Open Access
Add to BibSonomy
Abstract
Brain/computer interfaces (BCIs) rely on the concurrent recording of many channels of electrical activity from excitable tissue. Traditionally such neural interfacing has been performed using cumbersome, channel-limited multielectrode arrays. We believe that BCIs can greatly benefit from using an optical approach based on simple yet powerful liquid-crystal based transducer technology. This approach potentially offers a technology platform that can sustain the necessary bandwidth, density of channels, responsivity, and conformability that are required for the long-term viability of such interfaces. In this paper we review the overall architecture of this approach, the challenges it faces, and the solutions that are being developed at UNSW Sydney.
Published by Optica Publishing Group under the terms of the Creative Commons Attribution 4.0 License. Further distribution of this work must maintain attribution to the author(s) and the published article's title, journal citation, and DOI.
1. INTRODUCTION
There is an imperative need for the recording of the electrical activity of neural, cardiac, and muscle cells to better understand them in their healthy states and to better treat them in their disease states. The field of electrophysiology has relied on a few well-established techniques [1] to address this need. However, despite the remarkable potential demonstrated by the application of these electrical methodologies in brain/computer interfacing (BCI), the advancement of their integration into clinical settings has been impeded by challenges associated with scalability, precision, and invasiveness [2]. Hence, an increasing trend is emerging towards the exploration of optical techniques as alternative approaches to overcome these challenges, including optogenetics, phase-sensitive plasmonics, and photothermal modulation [3,4].
Herein, we present a novel, and arguably advantageous, approach, which could yield, in the short term, better scientific instrumentation and, in the long term, viable BCIs with sensory feedback for restoring lost functions in disabled people, controlling prostheses, or even enhancing the capabilities of the human brain. We have demonstrated theoretically [5] and proven experimentally [6] that an optical approach to neural interfaces offers unique advantages over its traditional competitors operating in the electrical domains. Amongst those are the following:
- • Heat dissipation: optical neural interfaces can be implemented as passive devices and hence do not dissipate heat in the surrounding biological tissues;
- • Density and channel counts: by leveraging optical multiplexing and avoiding impedance matching issues, optical neural interfaces can potentially offer higher density and channel counts than is possible in the electrical domain;
- • Conformability: as opposed to CMOS electronics (or equivalent), optical circuitry can be built on a number of flexible substrates (polymer, sol-gels) suitable for implantation;
- • Bandwidth: optical fiber transmission affords unparalleled data throughput.
In this original work, we demonstrate the capability of a single optical electrode (a.k.a. optrode) as a substitute for standard micro-electrodes and thus prove its viability to form the basis for a new class of optical neural interfaces. We address herein the system-level requirements to support the development of a dense array of optrodes (e.g., ${\gt}{1000}$ channels) in the context of brain/machine interfaces (BCIs). First, in Section 2, we review the principle of operation of the optrode so as to be able to identify the supporting system-level infrastructure. We also review some of the experimental work proving the viability of the approach and in so doing point out the current limitations of the technology and identify the future challenges.
Fig. 1. Optrode transducer. (a) Cross section of a typical liquid-crystal based optrode used in the context of electrophysiological studies. An optical polarization maintaining (PM) fiber guides polarized incoherent white (broadband) light into a GRIN lens, which acts as a beam expander before entering the transducer per se. The light travels through the liquid crystal layer before being reflected by the back mirror and recaptured by the PM fiber. The polarization of the light is rotated during both the forward and backward directions so that when it reenters the PM fiber, part of it is extinguished by the PM fiber, which acts as a polarizer. (b) Packaged optrode as developed at UNSW.
2. SINGLE OPTRODE
A. Principle of Operation
Figure 1 offers a schematic of a single optrode that is to be understood as a passive three-port opto-electronic transducer. Two of the ports (metal pins) act as standard electrical terminals onto which a voltage can be applied. The third optical port is used to bring light into the device and recollect the portion of that light reflected by the transducer.
The net result of applying a voltage on the transducer is to modulate the amount of light reflected. In that sense, it is simply a variable reflectance mirror. Conceptually, the transducer thus acts as a passive device converting electrical signals applied to its terminals into an optical signal. In the context of electrophysiology, the transducer acts as an optical (micro-)electrode that converts biopotentials into optical signals and thus—put more simply—as an optrode.
Still with reference to Fig. 1, the optrode principle of operation can be described by a simple model, where we assume that the transparency of all layers, the reflectance of the mirror, and the extinction ratio of the polarizer are perfect. Birefringence is an essential optical property of the LC material in this system given that it has a refractive index that depends on the propagation direction and polarization of light. Under these conditions, applying a voltage on the optrode terminals results in the creation of an electric field $E$ perpendicular to the cell plane, which essentially amounts to (1) a change of the liquid-crystal effective birefringence, $\Delta n(E)$, and (2) a rotation of the cell optical axes by an angle $\Omega (E)$ around an axis parallel to the field. The reflectance (at parallel analyzer/polarizer) can then be expressed as
B. Experimental Results
The sensing capability of single optrodes was validated in a series of biological experiments [6,8]. The optrode system was able to detect and record spontaneous extracellular potentials from the pacemaker tissue of a rabbit heart preparation. Importantly for BCI applications, the optrode was able to sense extracellular potentials evoked by electrical stimulation of the sciatic nerve of rabbits under general sedation [6]. The sensed signal included components from nerve fibers responsible for transmitting a variety of sensory and motor information [8]. The temporal fidelity of signals recorded by the optrode was verified by comparison against recordings obtained by conventional electrophysiology recording systems [6]. Figure 2 illustrates typical biopotential recordings obtained with the above approach.
Fig. 2. Biological validation of optrode technology for biopotential recording. (a) Comparison of in vivo rabbit nerve responses (evoked compound action potentials) to electrical stimulation recorded by the optrode technology and conventional bioamplifier systems. The asterisks indicate the stimulation artefacts following each stimulus pulse. Shaded areas highlight the nerve responses. (b) Spontaneous cardiac electrograms recorded using the optrode transducer from a rabbit ex vivo sino-atrial node tissue preparation. The inset is magnification from (b) showing the cardiac extracellular potential for a single heart beat as recorded by the optrode. All signals were measured using single-channel optrode devices [Fig. 1(b)] as detailed in [6].
3. OPTRODE ARRAYS
In the long term, a viable neural interface—such as the one depicted in Fig. 3—relies on the development of optical circuitry, using techniques and components borrowed from optical telecommunication networks and integrated optics. Still, with reference to Fig. 3, an optrode array would require the delivery of light via one or more optical fibers followed by the distribution of light in the plane of the array using optical circuitry. The exact specification of this circuitry will vary according to the addressing mode used (see Section 3.B), but in all cases, the single optrode as described in Section 2 must be redesigned to be integrated onto a (preferably conformable) planar substrate.
Fig. 3. Artistic renderings. (Left) Conformable multioptrode array (MOA) interfacing with cerebral cortex. It consists of 64 (${8} \times {8}$) subarrays (right) each implementing a wavelength division multiplexing (WDM) scheme using, in this case, arrayed waveguide gratings thus enabling the addressing of specific individual optrode, altogether forming an entirely passive 4096-channel neural interface.
Fig. 4. Schematics of an optrode as implemented onto a planar substrate. This approach would allow the creation of optrode arrays by repeating its structure in the plane of the device and adding the required optical circuitry to deliver and collect the optical signals emitted by a broadband source (not shown).
A. Optrode Architecture
Figure 4 illustrates a typical implementation of such a redesigned optrode. The exact composition of the various layers may vary according to the requirements but remains conceptually the same. From the bottom up we have first the biological tissue in electrical contact with a metallic through-via. This via connects to a mirror (gold) capable of reflecting light upwards. A layer of liquid crystal is confined between the substrate and a transparent indium tin oxide (ITO) layer deposited onto the superstrate, which is engineered to guide light from a light source (not shown).
The superstrate is itself multilayered to contain the optical circuitry, which connects each optrode optically to the light source. The circuitry contains both the waveguide and optical elements—in this specific case a Bragg grating coupler—necessary to couple the light in and out of the plane of the array.
By grounding the conductive ITO layer to a distant ground (shown in green), the electrical activity in the biological tissues creates a load on the terminals of the optrode that can then be detected optically as in the case of a single optrode. This requires precise control of the polarization of the incoming light and can be managed via an appropriate design of the waveguiding geometry.
B. Array Addressing
The array architecture mainly depends on the strategy used to address individual optrodes. Ideally, such an approach would support thousands of individual optrodes, and dissipate little energy (i.e., heat) in the surrounding biological tissue as typical light sources use tens of milliwatts of optical powers, most of it being reflected back toward the data acquisition system. Such an approach can be adapted from the classic wavelength division multiplexing (WDM) scheme common in the telecommunications industry [9].
The basic idea here is to subdivide the array into groups of optrodes and then assign a specific wavelength to each optrode within a group. Using this approach it is then possible to use a passive multiplexer such as arrayed waveguide gratings (AWGs) (Fig. 3) to route a specific wavelength to its assigned optrode.
C. Array Scalability
In the long term, traditional neural interfaces will require densely packed arrays of electrodes to be developed, which normally raises a number of thorny technical issues. Two of them are of particular relevance here: (1) impedance mismatch and (2) cross-talk. Impedance mismatch stems from an inverse relationship between channel size and signal quality, while cross-talk occurs when biopotentials generated by localized group of cells are recorded by multiple adjacent recording channels, leading to interference or contamination of the recorded data. The working principle of the optrode technology addresses these issues as the LC component within the optrode plays a critical role in the system’s structure and transduction mechanism.
Traditional bio-amplifiers require a high input impedance to preserve the fidelity of the recorded signals, especially when the electrode (or interface) impedance increases. However, this is associated with complex circuit designs and hardware that are not adaptable to different electrode shapes and sizes. Reducing the size of the electrode increases the interface impedance and hence requires an even higher input impedance. Our benchtop characterizations showed quantitative evidence of automatic scaling of input impedance within the optrode device in relation to interface impedance, preserving the recorded signal quality regardless of channel size [10]. The LC region located right below the sensing channel is equivalent to the system input impedance of conventional electrophysiology recording systems. Hence, when reducing the size of the recording site, both the interface impedance and the input impedance increase. This results in an automatic scaling mechanism that maintains a consistent ratio between input impedance and total system impedance.
Similarly, the simulated spatial performance of optrode arrays showed that the changes in the polarization of the LC helical structure by an applied electric field are restricted to LC regions directly beneath the sensing channels without the need to confine the LC layer to small regions [5]. This feature ensures minimal electrical or optical cross-talk between contiguous channels, contributing to the system’s detection accuracy and resolution.
Fig. 5. Multioptrode array (MOA) system overview. A superluminescent diode (SLD) emits a broadband signal comprising a number of wavelengths $\{{\lambda _1},{\lambda _2},\ldots {\lambda _n}\}$, which are then demultiplexed into physical channels via a demultiplexer (DE/MUX) to reach the multioptrode array. The circulator guarantees that no light is returned to the light source by rerouting the reflected optical signal from the array to the signal acquisition unit (ACQ) where the optical signal is demultiplexed (not shown) and converted into analog electrical signals to be finally digitized by the data acquisition system (DAQ).
D. Array Material System
The long-term success of the proposed approach relies on the development of a material system capable of light guidance and transduction. In addition, a viable physical interface adjacent to biological tissues must offer both conformability and biocompatibility. These issues are not specific to multioptrode arrays (MOAs), but most, if not all, competing technologies require substantial engineering to realize a truly viable implementation.
This section focuses on the requirements for both light guiding and transduction imposed on potential candidates, while Section 4.A discusses the addressing requirement—in our view a system-level requirement—on the material system when implementing a WDM [9] addressing scheme.
Designing a system that can seamlessly conform to various organ geometries and irregularities while channelling the light to and from the system necessitates the utilization of isotropic flexible materials. These materials must possess specific optical and mechanical properties meticulously tailored to minimize optical losses [11]. However, this task presents challenges, especially when dealing with soft and sensitive materials, given the intrinsic optical losses, artefactual signals in response to bending and movement, and complex fabrication processes. Fabricating optical systems with soft materials involves numerous steps and unconventional planning to preserve their optical and mechanical properties while ensuring optimal performance. The biocompatibility of these materials is equally critical for their safe, long-term implantation, with factors such as fluid absorption, delamination, mechanical degradation, and biostability warranting careful consideration in material design and testing processes.
Currently, the material design of MOAs is undergoing extensive testing and characterization according to the demands and requirements of the system. This includes enhancing the flexibility of the interfacing system to enable direct contact with tissues, maintaining a stable LC alignment, mitigating light scattering under mechanical stresses and strains, and improving the electro-optical performance [12]. Our work employs flexible and transparent conductive polymers to offer a versatile platform for advancing MOAs’ practical implementation.
4. SYSTEM-LEVEL CONSIDERATION
A. System Architecture
Of course, a complete BCI consists of more than an optrode array. Figure 5 illustrates one possible implementation of such a system using the WDM addressing approach detailed above. The system-level characteristics, such as the number of channels, sensitivity, and signal-to-noise per channel, depends inter alia on the stability (in time) of the optical source, the optical loss, detector noise, and many other factors pertaining to the electronics.
It is given that sensory signals from nerves could be as low as tens of microvolts in amplitude. A typical goal would be to reach sensitivity as low as 10 µV or a peak-to-peak baseline noise level of 20 µV. The duration of nerve impulses is in the order of a few milliseconds. A typical bandwidth requirement per channel is 5–10 kHz. In most cases, it is possible to implement noise reduction strategies that will help increase signal-to-noise. We will discuss two of those in the following sections.
B. Active Noise Canceling
When designing the array, it is possible to use the fact that optrodes can be used in pairs to implement an active noise-canceling scheme. In one possible implementation, the signal received from two nearest-neighbour optrodes, one active and one inactive, can be compared. With this setup, all things being equal, one channel contains the biopotential signal and system noise, while the other only contains the latter. By passing the signals from the two channels to an active noise-cancelation algorithm [13,14], the noise in the message channel ${x_1}(i)$ can be substantially reduced based on the information contained in the noise channel ${x_2}(i)$.
The end goal is to remove the part of the noise in ${x_1}(i)$ that is correlated to the noise in ${x_2}(i)$. In order to quantify this statement, we use the traditional definition of cross correlation between two signals $x(i)$ and $y(i)$:
noting that ${R_{\textit{xx}}}(i)$ is the auto-correlation. We also posit that the signals are well represented by where $n(i)$ is the part of the noise assumed present and correlated in both ${x_1}(i)$ and ${x_2}(i)$, $s(i)$ is the signal (plus potentially other random noise) in ${x_1}(i)$, $m(i)$ is the random noise in ${x_2}(i)$, and $k$ is an unknown scale factor that needs to be estimated. All of $s(i)$, $n(i)$, and $m(i)$ are random signals and assumed to be uncorrelated with each other.Experimentally, we have observed that the dominant noise term in ${x_2}(i)$ is the correlated term, such that $kn(i) \gg m(i)$, and under all these assumptions, the combined output
should have a much lower noise level than ${x_1}(i)$ itself. Lastly, the $k$ factor can be estimated from the cross correlation between ${x_1}(i)$ and ${x_2}(i)$, namely that it is given by the ratio of the cross correlation to the auto-correlation for zero lag, i.e., for $i = 0$, Figure 6(a) shows the captured, digitized signals before (“Active Optrode”) and after (“Noise-Canceling Output”) the noise-canceling process is applied. The Active Optrode channel clearly has a substantially larger amplitude than the Noise-Canceling Output channel. However, these two signals still have a normalized cross correlation of 0.4, implying that there is still room to improve the performance. In Fig. 6(b), a large 200 mV sinusoidal signal is applied to the optrode. Again, the Noise-Canceling Output channel has substantially less noise than the Active Optrode channel, confirming that the algorithm is working in the case of sinusoidal signals, even when these are much larger than the noise level.Fig. 6. (a) 500 mA light source current with zero input signal at the active optrode. (b) 500 mA light source current with 200 mV peak-to-peak 1 kHz sinusoidal input signal at the active optrode.
Figure 7 shows the signal-to-noise ratio before and after the active noise-canceling process, with a 10 mV sinusoidal inputs applied to the optrode for different light source intensities, here parameterized by the diode driving current. There is an outlying data point at the 300 mA light source current, but overall the data expressed the common-sense conclusions that (i) at lower light source currents the signal-to-noise ratio improvement is modest; (ii) as the light source current increases, the signal-to-noise ratio improvement increases; and (iii) as the light source drive current exceeds 500 mA, the improvement plateaus.
Fig. 7. Signal-to-noise-ratio at different light source current with 200 mV peak-to-peak 1 kHz sinusoidal input signal at the active optrode.
It is important to notice that the noise-canceling approach does not require a strict one-to-one pairing of an active to an inactive optrode. For example, in a $3 \times 3$ array, the central inactive optrode can be paired to its eight nearest neighbors or, likewise in a $4 \times 4$ array, to its 15 closest neighbors.
C. Lock-In Detection
Lock-in detection is a standard approach developed in the 1930s capable of extracting signal amplitudes and phases in extremely noisy environments. With lock-in detection, the signal of interest is modulated by a carrier frequency, sensed and amplified at this carrier frequency, and subsequently demodulated back to the base-band [15].
The frequencies of the optrode signals lie in a frequency band where substantial noise from components and other sources is present. Examples of such noise sources are pink noise in the photodetector and amplifier chain, mains interference, and movement artefacts. Since the optrode modulates the light from the light source, the optrode can be used as a mixer: by modulating the light from the light source, the optrode signal can be moved to a frequency band in which the noise from components and other sources is substantially lower than the noise in the optrode signal band. After the photodetector, the signal can be subsequently demodulated and filtered, removing a substantial part of the system noise present in the optrode signal frequency band.
The modulator can be implemented by electronically modulating the light source drive current or can be implemented in the optical domain by means of a shutter or a mirror. The demodulator can be implemented as a quadrature demodulator, or it can be implemented as a synchronous demodulator if the modulator implementation allows. The demodulator can be implemented directly with electronic components such as mixers and filters, or it can be implemented as a digital state-machine, micro-controller code, or any other digital signal processing method if the photodetector first digitizes the received light signal [16]. The removal of in-band noise by lock-in detection can be applied in conjunction with other noise reduction methods such as the active noise-canceling method described above.
D. Hardware Requirements
Electronic hardware resources, required to implement a specific BCI architecture, are an important system consideration. These resources—including power dissipation, volume, and cost—scale with the number of channels, and thus for practical implementations there will be a design trade-off between the number of channels, channel sensitivity, and system volume: to improve channel sensitivity, for instance, one or both proposed noise-canceling methods can be employed, each increasing the amount of electronics resources required per channel. Electronic resources residing internal to the body are restricted in volume and power dissipation. Therefore, in BCIs residing wholly within the body, a careful channel count/channel sensitivity trade-off must be done. Electronics resources residing external to the body on the other hand are comparatively affordable. Thus, in BCIs with parts external to the body high channel count with good sensitivity can be achieved at the expense of external system volume, power dissipation, and cost.
5. CONCLUSION
We have reviewed some aspects of the research on optical neural interfaces conducted at UNSW Sydney over the past few years. Although we have demonstrated the central claims of the approach—inter alia sensitivity, passivity, and scalability—much R&D remains to be done in order to yield a viable neural interface and to see it translated into a commercial reality.
Amongst those are issues related to conformability, bio-compatibility, and the cost of fabrication. Nevertheless, due to the intrinsic advantages our approach offers over competing ones, we believe that in the long run, we can provide a solution that will be applicable both for the development of novel scientific instruments and, ultimately, for embeddable in vivo neural implants.
Funding
National Health and Medical Research Council (2002282); Australian Research Council (DP200102825); Office of Naval Research (ONR).
Acknowledgment
This work was performed in part at the NSW Node of the Australian National Fabrication Facility.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
REFERENCES
1. S. Sadaghiani, M. J. Brookes, and S. Baillet, “Connectomics of human electrophysiology,” NeuroImage 247, 118788 (2022). [CrossRef]
2. C. A. Scholl, E. Bar-Kochba, M. J. Fitch, et al., “Optical noninvasive brain–computer interface development: challenges and opportunities,” Johns Hopkins APL Tech. Dig.-Appl. Phys. Lab. 35, 288–295 (2021).
3. N. T. Ersaro, C. Yalcin, and R. Muller, “The future of brain–machine interfaces is optical,” Nat. Electron. 6, 96–98 (2023). [CrossRef]
4. R. M. Almasri, F. Ladouceur, D. Mawad, et al., “Emerging trends in the development of flexible optrode arrays for electrophysiology,” APL Bioeng. 7, 031503 (2023). [CrossRef]
5. A. Al Abed, H. Srinivas, J. Firth, et al., “A biopotential optrode array: operation principles and simulations,” Sci. Rep. 8, 2690 (2018). [CrossRef]
6. A. A. Abed, Y. Wei, R. M. Almasri, et al., “Liquid crystal electro-optical transducers for electrophysiology sensing applications,” J. Neural Eng. 19, 056031 (2022). [CrossRef]
7. Z. Brodzeli, L. Silvestri, A. Michie, et al., “Reflective mode of deformed-helix ferroelectric liquid crystal cells for sensing applications,” Liq. Cryst. 40, 1427–1435 (2013). [CrossRef]
8. R. M. Almasri, Y. Wei, F. Ladouceur, et al., “A comparative assessment of evoked compound action potentials measured by optrode and conventional bioamplifier systems,” in 11th International IEEE/EMBS Conference on Neural Engineering (NER) (2023), pp. 1–4.
9. C. Siva Ram Murthy and M. Guruswamy, WDM Optical Networks, Concepts, Design, and Algorithms (Prentice Hall, 2002).
10. R. M. Almasri, A. Al Abed, Y. Wei, et al., “Impedance properties of multi-optrode biopotential sensing arrays,” IEEE Trans. Biomed. Eng. 69, 1674–1684 (2021). [CrossRef]
11. R. M. Almasri, A. Al Abed, D. Esrafilzadeh, et al., “Electromechanical stability and transmission behavior of transparent conductive films for biomedical optoelectronic devices,” in 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) (2022), pp. 5–8.
12. R. Almasri, “Flexible liquid-crystal based optrode for electrophysiology applications,” Ph.D. thesis (UNSW Sydney, 2023).
13. J. Jeong, “Real-time acoustic noise canceling technique on innovations-based inverse kepstrum and FIR RLS,” in IEEE International Symposium on Intelligent Control (2010).
14. A. Yelwande, S. Kansal, and A. Dixit, “Adaptive wiener filter for speech enhancement,” in International Conference on Information, Communication, Instrumentation and Control (ICICIC) (2017).
15. M. Meade, “Advances in lock-in amplifiers,” J. Phys. E 15, 395 (1982). [CrossRef]
16. J. Gaspar, S. F. Chen, A. Gordillo, et al., “Digital lock in amplifier: study, design and development with a digital signal processor,” Microprocess. Microsyst. 28, 157–162 (2004). [CrossRef]
