How is AI Changing Ophthalmology Diagnosis and Treatment?

In recent years, the intersection of artificial intelligence (AI) and life sciences has led to advancements across the medical and clinical research fields. Though AI is serving many therapeutic areas, the field of ophthalmology has seen the potential to apply AI to advance diagnosis, drug efficacy, and other treatment-related interventions.

In the past, the focus of AI research in ophthalmology centered on screening and diagnosing fundus diseases, e.g., diabetic retinopathy (DR), age-related macular degeneration (AMD), and glaucoma. According to Zhang et al. (2023), AI research in ocular surface diseases is on the rise.

Now, with the use of deep learning (DL), computers are able to automatically learn hidden features in images and pick out patterns invisible to the naked eye. The development and clinical application of innovative imaging techniques include incorporating AI, e.g., artificial neural networks (ANNs), which increasingly contribute to the detection and customized control of ophthalmic diseases.

In this article, we explore how AI, particularly deep learning, is changing ophthalmology by enhancing the detection and treatment of eye diseases such as diabetic retinopathy and ocular surface conditions. We dive into AI’s role in early diagnosis, tailored treatments, and clinical decision-making while highlighting challenges and the need for ongoing research to optimize AI’s impact in ophthalmology.

Enhancing Early Detection

Traditionally, ophthalmology heavily relied on manual assessment and interpretation of medical images for detection, often leading to delays in diagnosis and treatment. Early detection of ophthalmic diseases plays a pivotal role in preserving visual health. Timely identification allows for swift intervention and tailored treatment strategies that can halt or slow down disease progression.

For conditions like diabetic retinopathy, age-related macular degeneration, and glaucoma, where irreversible vision loss is a potential outcome, early diagnosis empowers medical professionals to implement targeted therapies, reducing the impact of the disease and improving long-term outcomes for patients.

Recent studies have indicated that AI increases patient access to screening and clinical diagnosis while decreasing healthcare costs. In the U.S., two autonomous AI-enabled systems received regulatory approval for screening mid-level or advanced DR and macular edema (ME).

AI approaches like deep convolutional neural networks (DCNNs) and vision transformers for Computer-Aided Diagnosis (CAD) have performed well in automating disease detection. A recent review on the use of DL techniques in retinal disease detection and grading of glaucoma, DR, AMD, and other retinal diseases found that CAD, through DL, will increasingly become vital in ophthalmology. Sheng et al. (2022) studied a trained AI system and demonstrated that its performance in classifying AMD and diabetic macular edema (DME) was comparable to human experts.

AI and Ocular Surface Diseases

A study by Zhang et al. (2023) revealed that different AI models are used for different ocular surface diseases. Most AI applications for pterygium, keratitis, and dry eye disease (DED) are convolutional neural networks (CNN). At the same time, the random forest (RF) model is accurate in predicting healthy eyes and keratoconus.

DED is one of the most underdiagnosed and undertreated ophthalmology conditions. The diagnosis of DED relies on tests, many of which depend on an experienced observer for image interpretation. This may be subjective and result in variation in diagnosis. Further, the fluorescein tear breakup time test, which is currently widely used to diagnose DED, is an invasive and subjective method, contributing to variability in diagnostic results.

Kikukawa et al. (2023) developed an objective approach to detect tear breakup using CNNs on tear film images taken by a non-invasive device. Wan et al. (2023) developed a segmentation algorithm based on DL and image processing to automatically measure tear meniscus height (TMH), a vital reference parameter. The researchers found that their proposed method was highly consistent with the manual measurement of TMH.

AI-Based Fundus Screening

Cao et al. (2023) investigated the application effect of an AI-based fundus screening system in a real-world clinical environment. The system demonstrated superior diagnostic effectiveness for DR, retinal vein occlusion (RVO), and pathological myopia (PM). In addition, a systematic review and meta-analysis on AI in detecting PM found that current AI algorithms based on fundus and OCT images perform well.

Diabetic Retinopathy Screening in Different Settings

Diabetic retinopathy screening allows ophthalmologists the ability to detect early signs of a potentially sight-threatening complication of diabetes. The condition often develops without noticeable symptoms in its initial stages, making regular screening crucial. However, there are challenges with implementing effective screening programs.

Accessibility and coverage pose an issue as not all diabetes patients have equal access to screening services, especially in rural or underserved areas. This can result in delayed diagnosis and treatment, jeopardizing visual health. Additionally, the accuracy of screening methods can vary, leading to false outcomes, which could impact patient management decisions.

The role of AI in diabetic retinopathy screening may offer solutions to these challenges across diverse settings.

Community hospital

A study by Liu et al. (2022) assessed the feasibility and clinical utility of AI-based screening for DR and ME, combining fundus photography and OCT. Ophthalmologists graded these photos in line with the International Clinical Diabetic Retinopathy (ICDR) Severity Scale. The study revealed that AI-based DR screening had high sensitivity and specificity.

AI and smartphone-based retinal cameras

Malerbi et al. (2022) evaluated the diagnostic accuracy of a DL algorithm of a DR screening program in a real-world, high-burden setting. They found the performance of the retinal camera adequate, and the DL algorithm achieved excellent diagnostic accuracy in a real-world setting. A meta-analysis of similar smartphone-based AI programs for DR detection found that such programs demonstrate high diagnostic accuracy for detecting DR and RDR.

Rural setting

In light of past studies using standardized data sets in urban outpatient settings, Pawar et al. (2021) sought to validate an AI algorithm in DR detection and staging through fundus photography in a real-time rural setting. The study revealed that the algorithm showed sensitivity and specificity in RDR detection, comparable with the performance of ophthalmologists.

Addressing AI’s Limitations Within Ophthalmology

While AI presents immense potential, there are several limitations that warrant consideration. Zhang et al. (2023) identified several limitations in applying ML in ophthalmology.

  • ML approaches need more training and validation sets to improve accuracy, sensitivity, and specificity.
  • Inspection equipment used by different countries, regions, and medical institutions differ, e.g., disparities in color and resolution, which affect the accuracy of diagnosis.
  • Current ML methods cannot explain disease diagnosis.
  • AI cannot learn effectively for rare diseases with insufficient data.

The Role of AI in Clinical Decision Making

The integration of AI into clinical decision-making processes within ophthalmology presents both opportunities and challenges. As highlighted by Wawer et al. (2023), the incorporation of AI into routine clinical practices is constrained by the diversity of imaging techniques and the heterogeneity of AI methods used across different healthcare settings. The medical field’s existing infrastructure, which has evolved over decades, poses challenges in seamlessly integrating AI technologies.

However, amidst these challenges, AI emerges as a valuable tool for augmenting doctors’ expertise. By processing and analyzing large volumes of complex data quickly and accurately, AI offers an objective and data-driven perspective that can complement clinical judgment. This synergy between human insight and AI-driven analytics has the potential to improve diagnostic accuracy, treatment planning, and patient outcomes.

AI Changing Ophthalmology Treatment and Clinical Routine

The transformative influence of AI extends beyond diagnostics, reshaping the landscape of ophthalmology treatment and clinical practices. With advancements in AI algorithms, the field is witnessing a paradigm shift towards designing tailored treatments and optimizing patient care.

Designing tailored treatments

AI’s integration into treatment design is particularly evident in DR. Automated fine-grained DR grading can assist ophthalmologists in monitoring DR and designing tailored interventions. Li et al. (2022) developed a simple deep attentive CNN (DACNN) for DR grading and lesion discovery. Study results indicated that the approach achieved competitive performance in DR grading and lesion discovery.

Clinical applications

Machine learning (ML) and DL algorithms have significantly progressed in ophthalmology, which can be witnessed in the diagnosis and management of anterior segment diseases. Xu et al. (2023) provide compelling evidence of AI’s capacity to diagnose and manage conditions affecting the front portion of the eye, further expanding its clinical applications.

Future Research Directions

As AI tools continue to demonstrate their value in diagnosing eye diseases and guiding treatment decisions, the road ahead is brimming with possibilities. The ongoing collaboration between AI experts, clinicians, and researchers is key for harnessing AI’s full potential. Future research endeavors should aim to refine and optimize AI and deep learning algorithms, enabling them to evolve alongside the dynamic landscape of ophthalmic care.

Vial Ophthalmology CRO: Pioneering Innovation in Ophthalmic Research

At Vial, we are at the forefront of driving innovation in ophthalmic research through our specialized Contract Research Organization (CRO) – the Vial Ophthalmology CRO. Powered by cutting-edge technology, we are committed to reimagining the way clinical trials are conducted in the field of eye health. Our mission is rooted in accelerating the development of groundbreaking therapeutics that enhance vision and improve the lives of patients worldwide.

Our approach is technology-forward, integrating intuitive, end-to-end technology solutions into every facet of our operations. Our intuitive technology platform empowers us to conduct faster, more efficient trials, significantly lowering costs for biotech sponsors. By leveraging technology, we ensure that each trial is conducted with the highest standards of quality, accuracy, and ethical considerations.

Through our Vial Ophthalmology CRO and technology platform, we are reimagining the future of eye health by bringing novel therapeutics to patients faster and more effectively than ever before.

Ready to run faster, more affordable clinical trials? Connect with a Vial team member today!

Contact Us

By submitting, you are agreeing to our terms and privacy policy
This field is for validation purposes and should be left unchanged.