Purchase this article with an account.
N.P. Cottaris, S.D. Elfar, R. Iezzi, G.W. Abrams; Development and Computational Evaluation of a Neurophysiologically–Based, Real–Time Image Encoder for Retinal and Cortical Visual Prostheses. . Invest. Ophthalmol. Vis. Sci. 2004;45(13):4203.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose:Previous studies related to the design and performance of image encoders for visual prostheses have assumed that the percepts evoked by electrical stimulation are pixelized versions (phosphenes) of the input image. However, such pixel–based image encoders are of limited use for visual prostheses that strive for pattern vision and target neurons downstream from the photoreceptors, e.g., retinal ganglion cells and, in particular, visual cortex neurons. In the present work we engineer and evaluate a real–time image encoder for artificial vision that incorporates the functional properties of the targeted neurons. Methods:To achieve this goal, our encoder computes a family of stimulating signals based on a quasi–linear relationship between the spatiotemporal image structure and the targeted neurons' estimated spatiotemporal receptive field (RF) structure. To evaluate the encoder's performance, we compute reconstructions of the input image from the mosaic of activated neurons. In these reconstructions, we assume that each neuron's contribution is determined by its RF structure and the amount of stimulation it receives. Results:We implemented the image encoder on a laptop computer (1GHz, Apple PowerBook) interfaced with a webcam (iSight). For frame sizes of 64x64 pixels, we obtained a real–time video encoding of 15–20 frames/sec, depending on the number of encoding filters. For objects of medium complexity (e.g., letters), we obtained recognizable reconstructions for electrode array sizes greater than 8x8. However, the quality of these reconstructions is heavily influenced by the number of stimulated neurons per electrode and by the RF characteristics and functional organization of the targeted tissue. Conclusions:We have engineered a real–time, off–the–shelf–hardware–based, neurophysiologically–inspired image encoder for retinal and cortical prostheses. We are currently testing the dependence of the quality of image reconstruction on a variety of factors, such as RF structure and peri–electrode spatial spread of stimulation, with special focus on factors related to the functional organization of the visual cortex (orientation and spatial frequency maps, RF size & RF position scatter, and magnification factor). Our results are instrumental in optimizing the stimulating parameters and electrode design for retinal and cortical visual prostheses that endeavor to achieve better than phosphene–based vision.
This PDF is available to Subscribers Only