Navigation outcomes with multimodal vision processing in a suprachoroidal retinal prosthesis.

Abstract

Purpose : The second-generation (44-channel) suprachoroidal retinal prosthesis has been shown to improve object localisation and navigation in recipients with end stage retinitis pigmentosa (RP) using a comprehensive vision processing method (Lanczos2; L2). However, detecting objects of low contrast and identifying them remains difficult. A multimodal (MM) vision processing approach has been developed to enhance face (FaD), empty chair (ChD) and depth detection (LBE). We assessed whether using a combination of MM and L2 improves navigation and object identification during obstacle course testing in comparison to using L2 alone.

Methods : Four implant recipients (#NCT05158049) with profound vision loss due to RP were acclimatised to all vision processing methods (FaD, ChD, LBE 4m with depth attenuation and Lanczos 2 Dark Mode) prior to being asked to detect objects and navigate through an obstacle course. The seeded obstacles included mannequins, overhanging boxes, bins and chairs with each object was either in black or white colour. One object was positioned randomly in each of six rows in one of three positions, six objects were used per trial. The obstacle course background was white. Participants were masked to the number and type of obstacles. During the MM trials, participants were permitted to select from the range of vision processing algorithms available to them. Performance of visual function tasks were compared between MM and L2 alone.

Results : The total number of obstacles detected in the obstacle course with L2 was 45.4%, compared to 72.0% with MM (p<0.001). The total number of obstacles contacted with L2 was 39.2% compared to 23.7% with MM (p<0.001). Correct identification of mannequin faces (p<0.001) and empty chairs (p<0.001) were both significantly better with MM compared to L2.
Correct identification of white objects with MM (66%) was significantly better than L2 (5%, p<0.001). Identification of black objects with MM (71%) was also significantly better than L2 (53%, p<0.001).

Conclusions : The MM vision processing approach improved navigation through a standard obstacle course with black and white objects with more obstacles detected and less contacted than with L2 alone. Improved detection and identification of low contrast obstacles (white obstacles was against white background) with MM provided the participants better orientation and mobility with their suprachoroidal retinal prosthesis.

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