AI Telemedicine-Based Screening Tool may successfully Identify Glaucoma Suspects: Study
Glaucoma is a group of diseases that damage the eye's optic nerve and result in vision loss and blindness. Glaucoma, with age-related macular degeneration (AMD) and diabetic retinopathy (DR), is one of the three leading causes of blindness in developed countries and is now the second leading cause of blindness globally, after cataracts.
Glaucoma is characterized by loss of retinal ganglion cells (RGCs), which results in visual field impairment and structural changes to the retinal nerve fiber layer (RNFL) and optic disc. Most of the time, when detected, it is already late, i.e., with irreversible visual field loss. Therefore, it is essential to identify individuals at the early stages of this disease for treatment.
The relationship between estimated RGC counts and CDR suggests that assessment of change in CDR is a sensitive method for the evaluation of progressive neural losses in glaucoma; specifically, the retinal cup-disc ratio (CDR) is highly correlated with glaucoma.
The CDR can be an effective tool to identify the glaucoma suspect, and the focus is mainly to identify the glaucoma-suspect individuals (as a screening process from primary-care settings), who can be further tested to determine glaucoma and its progression.
A larger or abnormal CDR is mentioned and categorized as CDR > 0.5. Small changes in CDR may be associated with significant losses of RGCs, especially in eyes with large CDRs. Enlarged CDR is one indicator of the risk of glaucoma.
Alauddin Bhuiyan and team utilized the quantified vertical CDR when other research schemes used the qualitative assessment (e.g., small, medium, and large). They developed and validated vertical CDR quantification software to perform this quantified grading. The software demonstrates high repeatability and reliability.
- (i)The paper describes a method for glaucoma-suspect screening that utilizes a cloud-based system and incorporates telemedicine facilities. Thus, the screening will be available in remote clinics and primary-care settings.
- (ii)The paper describes results on a novel automated method that addresses the early screening of glaucoma suspects, which is a major public health concern.
- (iii)Therefore, an accurate and efficient screening in remote primary-care settings can provide a mass screening of the population currently dropping from yearly visits to the ophthalmologist.
1546 disc-centered fundus images were selected, including all 457 images from the Retinal Image Database for Optic Nerve Evaluation dataset, and images were randomly selected from the Age-Related Eye Disease Study and Singapore Malay Eye Study to develop the system.
First, a proprietary semiautomated software was used by an expert grader to quantify vertical CDR. Then, using CDR below 0.5 (nonsuspect) and CDR above 0.5 (glaucoma suspect), deep-learning architectures were used to train and test a binary classifier system. The binary classifier, with glaucoma suspect as positive, was measured using sensitivity, specificity, accuracy, and AUC.
The CDR carries three advantages:
- It is a single variable that is known to be strongly correlated with the disease, in particular with losses of RGCs
- Measurement of CDR can be accomplished from a single retinal color photograph obtained by an automated, non mydriatic camera in a primary-care office and forwarded on a telemedicine platform for expert interpretation with semiautomated methods
- Expert interpretation, which is still time consuming and expensive for humans, can be replaced by AI for efficiently and accurately evaluating the images
In the future, the authors proposed to use the Software Tool 'iPredict-glaucoma' at the iPredict platform. The AI-based telemedicine platform iPredict developed by iHealthscreen Inc. integrates the server-side programs and local remote computer/ mobile devices (for collecting and uploading patient data and images). The images are first checked for gradability automatically by an artificial-intelligence-based system developed in house from 3000 fundus images manually graded for gradability, and the system achieved over 99% accuracy. The server analyzes the images, and a report will be sent to the remote clinic with an individual's screening results and further recommendations.
The two-class glaucoma model (CDR ≤ 0.5 and above 0.5) achieved an accuracy of 89.67% with a sensitivity of 83.33% and a specificity of 93.89%. The AUC for the same data was 0.93. On the external validation dataset, the two-class model achieved a sensitivity of 80.11% and specificity of 84.96%. The AUC for the same data was 0.85.
The cloud-based and HIPAA-compliant telemedicine platform 'iPredict' has been validated for image and data transfer accuracy. The authors had transferred and analyzed nearly 850 images for AMD screening and DR screening from 4 primary-care clinics in Queens and Manhattan, New York, USA. They found a 100% correlation between the results obtained from directly evaluated images and the images transferred and processed by iPredict. They also tested 100 images for vertical CDR computation and received the same accuracy.
In this study, the authors demonstrated an accurate and fully automated deep-learning screening system for glaucoma suspects through retinal photography that may be effective for the identification of glaucoma suspects in primary-care settings.
They showed on several large datasets that the cup/ disc ratio (CDR) can be measured automatically from retinal photography with sufficient accuracy to discriminate suspects from nonsuspects and, thus, potentially facilitate referral of suspects to an ophthalmologist for specialized care.
Thus, a future, achievable goal is an AI telemedicine platform in which current methodology will be deployed in primary-care settings through remote image capture. A prospective trial will be needed to determine the feasibility of the system in clinical settings, with inexpensive, automated nonmydriatic retinal cameras and a telemedicine platform for image transfer to the deep-learning screening system.
Source: Alauddin Bhuiyan, Arun Govindaiah and R. Theodore Smith; Hindawi Journal of Ophthalmology