This study aims to improve the apparent motility of ocular prosthetic devices using technology. Prevailing ocular prostheses are acrylic shells with a static eye image rendered on the convex surface. A limited range of ocular prosthetic movement and lack of natural saccadic movements commonly causes the appearance of eye misalignment that may be disfiguring. Digital screens and computational systems may obviate current limitations in eye prosthetic motility and help prosthetic wearers feel less self-conscious about their appearance.

We applied convoluted neural networks (CNNs) to track pupil location in various conditions. These algorithms were coupled to a microscreen digital prosthetic eye (DPE) prototype to assess the ability of the system to capture full ocular ductions and saccadic movements in a miniaturized, portable, and wearable system.

The CNNs captured pupil location with high accuracy. Pupil location data were transmitted to a miniature screen ocular prosthetic prototype that displayed a dynamic contralateral eye image. The transmission achieved a full range of ocular ductions and with grossly undetectable latency. Lack of iris and sclera color and detail, as well as constraints in luminosity, dimensionality and image stability limited the real eye appearance. Yet, the digitally rendered eye moved in the same amplitude and velocity as the native, tracked eye.

Real-time image processing using CNNs coupled to microcameras and a miniscreen DPE may offer improvements in amplitude and velocity of apparent prosthetic eye movement. These developments, along with ocular image precision, may offer a next-generation eye prosthesis. NOTE: Publication of this article is sponsored by the American Ophthalmological Society.