Visual Prosthesis Learns to Identify Implant Signals

The encoder, which is being tested on sighted volunteers, learns in a continuous process how to change the camera output signal so that the respective patient can perceive the image. Initially, the Retina Encoder does not know which language the individual's virtual central visual system "speaks.” Therefore, the software translates the original picture--for example, a ring--in different, randomly selected dialects. The volunteer sees these variations on a small screen that is integrated in a frame of glasses. By means of head movements, he selects those variations that appear most similar to a ring.

From these choices, the learning software draws conclusions on how to improve the translation. In the next learning cycle, several new improved picture variations are presented, which are already more similar to the original: during this process, the Retina Encoder becomes adapted step-by-step to the language of the virtual central visual system. In normally sighted humans, a kind of natural retina encoder is already integrated in the retina: specifically, four layers of nerve cells are positioned in front of the photoreceptor cells.

The scientists added that in the brain, the complex information is decoded, and the brain acquires the corresponding ability within the first months of life. During this time, the central visual system becomes individually adjusted to the retinal signals: the brain learns how to interpret the data from the optic nerve. In adults, however, who become blind later in life, the central visual system has already matured, and is not as flexible.

"The retina is a transparent biocomputer,” said Prof. Rolf Eckmiller, from the department of computer science at Bonn University. "It transforms the electrical signals of rod and cone photoreceptors into a complex signal. This signal reaches the brain via the optic nerve. The artificial retina must learn to generate signals that are useful for the brain.”


Related Links:
Bonn University