Intra-cortical visual stimulation system : design and optimization

Jonathan Coulombe, Louis-Xavier Buffoni, Sylvain Carniguian, Jean-François Gervais, Mohamad Sawan*

PolyStim NeuroTechnology Laboratory, Dept. of Electrical Engineering, Ecole Polytechnique de Montreal, Canada

Introduction

Many researchers are working towards the realization of a visual substitution system based on intra-cortical stimulation of the brain using a wireless implantable device. Some teams have already shown encouraging results in this area [1]. However, before a system reaches performances high enough to provide users with real functional vision, several elements need to be optimized.

For example, on the external side, efficient image processing algorithms have to be implemented to facilitate motion and object identification despite the limited achievable resolution of an implant, both in terms of phophene number and precision. Also, some novel solutions have to be included to minimize the serious power consumption constraints of the implanted device, and performance and reliability of electromagnetic power and bidirectionnal data link must be improved.

We present a system that integrates convenient and flexible tools for research and optimization of a complete intra-cortical stimulator. This paper emphasizes on the integrated environment, and key system performance enhancement units performing tasks such as image processing and mapping, and stimulation ordering, intended for saving power at the implant level.

Method

In addition to including all elements required for a complete visual cortical stimulator in a modular and upgradable fashion (image acquisition and processing units, data management, power and data link, and implantable stimulator), the system presents configurable user interfaces intended for:

1– Performing numerous tasks for basic verification, such as stimulation sites sensitivity measurement, parameter optimization, and pattern recognition testing;

2– Testing image processing algorithms that could enhance the performance of the user by simplifying the image to be interpreted, compensating for the low achievable resolution of the stimulator compared to normal vision.

The system can be used either to generate an image in a visual field emulator presenting, as realistically as possible, the sensation experienced by a subject, or on live subjects to get direct subjective or measurable feedback. Both alternatives can be performed simultaneously, to provide the research team with a direct feedback about the visual perception of the subject.

To achieve this, a Visuotopic Database (VDB, set of phosphene positions and characteristics in the user’s visual field) of the blind subject has to be determined, since the relation between electrode placement and phosphene position in the visual field cannot be exactly know a-priori [2]. However, a time-wise constant mapping exists between each electrode, defined by a unique Stimulation Site Address (SSA), and specific Phosphene Visual Field Coordinates

(PVFC). This SSA-PVFC mapping may be unique for every user, and has to be determined experimentally for predicting accurately the sensation produced by whatever stimulation pattern.

When the system is used for image processing algorithm evaluation or to preview the result of pre-programmed pattern recognition tests while no VDB has been experimentally defined, an artificial VDB is generated, as explained in the emulator section hereafter.

Design

Figure 1 shows the main components of the system involved in stimulation. Note that upload telemetry components, although present in the system, are omitted from the figure.

A C++ coded graphical user interface (GUI) provides the user with a flexible and visual configuring and programming environment for controlling stimulation in detail. Input for stimulation can also come from an image acquisition device. In this latter case, the image acquired is processed and fitted to the specific VDB of the subject. Adequate processing before mapping can significantly ease image interpretation by the user, as shown in Figure 2.

Each pixel out of the image fitting block is defined by its intensity and PVFC, and the image crudely represents what the blind user is expected to perceive.

Figure 2 : Basic image enhancement example and fitting to a uniformly generated VDB. a) Input image, b) direct fit, c) input image processed for edge detection, and d) 2-level conversion and fitting of the processed image.

The pixels coordinates are then converted from PVFC to SSA. Following this, an ordering task can be performed in order to maximize efficiency of the implantable device.

This comes from the fact that the large number of stimulation sites of the implant necessitates many channels to be activated simultaneously. Thus, a large part of power consumption is associated with stimulation, rather than with control and peripheral circuitry. If the stimulation is done without paying attention to the content of the image, some combinations of dark and bright areas will lead to significant load variations. This tightens requirements on the regulator performances and reduces power efficiency, as its drop-out voltage needs to be increased. Furthermore, when Load Shift Keying (LSK) is used for telemetry, these consumption variations might cause data transfer errors. A special ordering algorithm optimizes the stimulation sequence, keeping the total supply current as constant as possible throughout stimulation, as detailed in [3].

Finally, pixel intensity to stimulation parameter conversion, low-level pulse sequencing, encoding and timing of stimulation instructions is performed, before they are sent to the implanted device. In the present version of the system, amplitude modulation and a class E amplifier are used for transcutaneous data transmission.

The implant is composed of a plurality of Stimulation Modules (SMs) integrated circuits connected directly to electrode matrices inserted in the cortex, and a single physically distinct Interface Module (IM), which main tasks are to provide the appropriate SMs with power and instructions, as received from the external controller [4]. The IM also sends monitoring data to the external controller through a LSK telemetry link (not shown in Figure 1).

The implant allows stimulation to be monopolar as well as bipolar, and uses a configurable communication protocol, minimizing the data transfer rate requirements according to the stimulation strategy used.

Safe stimulation is ensured by constantly verifying the validity of every programmed parameter, and characterization/trouble-shooting after implantation is possible by monitoring voltage/current at any electrode. Also, for enhanced reliability, communication data verification and validation from the external controller is used to ensure adequate configuration and stimulation.

The emulator is mainly used for displaying an approximation of the expected visual sensation experienced by the blind subject. During stimulation, whether a subject is actually being stimulated or not, each site activation is sent to the emulator display, with its intensity and SSA definition. According to the visuotopic and PVFC-SSA maps, position, size and brightness of each phosphene is computed and the visual field content is displayed.

When the system is used for visualizing programmed tests or for evaluating image processing algorithms, and when no specific VDB has been previously built, realistic visuotopic and SSA-PVFC maps have to be generated first. PVFCs in the generated maps follow probalistic relations that can be uniform or modulated by retinotopic map models, while other phosphenes are determined randomly. Details can be found in [5].

Summary

A complete and flexible visual intra-cortical stimulation system has been designed, and functionality of the system has been verified on a reduced size implantable stimulator prototype. The system facilitates execution of various tests for research and stimulation validation, and allows different image processing and data management techniques to be tested, both in real and emulated environments.

Current research on the implant concerns an efficient full-duplex data link and a flexible and reliable package for a chronically implanted device. On the external side, while different image enhancement algorithms are being explored, a flexible signal processor with specific mapping and fitting capabilities is being designed.

References

[1] The laboratory of neuroprosthetic research at Pritzker institute of medical engineering web page, http://neural.iit.edu

[2] Warren, D.J., Fernandez, E., Normann, R.A, High-resolution two-dimensional spatial mapping of cat striate cortex using a 100-microelectrode array. Neuroscience, 2001. 105(1): p.19-31.

[3] Carniguian, S. Coulombe, J., Sawan, M. New scanning technique for the power management of pixel array. IEEE Canadian Conf. in Elec. and Comp. Eng., Montreal, May 2003.

[4] Buffoni, L.-X., Coulombe, J., Sawan, M. An image processing system dedicated to cortical visual stimulators, IEEE Canadian Conf. in Elec. and Comp. eng., Montreal, May 2003.

[5] Coulombe, J., Gervais, J.-F., Sawan, M. A cortical stimulator with monitoring capabilities using a novel 1 Mbps ASK data link, IEEE Int. Symposium on Circuits and Systems, Bangkok, May 2003

Acknowledgement : Authors acknowledge financial support from NSERC and NATEQ, and microelectronic design tools from CMC.