This article was originally published in a sponsored newsletter.

Artificial intelligence (AI) is “the science and engineering of making intelligent machines through algorithms or a set of rules, which the machine follows to mimic human cognitive function such as learning and problem solving.”¹

I have to admit, I haven’t kept up with this technology. I tried one of the popular online chatbots for the first time only a couple of weeks ago to help me write a personal letter. Although it was nicely organized and well-written, it wasn’t personal enough. The chatbot obviously had no memories to draw from, and it just didn’t sound like “me.” I ended up using the AI-generated draft as nothing more than a starting point and spent a lot of time on revisions. In fact, it likely took me twice as long to write the letter with AI than it would have if I had just written it from scratch. As a first-time user, I didn’t know what I was doing when I created my prompt and I’m sure additional practice will improve the output; however, I’m not that excited to try again.

Of course, using online chatbots is not the focus of this week’s issue, but it is a light-hearted introduction to the topic of AI in eye care. Interestingly, the birth of AI began as a Summer Research Project at Dartmouth in the 1950s, and by the 1970s, its use in health care first began. Today, AI has multiple applications in health care such as roles in drug discovery; streamlining scheduling, billing, EHR entry and other administrative tasks; and medical imaging analysis, which is likely one of the fastest-growing areas of AI in the eye care sector. I came across one company recently that peaked my interest because it specifically advertises the use of AI to improve dry eye diagnosis and treatment.² The company promotes their business as “the first machine-learning, cloud-based software of its kind” that “eliminates biases in the diagnosing and treating process through evidence-based algorithms and machine learning processes.”² I’m looking forward to doing a deeper dive into what they offer.

There are many potential benefits to using AI in health care, but not unlike my first attempt at using a chatbot for letter writing, the output can only be as good as the input. AI could be a great addition to our clinical armamentarium for discovery, interpretation and detecting change over time, but it will also come with its challenges. Data privacy, patient safety and trust are just three that come to mind immediately. Read on in this issue to see how Dr. Ziemanski explores AI further.

References:

1. Bajwa J, Munir U, Nari A, Williams B. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J. 2021 Jul;8(2):e188-e194. doi:10.7861/fhj.2021-0095.
2. CSI Dry Eye. Accessed April 8, 2024. https://csidryeye.com/

Amber Gaume Giannoni, OD, FAAO
Editor

How to Look for Incomplete Blinking or Non-lid Seal and Discuss It with Patients

Assessing the eyelids is a valuable component of a comprehensive eye examination. Identifying incomplete blinking and non-lid seal not only saves chair time and hassle, but also helps identify the root culprit of chronic evaporative dry eye and ocular surface disease. Although there are several diagnostic instruments that assist with quantifying blink rate, the Look-Lift-Pull-Push (LLPP) test is a quick, free and helpful in-office evaluation to assess lid-seal, starting with simply looking at patients.

Look

From the moment patients arrive in the exam room, I look for signs of facial rosacea, collarettes or other eyelash debris, eyelash extensions, tattoo eyeliner, lagophthalmos, ptosis and scleral show, among other conditions. With the slit lamp, I typically start all routine exams with patients closing their eyes or looking down. Regardless of a patient’s age, observing the anatomical resting position of their closed eyes can be surprising and can further explain why certain ocular surface symptoms may worsen nocturnally. A transilluminator or light source placed on closed eyelids can help visualize eyelid seal (or lack thereof) quickly. Clinically, when I observe incomplete blinking or poor eyelid seal, recording the eyelids in action seems to be most impactful for patient understanding. Often, patients have no idea that they are blinking incompletely or that their eyes don’t close completely until they see it for themselves.

Lift and Pull

Lifting and pulling patients’ eyelids can also give hints about other possible ocular and systemic health conditions. Floppy eyelids with poor apposition and lid laxity, found by simply touching the eyelids, could indicate potential sleep apnea. Be sure to ask about a potential or known sleep apnea diagnosis, CPAP use or both. Acknowledging lid seal is important for these patients because some may also have an anatomical or acquired eyelid issue that may be discovered by lifting and pulling the eyelids.

Push

Pushing on the eyelids with an upward motion to assess the quality and quantity of meibomian gland oil expression is also an important factor in helping to diagnose dry eye disease. My goal is to assess for untreated meibomian gland dysfunction, incomplete blinking and poor lid-seal that could further worsen the desiccation stress, inflammatory factors and the overall chronic dry eye cycle. Taking note of the status of meibomian gland function is a necessary component of assessing the eyelids to develop a targeted treatment plan.

Additional Factors to Evaluate

Helping patients better understand the multifactorial complexity of treating exposure-related ocular surface disease can be difficult. I often use the analogy that our eyelids are like windshield wipers. I ask patients to think of an incomplete blink like windshield wipers that rarely clear a certain portion of the windshield. It is important to educate our patients about daily activities such as computer use that further decrease our blink rate and could further exacerbate symptoms of incomplete eyelid closure. Unfortunately, many other factors may also impact and contribute to incomplete blinks, including use of certain medications (i.e., Botox), Bell’s palsy and thyroid issues such as Graves’ disease.

Assessing incomplete blinking and lid-seal is easy, quick and affordable, and provides insight on how to proceed with navigating chronic ocular surface disease.

Dry eye disease (DED) is a multifactorial disease of the tears and ocular surface that results in discomfort, visual disturbances, tear film instability and, potentially, damage to the ocular surface. Blinking is an often-overlooked aspect of the condition. Incomplete blinking (IB) can be associated with pathological changes in the meibomian glands and a disturbance in meibomian lipid flow. IB increases ocular surface exposure due to prolonged interblink intervals, which leads to elevated tear osmolarity, tear film instability and decreased tear film break-up time.

The number of incomplete blinks and IB rate are the most important aspects of assessing blink patterns in patients with DED. The challenge with assessing blinking is that it is a complicated process which includes both physiological and psychological factors. Blinking is influenced by light sources and the intensity of the light source, as well as other surrounding environmental factors. It is also challenging to capture evidence of IB because high-speed equipment that can capture minute movements is required. Currently, LipiView (Johnson & Johnson Vision) is an option for researchers to perform an objective evaluation of blinking patterns. However, the live capture is limited to 20 seconds of video and the cost may be prohibitive for many clinical practices. Deep learning machine (DLM) has been proposed as an optional tool for blinking analysis, but its clinical practicability still needs to be proven. The current study aims to compare the novel DLM-assisted Keratograph 5M (K5M) with the currently available LipiView and assess whether blinking parameters can be applied in DED diagnosis.¹

A total of 140 eyes were examined, and 35 DED participants and 35 normal subjects were recruited in this cross-sectional study. A DED questionnaire and ocular surface signs were evaluated. Blinking parameters included the number of blinks, the number of incomplete blinks and IB rate, which were individually collected from the DLM-analyzed K5M videos and LipiView-generated results. The agreement and consistency of blinking parameters were compared between the two devices and the association of blinking parameters to DED symptoms and signs were evaluated.¹

The results demonstrated that LipiView presented a higher number of incomplete blinks and IB rate than those from DLM-assisted K5M. The DLM-assisted K5M captured significant differences in the number of blinks, the number of incomplete blinks and IB rate between DED and normal subjects. In all three parameters, DLM-assisted K5M also showed a better consistency in repeated measurements than LipiView. The authors concluded that the DLM-assisted K5M is a useful tool to analyze the blinking process and reveal abnormal blinking patterns. The K5M provides a recording condition that is physiologically more comfortable than LipiView, and the application of DLM made the analysis more objective and efficient, especially in distinguishing the DED patients from normal subjects. The authors recommended larger trials to determine efficacy before implementing DLM-assisted K5M into clinical practice.¹

References:

1. Ren Y, Wen H, Bai F, et al. Comparison of deep learning-assisted blinking analysis system and Lipiview interferometer in dry eye patients: a cross-sectional study. Eye and Vision. 2024 Feb;11:7. doi:10.1186/s40662-024-00373-6

Comparison of deep learning-assisted blinking analysis system and LipiView interferometer in dry eye patients: a cross-sectional study

Yueping Ren, Han Wen, Furong Bai, Binge Huang, Zhenzhen Wang, Shuwen Zhang, Yaojia Pu, Zhenmin Le, Xianhui Gong, Lei Wang, Wei Chen and Qinxiang Zheng
Eye and Vision. 2024 Feb 19;11:7. doi:10.1186/s40662-024-00373-6.

BACKGROUND: Abnormal blinking pattern is associated with ocular surface diseases. However, blink is difficult to analyze due to the rapid movement of eyelids. Deep learning machine (DLM) has been proposed as an optional tool for blinking analysis, but its clinical practicability still needs to be proven. Therefore, the study aims to compare the DLM-assisted Keratograph 5M (K5M) as a novel method with the currently available LipiView in the clinic and assess whether blinking parameters can be applied in the diagnosis of dry eye disease (DED).

METHODS: Thirty-five DED participants and 35 normal subjects were recruited in this cross-sectional study. DED questionnaire and ocular surface signs were evaluated. Blinking parameters including number of blinks, number of incomplete blinking (IB) and IB rate were collected from the blinking videos recorded by the K5M and LipiView. Blinking parameters were individually collected from the DLM analyzed K5M videos and LipiView generated results. The agreement and consistency of blinking parameters were compared between the two devices. The association of blinking parameters to DED symptoms and signs were evaluated via heatmap.

RESULTS: In total, 140 eyes of 70 participants were included in this study. LipiView presented a higher number of IB and IB rate than those from DLM-assisted K5M (P≤0.006). DLM-assisted K5M captured significant differences in number of blinks, number of IB and IB rate between DED and normal subjects (P≤0.035). In all three parameters, DLM-assisted K5M also showed a better consistency in repeated measurements than LipiView with higher intraclass correlation coefficients (number of blinks: 0.841 versus 0.665; number of IB: 0.750 versus 0.564; IB rate: 0.633 versus 0.589). More correlations between blinking parameters and DED symptoms and signs were found by DLM-assisted K5M. Moreover, the receiver operating characteristic analysis showed the number of IB from K5M exhibiting the highest area under curve of 0.773.

How and When to Use AI in Dry Eye

Dr. Ziemanski has no conflicts of interest to disclose.

A sage mentor once shared his perspective on AI with me. He said, “AI is expanding at the sacrifice of actual intelligence.” He might be right. When I look at the generative AI platforms out there, I wonder how younger generations will learn the art and science of composition. How will they feel the struggle of hard-fought paragraphs or the pride and triumph of finally discovering the right combination and sequence of words to capture our true thoughts and emotions? How will they learn perseverance if there is always a capable and willing machine to take on the work for them?

Shortly after writing the above paragraph, I had an idea. An awful, no-good, lazy idea. Why should I struggle to write this article when there is a capable and willing machine that can do it for me? I succumbed, leaned into the idea (with permission!) and submitted the following prompt into Gemini, Google’s generative AI platform:

“write a 300 word article on artificial intelligence in the care of patients with dry eye disease. include how it can be used, what technology currently uses it, and how it can help with efficiency in writing referral letters. include literature sources and risks of breaching HIPAA rights.”

It took me approximately 15 seconds to complete the brain dump for the prompt. After another three seconds, I received the output. I now introduce you to my guest writer, Gemini. All AI-generated text will appear in red. My voice will appear a few times in black and at the end. Read on.

Interesting, isn’t it? While I could expound on the metaphysics, ethics and general strengths and weaknesses of the technology to the extent my brain is capable, I don’t want to lose sight of the main purpose of this article: to inform you of the utility of AI in our field.

First, it is an immensely powerful time-saving tool. The above article was “completed” in under 20 seconds. Generative AI is being used to write referral, care-coordination or reference letters for students, residents and colleagues. Most use the output as a first draft that can be further edited for clarity or to add specific examples. If you choose to use it in this manner, please be advised that all text and inputs to generative AI are not secure. Never include identifying information for patients or other individuals. Also consider whether you should disclose that AI contributed to the development of the text.

As for this technology in clinical care, there are approved devices to aid in the detection of diabetic retinopathy, for example; however, there are currently no approved technologies to aid in the diagnosis of dry eye disease. Despite this gap, it’s inevitable that there eventually will be. Below is a summary of two areas that have been explored and could be the most meaningful to clinical practice:

1. Meibography: Despite how wide-spread meibography has become, image interpretation is still almost exclusively qualitative. Deep learning has been used to quantify meibomian gland dropout with accuracy exceeding clinical experts.¹ With improved accuracy, clinicians will be better able to detect change from baseline or, potentially, decelerated rates of progression after treatment.
2. Blink patterns: We all know that altered blink patterns are associated with dry eye disease, but most of us do not truly assess blink patterns in clinical practice. While partial blinks are tracked during LipiView II assessment, many providers are skeptical whether the value truly correlates with patterns in patients’ habitual environments. Capturing blink videos without bright, flickering lights and employing deep-learning or machine-learning approaches could help to differentiate dry eye patients from normal patients more efficiently.^2,3

Other areas have also been explored, but in my opinion, they fill less of a current need or are still in primitive stages of development. Nonetheless, the future is exciting, if a bit scary. While I think we will all welcome the clinical advancements that accompany the growth of AI, how will health care evolve? What will be our role as diagnosticians? Will actual intelligence suffer? I sincerely hope that our society will continue to appreciate, value and encourage human creations. On a lighter and more personal note, I also hope that our readership appreciates the few additional hours of struggle I endured to put these hard-fought paragraphs on your screens! Which do you prefer: the artificially composed ones, or mine?

References:

1. Wang J, Yeh TN, Chakraborty R, Yu SX, Lin MC. A deep learning approach for meibomian gland atrophy evaluation in meibography images. Trans Vis Sci Technol. 2019 Nov;8(6):37. doi:10.1167/tvst.8.6.37
2. Su Y, Liang Q, Su G, et al. Spontaneous eye blink patterns in dry eye: clinical correlations. Invest Ophthalmol Vis Sci. 2018 Oct;59(12):5149-5156. doi:10.1167/iovs.18-24690
3. Zhang ZZ, Kuang RF, Wei ZY, et al. [Detection of the spontaneous blinking pattern of dry eye patients using the machine learning method- Article in Chinese]. Zhonghua Yan Ke Za Zhi [Chinese J Ophthalmol]. 2022 Feb;58(2):120-129. doi:10.3760/cma.j.cn112142-20211110-00537