Camera Testing of an Intracortical Visual Prosthesis for Identifying Occupied and Vacant Chairs

Abstract

Purpose : Individuals who are blind face challenges with daily tasks that rely on vision. The intracortical visual prosthesis (ICVP) aims to restore some visual function by stimulating the visual cortex. The optimal camera input for certain tasks is unknown. This study evaluated an ICVP user’s ability to identify occupied and vacant chairs using various camera modalities.

Methods : An ICVP system of 25 wireless floating microelectrode arrays, each with 16 electrodes was implanted into the right dorsolateral visual cortex of a participant with bare light perception. Performance identifying occupied/vacant chairs was tested in 3 conditions: 1) visible-light camera, 2) infrared thermal camera, and 3) prosthesis input off. In 20 randomized trials per task/condition, the participant sat 2.7m from 3 white chairs against a dark background, with 1 chair occupied for the occupied task and 2 for the vacant task. Accuracy and response time were recorded.

Results : For occupied chair trials, accuracy was 95% for both camera modalities. Response times were significantly faster using the infrared (mean= 6.26 s) than visible-light camera (mean= 8.85 s) (p < 0.05). Accuracy was 25% with prosthesis input off (mean response time = 2.12 s). A Kruskal-Wallis test found a significant difference between conditions (χ2(2) = 31.64, p < .001). Post-hoc tests showed significantly better performance with cameras vs. no input, but no difference between cameras. For vacant chair trials, accuracy was 95% for both cameras, with no significant difference in response times between the visible-light (mean = 6.81s) and infrared (M = 6.60 s) cameras. Accuracy was 20% with prosthesis input off (mean response time =1.49 s). A Kruskal-Wallis test again found a significant difference (χ2(2) = 35.12, p < .001), with post-hoc tests showing significantly better performance with cameras vs. no input, but no differences between cameras.

Conclusions : Both camera modalities achieved highly accurate detection of occupied and vacant chairs, with the infrared camera enabling faster identification of occupied chairs. In contrast, removing camera input altogether resulted in much poorer performance, but faster response times due to guessing. Recognizing occupied/available spaces enables independent and safe navigation. Further testing with more participants is needed to better characterize capabilities across camera modalities and more complex tasks.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.