gms | German Medical Science

Objective: To simulate prosthetic vision in patients with the PRIMA subretinal implant and develop the image pre-processing for improved face recognition.

Materials and Methods: To simulate the sampling limit of resolution imposed by the implant’s pixel pitch, we lowpass filtered the spatial frequencies in the images projected onto the prosthesis. We minimized the ringing artifacts due to the Fourier lowpass filter by apodization using a Tukey window. To simulate the reduced contrast sensitivity and dynamic range of prosthetic vision compared to normal, we compressed the contrast range of the images using gamma correction.

Using such simulated prosthetic vision, we assessed recognition of letters and faces. To improve face perception, we applied image enhancement techniques, including background removal, contrast stretching, and machine learning-based facial landmarking.

Results: Recognition of letters, like the Landolt-C, with simulated prosthetic vision matched the visual acuity measured in AMD patients with PRIMA implants: letters had 100% contrast down to a single pixel per line (equivalent of 20/420 acuity), and recognizable with a reduced contrast up to 30% below the single pixel per line (equivalent of 20/300 acuity). When applied to faces, it reflected the difficulties patients reported in recognizing facial features and identities. Processing the images before application of the prosthetic vision simulator enhanced the visibility of key facial features, such as the nose and eyes.

Discussion: Simulator of prosthetic vision helps sighted individuals understand the patients' visual experiences. Unlike prior models that depict vision as dotted phosphenes, our simulator of prosthetic vision with PRIMA implants aligns with patients’ reports of smooth, continuous, albeit lower resolution, perception of lines and patterns. It also helps illustrate the extent of expected improvement in visual perception with future implants having larger width and smaller pixels.