Visual perfromance under simulated conditions of prosthetic vision

Purpose: To assess the ability of future retinal and cortical implant recipients to perform activities of daily living. Methods: Crude pixelized images, either computer generated or transformed from live video, were used to simulate prosthetic vision properties. Pixel number, size, density, jitter, random dropout, and contrast range and resolution were varied (from 4×4 (2° each) to 32×32 (15 arcmin each) dots, 10-70% dropout, 15-100% contrast, 2-8 gray levels), as well as spatiotemporal properties of background noise (SNR=0.2-5). Both real-world tasks and tasks in a virtual environment were presented to viewers in a video headset, and zoomed to optimize performance. Real-world tasks included 2D/3D object and face recognition, paragraph text reading, object manipulation, and mobility. VR tasks included mobility and object location and manipulation. Subjects were normally sighted or visually impaired. Error score and time to completion were used as task performance parameters; time constant towards plateau performance was used as an overall performance measure. Results: Using 2° rasters of 4×4 dots allowed only cumbersome recognition of simple shapes, and slow and awkward mobility, with frequent contacts; training improved performance, but it remained extremely slow. Increasing raster size to 6×10 greatly improved wayfinding, shape and letter recognition, and object manipulation, but extensive training was still required, especially for effective camera scanning. For rasters over 10×10 dots, all tasks could be learned, albeit that facility was not generally achieved with less than 16× 16 dots. For these larger rasters, dropouts <=50%, contrast >=15%, jitter RMS <=50% of dot diameter, and grayscale resolution >=4 levels do not appreciably hamper most tasks. Conclusions: Visual prostheses with as few as 100 electrodes may be effective, but their success will critically depend on task-specific rehabilitation programs. Moreover, these simulations may overestimate prosthesis wearers' ability to process information from many individual dots in parallel.