Three-dimensional electrode arrays for retinal prostheses: modeling, geometry optimization and experimental validation.

Three-dimensional electrode geometries were proposed to increase the spatial resolution in retinal prostheses aiming at restoring vision in blind patients. We report here the results from a study in which finite-element modeling was used to design and optimize three-dimensional electrode geometries. Proposed implants exhibit an array of well-like shapes containing stimulating electrodes at their bottom, while the common return grid electrode surrounds each well on the implant top surface. Extending stimulating electrodes and/or the grid return electrode on the walls of the cavities was also considered. The goal of the optimization was to find model parameters that maximize the focalization of electrical stimulation, and therefore the spatial resolution of the electrode array. The results showed that electrode geometries with a well depth of 30 µm yield a tenfold increase in selectivity compared to the planar structures of similar electrode dimensions. Electrode array prototypes were microfabricated and implanted in dystrophic rats to determine if the tissue would behave as hypothesized in the model. Histological examination showed that retinal bipolar cells integrate the electrode well, creating isolated cell clusters. The modeling analysis showed that the stimulation current is confounded within the electrode well, leading to selective electrical stimulation of the individual bipolar cell clusters and thereby to electrode arrays with higher spatial resolution.

A modular 256-channel micro electrode array platform for in vitro and in vivo neural stimulation and recording: BioMEA.

In order to understand the dynamics of large neural networks, where information is widely distributed over thousands of cells, one of today's challenges is to successfully monitor the simultaneous activity of as many neurons as possible. This is made possible by using the Micro-Electrode Array (MEA) technology allowing neural cell culture and/or tissue slice experimentation in vitro. Thanks to development of microelectronics' technologies, a novel data acquisition system based on MEA technology has been developed, the BioMEA™. It combines the most advanced MEA biochips with integrated electronics, and a novel user-friendly software interface. To move from prototype (result of the RMNT research project NEUROCOM) to manufactured product, a number of changes have been made. Here, we present a 256-channel MEA data acquisition system with integrated electronics (BioMEA™) allowing simultaneous recording and stimulation of neural networks for in vitro and in vivo applications. This integration is a first step towards an implantable device for BCI (Brain Computer Interface) studies and neural prosthesis.

BioMEA: a 256-channel MEA system with integrated electronics.

In order to understand the dynamics of large neural networks, where information is widely distributed over thousands of cells, one of today's challenges is to successfully record the simultaneous activities of as many neurons as possible. This is made possible by using microelectrodes arrays (MEAs) positioned in contact with the neural tissue. Thanks to microelectronics' microfabrication technologies, it now becomes possible to build high density MEAs containing several hundreds of microelectrodes. However, increasing the number of electrodes using conventional electronics is difficult to achieve. Moreover, high density devices addressing all channels independently for simultaneous recording and stimulation are not readily available. Here, we present a 256-channel in vitro MEA system with integrated electronics allowing simultaneous recording and stimulation of neural networks. Both actions are performed independently on all channels.

A 64-channel ASIC for in-vitro simultaneous recording and stimulation of neurons using microelectrode arrays.

A 64 channels CMOS chip dedicated to in-vitro simultaneous recording and stimulation of neurons using microelectrode arrays has been developed. It includes, for each channel, a low noise, variable gain (10, 75 or 750), 0.08 Hz-3 kHz bandwidth measurement path with unity-gain for lower frequencies to allow measurement of the electrochemical potential. A snapshot style Sample & Hold circuitry allows to have "images" of the 64 channels at a maximum sampling frequency of 50 kHz. Input-referred noise of the measurement path is 4.3 microV RMS integrated from 0.08 Hz to 3 kHz. To get rid of the random, slowly varying, DC offset potential that exists at the electrode-electrolyte interface, the ASIC can be supplied with floating VSS, VDD. Circuit size is 2.4 mm per 11.2 mm (0.35 microm CMOS process) and its power consumption is about 125 mW.

Fast realistic modeling in bioelectromagnetism using lead-field interpolation.

The practical use of realistic models in bioelectromagnetism is limited by the time-consuming amount of numerical calculations. We propose a method leading to much higher speed than currently available, and compatible with any kind of numerical methods (boundary elements (BEM), finite elements, finite differences). Illustrated with the BEM for EEG and MEG, it applies to ECG and MCG as well. The principle is two-fold. First, a Lead-Field matrix is calculated (once for all) for a grid of dipoles covering the brain volume. Second, any forward solution is interpolated from the pre-calculated Lead-Fields corresponding to grid dipoles near the source. Extrapolation is used for shallow sources falling outside the grid. Three interpolation techniques were tested: trilinear, second-order Bézier (Bernstein polynomials), and 3D spline. The trilinear interpolation yielded the highest speed gain, with factors better than x10,000 for a 9,000-triangle BEM model. More accurate results could be obtained with the Bézier interpolation (speed gain approximately 1,000), which, combined with a 8-mm step grid, lead to intrinsic localization and orientation errors of only 0.2 mm and 0.2 degrees. Further improvements in MEG could be obtained by interpolating only the contribution of secondary currents. Cropping grids by removing shallow points lead to a much better estimation of the dipole orientation in EEG than when solving the forward problem classically, providing an efficient alternative to locally refined models. This method would show special usefulness when combining realistic models with stochastic inverse procedures (simulated annealing, genetic algorithms) requiring many forward calculations.

Multiple supratemporal sources of magnetic and electric auditory evoked middle latency components in humans.

The supratemporal sources of the earliest auditory cortical responses (20-80 ms) were identified using simultaneously recorded electroencephalographic (EEG) and magnetoencephalographic (MEG) data. Both hemispheres of six subjects were recorded two or three times in different sessions in response to 8000 right-ear 1 kHz pure tones stimuli. Four components were identified: Pa (28 ms), Nb (40 ms), and two subcomponents of the Pb complex, termed Pb1 (52 ms) and Pb2 (74 ms). Based on MEG data, the corresponding sources were localized on the anatomy using individual realistic head models: Pa in the medial portion of Heschl's gyri (H1/H2); Nb/Pb1 in the lateral aspect of the supratemporal gyrus (STG); and Pb2 in the antero-lateral portion of Heschl's gyri. All sources were oriented antero-superiorly. This pattern was clearest in the contralateral hemisphere, where these three activities could be statistically dissociated. Results agree with previous invasive human intracerebral recordings, with animal studies reporting secondary areas involved in the generation of middle latency auditory-evoked components, and with positron emission tomography and functional magnetic resonance imaging studies often reporting these three active areas although without temporal information. The early STG activity may be attributed to parallel thalamo-cortical connections, or to cortico-cortical connections between the primary auditory cortex and the STG, as recently described in humans.

An evaluation of dipole reconstruction accuracy with spherical and realistic head models in MEG.

MEG forward problem has been solved for about 2000 dipoles placed on the brain surface using a very fine 3-layer realistic model of the head and the boundary element method (BEM). For each dipole, spherical models, one-layer realistic BEM models and coarser 3-layer realistic BEM models, were used to reconstruct the dipole. It was found that the localization bias induced by using a spherical model of the head increased from 2.5 mm in the upper part of the head to 12 mm in the lower part, on average. It was also found that, for the same computing time, a 3-layer model of the head gave on average 2 mm better localization errors than a one-layer model of the head. Orientation errors of less than 20 degrees could only be retrieved with a 3-layer realistic model. Localization and orientation errors highly depended on the dipole position in the brain.

Human cortical responses evoked by dichotically presented tones of different frequencies.

Behavioral and patient studies have suggested that during dichotic listening the ipsilateral auditory pathways are strongly inhibited, so that each hemisphere is treats the sound coming to the contralateral ear. We analysed the auditory N100m neuromagnetic evoked response following passive listening of dichotic tones of different frequencies. We found that the N100m in each hemisphere depended on both ipsilateral and contralateral stimuli, revealing no strong inhibition of ipsilateral pathways. The N100m increased with the interaural frequency disparity and was reduced as both ears received identical stimuli. The results can be explained by the existence of a frequency-dependent excitatory/inhibitory organization of the auditory cortex, as has been described in the cat. We suggest that the N100m might also reflect automatic processes involved in multiple-stream perception.

A systematic evaluation of the spherical model accuracy in EEG dipole localization.

This paper presents a study of the intrinsic localization error bias due to the use of a spherical geometry model on EEG simulated data obtained from realistically shaped models. About 2000 dipoles were randomly chosen on the segmented cortex surface of a particular subject. Forward calculations were performed using a uniformly meshed model for each dipole located at a depth greater than 20 mm below the brain surface, and locally refined models were used for shallower dipoles. Inverse calculations were performed using four different spherical models and another uniformly meshed model. It was found that the best spherical model lead to localization errors of 5-6 mm in the upper part of the head, and of 15-25 mm in the lower part. The influence of the number of electrodes upon this intrinsic bias was also studied. It was found that using 32 electrodes instead of 19 improves the localization by 2.7 mm on average, while using 63 instead of 32 electrodes lead to improvements of less than 1 mm. Finally, simulations involving two simultaneously active dipoles (one in the vicinity of each auditory cortex) show localization errors increasing by about 2-3 mm.

Matching of digitised brain atlas to magnetic resonance images.

A method has been developed to match a standard digitised brain atlas onto MR images for identification of cerebral structures in anatomical images. This method uses, first, a three-dimensional crude registration based on the proportional system of Talairach. Then, a two-dimensional refined registration is performed using a deformation function based on a set of homologous landmarks on both images (MR and atlas). Displacements vectors are computed between each corresponding landmark. These vectors are interpolated by thin-plate splines, generating an unwarping function defined on the whole image. This function can then be applied on any structure of the atlas. An evaluation of the matching procedure has been performed. First, the influence of the choice of the landmarks has been evaluated for the fine registration method. The latter has been then compared to the crude registration method considered as a classical reference method. These results show the advantages of the fine registration approach.

Improved dipole localization using local mesh refinement of realistic head geometries: an EEG simulation study.

A systematic evaluation of dipole localization accuracy using the boundary element method is presented. EEG simulations are carried out with dipoles located in the right parietal and temporal regions of the head. Uniformly meshed and locally refined head models are considered in both spherical and realistic geometries. An initial study determines the influence upon the localization accuracy of the dipole depth below the brain surface, of its orientation (radial and tangential), and of the global and local mesh densities. Simulated potential data are computed analytically in the spherical case, and numerically using a very fine (locally refined) model in the realistic case. Results in both geometries show that in order to get localization errors of about 2-4 mm, uniformly meshed models may be used for dipoles located at depths greater than 20 mm, whereas locally refined models should be used for shallower dipoles. Two other studies show how the localization accuracy depends upon the size of the local refinement area and upon the number of electrodes (19, 32, 63). Results show that a large number of electrodes brings significant improvements, especially for shallow and tangential dipoles.

Improved forward EEG calculations using local mesh refinement of realistic head geometries.

A method for semi-automatically constructing realistic surface meshes of 3 head structures--scalp, skull and brain--from a stack of MR images is described. Then an evaluation is given for both spherical and realistic dipolar models, using the boundary element method (BEM). In both cases, locally refined models were considered. Two characteristic mesh parameters were defined: the global and the local mesh densities (in triangles per cm2). In spherical geometries, numerical and analytical solutions were compared, and in the realistic case, all models were compared to a highly refined one, considered as a reference. Both geometries gave comparable results. It was found that for "deep dipoles" located at more than 20-30 mm under the brain surface, meshes with a global density of 0.5 tri/cm2 gave "acceptable" results, whereas for more superficial dipoles (2-3 mm < depth < 20-30 mm), it was necessary to locally refine meshes near the source location up to a local density of about 5-8 tri/cm2, to get comparable results.