NON-INVASIVE SINGLE NEURON ELECTRICAL MONITORING (NISNEM Technology)

Lead Research Organisation: Imperial College London
Department Name: Bioengineering

Abstract

We propose the development of a new technology for Non-Invasive Single Neuron Electrical Monitoring (NISNEM). Current non-invasive neuroimaging techniques including electroencephalography (EEG), magnetoencephalography (MEG) or functional magnetic resonance imaging (fMRI) provide indirect measures of the activity of large populations of neurons in the brain. However, it is becoming apparent that information at the single neuron level may be critical for understanding, diagnosing, and treating increasingly prevalent neurological conditions, such as stroke and dementia. Current methods to record single neuron activity are invasive - they require surgical implants. Implanted electrodes risk damage to the neural tissue and/or foreign body reaction that limit long-term stability. Understandably, this approach is not chosen by many patients; in fact, implanted electrode technologies are limited to animal preparations or tests on a handful of patients worldwide. Measuring single neuron activity non-invasively will transform how neurological conditions are diagnosed, monitored, and treated as well as pave the way for the broad adoption of neurotechnologies in healthcare.
We propose the development of NISNEM by pushing frontier engineering research in electrode technology, ultra-low-noise electronics, and advanced signal processing, iteratively validated during extensive tests in pre-clinical trials. We will design and manufacture arrays of dry electrodes to be mounted on the skin with an ultra-high density of recording points. By aggressive miniaturization, we will develop microelectronics chips to record from thousands of channels with beyond state-of-art noise performance. We will devise breakthrough developments in unsupervised blind source identification of the activity of tens to hundreds of neurons from tens of thousands of recordings. This research will be supported by iterative pre-clinical studies in humans and animals, which will be essential for defining requirements and refining designs.
We intend to demonstrate the feasibility of the NISNEM technology and its potential to become a routine clinical tool that transforms all aspects of healthcare. In particular, we expect it to drastically improve how neurological diseases are managed. Given that they are a massive burden and limit the quality of life of millions of patients and their families, the impact of NISNEM could be almost unprecedented. We envision the NISNEM technology to be adopted on a routine clinical basis for: 1) diagnostics (epilepsy, tremor, dementia); 2) monitoring (stroke, spinal cord injury, ageing); 3) intervention (closed-loop modulation of brain activity); 4) advancing our understanding of the nervous system (identifying pathological changes); and 5) development of neural interfaces for communication (Brain-Computer Interfaces for locked-in patients), control of (neuro)prosthetics, or replacement of a "missing sense" (e.g., auditory prosthetics). Moreover, by accurately detecting the patient's intent, this technology could be used to drive neural plasticity -the brain's ability to reorganize itself-, potentially enabling cures for currently incurable disorders such as stroke, spinal cord injury, or Parkinson's disease. NISNEM also provides the opportunity to extend treatment from the hospital to the home. For example, rehabilitation after a stroke occurs mainly in hospitals and for a limited period of time; home rehabilitation is absent. NISNEM could provide continuous rehabilitation at home through the use of therapeutic technologies.
The neural engineering, neuroscience and clinical neurology communities will all greatly benefit from this radically new perspective and complementary knowledge base. NISNEM will foster a revolution in neurosciences and neurotechnology, strongly impacting these large academic communities and the clinical sector. Even more importantly, if successful, it will improve the life of millions of patients and their relatives

Planned Impact

Our vision is to develop clinically applicable, non-invasive single neuron monitoring (NISNEM) technology to advance the frontiers of neural interfacing. NISNEM has the potential to transform healthcare and directly impact the healthcare sector, the research and innovation sector, the medical technology industry as well as the national economy and social wellbeing. We aim to create a broad scientific framework for a revolutionary technology that would potentially impact virtually every field of the medical industry. The advances in understanding, monitoring and treating neurodegenerative diseases that NINSEM would enable will set the basis for a new form of technology-based healthcare. This technology, together with ongoing efforts on artificial intelligence (AI), would position the UK as a world-leader in next generation healthcare.
NISNEM would allow studies that increase our understanding of neurological diseases. Many neurological diseases (e.g., Parkinson's, Alzheimer's, Huntington's) are unique to humans; animal models are extremely useful, but, in reality, limited due to the lack in complexity. Having the means to record from large neural populations in patients would constitute a cornerstone to advance our knowledge on these diseases. This improved understanding would enable new targeted pharmacological and neurotechnology-based treatments that are more effective than current methods at treating neural conditions and have minimal side effects. Neurological diseases and conditions are a vast socioeconomic drain on society. For example, currently in UK there are 1.2 million stroke survivors and >100,000 new stroke cases every year; this incidence rate is forecasted to increase by approximately 60% over the next 20 years (stroke.org.uk, 2019) because of population ageing.
As NISNEM is a non-invasive technology, it holds the potential to extend healthcare from the hospital to the home, enabling continued and personalised healthcare. These achievements will have a tremendous social and economic impact, and imply a radical cultural change in how healthcare is provided.
The proposed research will have dramatic impact on the academia and industry. In particular, it will bring a completely new perspective, new technologies/tools and a complementary knowledge base to the biomedical/neural engineering community. The program will also be extremely relevant for the neuroscience community at large, since the proposed methods will enable a breadth of novel approaches to study the nervous system, especially in humans. The technological advances required to produce NISNEM in conjunction with the diverse applicability of the proposed technology are strong indicators that this project could foster a revolution in neurosciences and neurotechnology, strongly impacting and bridging these large academic communities. The research would also be extremely relevant for academics in the broader healthcare engineering and clinical neurology fields since we will develop and progress new techniques and applications of neural interfacing for clinical use, ultimately conditioning all aspects of healthcare.
The potential impact of the proposed research programme has been recognised by a number of leading companies, which have provided their strong support (~£2M in-kind contribution), and expressed a strong interest in discussing the co-creation phase following the project (see LoS from industries). Moreover, key patients' associations have enthusiastically supported the vision of the project and its focus on clinical viability (see LoS).

Publications

10 25 50

publication icon
Tanzarella S (2023) Neuromorphic Decoding of Spinal Motor Neuron Behaviour During Natural Hand Movements for a New Generation of Wearable Neural Interfaces. in IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society

publication icon
Tageldeen MK (2023) Analogue circuit realisation of surface-confined redox reaction kinetics. in Biosensors & bioelectronics

publication icon
Steenbergen N (2023) Surface electromyography using dry polymeric electrodes. in APL bioengineering

 
Description The project NISNEM is now at the end of its Phase I. This Phase was planned with the aim of developing dense grids of electrodes to be applied on the skin to record electric potentials generated by muscle activations and of designing algorithms to process these electric potentials. The purpose of the processing was to identify the neural activity in the spinal cord that generated the muscle electric fields. Therefore, NISNEM ambitiously aimed for Phase I to identify neural information in the spinal cord from wearable electrodes mounted on the skin. Phase I was successful with all objectives met. The teams of NISNEM developed, in parallel and in close collaboration, grids of hundreds of electrodes, amplifiers able to enhance the signals recorded by these grids, artificial neural networks that could interpret the signals and could generate realistic simulations. NISNEM has now entered in Phase II, which is the most ambitious and risky. In Phase II, the goal will be the development of non-invasive methods to decode the neural activity of the brain (while Phase I focused on the spinal cord).
Exploitation Route The developed methods provide fundamental techniques for interfacing the human spinal cord with external devices, such as assistive robots or virtual objects. We are heavily working on making the developed methods publicly available (e.g., see the deep learning model) so that they can be more extensively used by the scientific community.
Sectors Digital/Communication/Information Technologies (including Software),Electronics,Healthcare,Leisure Activities, including Sports, Recreation and Tourism

 
Description The methods proposed in NISNEM provides non-invasive access to neural structures in humans. This approach can be used in consumer electronics devices for connecting users with technologies, such as smart phones, virtual reality, the metaverse. Therefore, there are currently relevant industry efforts in the development of wearable EMG systems, for example wrist-mounted devices, for this type of interaction. The teams in NISNEM are for example collaborating with Facebook Reality Labs for the development of interfaces for gaming and virtual reality.
First Year Of Impact 2021
Sector Electronics,Leisure Activities, including Sports, Recreation and Tourism
Impact Types Societal,Economic

 
Description ERC Proof of Concept Grant U-Wear, #101069444
Amount € 150,000 (EUR)
Funding ID 101069444 
Organisation European Research Council (ERC) 
Sector Public
Country Belgium
Start 01/2023 
End 07/2024
 
Description HybridNeuro
Amount € 380,000 (EUR)
Funding ID 101079392 
Organisation European Commission 
Sector Public
Country European Union (EU)
Start 01/2023 
End 12/2025
 
Title 400 channel flexible grid for surface EMG detection 
Description The previous version of 400-channel rigid EMG grid developed in NISNEM had some drawbacks due to lack of flexibility and therefore adhesion with the skin surface. Therefore, to facilitate testing, refining and validation of our algorithms and models, a new flexible HD-EMG grid of 20x20 electrodes with an inter-electrode distance of 2mm and an electrode diameter of 500µm has been fabricated. The grid has been tested in experimental studies aimed at identifying a large and representative number of sources from EMG recordings (see separate item on dense grids in this section). 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? Yes  
Impact Experiments are being conducted to identify the minimum required inter-electrode distance and an estimation of the maximum identifiable number of MUs from a given muscle. We have shown that denser grids allow for the identification of smaller (deeper) MUs, which are usually overpowered by the neighbouring larger ones. 
 
Title A Soft Exoskeleton for Single/Multi-DoF Ankle Assistance/Rehabilitation 
Description NISNEM activities towards translation are focused on interfacing for exoskeletons, prostheses, and virtual reality. The ankle soft exoskeleton developed in NISNEM comprises an embodiment and an actuation pack. The embodiment provides an interface to transmit the force from the actuators to the limb. Two actuation packs were developed. The first is a test platform that can house 4 motors and their controllers, force sensors and string-pot sensors. This allows for simultaneous 2-DoF torque/stiffness control. The test actuation platform also includes safety limitations, such as limit sensors allowing for the testing of novel controller strategies. The second actuation pack is portable and consists of a Maxon motor and controller, string-pot sensor and a force sensor. The portable actuation pack was deployed and tested on a treadmill for drop foot application. Moreover, this exoskeleton technology will be interfaced with the NISNEM interfacing systems (see the description of stretchable grids developed in NISNEM and reported as separate item in this section). 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? No  
Impact The developed ankle soft exoskeleton platform is modular and can be extended from 1-DoF (plantar/dorsi-flexion) to 2-DoF (plantar/dorsi-flexion and inversion/eversion), and will be extended in the near future. This technology has been interfaced with a IMU+sEMG controller for applications in drop-foot impairment. 
 
Title A high TRL stretchable HD-sEMG sleeve for the control of prosthetics and exoskeletons 
Description The high-density non-invasive grids of electrodes developed for laboratory testing are progressively translated in NISNEM into viable interfacing systems for assistive technologies as well as for consumer electronics systems. The fabrication strategy for the development of a stretchable HD-EMG previously detailed has since been developed further, resulting in a third iteration of the sleeve. The fabrication strategy involves the use of a flexible PCB or other substrates consisting of sensors (sEMG, IMU, SMG, MMG, ultrasound ...) or stimulators (vibrotactile, electrotactile, FES,...). Specifically, we have used the fabrication method with surface EMG (sleeve with 64 electrodes) and ultrasound (sleeve with 16-20 transducers). The substrate is then pre-compressed and constrained within a mould. The mould may also have an additional layer of a stretchable fabric that acts as reinforcement. The moulds are then filled with a stretchable polymer, silicone, and rubber. 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? No  
Impact The current iteration of the method has been tested with a HD-EMG flexible PCB and ultrasound transducers, and has resulted in sleeves allowing for easy donning/doffing. We have combined the developed HD-EMG sleeves with commercial and lab-based amplification hardware, and they have resulted in comparable/better performance than commercially available wet electrode systems (while our system is dry electrode-based). The developed sleeves have been used in experiments for control of wheelchairs for hemiplegics (sleeve worn on the ankle) (assistive tehcnology) and worn on the forearm for a framework to identify typing based on HD-EMG activity while typing (consumer electronics application). 
 
Title A platform to study and restore high-dimensional behaviour in rodents 
Description This novel experimental paradigm enables the investigation of brain-wide dynamics, muscle activity, and movement kinematics during high dimensional behaviour in head-fixed mice. This setup consists of: i) a 3D treadmill enabling unrestricted 2D running behaviour in head-fixed mice; ii) an odor delivery system for cued running; iii) an "earthquake" machine for investigation of automated corrective behaviour in response to rapid positional perturbations on the treadmill; iv) a multi-camera setup enabling extracting of movement kinematics via pose estimation algorithms; v) a multi-probe micromanipulator system for precise positioning of up to 4 Neuropixels probes; vi) a 32 channel EMG acquisition system; and vii) a closed-loop system for auditory-based BCI tasks. 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? No  
Impact This experimental setup will enable the investigation of brain-wide contributions to motor control during both cued and spontaneous behaviour, and in response to mechanical perturbations. It will also serve as a platform to develop BCI tasks. Additionally, it will be used to validate the novel electrodes that are being developed in NISNEM, via decoding of movement kinematics and muscle activity from brain recordings during running behaviour, and in performance of the BCI task. 
 
Title A stretchable HD-sEMG sleeve for control of wearable devices such as prosthetics and exoskeletons 
Description We have developed a fabrication strategy that can convert a regular sEMG grid (available commercially) in which the individual electrodes are connected via a flexible PCB into a single stretchable HD-sEMG sleeve. The resultant fabrication allows for the obtained grid to be applied as a patch (e.g. across the chest muscles) or as a sleeve (e.g. around the forearm or shin) and the stretchable nature of the sleeve allows for the entire grid to compression fit across the body, allowing for better quality of signals and for the same grid to be easily applied across individuals of varying anatomy without the need for customised solutions. This development is part of the translational area of NISNEM that aim at translating the NISNEM interfacing technologies into assistive and rehabilitation devices. 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? No  
Impact While the system is under development, once completed, the developed stretchable and wearable grid will be integrated into a textile base resulting in a textile-based structure that can be worn along with a soft/rigid exoskeleton or as a layer in a prosthetic fitting. Outside applications for assistive devices and prosthetics, it could also form part of commercial solutions in research applications as well requiring for conformable HD-EMG systems. The same fabrication strategy can be even scaled outside HD-EMG towards other high-density sensing systems, such as accelerometers to sense tactile feedback, propagation of sensory information across the skin etc. 
 
Title An actuation pack for simultaneous assistance of coupled multi-DoF joints 
Description Exosuits (soft exoskeletons) primarily use a soft embodiment coupled with actuation systems like motors, pneumatic, hydraulic or other systems to provide assistance or resistance to the wearer with an aim to achieve therapeutic or assistive outcomes. Systems driven by motors through cable-based power transmission are widely accepted as a design archetype capable of providing the highest power-to-weight ratio, enabling the system to become wearable and portable. However, one downside of current systems is that one motor is used for either providing the agonist and antagonist movement for a single-DoF joint or actuating a single cable for a multi-DoF joint. This is because a single set of pulleys cannot span the entire range of values required for assisting a movement of coupled joints such as the shoulder, wrist or ankle. Therefore, to assist a complex system such as our upper-limbs, the current state-of-the-art systems would end up requiring a large number of actuators and sensors making the system bulky and impractical. We are in the process of developing a new actuation pack system which requires the theoretically minimum number of actuators and sensors (i.e. one each) for a single DoF irrespective of whether the joint is part of a coupled multi-DoF joint such as the wrist or ankle or a simpler 1-DoF joint such as the knee or elbow. This development is part of the translational area of NISNEM that aim at translating the NISNEM interfacing technologies into assistive and rehabilitation devices. 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? No  
Impact While the system is currently still under development, once tested and finalised, we should be able to considerably reduce the size, complexity and cost impact by close to 50%. This would allow for many of the currently available assistive devices such as exoskeletons to feature this design approach and reduce the mechatronic requirements considerably. 
 
Title High-density (hd) stainless-steel arrays for surface EMG detection (100-400 electrodes) 
Description One of the major goals of the project is to fabricate a polymer-based high-density (HD) non-invasive interface which is aimed to be flexible and bio-compatible. While the fabrication of this novel HD-EMG matrix is underway, comprehensive trial and testing are required for high-density systems in order to design the processing and information extraction algorithms. Therefore, to facilitate development, testing, refining and validation of our algorithms and models, a HD-EMG grid of 10x10 stainless steel electrodes with an inter-electrode distance of 2mm and an electrode diameter of ~500µm has been fabricated. The developed HD-sEMG stainless steel arrays will be used to record and decompose sEMG signals and will also be used to benchmark the polymer-based electrode technology. Concurrently, the obtained HD-sEMG data will be used to train and inform our models and decomposition algorithms. Currently, this 100-channel grid is being used to record sEMGs from the abductor pollicis brevis of the palm and flexor and extensor muscles of the forearm. As a next step, we are currently working on a 20x20 matrix of 400-channels with an inter-electrode distance of 1.25 mm. This new matrix which is currently under development is aimed at understanding the theoretical and practical limit of the electrode density which will further allow us to inform and optimise the design of the polymer-based HD-sEMG matrix to enable us to optimize the detection of MUs. 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? No  
Impact Experiments have been scheduled to extract information about the minimum required inter-electrode distance and an estimation of maximum identifiable MUs from a given muscle over. Simultaneously, the data recorded from these electrodes will be also used to improve EMG generation models and decomposition algorithms. 
 
Title Methods for low-noise, high-gain and high-precision biosignal recording in the presence of strong electrochemical DC offsets & stimulation artefacts. 
Description The novel method allows for the high-quality recording of EMG (and other bioelectric, e.g. EEG, ECG) signals even in the presence of very strong offsets; such offsets are often introduced by dry electrodes fabricated by means of experimental technologies and processes. The method: i) can be adopted either in low area profile PCB designs and/or monolithic (microchip) designs, ii) has led to the realisation of novel, "offset-robust" front-ends for the acquisition of EMG (but also of ECG and EEG) signals and iii) has been successfully tested with experimental/prototyped PEDOT PSS electrode arrays. 
Type Of Material Improvements to research infrastructure 
Year Produced 2021 
Provided To Others? No  
Impact Realisation of compact (area of ~ 2 business cards), high-count (256 channels), wearable/portable, battery-operated EMG recording devices/modules. 
 
Title Myolink-2: Miniaturised offset-robust 256-channel wearable EMG recording device 
Description Wearable battery-operated 256-channel micro-instrument benefiting from novel offset-robust recording channel topologies and state-of-art performance. Enables high-density sEMG recordings in a mobile setting. SUMMARY Number of channels: 256 Volume: 10.2cm x 6.7cm x (<1cm) Midband gain: 46dB (or 200V/V) up to 67 dB (or 2400 V/V) Bandwidth: 10Hz-500Hz Power supply: 3V (single) DC offset rejection: rail-to-rail Total power consumption: 32 mA drawn from (single) +3V power supply: 96 mWs Integrated noise: 600nV rms (10-500Hz) Input impedance: 5G? 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? No  
Impact The methods and architectures adopted for the realisation of Myolink-2 enable the de-risked migration to a third generation of low area profile and lower power EMG micro-instruments comprising 1024 high-performance recording channels. 
 
Title Numerical EMG model 
Description Acquiring experimental EMG data in sufficient quantity and quality is expensive, time-consuming, and in many cases, impossible. Realistic EMG signals can be generated by solving the electrostatic equations of the volume conductor. For this purpose, a suitable measure of the anatomy, e.g. by MRI, can be used as a basis for a numerical (finite element model, FEM) solution. The issue with FEM solutions is that the computational time required to obtain the solution is extremely high. Here, we proposed a realistic numerical EMG model that decreases the needed computational time by orders of magnitude with respect to previous solutions. This allows to generate a very large number of simulated signals that can be used for training neural networks for signal analysis and generation. This numerical model is at the basis of the AI-based simulation of EMG, also developed in this reporting period (described as separate item in this section). The numerical model has been accepted for publication in Nature Communications (see publications). 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? Yes  
Impact It was possible to simulate arbitrary large datasets of realistic electromyography signals with high internal variability and leverage it to train deep learning algorithms. Moreover, the computational model provides access to all the diverse hidden parameters of the simulation and allows to use some annotation strategies that are impossible with experimental data. 
 
Title Real time biofeedback based on the decoding of motor neuron activity 
Description The software was coded as a MATLAB app (version 2021a, The MathWorks, Inc, USA). It allows researchers to record and process signals from one to four grids of surface electrodes or intramuscular arrays with up to 64 channels. Of note, the current version has been developed to interface with only one commercialised multichannel EMG system (EMG-Quattrocento, 400 channel EMG amplifier; OT Bioelettronica, Italy). However, the blind source separation algorithm in the code can be used with all the systems recording EMG signals with grids of surface electrodes or intramuscular arrays of electrodes. The framework to perform real-time identification of motor neuron activity has four steps. First, a segment of EMG signals is recorded while participants perform a contraction at the requested intensity such that a mask is manually generated to remove channels with artifacts or low signal-to-noise ratio for the rest of the session. Second, the force offset is measured and removed before performing MVC. The measured MVC is used to standardize all the submaximal isometric contractions. Third, a baseline contraction is performed such that the motor unit filters are identified using offline blind source separation of EMG signals. Fourth, the motor unit filters are applied over incoming segments of EMG signals during a test contraction to identify motor units firing activity in real-time. Four forms of feedback can be displayed to the participant: a raster plot of all the motor units discharge times for a given muscle, a quadrant displaying the cumulative spike trains (CST) of two groups of motor units for a given muscle, a quadrant displaying the firing rate of two motor units, and the smoothed discharge rates of all the identified motor units for a given muscle. We demonstrated that: i) the computational delay to decompose the EMG signals in real-time is low, i.e., 0.37 ms per motor unit when high-density EMG electrodes are used, ii) the smoothed motor unit discharge rates displayed to the participant was accurately estimated, with a RMSE value lower than 2 pulses per second. Overall, this open-source software provides a set of tools for neuroscientists to design experimental paradigms where participants can modulate the neural drive to their motor units in real-time using visual feedback. 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? Yes  
Impact This open-source framework provides data, codes, and guidelines for researchers who aim to study the neural control of movement at the motor neuron level. This thread of research has been peaking recently, especially in the field of behavioural and computational neuroscience, where research teams start using concurrent recordings of neurons at the cortex and spinal level to link the neural activity to the motor output of the human body. 
 
Title Real-time censored decomposition of multi-channel EMG signals with limited computational resources 
Description Electromyography (iEMG) decomposition identifies motor neuron (MN) discharge timings from interference EMG recordings. When this is performed in real-time, the extracted neural information can be used for establishing human-machine interfaces, which are crucial for the translational activities of NISNEM. In our previous work, we developed a real-time decomposition algorithm of multi-channel EMG signals, based on a Bayesian modelling and filtering framework and a decomposed-checked channel strategy. All the decompositions were achieved under the available computational resources. However, for very complex multi-channel signals, containing a large number of MUs, the mathematical scenarios in the decomposition may exceed the limit of computational resources. In this case, we proposed the censored decomposition, in which the motor units (MUs) with low energy action potentials are considered as noise and only MUs with high energy potentials are processed with the Bayesian modelling and filtering algorithm. The new algorithm was validated in a set of 16-channel experimental signals acquired from the muscle abductor digiti minimi. The decomposition accuracy was more than 90%. The results are currently under review in a Journal paper submission. 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? No  
Impact Our proposed censored decomposition method enlarged the range of processible signals of the previous proposed real-time decomposition algorithm of multi-channel EMG signals recorded by thin-film electrode arrays in human. It lays a good foundation for further research of novel spike sorting algorithm. It will allow the design of viable human-machine interfacing systems. 
 
Title Simulation and experimental analysis of ultra-dense electrode grids 
Description We combined computational and experimental approaches to assess how the design parameters of surface grids of EMG electrodes influence the number and type of motor units identified by EMG source separation algorithms. We first simulated a pool of motor unit action potentials and computed the percentage of these action potentials that could be discriminated from all others when recorded with grids of electrodes of various sizes and densities. We then identified motor units from experimental EMG signals recorded during isometric contractions with grids of electrodes with a range of densities. We also tested a novel grid with 400 electrodes and 2-mm interelectrode distance. 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? Yes  
Impact The design parameters of EMG electrode grids have never been discussed and are usually arbitrarily fixed, often based on commercial availability. In this study, we showed that using larger and denser grids of electrodes than conventionally applied can drastically increase the number of motor units identified from surface EMG signals. Moreover, the identified motor units are more balanced between high- and low-threshold ones which provides a more representative sample of the neural drive to muscles. Gathering large datasets of motor units using large and dense grids, with the parameters proposed in this study, will impact the study of motor control, neuromuscular modelling, and human-machine interfacing. The study also fully justify the NISNEM approach of substantially increasing the density and number of electrodes. 
 
Title Techniques for neural recording and stimulation under anaesthesia 
Description We have set up the methods to perform brain (intracortical, ECoG) and muscle (EMG) recordings in anaesthesised rodents. These methods have been combined with a robotic platform to provide controlled mechanical perturbations, and a neurostimulation device to artificially activate the nervous tissue. 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? No  
Impact We have collected the first results to validate the NISNEM EMG electrodes in mice during anaesthesia. This approach avoids the spatial and temporal filtering imposed by the soft tissues that affects surface recordings, and will permit refining our signal decomposition algorithms. By applying nerve or brain stimulation in the anaesthesised animal, we can also test decomposition algorithms under various controlled recruitment strategies. 
 
Title Ultra-fast simulation of MUAPs during dynamic movement with a deep generative model 
Description Simulations of biophysical systems are fundamental for studying physiological mechanisms and developing information extraction algorithms. Whilst numerical methods provide such simulations, they are rarely used in simulating dynamic events because of the inhibitive computational cost. However, all biophysical systems have dynamic behaviors. We propose that an alternative method is to use deep neural networks to replicate the dynamic system by transfer learning. We present a hybrid-structured generative model that can rapidly predict the outputs of a dynamic volume conductor system by implicitly interpolating the system states in an arbitrary temporal resolution. This development is part of the signal processing and AI activities of NISNEM that generates the dataset for validating the decoding of multichannel signals during dynamic movements. 
Type Of Material Improvements to research infrastructure 
Year Produced 2023 
Provided To Others? Yes  
Impact The proposed transfer learning method with conditional generative models is a viable solution for dynamic simulation with any numerical models. This allows for simulating system states during dynamic events in high temporal resolution, for example analysing prosthesis contact forces during movement and electro-magnetic fields with volume conductor deformations. 
 
Title Wearable, low-noise, 512-channel HD sEMG interface. 
Description The acquisition system is made up of several customised 7cm x 10cm PCB modules featuring 128 channels each. A custom miniaturized analogue-front-end (AFE) has been designed for this purpose which delivers <1µV¬RMS input-referred noise whilst consuming <4mW power per channel. The AFE can acquire up to 15mVpp AC signals at a 10-500Hz bandwidth whilst suppressing DC offsets of up to 1.2V. Analogue to digital conversion is performed at 2KHz using 16-bit resolution. Two modules (256-channels) plus one custom controller PCB, to collect/transmit the data to the PC via USB, have been used to build the acquisition system (total volume 14cm x 9cm x3cm). Two acquisition systems can be used in tandem to enable recording of 512-channels. 
Type Of Material Improvements to research infrastructure 
Year Produced 2022 
Provided To Others? No  
Impact During the development of this system we have realized several low-noise sEMG acquisition prototypes. One such prototype was used to practically investigate the effect of amplifier input-referred-noise on the performance of HD sEMG decomposition. Another such prototype was utilized to enable acquisition of surface EEG/EMG/ECG from novel PEDOT nanowire electrodes developed by our NISNEM collaborators. Moreover, knowhow acquired during these studies has been applied to optimize power/performance/area trade-offs for the final AFE design. 
 
Title An AI-based EMG Generation Model - MorphWave 
Description MorphWave is an AI-based EMG generation model, that takes a novel approach to traditional analytical and numerical methods. The developed method (MorphWave) simulates motor unit action potentials (MUAPs) under a great number of conditions in a short time. MorphWave achieves this performance using conditional generative adversarial networks (GANs) to transform the input MUAPs to synthetic new ones with modified attributes. The model is broadly made up of two components - the generator and the discriminator. The generator component of the model is trained with source MUAPs (obtained from experimental data) and conditions to change (such as position of the motor unit, number of fibres etc.) to generate simulated MUAPs, representative of the changed conditions. Another module named discriminator is trained to distinguish between the artificially generated MUAPs from the real ones. Together, the generator and the discriminator compete with each other such that the generator trains itself to output more and more realistic MUAPs that are indistinguishable from the real experimentally-obtained source MUAPs. In MorphWave, the model has been trained while changing a total of six conditions including: 1). the depth of the motor unit, 2). the medial-lateral position of the motor unit, 3). the number of fibres, 4). the location of the innervation zone, 5). the propagation velocity of the current source, and 6) the fibre length. This development is part of the signal processing and AI activities of NISNEM. These modelling developments allow us, on the one hand, to simulate realistic signals to test/validate the processing algorithms and, on the other hand, to be the basis of new decoding algorithms that use the forward modelling information to solve the inverse problem. 
Type Of Material Computer model/algorithm 
Year Produced 2022 
Provided To Others? No  
Impact To the best of our knowledge, this is the first AI-based EMG generation model. Even with only one MUAP, MorphWave is capable of generating a large number of variations of this MUAP in a short time by feeding the model with diverse conditions. We see our developed model to be used for data augmentation, inverse modelling, and to test the performance of new EMG decomposition algorithms. 
 
Title BioMime, a deep conditional generative model for learning biophysical systems 
Description BioMime, the AI-based EMG simulation model, presented in the research tools has been made publicly available, together with example of simulations. 
Type Of Material Computer model/algorithm 
Year Produced 2023 
Provided To Others? Yes  
Impact Our proposed generative model, BioMime, can rapidly simulate realistic action potentials of thousands of motor units during dynamic movement. This allows a multi-domain simulation that combines the simulation of mechanical movement and neural activities. The publicly available version of this model (see URL below) allows all EMG researchers to simulate the most realistic signals to date with extremely low computational load and time. 
URL https://github.com/shihan-ma/BioMime
 
Title Real-time decomposition of multi-channel EMG signals recorded by thin-film electrode arrays in humans 
Description Electromyography (iEMG) decomposition is a mathematical technique that aims to identify motor neuron (MN) discharge timings from interference EMG recordings. When performed in real-time, the extracted neural information can be used for establishing high information transfer human-machine interfaces capable of achieving natural and intuitive control of devices, such as prosthetics and exoskeletons. We developed a real-time decomposition algorithm based on a Hidden Markov Model of EMG. This pattern analysis and classification method uses GPU-implemented Bayesian filter to estimate the spike trains of motor units (MU) and their action potentials (MUAPs). Moreover, parallel implementation in multiple GPU clusters was enable to maintain real-time performance in a multi-channel case. The Bayesian modelling and filtering was implemented through a multi-channel framework that processes neighbouring channels which share the activity of the same MUs. Additionally, a decomposed-checked channel strategy handles MU activities appearing in distant channels, and assigns channels into different groups to be processed in related GPU clusters. The algorithm was validated on six 16-channel simulated signals generated by an accurate model of EMG generation, and three 32-channel experimental signals acquired from the tibialis anterior muscle with thin-film array electrodes. All signals were decomposed in real time with an average decomposition accuracy >90%. This development is part of the signal processing and AI activities of NISNEM that aim at accurate decoding of multichannel signals into individual neural sources (in this case, individual spinal motor neuron activities). 
Type Of Material Computer model/algorithm 
Year Produced 2022 
Provided To Others? No  
Impact Our proposed multi-channel EMG decomposition algorithm can process multi-channel signals and is capable of tracking 20-30 motor units (MUs) simultaneously in real-time with high accuracy. This performance enables our method to be applied to multi-channel electrode arrays to establish human-machine interfaces with high-information transfer, allowing for accurate and real-time control of robotic systems such as exoskeletons and prosthetics. The proposed multi-channel EMG decomposition algorithm establishes a firm foundation for further research in novel spike sorting algorithms. Based on the proposed Bayesian modelling and filtering, the team will aim to develop new AI-based methods capable of efficiently discarding unnecessary scenarios, reducing computational time, and increasing convergence speed. 
 
Title Ultra-fast and highly realistic numerical modelling of surface EMG 
Description A fundamental component of the development of methods for information extraction from biosignals is the availability of forward generation models. Models provide the means to benchmark processing algorithms, to augment data for training deep learning networks, and to solve the inverse problem. However, current biosignal models are either simplifications of the experimental conditions or require impractical computational resources and time. We have developed a computationally efficient and realistic model of electric fields in body tissues (volume conductor), with a specific application to electromyography (EMG). The volume conductor is described from magnetic resonance images (MRI) of tissue surfaces by numerical discretization. The numerical solution of the forward model is optimized by reducing the main calculations to the solutions in a minimal number of basis points, from which the general (scalable) solution is then derived. This new numerical method allows for almost real-time simulations, without any constraints on the complexity of the volume conductor or of the electric sources. These numerical modelling developments have been fundamental for advancing AI-based modelling (this numerical model served as training for the AI-based models), here described as another output of the project. 
Type Of Material Computer model/algorithm 
Year Produced 2021 
Provided To Others? Yes  
Impact The computational time magnitude of previous state-of-the-art models is, in the best cases, in the order of hours for a single simulation (with a fixed set of model parameter values, approximately 50000 muscle fibres, and 5 recording electrodes. By exploiting the mathematical properties of the forward equations and source model, we were able to achieve a computational performance of the order of minutes per simulation. Moreover, in our model, changing most of the simulation parameters does not require recomputing the whole model and reduces computational time of new simulations to the order of seconds, if the volume conductor remains constant. As a result, it becomes practically possible to simulate arbitrary large datasets of highly realistic EMG signals with high variability in the simulation parameters. This is fundamental for data augmentation or even for building digital twins and directly train decoding algorithms on simulated data. 
URL https://www.biorxiv.org/content/10.1101/2021.06.07.447390v1.full.pdf
 
Description EPSRC Network+: Neurotechnology for enabling community-based diagnosis and care 
Organisation University of Birmingham
Country United Kingdom 
Sector Academic/University 
PI Contribution Because of the visibility of the work performed in NISNEM, my team has been invited to participate to a EPSRC Network+ application on neurotechnology for diagnosis and care, led by the University of Birmingham. We have just been informed that this network has been funded. Diagnostics is one of the future core applications of NISNEM technology and therefore this network will be fundamental for disseminating and further expanding the potential applications of the output of the project. My team fully participated to the conceptualization of the network focus, contributing specifically with the inclusion of non-invasive technologies for diagnostics in the network. Moreover, the focus of our network is the development of technologies that shift the emphasis of diagnosis and management of neurological conditions from hospital to non-hospital environments, which is also a core target of NISNEM.
Collaborator Contribution The partners in NISNEM contributed to the concepts included in the network proposal. The network is fully within the scope of NISNEM and will serve to broaden the dissemination of NISNEM technologies.
Impact The EPSRC Network+ on Neurotechnology for enabling community-based diagnosis and care will commence its activities this year. Updates will be reported in the coming reports.
Start Year 2021
 
Description EPSRC-Facebook/Meta co-sponsored PhD scholarship on human augmentation by wearable systems 
Organisation Facebook
Country United States 
Sector Private 
PI Contribution This is a joint sponsorship of PhD scholarship by the EPSRC CDT AI4Health (Imperial College) and Facebook Reality Labs for a 4-year project dedicated to human augmentation. Facebook is a partner of NISNEM and agreed to support the project in both direct financial contribution and in-kind. For this scholarship, in addition to the co-sponsoring of the scholarship, Facebook also provided wearable prototype technology from their labs (division of Ctrl-Labs) as well as time of their R&D team for regular meetings and discussion. The contribution of my team to this collaboration is in the provision of NISNEM methods for extracting information at the spinal motor neuron level from skin sensors (algorithms, AI) and the development of approaches that would allow augmenting human motor functions (neuroscience research).
Collaborator Contribution Etienne Burdet, co-I in NISNEM, is providing his expertise in human augmentation and robotics for this collaboration. Juan Gallego, co-I in NISNEM, is also collaborating on this topic. Juan Gallego and my team have secured a PhD student to extend on these topics (Imperial President's scholarship confirmed and offer accepted by the student to start in the academic year 2022-2023).
Impact A first scientific paper resulting from the first year of this collaboration is ready for submission. The PhD student has successfully passed the Early Stage Assessment and preparing for her Late Stage Assessment.
Start Year 2021
 
Description EPSRC-Huawei co-sponsored PhD scholarship on minimally invasive wearable systems 
Organisation Huawei Technologies Research and Development UK Ltd
Country United Kingdom 
Sector Private 
PI Contribution This is a co-sponsored PhD scholarship by Huawei within the CDT in AI4Health (Imperial). The project is on the development of thin-film subcutaneous (epimysial) electrode arrays and decoding algorithms for next generation minimally invasive interfaces with VR and AR. The project is well in line with the developments of NISNEM (Huawei is a partner of NISENM). My team is providing the technology (electrodes and hardware), as well as the algorithms for signal decoding.
Collaborator Contribution Juan Gallego, co-I of NISNEM, will contribute to the animal tests of the proposed electrodes. The team of Rylie Green is testing PEDOT coating of the thin-film electrodes.
Impact The PhD project has started in 2021. The student has successful passed the early stage assessment and is preparing for the late stage assessment. The first human experimental implants will be performed in spring 2022 (ethical approval already secured).
Start Year 2021
 
Description Imperial-Meta Wearable Neural Interfaces Research Centre 
Organisation Facebook
Country United States 
Sector Private 
PI Contribution Facebook Reality Labs supported the project NISNEM at the time of application. Given the progresses of NISNEM in the first 2 years, Facebook (Meta) decided to commit substantially more than initially foreseen. The PI of NISNEM, Dario Farina, is the Director of the new Centre, fully sponsored by Facebook for >£5m. The Centre activities have started on the 1st of January 2023.
Collaborator Contribution The research Centre is hosted by Imperial College. The NISNEM PIs Gallego and Drakakis are also part of the management committee of the new Centre, together with Farina. The Centre activities will focus on the translation of NISNEM technology into consumer electronics products.
Impact The Centre activities have started this year, therefore the first results will be reported at the next reporting period.
Start Year 2022
 
Description NeuroMod+: Co-creation for next-generation neuromodulation therapeutics 
Organisation University of Oxford
Country United Kingdom 
Sector Academic/University 
PI Contribution My research team has contributed to the concept of the network, with discussions on the content, mainly in relation to NISNEM technologies.
Collaborator Contribution Rylie Green, co-I in NISNEM, is co-I in this network proposal and extensively contributed to the conceptualization of the network focus. The network will focus on addressing the challenge of minimally invasive treatments for brain disorders, including non-invasive methods developed in NISNEM for monitoring and treating. The network has just been accepted for funding and includes new collaborations with the University of Oxford, the University of Nottingham and the University of Edinburgh. This network and the other newly funded EPSRC Network+ on Neurotechnology for enabling community-based diagnosis and care (see separate item in this section of the report) cover two fundamental core areas of NISNEM - treatment and diagnosis of neurological disorders - and will be essential in accelerating the dissemination of NISNEM technologies.
Impact This network will commence this year. We will update on the outputs in the following reports.
Start Year 2021
 
Title 3D PEDOT:PSS / PU electrodes 
Description Poly(3,4-ethylenedioxythiphene):polystyrene sulfonate (PEDOT:PSS) and polyurethane (PU) dry electrodes with hair-like microstructures were developed to overcome the limitations of conventional flat electrodes which are limited by their ability to penetrate hair to make appropriate skin contact. Casting methods were used to produce these 3D electrodes, which is preferable to the more expensive etching and lithography techniques. 
Type Of Technology New/Improved Technique/Technology 
Year Produced 2022 
Impact Conventional wet electrodes use an electrolytic gel, known to cause dermatological reactions and dry out, requiring continuous replacements. Dry electrodes are a promising replacements but often have high impedance especially on hairy skin, for example when used for electroencephalography (EEG). 3D PEDOT:PSS/PU electrodes have the potential to revolutionise the field of wearable electronics and permit long-term and at-home health monitoring. 
 
Title PEDOT:PSS / PU microelectrode arrays 
Description Poly(3,4-ethylenedioxythiphene):polystyrene sulfonate and polyurethane (PEDOT:PSS / PU) microelectrode arrays were developed by solvent-casting and laser machining. These microelectrodes have good biocompatibility, low Young's modulus, and good mechanical and conductive properties, improving on common metal-based dry electrodes. The fabricated electrode arrays were tested as EMG, EOG, ECoG, ECG and EEG devices showing comparable performance to commercial electrodes. 
Type Of Technology New/Improved Technique/Technology 
Year Produced 2022 
Impact Conventional wet electrodes have severe limitations as electrolytic gel dries over time and is unfit for long-term continuous monitoring. Dry electrodes are a suitable alternative for long-term recording. However, metal based dry electrodes have higher impedance due to the direct contact with the skin and due to the stiffness, it does not conform to the skin well and producing irreproducible results. PEDOT:PSS / PU microelectrodes perform as well as conventional electrodes and can be easily shaped for a multitude of applications via laser machining. 
URL https://onlinelibrary.wiley.com/doi/full/10.1002/anbr.202100102