AI Smart Glasses Show Promise, Limitations as Aid for Patients With Low Vision

Artificial intelligence (AI) smart eyeglasses as a potential aid for people with low vision produced mixed results in a small pilot case series.

Using verbal prompts to guide the eyeglasses, study participants, all healthy volunteers, easily recognized common objects on a white background and could tell whether an object was horizontal, but color discrimination proved to be more challenging. Accuracy for reading standard text fell below 60% but increased to about 90% for handwriting and children’s books. The participants were mostly spot on for recognizing paper money, but coin recognition was poor.

Overall, the results suggest that smart glasses might have a place in helping people with low vision, reported Carol Shields, MD, of Wills Eye Hospital and Thomas Jefferson University in Philadelphia, and colleagues in JAMA Ophthalmology.

“AI smart eyeglasses may offer a unique intervention for patients with low to no vision, performing best for identifying common objects, identifying neutral colors, and reading children’s books,” the authors concluded. “AI smart eyeglass users should be aware of current limitations, which might improve as technology evolves in this field. Further studies are needed to define these benefits and limitations with patients who have low to no vision.”

The authors of an accompanying commentary agreed with the authors’ assessment of the smart glasses’ limitations and added some of their own, but nonetheless found the study and results compelling.

“These findings represent a potentially transformative advance in assistive technology for patients with low or no vision,” wrote Benjamin K. Young, MD, of Casey Eye Institute at Oregon Health & Science University in Portland, and Peter Y. Zhao, MD, of the Kellogg Eye Center at the University of Michigan in Ann Arbor. “Yet several barriers to clinical endorsement remain.”

“The study participants were fully sighted, and task-specific accuracy was measured rather than real-world function. The participants were authors of the investigation; in addition, they did not test whether patients with true low or no vision may perceive enough to know when to prompt the glasses or to orient the camera appropriately,” they noted. “Further, there was no detailed description of the types of errors the AI made, which could range from inconsequential to dangerous.”

Poor color discrimination of green, blue, and violet came as a surprise, Young and Zhao continued. Whether the issue reflected hardware or software limitations remained unclear.

The study was conducted with Ray-Ban Meta AI glasses, which are not the only ones on the market. Moreover, some of the advances of smart eyeglasses can be accomplished with smartphones, Young and Zhao pointed out. Unlike smart glasses, smartphones do not have the advantage of being inconspicuous, wearable, and hands free, but they may reduce the cost barriers associated with purchasing an entirely separate device. Smart glasses pose privacy issues for patients and bystanders, as the data associated with photos or videos taken with the glasses are sent to Meta’s cloud for processing.

Limitations notwithstanding, “patients living with low vision or blindness may not be waiting for a perfect device,” Young and Zhao wrote in conclusion. “They are navigating a world designed for the sighted, sometimes without the tools they need to do so safely and independently. If the promise of wearable AI can be validated, refined, and made equitably accessible in a manner with strong ethical guardrails, it has the potential to be among the most meaningful advances for those patients for whom ophthalmologists have fallen short.”

Smart glasses can perform facial detection, object identification, and virtual reality interaction, Shields and colleagues noted in their introduction. The Meta smart eyeglasses capture images and video, record audio, and respond to voice commands. Lens options include single and progressive prescription lenses in clear, blue-violet light filtered, tinted, or transition options.

In their pilot study, the authors evaluated the Meta AI smart eyeglasses for the identification and description of stationary objects, as a reading aid, and identification of U.S. currency. Six of the authors served as study participants — three male and three female.

All of the tasks were novel and designed to simulate everyday tasks that can be performed from a seated position. None of the tasks have been validated for specific testing categories but served as piloting for feasibility in later studies. Trials, tasks, and objects were randomized for each participant, and participants completed at least five trials of each task. The results showed the following accuracy for each task:

• Common object identification (700 trials): 99%
• Color discrimination (450 trials): 64%
• Object directionality (300 trials): 83%
• Object counting (400 trials): 50%
• Primary colors (150 trials): 59% (red and yellow 86%; blue 6%)
• Secondary colors (150 trials): 44% (orange 82%, green 50%, violet 0%)
• Neutral colors (150 trials): 87% (white 96%, black 92%, brown 74%)
• Pencil counting (200 trials): 73% (decreasing as numbers increased from 100% for one to 32% for 10)
• Rice grains counting (200 trials): 27% (78% for one grain, 6% for 50 grains, 0% for 100)
• Reading standard text (100 trials): 59% (36% for medication labels, 82% for food labels, 100% for medication bottles)
• Handwriting reading, neat cursive (50 trials): 88%
• Counting money (350 trials): 53% (91% for paper; 0-6% for dimes, nickels, quarters)

“This study can provide a basis for the applicability of AI eyeglasses to the field of ophthalmology and improve on the understanding of how this technology can assist patients who experience low vision and blindness,” the author concluded.