New York Patent of the Month – December 2025

A breakthrough in medical artificial intelligence is set to transform how doctors track and treat vision loss. AEYE Health Inc. recently secured U.S. Patent No. 12493958, titled ‘Method and system for predicting manifestation or progression of a retinal malady and method for training machine learning (ML) models for the same’. This invention earned “Patent of the Month” honors for its potential to prevent blindness through early, automated detection.

Predicting Eye Disease with AI

The system uses advanced machine learning to analyze retinal images and predict how eye diseases will progress. By identifying subtle patterns in the eye, the software can alert clinicians to risks before physical symptoms appear. This proactive approach allows for faster intervention and more personalized care. The technology simplifies a complex diagnostic process into a streamlined digital assessment.

Real-World Impact and Accessibility

This innovation addresses a critical gap in global healthcare by making specialist-level screenings available at the primary care level. Millions of patients at risk for conditions like diabetic retinopathy often lack access to timely eye exams. The patented method provides a scalable solution that works quickly and accurately. Because it integrates easily into existing clinics, it removes the common barriers of cost and travel for patients. This leap in R&D marks a significant milestone in the fight against preventable vision loss.

The trajectory of medical artificial intelligence has historically been defined by the pursuit of diagnostic parity—engineering algorithms capable of matching the acute observational skills of human clinicians. However, the recent issuance of US Patent 12,493,958 marks a decisive inflection point, signaling the industry’s transition from static detection to temporal prognostication. This seminal intellectual property, formally titled “Method and system for predicting manifestation or progression of a retinal malady and method for training machine learning (ML) models for the same,” was applied for on March 2, 2022, and lists inventors Zack Dvey-Aharon, Danny Margalit, Amit Wohl, and Yovel Rom. Assigned to AEYE Health, Inc., a New York-based innovator in computational ophthalmology, this patent establishes the proprietary framework for a new class of diagnostic architecture: one that does not merely identify the presence of pathology but calculates the probability of its future manifestation. This report provides an exhaustive analysis of the invention, elucidating the technical and clinical factors that precipitated its selection as the “Patent of the Month” by Swanson Reed, and rigorously contrasting its capabilities against incumbent technologies to demonstrate its superior utility in the landscape of preventative medicine.

1. Technical and Clinical Foundations

To fully appreciate the magnitude of the innovation described in Patent 12,493,958, it is necessary to first deconstruct the clinical landscape of Diabetic Retinopathy (DR), the primary malady targeted by this technology. DR acts as a silent epidemic; it is the leading cause of blindness in working-age adults, a statistic that persists despite the existence of effective treatments. The persistence of this public health failure is not due to a lack of therapeutic options—anti-VEGF injections and laser photocoagulation are highly effective when administered in time—but rather a failure of timely identification.

The pathophysiology of DR is progressive. It begins with micro-vascular changes: the tiny blood vessels in the retina weaken, bulge (microaneurysms), and leak fluid or blood. This stage, Non-Proliferative Diabetic Retinopathy (NPDR), is often asymptomatic. Without intervention, it progresses to Proliferative Diabetic Retinopathy (PDR), where the retina, starved of oxygen, triggers the growth of abnormal, fragile new blood vessels that bleed into the vitreous, causing catastrophic vision loss.

2.1 The Limitations of Current Screening Paradigms

The current standard of care relies on annual screening. A patient visits an optometrist or ophthalmologist, or perhaps utilizes a teleretinal screening service. The output of this interaction is a binary or categorical diagnosis based on the current state of the retina: “No DR,” “Mild DR,” “Moderate DR,” or “Severe/Proliferative DR.”

This approach suffers from a fundamental “reactive” flaw. A patient diagnosed with “Mild NPDR” is typically sent home and told to return in one year. However, the rate of disease progression is highly variable and idiosyncratic. One patient with mild disease may remain stable for a decade, while another may progress to sight-threatening PDR in six months due to unmanaged glycemic variability, hypertension, or genetic predisposition. The static “snapshot” diagnosis fails to capture this velocity of disease. It tells the clinician where the patient is, not where they are going. Consequently, “rapid progressors” often fall through the cracks of the annual screening interval, presenting with irreversible damage before their next scheduled exam.

2.2 The Economic Burden of Late Diagnosis

From a health economics perspective, the cost differential between prevention and cure is staggering. Managing early-stage DR involves inexpensive lifestyle interventions and glucose control. Managing late-stage PDR involves complex vitreoretinal surgeries, expensive monthly intravitreal injections, and the immense societal cost of blindness (lost productivity, disability care). Technologies that can predict progression—shifting patients from the high-cost “cure” bucket to the low-cost “prevention” bucket—are the “Holy Grail” of value-based care. This is the precise clinical and economic void that Patent 12,493,958 addresses, moving the technological goalpost from “What do I have?” to “What will happen to me?”.

3. Architectural Analysis of US Patent 12,493,958

The innovation claimed in US Patent 12,493,958 is not merely an incremental improvement in image recognition accuracy; it is a fundamental architectural divergence. While the full text of the patent describes complex methodologies, the core abstract and associated claims reveal a system designed to ingest retinal imagery and output a probabilistic vector regarding future states.

3.1 The Algorithmic Shift: Detection vs. Prediction

Standard AI in ophthalmology utilizes Convolutional Neural Networks (CNNs) trained on cross-sectional data.

• The “Detection” Model (Prior Art): Input: Image at Time . Label: Diagnosis at Time . The model learns to map pixels to the current label. .
• The “Prediction” Model (Patent 12,493,958): Input: Image at Time . Label: Diagnosis at Time (e.g., 12 months later). The model learns to identify latent biomarkers in the current image that correlate with future deterioration. .

This distinction is profound. The features required to detect a hemorrhage are distinct (red blobs, distinct edges). The features required to predict a future hemorrhage are likely sub-perceptual—minute changes in vessel tortuosity, subtle variations in the caliber of the arteriolar-venular ratio, or texture changes in the retinal nerve fiber layer that are invisible to the human eye. The patent covers the method of training specialized Machine Learning (ML) models to extract these prognostic features.

3.2 Longitudinal Training Methodologies

The patent title explicitly mentions “method for training machine learning (ML) models.” This implies the curation of massive longitudinal datasets. To train such a system, AEYE Health would have had to aggregate patient data over years, linking baseline images with outcomes years down the line. The “training method” likely involves complex loss functions that penalize the model not just for misidentifying the current disease state, but for failing to rank the risk of future progression accurately. This represents a significantly higher barrier to entry than simple detection, as longitudinal data is exponentially scarcer and more difficult to curate than cross-sectional data.

3.3 Handheld Agnosticism and Image Quality

A critical, often overlooked component of real-world AI deployment is image quality. Retinal images taken with handheld cameras (which AEYE Health specializes in) are prone to motion blur, poor lighting, and artifacts. The patent likely incorporates advanced “binarizing sections” and image normalization techniques. These pre-processing steps are essential for “prediction,” as subtle prognostic biomarkers could easily be obscured by noise. The ability to run high-fidelity predictive models on low-fidelity handheld input is a technical marvel that democratizes access, allowing sophisticated risk stratification to occur in primary care or pharmacy settings rather than specialized imaging centers.

4. Competitive Landscape Analysis: AEYE Health vs. The Incumbents

To validate the “superiority” claims associated with Patent 12,493,958, a rigorous comparative analysis against the established market leaders—Digital Diagnostics and Eyenuk—is essential. Both competitors hold FDA clearances, but their technological foundations are rooted in the “detection” paradigm of the previous decade.

4.1 Digital Diagnostics (Formerly IDx)

Digital Diagnostics is the pioneer of autonomous AI, achieving the first De Novo FDA clearance for their product, IDx-DR (now LumineticsCore).

• Technological Philosophy: Their system is designed as a rigid “medical device.” It is strictly coupled with a specific high-end robotic camera, the Topcon NW400.
• Operational Limitation: This hardware-dependency creates a bottleneck. The camera is expensive, large, and requires a dedicated darkroom. It is suited for large health systems but fails to penetrate the “last mile” of care (e.g., rural clinics, pharmacies).
• Analytical Scope: IDx-DR is a binary classifier. It detects “More than Mild Diabetic Retinopathy.” It does not claim to predict which “Mild” cases will progress. It is a lagging indicator of disease.

4.2 Eyenuk (EyeArt)

Eyenuk’s EyeArt system is a powerful screening tool widely used in Europe and the US.

• Technological Philosophy: Eyenuk prioritizes sensitivity (catching every case). In pivotal trials, it demonstrated sensitivity exceeding 95%.
• The Specificity Trade-off: The Achilles’ heel of high sensitivity is lower specificity (approx. 81%). This results in a high rate of false positives—patients flagged as sick who are actually healthy. In a value-based care model, this is costly, as it triggers unnecessary specialist referrals.
• Analytical Scope: Like Digital Diagnostics, EyeArt detects current pathology. It grades the image based on visible lesions. It lacks the temporal predictive dimension claimed in AEYE Health’s patent.

4.3 The AEYE Health Advantage

AEYE Health’s technology, underpinned by Patent 12,493,958, disrupts this duopoly through three vectors of superiority:

4.3.1 The Prognostic Vector

By offering “Disease Prediction,” AEYE Health moves upstream. While competitors identify patients who need treatment now, AEYE identifies patients who need intervention to prevent future treatment. This capability aligns perfectly with the preventative mandates of insurers and Accountable Care Organizations (ACOs). The patent secures the exclusive right to the method of this prediction, creating a significant competitive moat.

4.3.2 The Accessibility Vector

AEYE Health is the only company with FDA clearance for a portable, handheld solution (using the Optomed Aurora camera).

• Impact: This decouples AI screening from the “eye clinic.” A nurse practitioner visiting a home-bound diabetic patient can perform a scan in the living room. The predictive AI can then analyze that image to forecast risk. This “point-of-care” capability is unique to AEYE and is enabled by the robust algorithms described in their intellectual property which handle the variable quality of handheld imaging.

4.3.3 The Efficiency Vector (Imageability)

A major metric in AI screening is “gradability” or “imageability”—the percentage of patient images the AI can actually read. In competitor trials, ungradable rates often hovered between 10-20%, requiring those patients to be referred anyway. AEYE Health reports a success rate of >99%. This implies that the “training methods” described in Patent 12,493,958 include novel data augmentation or domain adaptation techniques that make the model exceptionally robust to real-world imperfections.

4.4 Comparative Summary Matrix

The following table crystallizes the technological stratification between AEYE Health and its competitors, utilizing data derived from clinical trial reports and patent disclosures.

Feature Set	AEYE Health (AEYE-DS)	Digital Diagnostics (IDx-DR)	Eyenuk (EyeArt)
Core IP Claim	Prediction of Manifestation/Progression (Patent 12,493,958)	Detection of Referable DR	Detection of Referable DR
Temporal Capability	Prognostic (Future Risk Assessment)	Diagnostic (Current State)	Diagnostic (Current State)
Hardware Dependency	Agnostic / Handheld (Optomed Aurora)	Specific Tabletop (Topcon NW400)	Specific Tabletop (Canon/Topcon)
Point of Care	High (Home, Retail, Primary Care)	Low (Dedicated Clinic Space)	Low (Dedicated Clinic Space)
Imageability	> 99% (Superior Robustness)	Lower (High image quality required)	Variable (Significant ungradable rate)
Specificity	89% – 94% (Low False Positives)	~90%	~81% (High False Referrals)
Systemic Scope	Yes (Cardiovascular Prediction)	No (Ocular only)	No (Ocular only)

5. “Patent of the Month”: Rationale for Selection

The designation of US Patent 12,493,958 as the New York Patent of the Month for December 2025 by Swanson Reed is a rigorous endorsement rooted in specific evaluative criteria. Swanson Reed, as a specialized R&D tax advisory firm, reviews thousands of patents to identify those that demonstrate “exceptional novelty, technical advancement, and potential market influence”. The selection of this patent was driven by three converging factors:

5.1 Novelty in “Zero-to-One” Innovation

Most patents in this domain are incremental—improving the accuracy of a detector by 1% or speeding up processing time. AEYE Health’s patent represents a “zero-to-one” innovation. It introduces a capability (prediction) that simply did not exist in the prior commercial art. The shift from categorizing pixels to forecasting biological entropy is a fundamental leap in the application of machine learning to biology. Swanson Reed’s analysts likely recognized that this patent defines a new category of “Prognostic Medical Devices.”

5.2 Technical Complexity and Uncertainty

The “elimination of uncertainty” is a core component of R&D eligibility. Developing a predictive model is fraught with technical peril. “Ground truth” for the future is hard to establish. How do you label an image today based on what happened three years later, while accounting for confounding variables like patient diet or medication changes? The fact that AEYE Health successfully developed a method to train these models implies they solved significant data science and engineering challenges. This high degree of technical difficulty makes it a prime example of R&D excellence.

5.3 Societal and Market Impact

The patent’s abstract touches on “malady in a retina” broadly. AEYE Health’s roadmap includes the prediction of systemic diseases like cardiovascular disease, stroke risk, and hypertension via retinal imaging. This transforms the eye exam from a niche specialist procedure into a universal triage tool for population health. Swanson Reed’s selection highlights inventions with the potential to reshape industries; a technology that can non-invasively predict heart attacks via a handheld eye scanner essentially rewrites the playbook for preventative cardiology and primary care screening.

6. Systemic Implications: The Retina as a Window to the Body

While the immediate application of Patent 12,493,958 is Diabetic Retinopathy, the “real-world impact” extends much further. The retina is embryologically part of the brain and is the only location in the human body where the central nervous system’s vasculature can be visualized non-invasively.

The patent’s methodology for “predicting manifestation” is platform-agnostic regarding the specific disease. If the ML model can be trained to predict DR progression based on vascular changes, it can theoretically be trained to predict:

• Cardiovascular Events: Narrowing of retinal arterioles is a known biomarker for hypertension and stroke risk. AEYE Health’s predictive AI could flag a patient for a cardiology workup years before a cardiac event occurs.
• Neurodegenerative Disease: Emerging research suggests retinal thinning correlates with Alzheimer’s and Parkinson’s. The longitudinal analysis methods protected by this patent could become the foundation for early screening of dementia.

By securing the IP for the method of prediction using these images, AEYE Health effectively positions itself as the gatekeeper for this entire class of “Oculomics” (ocular-genomics/systemic) diagnostics.

7. Strategic Financials: R&D Tax Credit Eligibility and Application

The development of US Patent 12,493,958 is a textbook case for the utilization of the Research and Development (R&D) Tax Credit under Internal Revenue Code (IRC) Section 41. For a high-growth technology company like AEYE Health, these credits are not merely accounting bonuses; they are non-dilutive capital that fuels further innovation.

7.1 The Four-Part Test Analysis

To qualify for the credit, the activities undertaken to develop this patent must satisfy the IRS Four-Part Test. A detailed application of this test to the patent’s development process follows:

1. Permitted Purpose: The activity must relate to a new or improved business component (product, process, software).

Application: The development of the “AEYE-DS” predictive algorithms and the handheld camera integration constitutes a new product with improved functionality (prediction vs. detection) and performance (higher imageability).

1. Elimination of Uncertainty: The taxpayer must intend to discover information that would eliminate uncertainty concerning the capability or method of development.

Application: Predicting disease progression is not standard engineering. AEYE Health faced significant uncertainty: Can a neural network reliably predict progression from a single image? What is the optimal model architecture? How do we handle image noise? The research process was designed to eliminate these fundamental uncertainties.

1. Process of Experimentation: Substantially all of the activities must constitute a process of experimentation (simulation, trial and error).

Application: The “method for training machine learning models” described in the patent inherently involves experimentation. It requires iterative cycles of data splitting, model training, hyperparameter tuning, validation against test sets, failure analysis, and retraining. Each version of the AI (e.g., v1.0 to v2.0) represents a hypothesis tested and refined.

1. Technological in Nature: The experimentation must rely on the principles of the hard sciences.

Application: The innovation is rooted in computer science (machine learning, computer vision), mathematics (statistical probability, linear algebra), and biology (ophthalmology, pathology).

7.2 The Payroll Tax Offset for Startups

For companies like AEYE Health, which may be in the pre-revenue or early-revenue phase of commercialization, the federal R&D tax credit offers a critical liquidity mechanism: the Payroll Tax Offset.

• Mechanism: Qualified Small Businesses (QSBs)—defined as having less than $5 million in gross receipts for the credit year and no gross receipts for more than five years—can elect to apply up to $500,000 of their R&D credit annually against the employer portion of Social Security taxes.
• Impact: This effectively reduces the company’s burn rate. Instead of waiting for income tax profitability to use the credit, the company receives immediate cash flow relief, allowing them to hire more engineers or invest in further clinical trials.

7.3 State-Level Incentives (New York)

Since the patent was recognized as the “New York Patent of the Month,” it highlights AEYE Health’s eligibility for New York’s specific R&D incentives. New York offers its own R&D tax credit, which can often be claimed in conjunction with the federal credit, providing a “double-dip” benefit that significantly lowers the effective cost of research in the state.

8. Swanson Reed: The Strategic Partner for Innovation

Navigating the complexities of the R&D tax credit—particularly for software and AI companies where the line between “routine coding” and “qualified research” can be blurred—requires specialized expertise. Swanson Reed, as one of the largest specialist R&D tax advisory firms in the US, plays a pivotal role in helping companies like AEYE Health monetize their innovation.

8.1 Specialized Services for AI Companies

Swanson Reed offers a suite of services tailored to the nuances of high-tech R&D:

• Eligibility Assessment & “The Nexus”: Swanson Reed’s experts work to establish the “nexus” between the qualified expenses (wages of data scientists, cloud computing costs for model training) and the qualified activities (the experiments detailed in Patent 12,493,958). They segregate eligible “core R&D” (algorithm development) from ineligible activities (routine UI maintenance or market research).
• Audit Defense (CreditARMOR): The IRS scrutinizes software claims heavily. Swanson Reed provides Audit Insurance, ensuring that if a claim is challenged, they will defend it. Their preparation methodology involves creating contemporaneous documentation—linking git commits and Jira tickets to the 4-Part Test—to create an audit-proof paper trail.
• TaxTrex Software: To streamline the process, Swanson Reed utilizes TaxTrex, an AI-driven platform that helps companies identify and document R&D activities in real-time throughout the year, rather than scrambling at tax time. This ensures higher claim accuracy and compliance.

Final Thoughts

US Patent 12,493,958 stands as a testament to the transformative power of artificial intelligence. By successfully harnessing the predictive potential of the retina, AEYE Health has pioneered a technology that does not just see the present, but foresees the future. This “Patent of the Month” represents the pinnacle of R&D excellence—a high-risk, high-reward technical endeavor that solves a massive human problem. Through the support of the R&D Tax Credit and the expert guidance of Swanson Reed, such innovations are not only technically feasible but financially sustainable, ensuring that the next generation of life-saving diagnostics can move from the laboratory to the patient.

Swanson Reed invites innovative companies to explore their eligibility for the R&D Tax Credit. With a focus on audit compliance and technical expertise, Swanson Reed ensures that your groundbreaking work receives the financial recognition it deserves. Contact Swanson Reed today to learn how your patent could be the next “Patent of the Month.”

 

Learn More:

Are you developing new or improved products, or thinking of applying for a patent? Did you know your research work could be eligible for the R&D Tax Credit and you can receive up to 14% back on your expenses? To find out more, please check out our free online eligibility test.

Feel free to book a quick teleconference with one of R&D Tax specialists if you would like to learn more about R&D Tax Credit opportunities.

Who We Are:

Swanson Reed is one of the largest Specialist R&D Tax Credit advisory firm in the United States. With offices nationwide, we are one of the only firms globally to exclusively provide R&D Tax Credit consulting services to our clients. We have been exclusively providing R&D Tax Credit claim preparation and audit compliance solutions for over 30 years.

Swanson Reed’s New York office provides R&D tax credit consulting and advisory services to New York City, Buffalo, Rochester, Yonkers, Syracuse, Albany, New Rochelle, Mount Vernon, Schenectady, Utica, White Plains, Troy, Niagara Falls, Binghamton, Rome, Long Beach, Poughkeepsie, North Tonawanda, Jamestown, and Ithaca.

Swanson Reed hosts daily free webinars and provides free IRS CE and CPE credits for CPAs. For more information please click here or contact your usual Swanson Reed representative.

Keywords: AEYE Health Inc., Retinal Diagnostic, Machine Learning, Diabetic Retinopathy, Swanson Reed