar performance to senior paediatric ophthalmologists for discriminating healthy retinal images from ones with features of pre-plus or plus disease; however, CFDL models might generalise less well when considering minority classes. Care should be taken when testing on data acquired using alternative imaging devices from that used for the development dataset. Our study justifies further validation of plus disease classifiers in ROP screening and supports a potential role for code-free approaches to help prevent blindness in vulnerable neonates.

FUNDING

National Institute for Health Research Biomedical Research Centre based at Moorfields Eye Hospital NHS Foundation Trust and the University College London Institute of Ophthalmology.

TRANSLATIONS

For the Portuguese and Arabic translations of the abstract see Supplementary Materials section.

Artificial Intelligence in Ophthalmology - Status Quo and Future Perspectives

Artificial intelligence (AI) is an emerging technology in healthcare and holds the potential to disrupt many arms in medical care. In particular, disciplines using medical imaging modalities, including e.g. radiology but ophthalmology as well, are already confronted with a wide variety of AI implications. In ophthalmologic research, AI has demonstrated promising results limited to specific diseases and imaging tools, respectively. Yet, implementation of AI in clinical routine is not widely spread due to availability, heterogeneity in imaging techniques and AI methods. In order to describe the status quo, this narrational review provides a brief introduction to AI ("what the ophthalmologist needs to know"), followed by an overview of different AI-based applications in ophthalmology and a discussion on future challenges.Abbreviations: Age-related macular degeneration, AMD; Artificial intelligence, AI; Anterior segment OCT, AS-OCT; Coronary artery calcium score, CACS; Convolutional neural network, CNN; Deep convolutional neural network, DCNN; Diabetic retinopathy, DR; Machine learning, ML; Optical coherence tomography, OCT; Retinopathy of prematurity, ROP; Support vector machine, SVM; Thyroid-associated ophthalmopathy, TAO.

Research progress on diagnosing retinal vascular diseases based on artificial intelligence and fundus images

As the only blood vessels that can directly be seen in the whole body, pathological changes in retinal vessels are related to the metabolic state of the whole body and many systems, which seriously affect the vision and quality of life of patients. Timely diagnosis and treatment are key to improving vision prognosis. In recent years, with the rapid development of artificial intelligence, the application of artificial intelligence in ophthalmology has become increasingly extensive and in-depth, especially in the field of retinal vascular diseases. Research study results based on artificial intelligence and fundus images are remarkable and provides a great possibility for early diagnosis and treatment. This paper reviews the recent research progress on artificial intelligence in retinal vascular diseases (including diabetic retinopathy, hypertensive retinopathy, retinal vein occlusion, retinopathy of prematurity, and age-related macular degeneration). The limitations and challenges of the research process are also discussed.

Usability and Clinician Acceptance of a Deep Learning-Based Clinical Decision Support Tool for Predicting Glaucomatous Visual Field Progression

PRCIS

We updated a clinical decision support tool integrating predicted visual field (VF) metrics from an artificial intelligence model and assessed clinician perceptions of the predicted VF metric in this usability study.

PURPOSE

To evaluate clinician perceptions of a prototyped clinical decision support (CDS) tool that integrates visual field (VF) metric predictions from artificial intelligence (AI) models.

METHODS

Ten ophthalmologists and optometrists from the University of California San Diego participated in 6 cases from 6 patients, consisting of 11 eyes, uploaded to a CDS tool ("GLANCE", designed to help clinicians "at a glance"). For each case, clinicians answered questions about management recommendations and attitudes towards GLANCE, particularly regarding the utility and trustworthiness of the AI-predicted VF metrics and willingness to decrease VF testing frequency.

MAIN OUTCOMES AND MEASURES

Mean counts of management recommendations and mean Likert scale scores were calculated to assess overall management trends and attitudes towards the CDS tool for each case. In addition, system usability scale scores were calculated.

RESULTS

The mean Likert scores for trust in and utility of the predicted VF metric and clinician willingness to decrease VF testing frequency were 3.27, 3.42, and 2.64, respectively (1=strongly disagree, 5=strongly agree). When stratified by glaucoma severity, all mean Likert scores decreased as severity increased. The system usability scale score across all responders was 66.1±16.0 (43rd percentile).

CONCLUSIONS

A CDS tool can be designed to present AI model outputs in a useful, trustworthy manner that clinicians are generally willing to integrate into their clinical decision-making. Future work is needed to understand how to best develop explainable and trustworthy CDS tools integrating AI before clinical deployment.

Global Disparities in Retinopathy of Prematurity: A Literature Review

PURPOSE

To provide an overview of the impact of retinopathy of prematurity (ROP), and the challenges in the screening, diagnosis, and treatment of ROP worldwide.

METHODS

A comprehensive search was conducted using the PubMed database from January 2011 to October 2021 using the following keywords: retinopathy of prematurity, laser, and anti-vascular endothelial growth factor (VEGF). Data on patient characteristics, ROP treatment type, and recurrence rates were collected. The countries included in these studies were classified based on 2021-2022 World Bank definitions of high, upper-middle, lower-middle, and low-income groups. Moreover, a search for surgical outcomes for ROP and screening algorithms and artificial intelligence for ROP was conducted.

RESULTS

Thirty-nine studies met the inclusion criteria. ROP treatment and outcomes showed a trend towards intravitreal anti-VEGF injections as the initial treatment for ROP globally and the treatment of recurrent ROP in high-income countries. However, laser remains the treatment of choice for ROP recurrence in middle-income countries. Surgical outcomes for ROP stage 4A, 4B and 5 are similar worldwide. The incidence of ROP and ROP-related visual impairment continue to increase globally. Although telemedicine and artificial intelligence offer potential solutions to ROP screening in resource-limited areas, the current models require further optimization to reflect the global diversity of ROP patients.

CONCLUSION

ROP screening and treatment paradigms vary widely based on country income group due to disparities in resources, limited access to care, and lack of universal guidelines.

Tackling prediction uncertainty in machine learning for healthcare

Predictive machine-learning systems often do not convey the degree of confidence in the correctness of their outputs. To prevent unsafe prediction failures from machine-learning models, the users of the systems should be aware of the general accuracy of the model and understand the degree of confidence in each individual prediction. In this Perspective, we convey the need of prediction-uncertainty metrics in healthcare applications, with a focus on radiology. We outline the sources of prediction uncertainty, discuss how to implement prediction-uncertainty metrics in applications that require zero tolerance to errors and in applications that are error-tolerant, and provide a concise framework for understanding prediction uncertainty in healthcare contexts. For machine-learning-enabled automation to substantially impact healthcare, machine-learning models with zero tolerance for false-positive or false-negative errors must be developed intentionally.

Evaluation of an Artificial Intelligence System for Retinopathy of Prematurity Screening in Nepal and Mongolia

PURPOSE

To evaluate the performance of a deep learning (DL) algorithm for retinopathy of prematurity (ROP) screening in Nepal and Mongolia.

DESIGN

Retrospective analysis of prospectively collected clinical data.

PARTICIPANTS

Clinical information and fundus images were obtained from infants in 2 ROP screening programs in Nepal and Mongolia.

METHODS

Fundus images were obtained using the Forus 3nethra neo (Forus Health) in Nepal and the RetCam Portable (Natus Medical, Inc.) in Mongolia. The overall severity of ROP was determined from the medical record using the International Classification of ROP (ICROP). The presence of plus disease was determined independently in each image using a reference standard diagnosis. The Imaging and Informatics for ROP (i-ROP) DL algorithm was trained on images from the RetCam to classify plus disease and to assign a vascular severity score (VSS) from 1 through 9.

MAIN OUTCOME MEASURES

Area under the receiver operating characteristic curve and area under the precision-recall curve for the presence of plus disease or type 1 ROP and association between VSS and ICROP disease category.

RESULTS

The prevalence of type 1 ROP was found to be higher in Mongolia (14.0%) than in Nepal (2.2%; P < 0.001) in these data sets. In Mongolia (RetCam images), the area under the receiver operating characteristic curve for examination-level plus disease detection was 0.968, and the area under the precision-recall curve was 0.823. In Nepal (Forus images), these values were 0.999 and 0.993, respectively. The ROP VSS was associated with ICROP classification in both datasets (P < 0.001). At the population level, the median VSS was found to be higher in Mongolia (2.7; interquartile range [IQR], 1.3-5.4]) as compared with Nepal (1.9; IQR, 1.2-3.4; P < 0.001).

CONCLUSIONS

These data provide preliminary evidence of the effectiveness of the i-ROP DL algorithm for ROP screening in neonatal populations in Nepal and Mongolia using multiple camera systems and are useful for consideration in future clinical implementation of artificial intelligence-based ROP screening in low- and middle-income countries.

Applications of natural language processing in ophthalmology: present and future

Advances in technology, including novel ophthalmic imaging devices and adoption of the electronic health record (EHR), have resulted in significantly increased data available for both clinical use and research in ophthalmology. While artificial intelligence (AI) algorithms have the potential to utilize these data to transform clinical care, current applications of AI in ophthalmology have focused mostly on image-based deep learning. Unstructured free-text in the EHR represents a tremendous amount of underutilized data in big data analyses and predictive AI. Natural language processing (NLP) is a type of AI involved in processing human language that can be used to develop automated algorithms using these vast quantities of available text data. The purpose of this review was to introduce ophthalmologists to NLP by (1) reviewing current applications of NLP in ophthalmology and (2) exploring potential applications of NLP. We reviewed current literature published in Pubmed and Google Scholar for articles related to NLP and ophthalmology, and used ancestor search to expand our references. Overall, we found 19 published studies of NLP in ophthalmology. The majority of these publications (16) focused on extracting specific text such as visual acuity from free-text notes for the purposes of quantitative analysis. Other applications included: domain embedding, predictive modeling, and topic modeling. Future ophthalmic applications of NLP may also focus on developing search engines for data within free-text notes, cleaning notes, automated question-answering, and translating ophthalmology notes for other specialties or for patients, especially with a growing interest in open notes. As medicine becomes more data-oriented, NLP offers increasing opportunities to augment our ability to harness free-text data and drive innovations in healthcare delivery and treatment of ophthalmic conditions.

Performance of deep-learning artificial intelligence algorithms in detecting retinopathy of prematurity: A systematic review

PURPOSE

Artificial intelligence (AI) offers considerable promise for retinopathy of prematurity (ROP) screening and diagnosis. The development of deep-learning algorithms to detect the presence of disease may contribute to sufficient screening, early detection, and timely treatment for this preventable blinding disease. This review aimed to systematically examine the literature in AI algorithms in detecting ROP. Specifically, we focused on the performance of deep-learning algorithms through sensitivity, specificity, and area under the receiver operating curve (AUROC) for both the detection and grade of ROP.

METHODS

We searched Medline OVID, PubMed, Web of Science, and Embase for studies published from January 1, 2012, to September 20, 2021. Studies evaluating the diagnostic performance of deep-learning models based on retinal fundus images with expert ophthalmologists' judgment as reference standard were included. Studies which did not investigate the presence or absence of disease were excluded. Risk of bias was assessed using the QUADAS-2 tool.

RESULTS

Twelve studies out of the 175 studies identified were included. Five studies measured the performance of detecting the presence of ROP and seven studies determined the presence of plus disease. The average AUROC out of 11 studies was 0.98. The average sensitivity and specificity for detecting ROP was 95.72% and 98.15%, respectively, and for detecting plus disease was 91.13% and 95.92%, respectively.

CONCLUSION

The diagnostic performance of deep-learning algorithms in published studies was high. Few studies presented externally validated results or compared performance to expert human graders. Large scale prospective validation alongside robust study design could improve future studies.

Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence: Application to Retinopathy of Prematurity Diagnosis

Purpose

Developing robust artificial intelligence (AI) models for medical image analysis requires large quantities of diverse, well-chosen data that can prove challenging to collect because of privacy concerns, disease rarity, or diagnostic label quality. Collecting image-based datasets for retinopathy of prematurity (ROP), a potentially blinding disease, suffers from these challenges. Progressively growing generative adversarial networks (PGANs) may help, because they can synthesize highly realistic images that may increase both the size and diversity of medical datasets.

Design

Diagnostic validation study of convolutional neural networks (CNNs) for plus disease detection, a component of severe ROP, using synthetic data.

Participants

Five thousand eight hundred forty-two retinal fundus images (RFIs) collected from 963 preterm infants.

Methods

Retinal vessel maps (RVMs) were segmented from RFIs. PGANs were trained to synthesize RVMs with normal, pre-plus, or plus disease vasculature. Convolutional neural networks were trained, using real or synthetic RVMs, to detect plus disease from 2 real RVM test datasets.

Main Outcome Measures

Features of real and synthetic RVMs were evaluated using uniform manifold approximation and projection (UMAP). Similarities were evaluated at the dataset and feature level using Fréchet inception distance and Euclidean distance, respectively. CNN performance was assessed via area under the receiver operating characteristic curve (AUC); AUCs were compared via bootstrapping and Delong's test for correlated receiver operating characteristic curves. Confusion matrices were compared using McNemar's chi-square test and Cohen's κ value.

Results

The CNN trained on synthetic RVMs showed a significantly higher AUC (0.971; P = 0.006 and P = 0.004) and classified plus disease more similarly to a set of 8 international experts (κ = 0.922) than the CNN trained on real RVMs (AUC = 0.934; κ = 0.701). Real and synthetic RVMs overlapped, by plus disease diagnosis, on the UMAP manifold, showing that synthetic images spanned the disease severity spectrum. Fréchet inception distance and Euclidean distances suggested that real and synthetic RVMs were more dissimilar to one another than real RVMs were to one another, further suggesting that synthetic RVMs were distinct from the training data with respect to privacy considerations.

Conclusions

Synthetic datasets may be useful for training robust medical AI models. Furthermore, PGANs may be able to synthesize realistic data for use without protected health information concerns.

Federated learning for multi-center collaboration in ophthalmology: improving classification performance in retinopathy of prematurity

OBJECTIVE

To compare the performance of deep learning (DL) classifiers for the diagnosis of plus disease in retinopathy of prematurity (ROP) trained using two methods of developing models on multi-institutional datasets: centralizing data versus federated learning (FL) where no data leaves each institution.

DESIGN

Evaluation of a diagnostic test or technology.

SUBJECTS, PARTICIPANTS, AND/OR CONTROLS

DL models were trained, validated, and tested on 5,255 wide-angle retinal images in the neonatal intensive care units of 7 institutions as part of the Imaging and Informatics in ROP (i-ROP) study. All images were labeled for the presence of plus, pre-plus, or no plus disease with a clinical label, and a reference standard diagnosis (RSD) determined by three image-based ROP graders and the clinical diagnosis.

METHODS, INTERVENTION OR TESTING

We compared the area under the receiver operating characteristic curve (AUROC) for models developed on multi-institutional data, using a central approach, then FL, and compared locally trained models to either approach. We compared model performance (kappa) with label agreement (between clinical and RSD), dataset size and number of plus disease cases in each training cohort using Spearman's correlation coefficient (CC).

MAIN OUTCOME MEASURES

Model performance using AUROC and linearly-weighted kappa.

RESULTS

Four settings of experiment: FL trained on RSD against central trained on RSD, FL trained on clinical labels against central trained on clinical labels, FL trained on RSD against central trained on clinical labels, and FL trained on clinical labels against central trained on RSD (p=0.046, p=0.126, p=0.224, p=0.0173, respectively). 4/7 (57%) of models trained on local institutional data performed inferiorly to the FL models. Model performance for local models was positively correlated with label agreement (between clinical and RSD labels, CC = 0.389, p=0.387), total number of plus cases (CC=0.759, p=0.047), overall training set size (CC=0.924, p=0.002).

CONCLUSIONS

We show that a FL model trained performs comparably to a centralized model, confirming that FL may provide an effective, more feasible solution for inter-institutional learning. Smaller institutions benefit more from collaboration than larger institutions, showing the potential of FL for addressing disparities in resource access.

A Hybrid Deep Learning Construct for Detecting Keratoconus From Corneal Maps

Purpose

To develop and assess the accuracy of a hybrid deep learning construct for detecting keratoconus (KCN) based on corneal topographic maps.

Methods

We collected 3794 corneal images from 542 eyes of 280 subjects and developed seven deep learning models based on anterior and posterior eccentricity, anterior and posterior elevation, anterior and posterior sagittal curvature, and corneal thickness maps to extract deep corneal features. An independent subset with 1050 images collected from 150 eyes of 85 subjects from a separate center was used to validate models. We developed a hybrid deep learning model to detect KCN. We visualized deep features of corneal parameters to assess the quality of learning subjectively and computed area under the receiver operating characteristic curve (AUC), confusion matrices, accuracy, and F1 score to evaluate models objectively.

Results

In the development dataset, 204 eyes were normal, 123 eyes were suspected KCN, and 215 eyes had KCN. In the independent validation dataset, 50 eyes were normal, 50 eyes were suspected KCN, and 50 eyes were KCN. Images were annotated by three corneal specialists. The AUC of the models for the two-class and three-class problems based on the development set were 0.99 and 0.93, respectively.

Conclusions

The hybrid deep learning model achieved high accuracy in identifying KCN based on corneal maps and provided a time-efficient framework with low computational complexity.

Translational Relevance

Deep learning can detect KCN from non-invasive corneal images with high accuracy, suggesting potential application in research and clinical practice to identify KCN.

Deepfakes in Ophthalmology

Purpose

Generative adversarial networks (GANs) are deep learning (DL) models that can create and modify realistic-appearing synthetic images, or deepfakes, from real images. The purpose of our study was to evaluate the ability of experts to discern synthesized retinal fundus images from real fundus images and to review the current uses and limitations of GANs in ophthalmology.

Design

Development and expert evaluation of a GAN and an informal review of the literature.

Participants

A total of 4282 image pairs of fundus images and retinal vessel maps acquired from a multicenter ROP screening program.

Methods

Pix2Pix HD, a high-resolution GAN, was first trained and validated on fundus and vessel map image pairs and subsequently used to generate 880 images from a held-out test set. Fifty synthetic images from this test set and 50 different real images were presented to 4 expert ROP ophthalmologists using a custom online system for evaluation of whether the images were real or synthetic. Literature was reviewed on PubMed and Google Scholars using combinations of the terms ophthalmology, GANs, generative adversarial networks, ophthalmology, images, deepfakes, and synthetic. Ancestor search was performed to broaden results.

Main Outcome Measures

Expert ability to discern real versus synthetic images was evaluated using percent accuracy. Statistical significance was evaluated using a Fisher exact test, with P values ≤ 0.05 thresholded for significance.

Results

The expert majority correctly identified 59% of images as being real or synthetic (P = 0.1). Experts 1 to 4 correctly identified 54%, 58%, 49%, and 61% of images (P = 0.505, 0.158, 1.000, and 0.043, respectively). These results suggest that the majority of experts could not discern between real and synthetic images. Additionally, we identified 20 implementations of GANs in the ophthalmology literature, with applications in a variety of imaging modalities and ophthalmic diseases.

Conclusions

Generative adversarial networks can create synthetic fundus images that are indiscernible from real fundus images by expert ROP ophthalmologists. Synthetic images may improve dataset augmentation for DL, may be used in trainee education, and may have implications for patient privacy.