Abstract
Purpose :
Recent developments in artificial intelligence (AI) are being applied for assistive technology (AT) for people with vision impairment. Whereas evidence-based recommendations are imperative to increase the appropriateness of AI-AT referrals, the benefit for people with vision impairment has not been thoroughly studied. To aid this process, we compared the performance and usability of four AI-based devices and mobile apps (i.e., Orcam, Envision, Seeing AI and Lookout) to a baseline condition with no device.
Methods :
A cross-sectional, individually controlled, counterbalanced study was performed (n=30). Main outcome measures were objective performance on vision-related activities of daily living, with eight reading tasks, and seven searching and identifying tasks. Usability was assessed with the System Usability Scale. To analyze the data, Logistic Regression Analyses and Friedman One-way Repeated Measure Analyses of Variance by Ranks were used.
Results :
Thirty participants (65% female) with vision impairment due to a variety of eye diseases were included. For reading a newspaper article and invoice, participants performed significantly better with the devices/apps than baseline (p≤0.05). For reading a table of content, participants performed significantly better with Orcam and Seeing AI than baseline. For reading a handwritten grocery list, street sign, dollar bill and medicine bottle, participants performed significantly better with Envision, Lookout and Seeing AI compared to baseline. For reading a TV guide, none of the devices/apps performed better than baseline. Participants were not able to recognize faces, color or objects with Lookout and they were not able to recognize a landscape, object or identify a room with Orcam. For identifying a room, participants performed significantly better with Envision, Lookout and Seeing AI compared to baseline. For identifying a landscape and color, participants performed significantly better than baseline when using Seeing AI. Usability scores reflected performance outcomes.
Conclusions :
Whereas performance on various reading tasks was improved by several devices/apps compared to baseline, performance on several searching and identifying tasks was worse than baseline for Lookout and Orcam. Professionals and patients could use the outcomes of this study in clinical practice to determine when a particular device might be helpful.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.