AI advances vision science, identifies disease patterns faster
Click Here to Manage Email Alerts
“Artificial intelligence adds a new level of analysis and deeper intelligence to the health care profession by allowing machines to perform human-like specific tasks through processing large amounts of data, recognizing patterns in the data input and learning from experience,” Primary Care Optometry News Editorial Board Member Karen F. Perry, OD, FAAO, said in an interview.
“AI automates repetitive learning and adds accuracy through progressive learning algorithms,” she said.
Considering all the industries with the potential to evolve using AI, eye care will be profoundly affected, as much of the specialty involves imagery and visualization, according to PCON Editorial Board Member Louis J. Catania, OD, FAAO, who has spent years studying AI and its application in eye care.
“We look at the eye with a biomicroscope, we look at the inside of the eye with a binocular indirect ophthalmoscope, we use OCT, we use all sorts of photography, and these images are convertible into AI imagery,” Catania told PCON.
He wrote a chapter on AI and its application in vision and eye care in a book, Advances in Ophthalmology and Optometry, scheduled to be published in August.
The science of making things smart
Lily Peng, MD, PhD, specializes in applying deep learning to medical information at a Google research branch.
“AI is the science of making things smart. It’s a large discipline made up of many different sciences,” she said in a recent keynote address during the Focus on Eye Health National Summit in Washington.
“Machine learning teaches machines to be smarter, and deep learning is a kind of machine learning shown to be remarkably effective in the past few years ... it’s not magic, it’s just a lot of math,” Peng said.
The science is based on artificial neural networks, “neural nets,” that have been around since the 1960s, she said.
“Neural nets are a collection of simple, trainable units that are organized in layers. These units are mathematical equations that take in numbers, make a computation and output numbers that are taken in by another layer, another set of equations that are shrunk together,” Peng explained.
Neural nets appeared in scientific literature in the 1980s, and while they were interesting academically, “they weren’t performant,” she added.
As dataset sizes increased and computers became more powerful, the neural nets were more accurate and robust, Peng said.
“With the right data set that is large enough and relevant enough to the question being asked, a neural net can be trained to be quite accurate,” she said. “This is one major benefit to deep learning.”
Further, neural nets are efficient at learning by example.
To highlight this, she referred to spam email.
“You can show the neural net many examples of email that are spam and examples that are not spam. The neural net will find the rules itself and can identify which is spam,” Peng said.
In health care, AI is extraordinarily skilled at screening programs and it is also a good fit where expertise is limited, such as a lack of health care providers or with patients who have many demands, such as those with diabetic retinopathy, Peng said.
It is widely believed that AI will improve clinical outcomes, increase clinician efficiency and decrease health care costs, Perry said.
“I can see where, especially in optometry, if we aren’t embracing this technology from the outset, we may risk the profession being bypassed by legislation and insurance companies,” she said.
AI will revolutionize EHR
AI will revolutionize and improve the current drawbacks to EHR systems, according to PCON Editorial Board Member and blogger, John A. Hovanesian, MD, FACS.
Most current EHR systems require a great deal of work for doctors to input data, he said.
“The systems that are strong in one area are weak in another. A system that is uniquely created for ophthalmology might not work for an oculoplastic surgeon,” Hovanesian told PCON.
Other systems are flexible and can be used by anyone but require a lot of customization, he said.
“Then, once you’ve customized it, you run into problems with releases in the software because you’ve modified it, and what the company releases in the future may not be capable because you’ve modified it,” Hovanesian added.
One promise of EHR was that clinicians would gain real insight into how patients are doing, he said. AI will be able to streamline and produce faster results for what conditions relate to other conditions and how that data correlate to one another.
“The Office of the National Coordinator for Health Information Technology and Agency for Healthcare Research and Quality have embraced all of AI to identify the best possible opportunities to use this technology for their reasonings and outcomes and to help drive precision medicine,” Perry added.
Further, CMS is aggressively embracing AI, she said, using it for statistical analysis to improve patient care and outcomes in order to have more accurate population statistics to create benchmarks to better deliver evidence-based information to providers and drive cost-effective care.
AI changes the game because “it recognizes patterns and data that would take humans years to assimilate,” Hovanesian said.
The algorithms work at a high speed, looking at different charts with different data that is not entered into structured fields and can make sense of it, he said.
“A computer can understand what different physicians mean when they communicate, and with AI they can start to determine patterns,” Hovanesian said. “They can identify certain drugs or diseases that may be associated with cataracts that we didn’t recognize before because we never asked the question. The computer tells us what questions to ask and finds the patterns that we never thought about.”
AI is currently impacting optometry within diabetic retinopathy diagnosis, age-related macular degeneration diagnosis and progression, and glaucoma progression analysis.
ForeseeHome for AMD
“We have many patients with conditions and a lack of practitioners to take care of them. We need a streamlined approach to address the problems, to work more efficiently, faster and smarter, and AI is the cornerstone to make that happen,” Rishi P. Singh, MD, staff surgeon at the Cole Eye Institute and assistant professor of ophthalmology at the Lerner College of Medicine in Cleveland, told PCON in an interview.
The ForeseeHome at-home monitoring device has been shown to reduce the decrease in visual acuity in age-related macular degeneration patients who are identified as progressing at home vs. through standard office procedures, according to Singh.
The technology uses an AI platform to detect changes in a patient’s vision over time using a hyperacuity perimeter, he explained.
“It uses AI to determine changes or progression in that break in line over time and can determine whether a patient has transitioned from dry to wet disease,” he said.
This device helps improve the current standard of care in AMD, where there is typically a 4- to 6-month gap between patient visits, according to Damon Dierker, OD, FAAO, who practices consultative optometry in Indiana.
Singh said it has revolutionized AMD management.
“It allows you to monitor vision remotely, and you’ll notice patients coming to the office with better visual acuities,” he said.
Singh said that one big correlation between improvement with therapy and outcomes is the presenting visual acuity. Chances of preserving vision are much better when patients come into the office in an earlier time frame.
With at-home monitoring, patients can identify vision loss sooner, get a diagnosis and see a retina specialist for treatment, Dierker added.
In exudative (wet) AMD, most patients’ presenting visual acuity is typically 20/80 to 20/100, he said.
“If we can detect and monitor with ForeseeHome we can have 94% of patients maintain 20/40 vision at the time of their diagnosis, which was shown in the HOME study,” Dierker said.
An arm of AREDS2, the HOME study involved 1,500 patients with AMD, he said.
Half of the patients received the current standard of care for AMD: an in-office visit every 6 months and an OCT each visit looking for signs of conversion to wet AMD. These patients were also given an Amsler grid to monitor their vision at home and were directed to call the office if they notice any change in the grid.
The other half were given the ForeseeHome device for at-home monitoring.
In patients who progressed to wet AMD, only 60% in the Amsler grid group had 20/40 vision or better at documentation.
In those using the ForeseeHome device regularly, 94% of patients had 20/40 better at time of their diagnosis.
“They actually stopped the study early because of positive efficacy,” Dierker said. “ForeseeHome was so much better than standard care that the NEI researchers said it would be unethical to offer Amsler grid to these patients because ForeseeHome works so much better at detecting changes earlier.
“As an optometrist, I won’t be doing anti-VEGF injections, but if I can identify that patient when they present, at the earliest possible time, I can have a huge impact on their long-term visual prognosis,” he continued.
The device is covered by Medicare, and if a patient has supplemental insurance, they often do not pay anything, Dierker said. Patients with Medicare and no additional coverage would pay $10 to $15 per month, “which is reasonable in my opinion, and patients see that as being a relative bargain to preserve their vision.”
Dierker said he has taken some high-risk patients from a 4-month follow-up to a 6-month follow-up if they use the ForeseeHome.
“So, two visits per year instead of three,” he said. “In my opinion, minimum standard of care for intermediate AMD follow-up is 6 months exam with OCT, and I’m not quite ready to move away from that yet.”
IDx-DR in diabetic retinopathy
The University of Iowa is using AI to screen for diabetic retinopathy.
IDx-DR can be easily operated by a medical assistant and is a quick screening tool, Yumi Imai, MD, Fraternal Order of Eagles Professor for Diabetic Research and associate professor at the University of Iowa’s Carver College of Medicine, told PCON.
When patients with diabetes visit the diabetes clinic for glucose control, health care providers identify those who have not had a retinal exam in more than a year and those without a previous diagnosis of eye disease, and these patients are offered IDx, Imai said.
“Clinicians at the diabetes clinic have been very happy that we can make sure patients get appropriate screening for diabetic eye disease,” she said. “Patient feedback has been positive overall. It only takes an extra 10 minutes, and they get a diagnosis on the spot.”
Her group is expecting to increase availability of IDx to primary care and family medicine.
“We’ve had several patients who were found to have retinopathy, but those were usually ones who have had poorly controlled diabetes for many years,” she said. “It is reassuring that the diagnosis made by IDx matches our clinical suspicion of retinopathy.”
Eyenuk’s impact in Asia
Pattern recognition software from Eyenuk recognizes the natural configuration of the blood vessels in the back of the eye much like similar software uses facial recognition technology to identify a face to unlock an iPhone, according to Roger V. Ohanesian, MD, an anterior segment surgeon and founder and president of Harvard Eye Associates. He is also founder of the Armenian EyeCare Project.
By using photography and pattern recognition, the technology can identify a clot, hemorrhage, microaneurysm and fatty exudate.
“These are things that the pattern recognition looks for and sometimes sees even better than a human eye,” he said.
Eyenuk’s EyeArt technology combines automated image analysis tools with a user-friendly telemedicine and cloud-based interface that allows for faster screening of more diabetic patients, Ohanesian explained.
The Armenian Eye Project is a large endeavor with the goal of eliminating preventable blindness in the country. The project aims to photograph every patient in the country with diabetes through five regional eye centers and a mobile eye hospital. It takes the mobile eye hospital 2 years to go around the country, which it has done seven times thus far, Ohanesian said.
At the regional eye center, tabletop cameras are used, he said, and portable cameras are used when necessary. The photographs are uploaded to the cloud, and AI reads the images and identifies whether a patient has diabetic retinopathy.
Ohanesian said that practitioners from all over the world have volunteered for the project, using their own money for travel and room and board.
“Even though I bring doctors from all over to Armenia, the rule I have is if the case is unique, the international specialists operate. If it’s a more common case, I encourage the Armenian doctors to handle it and the international doctors assist,” Ohanesian said. “That way they learn, and it’s gotten to the point where I no longer need to operate.”
He said now numerous patients from other countries are sent to Armenia, and other doctors go there to learn about imaging, diagnosis and treatment. Aside from being a medical and social mission, the economic implication of preserving patients’ vision is astounding, he said.
“We’ve treated close to 6,000 people. Some have never had their eyes examined ... you can imagine the level of disease present,” Ohanesian said.
Aside from diabetic retinopathy screening, they also run programs for amblyopia screening for children, corneal transplant, eyeglasses, prosthetic eyes, a mobile eye hospital and subspecialty clinics.
“I look at this as medical diplomacy,” he added.
The U.S. Agency for International Development has identified the project as a Center of Excellence.
Zeiss pilot program for primary care physicians
Cynthia Matossian, MD, FACS, an ophthalmologist who practices in New Jersey and Pennsylvania, is working on a collaboration with Carl Zeiss Meditec for a pilot program to improve detection of diabetic eye disease in primary care offices using a handheld non-mydriatic fundus camera with AI software.
The portable camera is about the size of a trumpet, and one camera is adequate for an office with multiple physicians, she said in an interview.
The medical assistant or nurse would photograph the patient’s retina when taking vitals such as height, weight and blood pressure, she explained.
“An image can be captured in less than a minute,” Matossian said. “It’s very simple to do, and the learning curve is not steep.”
The image is then uploaded to a HIPPA-secure platform that Zeiss is developing, then sent to a reading center.
“It’s been done in India with great success in more rural areas that don’t have ophthalmologists or retinal specialists,” she said. “We are hoping to capture an entire subset of patients diagnosed with diabetes who are not getting adequate eye care despite recommendations to do so by their primary care providers. Instead of begging that they schedule eye care visits, we can screen them in the primary care offices.”
The first pilot program is underway at University of North Carolina at Chapel Hill.
Retina-AI releases mobile app
Retina-AI recently announced its development and release of an AI mobile app for eye care providers called Fluid Intelligence. Using machine learning AI algorithms in a cloud, the app detects macular edema and subretinal fluid on OCT retinal scans with greater than 90% accuracy, according to a press release from the company.
Retina-AI founder and CEO, Stephen G. Odaibo, MD, MS, is a retina specialist with advanced degrees in mathematics and computer science.
Moving forward, working more efficiently
Catania recommends that practitioners follow the literature, “in tabloids and scholarly journals, as changes are occurring on a weekly and monthly basis. It’s critical now to stay abreast of AI in your field.”
Dierker said he looks for three things when adopting a new technology in his practice: Is there evidence that it will impact patients in a positive way? Does it have reasonable costs and affordability? Will the patient buy into it and use it appropriately?
“I feel that we are at the threshold of AI in as much as health care is driving efficiency and lowering cost and at the same time focusing on outcomes and better patient care,” Perry said. “I see clear opportunity for this to fit into health care.”
Population management and population data will be streamlined and easier to understand, according to Perry. In addition to analyzing the full scope of chronic conditions, AI will connect how treatment options have increased outcomes and reduce the financial burden on patients.
It also empowers patients to take control of their own health by using home health equipment, Perry added.
“For those more skeptical of AI I think the fear and concerns can be set aside because we are using it to a limited degree already, like in OCT and normative databases,” she said.
AI takes the information, consolidates it to one database and allows the computer to do the work for providers so they have more information, faster, at the point-of-care, to reduce the amount of time spent in nonpatient activities so they can focus more on the care provided, she concluded. – by Abigail Sutton
- References:
- Chew EY, et al. Ophthalmology. 2014;doi:10.1016/j.ophtha.2013.10.027.
- Peng L. A glimpse into the future: Artificial intelligence and eye care. Presented at: Prevent Blindness Focus on Eye Health National Summit; Washington; July 18, 2018.
- For more information:
- Louis J. Catania, OD, FAAO, is in private practice at Nicolitz Eye Consultants in Jacksonville, Fla. He can be reached at: lcatania@bellsouth.net.
- Damon Dierker OD, FAAO, is director of optometric services at Eye Surgeons of Indiana. He can be reached at: damon.dierker@esi-in.com.
- John A. Hovanesian, MD, FACS, is faculty member at the UCLA Jules Stein Eye Institute and practices at Harvard Eye Associates in Southern California. He can be reached at: jhovanesian@harvardeye.com.
- Yumi Imai, MD, is Fraternal Order of Eagles Professor for Diabetes Research and associate professor in the Department of Internal Medicine at the Fraternal Order of Eagles Diabetes Research Center, University of Iowa Carver College of Medicine. She can be reached at: yumi.imai@uiowa.edu.
- Cynthia Matossian, MD, FACS, is founder and medical director at Matossian Eye Associates in New Jersey and Pennsylvania. She can be reached at: cmatossian@matossianeye.com.
- Roger V. Ohanesian, MD, practices at Harvard Eye Associates in Southern California. He can be reached at: rogerohanesian@gmail.com.
- Lily Peng, MD, PhD, is a non-practicing physician and product manager for a team at Google that works on applying deep learning and other Google technologies and expertise to medical imaging. She can be reached at: lhpeng@google.com.
- Karen F. Perry, OD, FAAO, is co-owner of the Vision Health Institute in Orlando and director of professional relations at Compulink Healthcare Solutions. She can be reached at: KFP@compulinkadvantage.com.
- Rishi P. Singh, MD, is staff physician at Cole Eye Institute in Cleveland, medical director at Cleveland Clinic and associate professor of ophthalmology at Case Western University. He can be reached at: drrishisingh@gmail.com.
Disclosures: Catania wrote a chapter on AI in Advances in Ophthalmology and Optometry, which will be published in August. Dierker served as a consultant for Notal Vision. Hovanesian consults with ForeseeHome and Zeiss and is founder of MDBackline. Imai reports no financial disclosures. Matossian consults for Zeiss. Ohanesian is founder of the Armenian EyeCare Project. Peng is employed by Google. Perry is director of professional relations at Compulink. Singh consults for Zeiss and Optos.