Validation and Implementation of an Artificial Intelligence Machine-to-Machine (M2M) Model for Glaucoma Screening - Revised PROJECT SUMMARY Glaucoma is a leading cause of irreversible blindness in the United States. Many individuals face barriers in accessing timely and effective diagnostic services, which often leads to late-stage presentation and a substantially increased risk of blindness. This underscores the need for the development of affordable and accessible screening approaches for glaucoma. Although imaging technologies such as spectral domain-optical coherence tomography (SDOCT) can provide highly reproducible and accurate quantitative assessment of glaucomatous damage, their use in widespread screening is limited by high cost and operational complexity. Fundus photography is a low-cost alternative that has been used successfully in teleophthalmology. However, subjective grading of fundus photos for glaucoma is poorly reproducible and frequently inaccurate. We propose a new paradigm for assessing glaucomatous damage by training a deep learning (DL) convolutional neural network to provide objective quantitative estimates of neural damage from fundus photographs. In our Machine-to-Machine (M2M) approach, a DL network analyzes fundus photos to predict quantitative measurements of glaucomatous damage provided by SDOCT, such as retinal nerve fiber layer (RNFL) thickness. Our preliminary results showed that the M2M predictions have very high agreement and correlation with original SDOCT observations. Most importantly, the M2M was shown to detect glaucoma, predict damage in suspects, and track longitudinal change with performance comparable to SDOCT. In this proposal, we aim to refine and validate the M2M model for opportunistic, population-based and community-based glaucoma detection. In Aim 1, we will refine the M2M model using large clinic- datasets, applying innovative frameworks such as vision transformers and generative AI to improve accuracy and robustness for screening applications. In Aim 2, we will implement the M2M model for opportunistic glaucoma screening at the Bascom Palmer Eye Institute (BPEI) eye-specific emergency room (ER), the highest volume eye-specific ER in the nation. This ER serves a large adult population that may not otherwise undergo routine screening. We hypothesize that the M2M model will be feasible to implement and will show high acceptance and diagnostic accuracy. In Aim 3, we will apply the M2M model for detecting glaucoma in population-based datasets and also in a community-based screening program that reaches individuals across several counties in South Florida. We hypothesize that the model will demonstrate high accuracy for glaucoma detection and will outperform clinician-based assessment. This proposal harnesses the power of AI to bridge current gaps in glaucoma care. If successful, it will validate a scalable tool that can be seamlessly integrated into various healthcare settings to support early and accurate glaucoma diagnosis.
