When implementing AI, it is useful to think of the technology as an adjunct to human staff.
Physicians have notoriously high workloads, and with a provider shortage on the horizon, that situation is not likely to improve.
Delegating responsibilities to staff and establishing partnerships with optometry can help, but the best hope for better balance in ophthalmology might be found in technology. With AI tools for scribing and screening, as well as digital surgical ecosystems, there are opportunities for ophthalmology practices to find efficiencies in everyday tasks, according to Eric D. Rosenberg, DO.
“I think AI scribing is important, but that AI integration into a lot of repetitive tasks we do is also helpful,” he said. “AI manning our front desks and the telephone lines is primed to do well. Callbacks after surgery is a great area for AI.”
Rosenberg said his practice recently started using AI for front desk processes such as managing appointments. He is working with a company called AVTR Med that has an AI solution for those processes and others.
“They’re using it for a personal assistant, too,” Rosenberg said. “Doctors, when they’re on the fly, are always managing. There are a lot of balls in the air, and the personal assistant helps coordinate schedules and other things like that. It’s an elegant solution.”
In addition, conversational AI phone systems can take the place of traditional number dial directories, freeing up valuable time for human staff to provide help where it is needed most. Companies such as OptiCall use AI to reduce long call queues that can be overwhelming for patients who want to book an appointment. Routine calls about a practice’s address or office hours can be answered directly by the AI program, keeping those calls out of the stream and preserving a place for higher-value appointment bookings. Roseberg said these phone systems can make a difference in call volume.
“Say a patient calls and wants to make an appointment for a pediatric ophthalmologist,” he said. “The system can comb the schedule and see when those openings are. If there is anything that the system feels unable to handle, then it prompts the actual front desk team to answer the phone. There are a lot of things that can be done without my team having to drop what they’re already doing.”
Rosenberg said AI systems for practice management are best used as an adjunct to human staff.
“It’s augmented intelligence, not artificial intelligence,” he said. “I don’t think it replaces anyone yet. It augments and streamlines the flow. If you look at it like that, it’s going to be a lot more useful to you.”
In his experience, Rosenberg said AI has been helpful in filling gaps. Training staff takes time and effort.
“There’s been a lot of turnover with the staff, and now that I’m training people more frequently than I used to, it’s hard to do that in large numbers. When I get one person up to speed, I feel confident in that,” he said. “Now, having this augmented technology, it helps that person do their job. Instead of training two people at the same time, I’m training one person to do it right, and the AI almost becomes an assistant for them, not just for me.”
Rosenberg said it is important to remember that like any new technology, AI — and how it will be used in medicine — is still a work in progress. Unlike in sci-fi movies, these systems are not going to immediately replace people.
“AI is going to refine our system. It is not going to replace our system,” he said.
Practices will not lay off staff if they implement AI. Rosenberg said to think of AI as a way to help staff while maintaining reasonable expectations.
“It’s important to realize that these things are just coming out onto the scene,” he said. “It’s like when we were dealing with multifocal IOLs when they first came out. It is a mistake to come out and promise that it’s going to do everything under the sun. It’s important that we learn what the limitations are going to be with these systems and where we still need to bolster our own practices to make sure we have good coordination.”
Other digital solutions beyond AI are designed to streamline workflow and improve outcomes. Rosenberg uses Alcon’s SMARTCataract system, an intraoperative solution that is compatible with Alcon’s Argos biometer and Ngenuity 3D visualization system.
“The digital suite allows you to take pictures when patients are in a seated position, and it finds where the visual axis is,” he said. “It marks on the patients, using the Ngenuity camera, where to place a toric lens or where to center a multifocal lens. It has taken these technologies that we already have and allows them to work together. It heightens the experience of providing premium services to patients.”
Zeiss also has digital workflow systems for cataract, glaucoma, retina and corneal refractive surgery.
For Rosenberg, the SMARTCataract system quickly became essential to his practice.
“I don’t know what I did before because picking out IOLs was very manual,” he said. “I needed the full charts. … Now, it’s just on my iPad.”
Rosenberg can access patient data and electronic health records and pick out IOLs all in the same place. The system also helps him make changes if patients change their mind about a lens.
“If I already had a lens picked out and they decided they wanted to do a multifocal, traditionally, I would have to go back into the EHR and a physical chart,” he said. “I’d have to call my surgical coordinator and make sure I had the IOL the patient wanted. Now, I just go into the software, select the IOL they want and change it on the fly. It automatically alerts my surgical coordinator, and it’s done. It’s a nonissue.”
In the future, Rosenberg hopes that AI can be a tool to level the playing field with insurance companies. As insurance companies started rolling out AI programs to cut costs, he said the onus was put on practices to figure out the process.
“Now, I think the time is ripe that doctors can start fighting back with technologies of their own,” he said. “We can start leveraging that in medical practices, let their AI talk to our AI. Why are we wasting our time as doctors trying to solve their issues? Insurance companies need to be prepared for that because their policyholders are going to demand better. Hopefully coming up in the next year or two, physicians will no longer bear that load. It’s going to be up to our insurance companies to be able to care for their patients and their policyholders.”
While AI is just taking hold in the office, AI screening programs are already making a difference in places where they are needed the most.
Richard Hill, MD, a veteran of the Armenian EyeCare Project, has made numerous trips to deliver care and elevate eye care over his career. A few years ago, the program started a partnership with a company called Eyenuk to bring AI diabetes and glaucoma screening to Armenia.
“Their AI obtains images from a Crystalvue camera,” Hill said. “We have 30 of them stationed around Armenia, and we’ve been happily screening patients for glaucoma and diabetic retinopathy.”
Hill and colleagues previously screened more than 40,000 people for diabetic retinopathy, and with the help of an engineer, he said the Armenian EyeCare Project was able to scan more than 11,000 images and start screening for glaucoma.
“When we first started looking at them, there was a 9% flag rate, which is good,” he said. “It greatly cuts down on patients needing to be seen, but 9% is still too high. Eyenuk tweaked the software, and we were able to get the flag rate down to about 6.5%.”
Once an image is flagged, a human looks at it to decide if a referral is needed. Hill said there is about a 50% referral rate on flagged images.
“You’re down to about 3% of 11,000 patients, which is a lot more manageable and preserves one of the core arts of medicine, diagnosis,” he said.
While this system has seen success on a small scale in Armenia, Hill thinks that it could be brought to a wider population in the United States. Simple screening devices could make the process more accessible.
“One of my favorite dreams was to have a little machine in a shopping mall where people could walk up and get screened,” Hill said. “If you could screen millions of people, that would be the ultimate goal and get this down to a manageable amount for referral.”
Cutting back on physician hours used for screening is more important than ever, Hill said. Time is limited, and funding might not always allow for a physician to reach every patient, but AI screening offers a way to optimize use of available resources.
“A technician does this in Armenia, and the whole idea is to free up MD and OD time,” Hill said. “A technician takes the photos and uploads them to the cloud. They get downloaded, looked at and analyzed. It’s basically an augmentation of an MD or OD to find the disease and get it funneled in.”
If he were asked to look at the 40,000 images from the Armenian EyeCare Project, Hill estimated that it could take 1 to 2 years on top of normal duties. Instead, there is a roving fleet of 30 cameras across the country to find patients with glaucoma, he said.
“I think this is where this can shine,” he said. “There are people who ordinarily wouldn’t get the attention or care they need, but AI can flag them and say, ‘Hey, wait a minute. This looks funny. Maybe we should send this back to a human for evaluation.’ You can save a lot of MD resources.”
Hill also envisions a place for AI in screening for other eye conditions, including cataracts. Although the program was not trained to look for cataracts, he noticed some degraded images that were getting flagged. Often, he would send the patient for cataract surgery.
“It might not be so critical for the Unites States and Europe, but in other countries, you could make an announcement and bring this screening to remote villages. Once a month, the eye picture machine is coming. People can get their pictures taken, and if they have a problem, they can be sent to a larger city for care,” he said.
There might be screening opportunities in less common diseases, but Hill said it is better to focus on the bigger problems first.
“The populations are best served to target diseases that are more common at first,” he said. “Then, when those are under control, you can program in more esoteric diseases. Any retinal disease that has fundus changes could be looked for.”
To be truly beneficial as an eye health care assisting tool, Michael Assouline, MD, PhD, believes that AI needs to combine both subjective data from a smartbot and multimodal objective data from (robotized) medical devices, as well as be full range in order to cover every part of the eye.
“In ophthalmology, we know that everything interacts,” he said. “Just because you have a cataract doesn’t mean that you cannot have something else. You can have glaucoma. You can have a retina condition. You can have dry eye. You can have keratoconus. If you want to deal with your cataract seriously, you need to have information on everything. Otherwise, you’re doing approximate work.”
With this in mind, Assouline started a company called Mikajaki that leverages AI, a smartbot and an all-in-one machine to provide a comprehensive visual exam and medical intelligence in only a few minutes. With the EyeLib station, patients put their face into an opening on a large cabinet. Screening devices are rotated in and provide a full analysis without the need for repositioning. The machine integrates several diagnostic modalities: anterior and posterior spectral domain OCT with high-definition epithelial mapping, biometry, retinal ganglion cell layer, retinal nerve fiber layer, retinal imaging, refractometry, aberrometry, elevation topography, Scheimpflug analysis, retroillumination and tonometry, according to the company website.
The results and objective symptoms collected by a pre-appointment chatbot are sent to the AI component, which provides a comprehensive readout of the results. The process takes about 6 minutes.
This project started in 2017 when Assouline discussed the future of ophthalmology with his future associate Jacky Hochner. The decades-old model of one-on-one, face-to-face interaction at a slit lamp seemed outdated, especially with the current and predicted physician shortage, he said.
“We both had in mind that the machines were doing more because a lot of diagnoses are dependent on machines, not on humans,” he said. “If you have a retina disease, the OCT will tell you, not the ophthalmologist. It seemed logical to harness the new technologies that were available off the shelf and use them in a new way.”
Once an analysis is complete, Assouline sits with the patient to go over the results, which are laid out on a panoramic screen. The output of 30+ algorithms based on this comprehensive dataset suggests assisting the doctor to prescribe glasses, identifying common and even rare diagnoses, recommending follow-up and selecting the most appropriate procedure from a list of 39 eye surgeries for each patient. The AI calculation module compares and combines the most advanced AI formulas to plan implantation of premium IOLs, intracorneal ring segments or phakic IOLs.
“We are full range, which means we cover every aspect of ophthalmology, including the anterior and posterior segment, cornea, cataracts, calculation of the IOL, retina and glaucoma as opposed to most of the AI projects that deal with just one disease,” Assouline said. “We’re also multimodal, which means that we use more than one source for each condition. For instance, for keratoconus and glaucoma detection and follow-up, we use up to seven different data sources.”
Mikajaki currently has 30 EyeLib stations installed in 14 countries. Assouline said the station can streamline the workflow process to ease patient backlog and get people the care they need.
One major Spanish eye center, La Paz hospital in Madrid, had a 2-year backlog of patients.
“The distributor of our machine didn’t sell the machine to the hospital,” Assouline said. “The distributor bought the machine, paid the technicians and sold the exam to the hospital. By doing that, they were able to send 43% of the patients back home without seeing a doctor and double the flow seen by specialists. Once the exam comes back normal, you don’t have to see the patient.”
Patients could be directed to optometry for new glasses, and even in cases of cataract surgery, 13% of patients could proceed to surgery without a second exam, Assouline said.
The prospect of AI in eye care is no longer just theory, Assouline said
“We cannot escape this problem,” he said. “A lot of people are afraid to use AI or machines because they say, ‘All right, I don’t have a job anymore.’ But in fact, those who don’t use machines and AI won’t have a job at all. I think it’s something we cannot avoid.”
For AI to be taken seriously in medicine, it must be comprehensive and transparent. EyeLib includes a feedback loop in which physicians can help train the AI.
“You have to give the doctor a way to control why AI made a suggestion. You cannot have a black box,” he said. “Having this feedback loop and this transparency of AI is an important way to build confidence for the doctors and the patients.”
Michael Assouline, MD, PhD, of Centre Iéna Vision in Paris and Mikajaki, can be reached at [email protected].
