Introducing Nicholas Riina, MS Focused on the Future: Growing Up in Optometry, Advancing Ophthalmology

Introducing Nicholas Riina, MS

Nicholas Riina, MS, had his eye on the prize early on, growing up in a family in which he was exposed to his mother’s optometry practice. As with many individuals, he was on a winding path, along which he learned about the eye and medicine. During his undergraduate years at Indiana University, his interests became sharply focused on neuroscience, artificial intelligence (AI), and robotics.

Riina is now an MD candidate at the Icahn School of Medicine at Mount Sinai, and a clinical research fellow, at Mount Sinai, under the supervision and mentorship of Alon Harris, MS, PhD, FARVO, Professor of Ophthalmology, Professor of Artificial Intelligence & Human Health, Interim Director of the Barry Family Center for Ophthalmic Artificial Intelligence & Human Health, Vice Chair of International Research and Academic Affairs and Director of Ophthalmic Vascular Diagnostic & Research Program at Mount Sinai Hospital.

Riina’s arrival at Mount Sinai did not happen in a straight line; he had some interesting detours along the way. He was initially in a premed program but shifted gears to concentrate on AI while working with a team in an autonomous driving lab where AI systems were created to drive robots through forests and lakes. This led to a master’s degree. “I was working with sensor data and training AI and applying some theories about AI to real-world data using similar principles that are now used in medicine. We had a way to investigate the environment using an intelligence system to learn based on what is measurable,” he said.

Acceptance into the Donald and Vera Blinken FlexMed Program at the Mount Sinai School of Medicine, an early admissions program, allowed him to pursue medicine and simultaneously use his AI and robotics knowledge. “I wanted to use these tools that I spent time learning and that held my fascination--imaging, use of AI to learn about health care, and noninvasive means of diagnosis,” he explained.

He discovered early that among the medical specialties, Ophthalmology was that special arena where he could combine his interests. However, he considered himself an iffy choice because of his robotics background. Harris, the interim director of the AI and Ophthalmology Department, thought otherwise.

After he got his feet wet in his first Ophthalmology research project, he realized first-hand that the environment would allow him to learn and use these skills, exactly as he desired. He also embraced the clinical side following the realization of how the specialty profoundly affected patients’ lives. “Ophthalmology checked all the boxes,” he commented.

In commenting on Riina’s acceptance into the program, Dr. Harris explained that while he has seen numerous excellent physician scientists over the years, he chose Nick because of his ability to address a hypothesis by introducing a technology developed during his robotics training to obtain another level of information for use with glaucoma patients. “Because of his work withoptical coherence angiography (OCTA) and neural networks,¹ we were able to see that diagnosing glaucoma can be accomplished by something that I had hypothesized for many years. Glaucoma, which has been diagnosed based on anatomic changes can now be recognized based on the retinal vasculature. Nick brought a very refreshing view and approach to addressing the same question in a different way,” he explained.

Research path

Imaging study. In his afore-mentioned first foray into an Ophthalmology project, the team evaluated different imaging devices, i.e., optical coherence tomography (OCT) to evaluate the structure and OCTA to evaluate the vasculature to determine what can be learned about glaucoma based on the vasculature in the same way that clinicians analyze the ocular anatomy. Riina used an algorithm that he had used previously to maneuver a robot to train the algorithm on blood vessel data. “The outcome was that the blood vessel data were just as good as the anatomical data,” he said. Riina was the lead author in the study published in Investigative Ophthalmology and Visual Science.¹

In this study, Riina and colleagues wanted to identify relevant ocular biomarkers to diagnose primary open-angle glaucoma (POAG). They trained multi-layer perceptrons (MLPs), which are neural network models, on a prospectively collected observational dataset that included 93 patients with confirmed glaucoma and 113 controls. The investigators explained that the base model used intraocular pressure (IOP), blood pressure (BP), heart rate, and visual field parameters to diagnose glaucoma. Other models then each were assigned additional factors including structural features such as optic nerve parameters, retinal nerve fiber layer (RNFL), ganglion cell complex (GCC), macular thickness; choroidal thickness; and the thickness of the RNFL + GCC only seen on OCT; and vascular features observed using OCTA.

The investigators reported that “the vascular and structural models both had significantly higher accuracies than the base model, with the hemodynamic area under the receiver operating characteristic curve (AUC) (0.819) insignificantly outperforming the structural set AUC (0.816). The GCC + RNFL model and the model containing all structural and vascular features were also significantly more accurate than the base model.”

They concluded that the neural network models indicated that the OCTA optic nerve head vascular biomarkers are nearly equally useful for machine learning diagnosisof POAG compared to OCT structural biomarker features alone.

This study was the steppingstone to others that evaluated the dynamics of the ocular vascular information using continuous time neural networks.

Prediction of RNFL thinning. The investigators evaluated the dynamics of the ocular vasculature. In one pilot study,² presented at the annual meeting of the Association for Research in Vision and Ophthalmology in 2025, the researchers developed a machine learning tool to predict progression of the thinning of the RNFL.

A model of the ocular vasculature was built based on patient data; a neural network that was trained on these data was built to look at trends in RNFL thinning over time from baseline and then every 5 years. The investigators then looked for trends in the patient hemodynamic information that predicted why they were losing vision.

“This remains an open question. We found certain significant correlations, but the cohorts were small and specific conclusions could not be drawn,” Riina said. However, he explained that AI is uncovering findings in the hemodynamic information from the ocular vasculature that predict the visual loss.

Riina and colleagues reported, “The best-performing model was tested with different values of systolic BP (SBP), IOP, and diastolic BP (DBP) to show how these values may impact progression.” They described a patient in whom the RNFL became thinner as the SBP and IOP increased and as the DBP decreased.

They concluded that a continuous-time neural model is useful for extracting information from irregularly sampled OCTA scans. A subset of patients who were not progressing was identified but who were at higher risk. The model found that higher SBP increased the rate of RNFL progression; however, the magnitude of this difference was variable between each patient,² they reported.

Heart rate and ocular imaging. In a study published in Scientific Reports,³ the Harris team described and tested a new system that integrated electrocardiography (ECG) with ocular imaging.

They explained that the retinal microvasculature is one of few sites where the microcirculation can be directly and non-invasively visualized in vivo, offering a unique window into ocular disease and systemic health.³

The ECG included laser Doppler holography (LDH), which captures Doppler-induced phase shifts and provides real-time cardiac cycle-resolved assessment of the retinal hemodynamics, they explained.

The study included 25 healthy adults who underwent imaging. Data analysis showed that the “system reliably synchronizes cardiac and retinal signals, and provides repeatable metrics of real-time hemodynamics, with R-PSV½, ie,defined as the time from the ECG R-peak to 50% PSV amplitude, emerging as the most robust ECG-retina latency. Furthermore, this study provides preliminary normative data for healthy adults and demonstrates that retinal blood flow is relatively preserved across a physiologic range of ocular perfusion pressure. This integrated ECG-LDH platform offers a new approach for exploring the eye-heart relationship, with potential applications in both clinical and research settings,” the authors concluded.

BP variability over time. In a study of the variations in patients over time that has been accepted by the Journal of Glaucoma,⁴ the results showed that patients with a higher standard deviation of BP measurements had more vision loss, but the structural loss was not as great. It was fascinating that the structure was fine, but the vision was deteriorating. This led us to what is so exciting about studying the vasculature, i.e., that something is happening before the ocular anatomy changes measurably. The nerve is disappearing but before that, it dies, which may be something that OCT cannot capture in time. The efficiency of the autoregulatory system of perfusion in patients losing vision may be the next focus of research,” Riina said.

Take-home messages for ophthalmologists and optometrists

Riina had two messages to share with eye care clinicians.

“In glaucoma, based on what I have observed, it is unclear why patients with normal tension glaucoma are losing vision despite treatment. Research is revealing some amazing correlations, new therapies will be developed that match the research findings, and there is hope for these patients. The study of other neurodegenerative diseases and Oculomics, an emerging field, are also very exciting,” he commented.

“Moving forward, AI is the hot topic that will impact the future of the clinic in many ways. We are working with clinicians to use AI to better understand these diseases and to provide individualized treatments. On the research side we use AI to study disease mechanisms and for training models of diagnosis. We are now working to bring these research insights of glaucoma and the vasculature into the clinic in ways that will empower physicians and enhance patient care.”

References
1. Riina N, Harris A, Siesky BA, et al. Using multi-layer perceptron driven diagnosis to compare biomarkers for primary open angle glaucoma. Invest Ophthalmol Vis Sci. 2024;65:16. https://doi.org/10.1167/iovs.65.11.16
2. Riina N, Harris A, Pasquale LR, et al. Continuous-time neural ordinary differential equation predicts RNFL progression in primary open angle glaucoma. Presented at the Association for Research in Vision and Ophthalmology 2025 annual meeting.
3. Wood K, Schnorbus N, Muncharaz L, et al. A novel electrocardiogram-synchronized laser Doppler holography system for cardiac cycle-resolved retinal hemodynamics: development and validation. Sci Rep. 2026;16:20436. https://doi.org/10.1038/s41598-026-51137-2
4. Riina N, Harris A, Siesky BA, et al. Blood pressure variability, ocular perfusion pressure variability, and visual field progression in primary open angle glaucoma. J Glaucoma. 2026; in press.