Wearable devices and methods of manufacturing the same

Abstract

The present invention provides a wearable device for placement on the skin of a wearer and configured to receive and / or transmit signals from or to the wearer, the device having a plurality of layers including: a first, textile layer having a first surface and a second surface opposite the first surface and having a plurality of conductive electrodes formed on the first surface and electrically connected through the first layer to the second surface; and a second layer, formed on the first layer, wherein the second layer has a plurality of conductive vias formed therethrough and a plurality of conductive tracks, the conductive vias connecting the electrodes of the first layer to the conductive tracks, wherein the plurality of conductive vias define a first array, wherein each of the plurality of conductive tracks has a contact point distal from the via to which it is connected, the plurality of contact points defining a second array; and wherein the area covered by the second array is smaller than the area covered by the first array.

Description

[0001] WEARABLE DEVICES AND METHODS OF MANUFACTURING THE SAME

[0002] The present invention relates to devices and methods of manufacturing. The invention is of particular, but not exclusive, relevance to wearable devices, for example those configured to detect and monitor physiological signals from a human or animal body.

[0003] Electrophysiology, the study of the electrical properties of biological cells and tissues, has long been a cornerstone in both research and clinical settings. By measuring the electrical activity of neurons, muscle cells and other electrically excitable cells, electrophysiology provides invaluable insights into the functioning of the heart, brain, and other organs. Traditional electrophysiological techniques, such as electrocardiograms (ECGs), electromyograms (EMGs) or electroencephalograms (EEGs), have become essential tools in diagnosing and monitoring a variety of conditions, from cardiac arrhythmias to neurological disorders. However, the clinical landscape is witnessing a significant shift towards high-density body surface potential mapping (BSPM). This advanced approach allows for more precise spatial resolution and detailed characterisation of electrical activity, which is crucial for understanding complex physiological phenomena and improving patient outcomes.

[0004] Flexible substrate electrode array fabrication techniques employing polymeric substrates such as polyimide (Kapton) are at the forefront of this transition due to their excellent thermal and chemical stability, durability and flexibility. However, despite their advantages, polymeric- substrate-based high-density electrode arrays face significant limitations, particularly in terms of conformability. The rigid nature of these substrates can hinder their ability to closely adapt to the heterogeneous and dynamic surfaces of biological tissues, potentially compromising the accuracy and efficacy of measurements.

[0005] In response to these challenges, e-textiles have emerged as a promising alternative fabrication method. E-textiles combine the electrical functionality of traditional sensors with the flexibility and comfort of fabrics, offering a new level of adaptability and integration with the human body. However, current e-textile technologies have two significant limitations that need to be addressed: the limitations on achievable resolution (usually ~l mm) and the integration of reliable electrical connections. For example, for certain electrode arrays, in order to obtain sufficient resolution from the electrodes, it is desirable that the electrodes form an array covering a relatively large area on the subject. However, connecting such electrodes to processing electronics, whether as part of the wearable device itself, or via a wired or wireless connection normally necessitates individual wires from each electrode, often covering significant distances. This is cumbersome and can lead to devices being impractical outside a clinical setting and / or to reduced user acceptance.

[0006] In other electrode arrays it is desirable to have a high density of electrodes in a relatively small area (i.e. with the electrodes close to each other). In such configurations connecting the electrodes to processing electronics can also be challenging due to the high density of the electrodes and the small space available.

[0007] Furthermore, many current electrodes dry up over time, which limits their usefulness for longterm recording / monitoring.

[0008] There is therefore a need for improved wearable electrode arrays which are also desirably more convenient for the wearer.

[0009] It is an aim of the present invention to overcome one or more of these limitations. This could pave the way for a wide range of future applications in both clinical and research settings. These include, without limitation, Holter monitors, ECG, EEG, EMG, mechanical heart conditions, and the measurement of any other electrophysiology signals.

[0010] At their broadest, aspects of the present invention provide devices and methods of manufacturing devices, where the devices bring together the inputs from a disparate array of sensors distributed across the desired detection area to a convenient connection configuration having a smaller area.

[0011] A first aspect of the present invention provides a wearable device for placement on the skin of a wearer and configured to receive and / or transmit signals from or to the wearer, the device having a plurality of layers including: a first, textile layer having a first surface and a second surface opposite the first surface and having a plurality of conductive electrodes formed on the first surface and electrically connected through the first layer to the second surface; and a second layer, formed on the first layer, wherein the second layer has a plurality of conductive vias formed therethrough, the conductive vias connecting the electrodes of the first layer to a plurality of conductive tracks, wherein the plurality of conductive vias define a first array, wherein each of the plurality of conductive tracks has a contact point distal from the via to which it is connected, the plurality of contact points defining a second array; and wherein the area covered by the second array is smaller than the area covered by the first array.

[0012] The device of this aspect can enable an array of electrodes to be provided which are positioned in a known relationship to each other. The device can also enable these electrodes to be arranged on a scale which is desirable for the sensing of, for example, physiological signals. The arrangement of the electrodes may also be on a scale suitable for relative discrimination of signals from a wearer, whilst being connected to a common interface, which may be connected to the contact points, which is configured on a (generally much smaller) scale suitable for electronics (for example, for connection to electronic components and / or to a common communications link). This configuration can mean that it is not necessary to have (multiple) individual wires exiting the device which connect to the individual electrodes.

[0013] Preferably the first and second layers are flexible and / or the device as a whole is flexible. This can permit the device to be worn by a user and to conform to the portion of the wearer’s body to which it is applied.

[0014] In certain embodiments, the device is part of, or is connected to, other elements which are configured to enable and assist in the application of the device to a wearer and to keep the device in a stable and consistent position on the wearer. These may include stretchable fabric elements, fasteners (such as hook and loop type fasteners) and / or adjustable straps or ties.

[0015] The conductive tracks may be formed on or in the second layer, or may be formed separately, for example as part of printed circuit board which is connected to the second layer. The printed circuit board having the conductive tracks may be flexible, which facilitates the device being worn by a user and conforming to the portion of the wearer’s body to which it is applied.

[0016] Preferably the device is configured to conform to the skin of the wearer. The textile layer preferably makes the device comfortable for a user to wear against their skin. This can help to provide a device which has better user acceptance than the application of individual electrodes to the skin.

[0017] The textile layer is preferably porous in order to allow the electrodes to be formed by printing the conductive material of the electrodes onto one side of the textile material and for the conductive material to pass through and be incorporated into the textile layer without significant sideways spread, such that the electrodes formed on the first and second surfaces of the textile layer exhibit good conductivity across the textile layer and are of approximately similar dimensions (ideally substantially identical sizes) when viewed in the plane of the device. The porosity of the textile layer can also assist in rendering the device as a whole breathable (i.e. such that air and moisture will pass through the device). This can improve comfort and / or wearability for the user.

[0018] The textile layer is preferably made from a textile that is resilient to multiple heating processes at over 100 degrees without losing its shape, form or elasticity. This enables the various layers of the device to be formed on the textile layer and cured without damaging the desirable properties which the textile layer brings to the device.

[0019] The device may include one or more printed circuit boards electrically connected to a plurality of said contact points. The printed circuit board(s) may contain processing electronics for processing signal inputs from the electrodes. Alternatively or additionally the printed circuit board may provide for communication to an external device, either through a wired or wireless connection. In embodiments which provide for a wireless connection, the device may have an antenna, which may, for example, be formed on or attached to the printed circuit board. The printed circuit board may be flexible.

[0020] In certain embodiments the device further includes a third layer formed on the second layer, the conductive tracks and contact points being encapsulated by the third layer such that the conductive tracks are not accessible through the third layer, but the contact points are accessible. This ensures that the conductive elements are sealed within the device and therefore can reduce or eliminate adverse effects from the contact of these elements with the environment. A lamination approach such as this can also be effective to overcome wiring scalability limitations for example by permitting multiple layers of tracks and vias in order to incorporate the connection of a large number of electrodes to corresponding contact points.

[0021] The device may also include a coating on the electrodes on the first side of the textile layer. Many current electrodes dry up over time, which limits their usefulness for long-term recording / monitoring, but the provision of the coating can reduce or prevent this. In particular embodiments, the coating is a eutectogel.

[0022] The vias, tracks and contact points may be formed any conductive material. In certain embodiments they may be formed of one or more of silver (and / or a silver-based substance such as silver nanoparticle inks), graphite, copper, gold or any other commonly employed metals or conductive polymers (such as PEDOT:PSS).

[0023] The device may have any number of electrodes (and corresponding numbers of vias, tracks and contact points). In some embodiments, two electrodes are sufficient for useful physiological monitoring. However, in other embodiments, there are at least 6, at least 10, at least 15 or at least 20 electrodes, vias, tracks and contact points.

[0024] The electrodes may be arranged in a regular pattern but need not be. The regular pattern may be such that the electrodes are at substantially the same distance from each neighbouring electrode when viewed in the plane of the electrodes. This pattern may be polyhedral, for example hexagonal, which is a configuration which provides for a compact but regular distribution of the electrodes.

[0025] The area covered by the first array may vary considerably depending on the intended application (including the type of animal to which the device is intended to be applied). Typically the area covered by the first array is at least 1 cm x 1 cm (i.e. 1 cm2). In certain embodiments the area covered by the first array is at least 25 cm2, at least 100 cm2, or at least 1000 cm2. For application to larger animal subjects, the area covered by the first array may be at least 1 m2, at least 10 m2or at least 100 m2.

[0026] The contact points may be arranged in a regular pattern but need not be. For example, the contact points may be provided to connect to different printed circuit boards. The area covered by the second array is typically of the size of known printed circuit boards. Typically this area will be at least 0.25 cm2(0.5 cm x 0.5 cm) and at most 2500 cm2(25 cm x 25 cm). In some embodiments the area covered by the second array is no more than 1000 cm2, no more than 500 cm2, no more than 100 cm2or no more than 25 cm2.

[0027] The ratio of the area covered by the first array to the area covered by the second array may be 2:1, preferably at least 5:1, more preferably at least 10:1. In some embodiments the ratio may be considerably larger, for example at least 100:1 or at least 1000:1.

[0028] The devices of this aspect can have mm-size, low-impedance conducting polymer coated electrodes. Such a device can potentially achieve unprecedented spatial resolution on BSPM and other electrophysiological measurements.

[0029] The devices can also have long-term recording capabilities as the electrodes will not dry up in the manner of known polymer electrodes.

[0030] Prototype arrays with optimally chosen conducting polymer formulation and geometrical parameters have been fabricated and measured on polymeric substrates to demonstrate the detection of physiological signals (including ECG, EMG, EEG, etc.) with comparable signal- to-noise ratio (SNR) to larger area commercial Ag-Ag chloride (Ag / AgCl) electrodes. Alternatively, for similar area arrays, improved SNR can be achieved.

[0031] Furthermore, the devices according to this aspect can provide high-density electrode arrays which allow additional spatial information to be obtained from physiological signals.

[0032] The device of the above aspect may include some, all or none of the above-described optional and preferred features in any combination.

[0033] A second aspect of the present invention provides a method of fabricating a wearable device, the method including the steps of: forming a plurality of conductive electrodes on a textile layer, wherein the electrodes extend from a first side of the textile layer to a second, opposite, side; forming a second, insulating layer on the second side of the textile layer, overlaying the electrodes; forming through-holes in the second layer at the locations of the electrodes; applying conductive material to the through-holes to form conductive vias connected to the electrodes, the plurality of conductive vias defining a first array; and forming a plurality of tracks on the second layer, each track connecting one of the conductive vias to a contact point distal from the conductive via, the plurality of contact points forming a second array, wherein the area covered by the second array is smaller than the area covered by the first array.

[0034] The method preferably further includes the step of forming a third layer over the tracks such that the conductive tracks are encapsulated between the second and third layers, but the contact points are accessible through the third layer. This ensures that the conductive elements are sealed within the device and therefore can reduce or eliminate adverse effects from the contact of these elements with the environment. A lamination approach such as this can also be effective to overcome wiring scalability limitations for example by permitting multiple layers of tracks and vias in order to incorporate the connection of a large number of electrodes to corresponding contact points.

[0035] The method preferably further includes the step of connecting a printed circuit board or other electronic component to the contact points.

[0036] The step of forming the electrodes may use blade-coating or screen-printing. Blade-coating has been found to be particularly effective in causing the applied conductive material to pass through the textile layer in order to be incorporated into the textile layer without significant sideways spread, such that the electrodes formed on the first and second surfaces of the textile layer exhibit good conductivity across the textile layer and are of approximately similar dimensions (ideally substantially identical sizes) when viewed in the plane of the device.

[0037] The method may include the further step of forming a eutectogel coating on the electrodes on the first side of the textile layer. Many current electrodes dry up over time, which limits their usefulness for long-term recording / monitoring, but the provision of the eutectogel coating can reduce or prevent this.

[0038] The method of this aspect may include some, all or none of the above-described optional and preferred features in any combination.

[0039] Preferably the method is a method of making the device of the above first aspect, but need not be. Where the method is a method of making the device of the above first aspect, it may include some, all or none of the above-described optional and preferred features of that aspect in any combination.

[0040] Unless indicated otherwise, any of the features (including the optional or preferred features) described in relation to one of the above aspects are equally applicable in combination with the devices, systems and methods of any of the other above-described aspects.

[0041] The invention is described below, by way of example, with reference to the accompanying figures in which:

[0042] Figure 1 shows a cross-sectional view through a device according to an embodiment of the invention, along with top views of the device at the layers indicated;

[0043] Figure 2 shows schematic views of various parts of a device according to an embodiment of the present invention;

[0044] Figure 3 shows views of a finished device according to an embodiment of the present invention;

[0045] Figure 4 shows flexible PCBs which can be used in the manufacture of a device according to an embodiment of the present invention;

[0046] Figure 5 shows a cross-section view through a device according to a further embodiment of the invention which includes a flexible PCB;

[0047] Figures 6(a), 6(b) and 6(c) show, respectively, a device according to an embodiment of the present invention, the installation and connection of that device to a wireless transmitter and electrophysiological signals across varying frequency ranges and amplitudes recorded from the device;

[0048] Figure 7 shows a schematic and simplified example of electrode and connector spacings;

[0049] Figure 8 shows the progression of the ratio between areas as a function of number of electrodes for a specific device design to illustrate the scalability approaches adopted in embodiments of the invention; Figure 9 shows the use of cardiac BSPM for posture recognition; Figure 9(a) shows a 16 electrode array design according to an embodiment of the present invention, its implementation and relative placement, Figure 9(b) shows interpolated maps of signal delay between individual electrodes and a reference electrode on the top left for different body postures as shown (from top left to bottom right: supine, upright, Fowler’s and left lateral recumbent postures);

[0050] Figure 10 shows performance metrics of posture identification from precordial ECG BSPM as measured by the device of Figure 9(a), with confusion matrices obtained from the evaluation of samples in the testing set from both standard heart rate features (left) and spatio-temporally rich delay features (right);

[0051] Figure 11 shows the use of forearm muscular BSPM for the distinction of grasped object shapes; Figure 11(a) shows a 16 electrode array design according to an embodiment of the present invention, its implementation and relative placement; Figure 11(b) shows interpolated BSPMs of rectified EMG envelopes for each individual channel when grasping objects with different shapes as illustrated (from left to right: spherical grasp, cylindrical grasp and unladen hand);

[0052] Figure 12 shows performance metrics of a grasped object shape recognition system from forearm EMG BSPM as measured by the device of Figure 11(a), with confusion matrices obtained from the evaluation of samples in the testing set from both single electrode SNR features (left) and spatio-temporally rich SNR features (right);

[0053] Figure 13 shows the use of cerebral BSPM for the distinction of sensorimotor stimuli; Figure 13(a) shows a 16 electrode array design according to an embodiment of the present invention, its implementation and relative placement, Figure 13(b) shows interpolated BSPMs of EEG frequency power bands from four canonical frequency bands, theta (4 Hz to 8 Hz), alpha (8 Hz to 13 Hz), beta (13 Hz to 30 Hz) and gamma (30 Hz to 40 Hz), obtained when exposing the participant to different sensorimotor cues as indicated above (from left to right: visual stimuli, auditory stimuli and hand closure);

[0054] Figure 14 shows performance metrics of a sensorimotor stimuli detection system from cerebral EEG BSPM as measured by the device of Figure 13(a), with confusion matrices obtained from the evaluation of samples in the testing set from both single electrode EEG power band features (left) and spatio-temporally rich EEG power band features (right);

[0055] Figure 15 shows a multi-modal high-density electrode array recording of muscular and cortical electrophysiology; two movements are compared: hand flexion (left) and wrist extension (right); Figure 15(a) schematically shows the multi-modality recordings with relative array placement; Figure 15(b) shows simultaneous cerebral power band mapping of beta wave activity and muscular BSPM of the forearm immediately preceding movement onset; Figure 15(c) shows simultaneous cerebral power band mapping of beta wave activity and muscular BSPM of the forearm during movement;

[0056] Figure 16 shows a spatial analysis of reaction time from multi-modal, simultaneous cortical and muscular high-density BSPM as produced from the configuration shown in Figure 15(a); Figure 16(a) is a descriptive illustration of the monitored reaction time: motor commands are planned, initiated, travel along the central and peripheral nervous system and reach the forearm muscles to produce a contraction; Figure 16(b) shows reaction time maps for hand flexion (top) and wrist extension (bottom), the values are on average 300 ms and the variation between channels reaches maximum values of approximately 6 ms; Figure 16(c) is a histogram of reaction times across all trials for hand flexion and wrist extension including Gaussian fittings;

[0057] Figure 17 shows a multimodal BSPM for the prediction of muscular activity maps from cortical mappings; Figure 17(a) is a descriptive illustration of the prediction pipeline: cortical EEG power band maps are given as input to a PLS regressor to predict muscle activity mapping, and the predicted maps are compared with real muscle activity mapping recordings; Figure 17(b) shows example predictions for hand flexion and wrist extension movements, alongside the corresponding ground truth muscle activity maps; and

[0058] Figure 18 shows a large-scale wireless BSPM for arm electrophysiology to demonstrate fabrication scalability; Figure 18(a) shows a 126-channel electrode array implementation according to an embodiment of the present invention; Figure 18(b) shows muscular BSPMs of forearm muscles during hand flexion and different arm movements engaging primarily the biceps, triceps and deltoid, including relative placement of the devices during each set of movements. Figure 1 shows a cross-sectional view through a device according to an embodiment of the present invention, and is also used to illustrate the method of manufacturing such a device according to a further embodiment of the present invention.

[0059] The electrode layer 20 is formed by means of a double-sided, Ag nanoparticle blade-coating technique. Initially, the textile layer 10 was pre-stretched against a flat surface and, whilst held in position, patterned with an adhesive mask formed by double-sided tape Causing tension on fabrics before deposition has been shown to minimise the absorption coefficient of textile fibres. 50% pre-stretching over the original length of the textile was thus implemented, causing a significant decrease in sideways spread of the Ag ink, the formation of a uniform conducting saturation layer and an increase in shape resolution. The blade-coating causes the Ag nanoparticle ink to penetrate the textile layer 10 so that contact electrodes are formed which extends through the textile layer from the body-contacting side (the lower side in Figure 1) to the contacts side (the upper side in Figure 1).

[0060] The double-sided tape mask is then removed from one of the textile sides and a new mask from the same material is attached to define the edges of the hexagonal profile of the device 1 when viewed from above. Non-cross-linked polydimethylsiloxane (PDMS) is poured over the electrodes and blade-coated with a squeegee to remove excess, generating a uniform insulation layer 30 moulded by the double-sided tape mask, see Figure 1(a). The PDMS layer 30 is subsequently cross-linked in an oven (105°C for one hour). The PDMS layer 30 penetrates into the textile layer 10 and thus bonds the device together.

[0061] In order to prevent lack of insulation at the edges of the Ag layer, a second double-sided tape mask with the same shape is attached over the first one and a second deposition of PDMS is performed. Once cross-linked, the PDMS is rastered in circular regions over the electrode areas by means of a laser cutter, see Figure 1(b).

[0062] A mask of silicon-based tape is then generated with the shape of the rastered holes and Ag epoxy is applied by means of a small swab to form vias 25. The silicon-based tape mask is removed and the epoxy is cured in the oven (105°C for 30 minutes), see Figure 1(c).

[0063] A second mask of silicon-based tape is produced with the track layer pattern (for example as shown in Figure 2(a)) and adhered to the PDMS substrate 30. Ag nanoparticle paste is then blade-coated on the PDMS surface 30 with a squeegee to form the tracks 50 connected to the vias 25. The mask is removed and the Ag is cured in the oven (105°C for 30 minutes), see Figure 1(d).

[0064] In order to prevent the cured Ag tracks 50 from peeling from the PDMS substrate 30, noncross-linked PDMS 80 is carefully applied to the tracks with a brush, leaving the areas of contact for the pads in the centre of the structure uncovered, see Figure 1(e). The PDMS is cross-linked in the oven (105°C for one hour) and a third silicon-based tape mask with the central pads pattern is attached to this PDMS layer 80. Ag epoxy is applied again by means of a small swab to form vias 60 and cured (105°C for 30 minutes) after removal of the mask, as shown in Figure 1(f).

[0065] The resulting structure is covered with a fourth silicon-based mask with the same central pads pattern and Ag nanoparticle paste 65 is blade-coated with a squeegee over the Ag epoxy contact pads 60 to form contact points for further connection, see Figure 1(g). While the Ag is uncured, the custom PCB 70 which is used to interface the electrodes 20 with external electronics is placed over the Ag paste 65 by aligning the bottom pads of the PCB 70 with the top pads of the device 1. The Ag paste is then cured in an oven (105 °C for 30 minutes) and a final doublesided tape mask with the shape of the device contour is attached over the previous two to allow for a final non cross-linked PDMS deposition 90 which both seals the PCB against the substrate and increases the robustness to bending of the Ag tracks.

[0066] The PDMS is cross-linked in the oven (105°C for one hour), followed by the removal of the double-sided tape masks used for contour, see Figure 1(h). Finally, a eutectogel (ETG) coating 15 is deposited at the other side of the textile layer 10 and the tape mask on that side is removed after ETG cool-down.

[0067] To improve the integration of the laminated layers on the textile and the overall robustness of the device 1 , a silicone rubber, such as PDMS, was applied in a layer 30 to surround and contain the conductive tracks 50. Applying PDMS to textile in its non-cross-linked, liquid state has been shown to allow for the creation of dyeing patterns embedded within a fabric. A similar approach was implemented to insert further track lamination layers over the blade-coated Ag electrodes 20, as shown in Figure 1 and Figure 2a. The device 1 achieves multi-layer lamination by consecutively alternating between PDMS insulation 30, 80, 90 and Ag nanoparticle paste wiring 20, 50, 65. The tracks connecting to electrodes further away from the external wire interface run above those of closer electrodes by means of intermediate insulation layers.

[0068] In order to retain a single interfacing plane, the contact points for the deeper tracks pierce through these layers by means of vias 25, 60 implemented with depositions of Ag epoxy (8330S-21G, MGChemicals). The external routing is performed by means of a custom-made interfacing printed circuit board (PCB) 70 inserted at the top of the upper PDMS layer, as shown in Figure 2(b). The bottom side of the PCB 70 consists of 19 via holes placed at specific locations designed to match those fabricated onto the device. The top side connects those holes to the pads of a surface-mounted device footprint corresponding to a standard vertical, 20 ways, 0.5mm pitch ZIF connector 100. The PCB 70 is connected to the device and sealed with an additional layer of PDMS 90.

[0069] The final device 1 manufactured using the above methods is shown in Figure 3. Figure 3(a) shows top and bottom views of the device 1 and illustrates the relative size of the PCB 70 and the pattern of electrodes 20. Figure 3(b) shows the device 1 incorporated into a textile strap which is sewn to the textile layer 10 of the fabricated device. Given that the PCB 70 allows for any footprint design to be included, it is possible to minimise the number of external wires required to interface with higher density modules.

[0070] The adhesion of the array to the body can be tuned for different individuals by employing a hook and loop fastening structure at the edge of the textile substrate 10. The use of a hook and loop fastening system is not only convenient for fitting the device to different subjects and / or to different portions of the subject anatomy, but also to avoid removal of the ETG coating material on the electrodes through shear stress.

[0071] The fabrication process outlined in Figure 1 can be adapted to reduce the number of fabrication steps and enhance mechanical robustness. Flexible printed circuit boards (FPCBs) 110 may be designed to replace the blade-coated silver tracks 50, as illustrated in Figure 4. These FPCBs 110 feature pads 120 on the bottom side (Figure 4(a)) with dimensions matching the Ag epoxy vias 25 formed above the textile electrodes, while the top side (Figure 4(b)) houses the connector interface 100 for external electronics. Beginning from the stage shown in Figure 1(c), an additional layer of Ag epoxy can be applied to both the electrodes and the bottom-side pads of the FPCB. Prior to curing, the electrodes and the FPCB are aligned, brought into contact and co-cured in an oven at 105°C for 30 minutes to establish electromechanical connections. Following this, a layer of uncured PDMS is applied across the entire assembly to seal the external connector.

[0072] A device fabricated using this modified method is shown in Figure 4(c).

[0073] Figure 5 shows a cross-sectional view through a device 1 ’ according to a further embodiment of the present invention which includes an FPCB such as that described above. Similar or identical features to those in the device shown in the cross-section of Figure 1 have been given the same reference numerals and the same or similar manufacturing methods can be used for the device of this embodiment and will not be further described.

[0074] In the device 1 ’ of this embodiment, the vias 25 connect to tracks 50 formed in a flexible PCB 40. The tracks 50 extend through the FPCB 40 to the contact points 65 which are variously connected to a connector 100 and to a PCB 70’.

[0075] Although this approach can reduce fabrication complexity and enhance connection reliability, the use of Kapton-based FPCBs with copper traces results in a structure that is less flexible and less breathable. This limitation can be partially addressed through careful PCB layout design, for instance by incorporating meandering tracks or segmented interconnects to improve bendability. Ultimately, the choice between the two approaches to fabricated the device will depend on the specific requirements and constraints of the intended application.

[0076] Performance evaluation via wireless multi-site electrophysiology

[0077] To evaluate the signal recording performance of the devices manufactured according to the approach described above with respect to Figure 1, the electrode array shown in Figure 6(a) was implemented. The array comprises five electrodes 15, each with a 5 mm radius and a IED of 5 cm, arranged in a single line. The tracks 50 carrying electrical signals from the electrodes to external electronics converge into a hexagonal pattern, to which an interface PCB 130 is bonded, as detailed above. As shown in Figure 6(b), the underside of the interface PCB 130 includes vias aligned with the converging track pattern. The top side houses a 20 -channel board-to-board connector. To process the analogue signals from the electrodes, a dedicated analogue front-end (AFE) PCB 130 was fabricated with a matching 20-channel connector on the lower face, allowing direct interfacing with the wearable e-textile array.

[0078] The top side of the AFE PCB 130 contains a set of ICs and passive components performing several functions. First, instrumentation amplifiers increase signal amplitude while maintaining high input impedance, which maximises SNR by preventing saturation and minimising distortion. The amplified signals are then filtered using active bandpass filters with fixed cutoff frequencies between 0.5 Hz and 100 Hz. Subsequently, the signals are digitised by an analogue-to-digital converter and routed to female header pins for connection to a wireless unit 150, which is a commercially available Bluetooth low energy (BLE) PCB. The wireless unit 150 transmits digitised signals to a receiver connected to a computer running a custom user interface, enabling real-time monitoring and recording onto user-friendly Excel files. The active components of the wireless transmission system are powered by a low-profile lithium- ion battery (not shown) positioned between the AFE PCB 140 and the Bluetooth module 150. The device supports continuous operation for up to five hours before requiring recharging via a USB-C port. The wireless system has an effective transmission range of approximately 20 m.

[0079] As shown in Figure 6(c), electrophysiological signals across varying frequency ranges and amplitudes were successfully recorded using the 5 -electrode linear array. For clarity, only a single channel is depicted. Figure 6(c) shows, from top to bottom: precordial ECG, forearm EMG, EOG and EEG signals corresponding to the alpha and theta bands. All signals were notch-filtered at 50 Hz to minimise power line noise. The ECG signals recorded from the chest clearly display the P wave, QRS complex and T wave. The subject’s heart rate during the trial was estimated from the R-peaks, yielding approximately 100 bpm. Muscle activity in the forearm was recorded during continuous wrist flexion.

[0080] Although EMG signals are typically approximately five times stronger than ECG signals, the recorded EMG envelope appears to be of similar magnitude. This discrepancy can be attributed to two factors: firstly, the electrode dimensions are optimised for capturing ECG signals, which have lower frequencies, resulting in partial averaging of the higher-frequency EMG signals; secondly, the AFE PCB filters, designed with cardiac applications in mind, impose a high cutoff frequency of 100 Hz, thereby excluding the higher-frequency components of the EMG signal (typically between 100 Hz and 400 Hz). Thus, the plotted EMG represents only a fraction of its full frequency spectrum.

[0081] Electrical activity due to blinking was also recorded when the electrode array was positioned on the forehead. Increased electrical activity observed during blinking corresponds to the EOG signal generated by eyelid muscle contractions. In contrast to forearm EMG signals, EOG signals occupy a lower frequency range (5 Hz to 100 Hz), which falls within the passband of the AFE filters. As a result, the recorded EOG signal appears nearly twice as strong as the filtered EMG signal. EOG and EEG signals were recorded simultaneously from the forehead. To isolate EEG components, the signals were band-pass filtered between 4-7 Hz (theta) and 8- 12 Hz (alpha). A median detrending filter was applied to remove EOG interference. As observed, closing the eyes led to an increase in both alpha and theta wave activity. This phenomenon occurs because the reduction in visual input promotes a more relaxed brain state with lower cognitive load. Reduced activity in the visual cortex enhances alpha wave expression, while the transition into a relaxed mental state increases theta wave activity.

[0082] External wiring scalability

[0083] Starting from the design specifications of both electrode arrays and PCB pads it is possible to determine the rate of size increase of the PCB connector area with respect to the electrode array area. The increase in the length of the main axis in the hexagonal structure of both electrodes and pads follows a linear relationship. Figure 7 shows a schematic and simplified example of electrode and connector spacings. Using these values and the general linear equation y = mx + n,

[0084] PADS

[0085] 11 = 2m + n 20 = 3m + n

[0086] => n = 11 - 2m 20 = m + 11

[0087] => m = 9, n = -7

[0088] ELECTRODES

[0089] 40 = 2m + n 70 = 3m + n

[0090] => n = 40 - 2m 70 = m + 40

[0091] => m = 30, n = -20 The ratio between electrode and pad length increase is given by,

[0092] The limit as x reaches infinity is given by L’Hopital’s rule as lim = 3.3 which indicates the length of the long axis of the electrodes tends to be three times larger than that of the PCB pads, regardless of the number of electrodes.

[0093] The area ratio between electrodes and pads can be obtained from the definition of the area of a hexagon,

[0094] 3A / 37A = - D2

[0095] 8 where D represents the long axis of the hexagon. Plugging in the expressions for D calculated for electrodes and pads yields the following expression,

[0096] Applying L’Hopital’s rule yields

[0097] 900 11. 1 which indicates the area covered by the PCB pads remains approximately 11 times smaller than that of the electrodes, regardless of the number of electrodes. The progression of the ratio between areas as a function of number of electrodes is depicted in Figure 8. This result demonstrates the external wiring scalability of the proposed e-textile, multi-layer lamination high-density electrode array fabrication process.

[0098] Examples

[0099] Devices according to embodiments of the invention have been tested in a range of potential applications which are detailed further below. Gravitational assessment of cardiac activity with BSPM

[0100] The force of gravity continuously interacts with the human body, and anatomically, adaptations have evolved to counteract its effects. For example, the muscles of the oesophagus perform autonomous peristaltic movements to transport ingested material into the stomach. This function operates independently of body orientation and can act against gravity. Similarly, a set of venous valves - commonly referred to as “swallow’s nest” valves - located in the lower limbs, ensure that blood from gravity-affected regions returns to the heart. These valves prevent retrograde flow and rely on leg muscle contractions to assist venous return. Such mechanisms illustrate the extent to which human physiology is adapted to the persistent influence of gravity, a principle that applies broadly across both animal and plant life. Astronauts in the International Space Station have enabled valuable observations regarding the physiological effects of microgravity on the human body. However, most existing data are collected post-mission, offering insight into the consequences of microgravity but limited information on continuous in situ changes. To address this gap, non-invasive, easy-to-deploy and functionally versatile monitoring devices are required.

[0101] High-density electrode arrays according to embodiments of the present invention have the ability to detect both electrical signals and morphological or mechanically induced changes in cardiac function reflected on electrical activity. The suitability of such devices for continuous, posture-sensitive monitoring has therefore been tested. In particular, a 16-electrode BSPM array was developed using the multi-layer lamination process described above, as illustrated in Figure 9(a). The array incorporates electrodes with diameters of 10 mm and a fixed IED of 15 mm, dimensions optimised for cardiac BSPM.

[0102] The array was integrated into a sports t-shirt and positioned over the left side of the chest. Cardiac electrical activity was recorded from all electrodes during two-minute trials in four distinct body positions: supine (lying face up), upright (standing), Fowler’s (seated at a 60° angle) and left lateral recumbent (lying on the left side). The signals from the 16 electrodes were band-pass filtered between 0.5 Hz and 20 Hz, notch filtered at 50 Hz and detrended using a median filter.

[0103] Heart rate was calculated by identifying R-peaks and measuring R-R intervals across each trace. This measure served as a reference ECG parameter to demonstrate the limitations of standard ECG in detecting morphological variations in cardiac activity caused by gravitationally induced pressure changes in different postures. Delays were computed by measuring the time difference between R peaks from each electrode and a reference electrode located at the top left of the array as an example of spatio-temporal feature containing information on propagation speed.

[0104] These delays, shown in Figure 9(b), varied consistently with electrode distance from the reference. In particular, delays were found to be greater in the downward direction than laterally, consistent with the slower conduction velocities of ventricular tissue compared to atrial tissue. Among the body positions studied, the supine posture exhibited the greatest delay magnitude, followed by upright, Fowler’s, and left lateral recumbent. The supine position increases venous return and preload, while also allowing abdominal organs to exert direct pressure on the heart, factors that may contribute to the observed delay. Despite also increasing cardiac pressure, the left lateral recumbent position demonstrated the smallest delays, an outcome that suggests additional physiological factors may be involved and warrants further investigation.

[0105] Features extracted from the electrode array were used to train two logistic regression classifiers with identical hyperparameters. Approximately 500 samples were obtained for each classifier (approximately 120 cardiac cycles in 2 minutes x four positions). The data was split into training and testing sets using an 80:20 ratio with stratification to ensure class balance. To evaluate model performance, accuracy, precision, recall and FI score metrics were computed. In order to ensure a robust model evaluation to data splitting, repeated cross-validation was employed by performing standard 10-fold cross-validation five times with different random data splits. In each repetition, the dataset was divided into ten sections, with each fold used once for validation while the remaining folds were used for training. Performance metrics were computed for every fold across all repetitions, and the final results are reported as the mean and standard deviation.

[0106] As shown in Figure 10(b), using only heart rate across all channels did not achieve class separability. The repeated cross-validation resulted in accuracy, precision, recall and FI scores of 0.49 ± 0.06, 0.48 ± 0.05, 0.49 ± 0.05, and 0.48 ± 0.06, respectively. While pure chance would yield an accuracy of 0.25, the model performed better due to identifiable heart rate variations across trials. In contrast, when spatio-temporally reach delay features were used to train the classifier, it could distinguish between postures effectively, achieving an accuracy, precision, recall and FI score of 0.97 ± 0.02, 0.98 ± 0.02, 0.98 ± 0.02, and 0.97 ± 0.02, respectively.

[0107] These results confirm that high-density electrical mapping of cardiac activity can detect morphological changes in the heart. Furthermore, delays encoding information about propagation speed and direction can accurately capture the effect of gravity on cardiac activity. This capability may support future continuous monitoring of microgravity’s effects on the heart.

[0108] Shape detection from BSP M for neuroprosthetics

[0109] The intricate musculature of the human forearm enables the complex and varied movements required for hand dexterity and these muscular patterns are crucial in the control of neural prostheses. Surface EMG can provide a non-invasive means of capturing electrical activity during voluntary muscle contractions and advances in high-density electrode arrays have allowed for more detailed spatial resolution of these signals. To study this, a 16-channel high- density EMG array according to an embodiment of the present invention was developed and deployed on the forearm of a subject to capture muscle activity patterns during functional hand grasps.

[0110] The ability to infer object shape based solely on forearm muscle activation has implications for next-generation neural prostheses, not only in enabling context-aware grasp execution, but also in potentially guiding sensory feedback systems to evoke object-specific tactile sensations. The array, fabricated using the multi-layer lamination process outlined above, consists of electrodes with 1 mm diameter and a fixed IED of 10 mm, dimensions optimised for EMG signals. As shown in Figure 11(a), the array was positioned longitudinally along the forearm over the primary muscle fibres involved in hand flexion.

[0111] Three grasping conditions were tested: spherical grasp, cylindrical grasp and an unladen hand (no object grasped) as shown schematically at the top of Figure 11(b). During each trial, the participant was initially instructed to relax the forearm for a duration of ten seconds to enable baseline noise recording. This was followed by a sequence of steady muscle contractions lasting five seconds, interleaved with five seconds of rest. Each contraction-rest cycle was repeated ten times within a single trial. A total of ten such trials were performed for each grasp condition, resulting in a comprehensive dataset capturing variability in muscle activation across repetitions and sessions.

[0112] The resulting EMG signals were rectified and band-pass filtered between 5 Hz and 400 Hz to obtain EMG envelopes. These were subsequently visualised as muscular BSPMs, depicting the absolute amplitude of electrical activity during contraction. As seen in Figure 11(b), each grasp produced a distinct spatial distribution of EMG activity. Notably, the no-object condition exhibited the most uniform and consistently high-amplitude EMG activity across the electrode array, indicating relatively unconstrained muscle contraction throughout the recorded region of the forearm. In contrast, both the cylindrical and spherical grasp conditions produced more spatially heterogeneous activation patterns, with notable similarity between them. The cylindrical grasp, in particular, demonstrated greater overall amplitude and a broader spatial distribution of activity within regions also engaged during the spherical grasp. This may reflect increased muscle recruitment enabled by the reduced mechanical constraints associated with holding a cylindrical object, compared to the spherical one, which likely imposes stricter postural constraints and limits compensatory movement strategies.

[0113] To quantify signal quality and assess the discriminative potential of EMG features, the signal - to-noise ratio (SNR) of the EMG envelope was computed for each electrode and used as the primary feature for classification. Each grasp type was performed across 10 trials, producing 10 envelopes of five-second duration per condition and yielding a total of 300 samples across all tasks. A five-second static baseline, recorded at the start of each trial during the relaxation phase was used as the noise reference for SNR computation. To demonstrate the limitations of conventional low-density EMG approaches, the SNR from a single, centrally located electrode was used to train a logistic regression classifier. As anticipated, the resulting model exhibited poor class separability among the three grasp types. In contrast, when the full set of SNR values from all 16 electrodes was used, thereby incorporating spatial information, classification performance improved significantly, highlighting the value of high-density EMG in capturing grasp-specific activation patterns.

[0114] The data was split into training and testing sets using an 80:20 ratio with stratification to ensure class balance. To evaluate model performance, accuracy, precision, recall and FI score metrics were computed. In order to ensure a robust model evaluation to data splitting, repeated cross - validation was employed by performing standard 10-fold cross-validation five times with different random data splits. In each repetition, the dataset was divided into ten sections, with each fold used once for validation while the remaining folds were used for training. Performance metrics were computed for every fold across all repetitions, and the final results are reported as the mean and standard deviation.

[0115] The results are shown in Figure 12. Using only the central electrode SNR feature (Figure 12(a)), the model achieved repeated cross-validation accuracy, precision, recall and FI scores of 0.40 ± 0.08, 0.40 ± 0.14, 0.39 ± 0.08, and 0.34 ± 0.08, respectively. Pure chance would yield an accuracy of 0.33, the model performed marginally better due to identifiable EMG envelope magnitude variations across trials with different objects.

[0116] When leveraging the spatially distributed SNR data (Figure 12(b)), the logistic regression classifier achieved significantly higher performance, with repeated cross-validation accuracy, precision, recall and FI scores of 0.98 ± 0.02, 0.98 ± 0.02, 0.98 ± 0.03, and 0.98 ± 0.02, respectively.

[0117] These findings support the hypothesis that grasp-specific EMG patterns exhibit sufficient spatial distinctiveness to enable accurate classification, even with simple amplitude -based features. This has significant implications for neural prosthetic control, suggesting that spatially distributed EMG signals can improve the resolution and reliability of intent decoding, potentially enabling more dexterous and intuitive prosthetic function.

[0118] Sensorimotor brain signal discrimination via BSPM

[0119] Understanding the cortical response to external stimuli is critical for developing non-invasive, EEG-based brain-computer interfaces (BCIs) capable of decoding sensory and motor events. To study this, a 16-channel high-density EEG array according to an embodiment of the present invention was designed and deployed to classify three distinct types of neural activation: visual input (a white square on a black background), auditory input (a 440 Hz tone) and voluntary motor activity (hand closure).

[0120] The ability to distinguish between these stimuli may support the development of non-invasive interfaces capable of recognising cognitive intent or perceptual state in real time. The array was fabricated using the multi-layer lamination process described above. It comprised 20 mm diameter electrodes with a 60 mm IED, dimensions selected to optimise spatial resolution for cortical signals in the 0 Hz to 40 Hz range. The array was configured to cover major cortical regions, including frontal, central, parietal, and occipital areas, as illustrated in Figure 13(a).

[0121] Each stimulus condition was presented across three trials with a 100 repetitions each, resulting in a total of 300 samples. During the visual stimulus condition, a white square was displayed for two seconds and removed for two seconds, generating an alternating visual input. The auditory condition followed the same temporal structure, with a 440 Hz tone replacing the visual cue. For the motor condition, the participant was instructed to close their hand for two seconds and relax for two seconds, introducing alternating periods of voluntary motor activation and rest.

[0122] The recorded EEG signals were band-pass filtered into four canonical frequency bands: theta (4 Hz to 8 Hz), alpha (8 Hz to 13 Hz), beta (13 Hz to 30 Hz) and gamma (30 Hz to 40 Hz). The fast Fourier transform (EFT) was then applied to one-second post-stimulus segments to compute power within each band. The resulting frequency-domain features formed the basis for stimulus classification.

[0123] The spatial distribution of band power varied significantly across stimuli, as shown in Figure 13(b). In the visual condition, alpha band power was especially prominent in the occipital region, consistent with alpha rebound following visual stimulus cessation, while gamma activity was suppressed, likely reflecting reduced visual perception, feature binding and conscious awareness. For auditory stimuli, theta power was large and widespread across the cortex, particularly in central and posterior regions, with minimal activity in other bands. This is consistent with the established association of theta band activity with auditory perception and integration. Motor-related activity was characterised by dominant beta band power with heterogeneous spatial distribution peaking over frontal and posterior regions, alongside a localised increase in gamma power over the left hemisphere. Occipital alpha activation was also present, likely due to background visual input, and theta band activity was uniform but low in magnitude. Increased beta activity, especially over the sensorimotor cortex, is expected following movement and may reflect event-related synchronisation.

[0124] To assess the ability of EEG features to discriminate between stimulus types, two classification pipelines were implemented using logistic regression with identical hyperparameters. In the first case, a single electrode located at the vertex of the scalp at position Cz was used, emulating a classical low-density EEG configuration. For each repetition, a feature vector consisting of five EFT power values (one per frequency band) was extracted. The resulting classifier demonstrated limited performance, reflecting the challenge of capturing stimulus-specific dynamics from a single location.

[0125] In the second case, EFT power values from all 16 electrodes across the five frequency bands were concatenated into an 80-dimensional feature vector per sample, thereby introducing spatial information. This spatially rich feature set enabled the classifier to better capture the distributed nature of stimulus-specific cortical activity.

[0126] To ensure robust evaluation, the dataset of 300 samples (100 per class) was split into training and testing sets using an 80:20 ratio with stratification to maintain balanced class representation. Repeated stratified 10-fold cross-validation was employed, repeated five times with different random seeds to minimise sampling bias and evaluate the model’s generalisation ability. In each repetition, the data was divided into ten folds, with each fold used once as a validation set and the remaining nine used for training. Accuracy, precision, recall, and FI- score were computed for each fold in each repetition, and the final metrics were reported as mean ± standard deviation across all runs.

[0127] Figure 14 shows the results. As anticipated, the single-channel model (Figure 14(a)) showed poor class separability, with repeated cross-validation accuracy, precision, recall and FI scores of 0.45 ± 0.08, 0.45 ± 0.09, 0.43 ± 0.07 and 0.41 ± 0.07, respectively. Similarly to cardiac and muscular counterparts, the classifiers perform only marginally above chance level (0.33) due to trial baseline value variations.

[0128] In contrast, the model trained on full-array spatio-spectral features (Figure 14(b)) exhibited robust performance, with repeated cross-validation accuracy, precision, recall and FI scores of 0.95 ± 0.04, 0.95 ± 0.04, 0.94 ± 0.04 and 0.94 ± 0.04, respectively.

[0129] These results confirm that EEG responses to sensory and motor stimuli are distributed across cortical regions and frequency bands and that high-density electrode arrays capturing this spatial and spectral complexity can greatly improve classification performance. This capability holds promise for the development of non-invasive BCIs capable of multi-modal cognitive state recognition.

[0130] Multi-modal BSPMfor muscle mapping prediction

[0131] The above sections have demonstrated that spatio-temporal features extracted from high- density cutaneous electrode arrays, when combined with classical ML algorithms, can facilitate novel analytical approaches across a range of electrophysiological applications. All of the presented results have relied on the deployment of a single electrode array, each specifically optimised for the acquisition of one electrophysiological signal modality.

[0132] In the next study, the two high-density electrode arrays previously used for the recording of muscular and cerebral BSPM are concurrently employed to enable the simultaneous recording of muscular and cortical activity at high spatio-temporal resolution (see Figure 15(a)). Beyond illustrating the adaptability and versatility of the e-textile fabrication method described above, these experiments demonstrate the unique capacity of this technology to support the spatiotemporal investigation of complex inter-organ and inter-system interactions. Furthermore, the ability to acquire synchronous, multimodal electrophysiological data can enable the application of more sophisticated ML techniques, facilitating a transition from conventional classification paradigms towards predictive modelling and integrative physiological inference.

[0133] As shown in Figure 15(b) and Figure 15(c), two distinct movements were performed during the multimodal experiments: hand flexion (i.e., closing the fist) and wrist extension. These movements are considered functionally antagonistic, despite targeting different anatomical joints. Hand flexion primarily engages the forearm’s finger flexor muscles, such as the flexor digitorum superficialis and flexor digitorum profundus, whereas wrist extension is mediated by muscles including the extensor carpi radialis and extensor carpi ulnaris. Although these muscle groups act on the fingers and wrist, respectively, their biomechanical actions can oppose each other during coordinated tasks.

[0134] Electrophysiological signals were recorded using the same electrode array placements for both movements, with a sampling rate of 30 kHz. An analog bandpass filter between 0.5 Hz and 400 Hz was applied to capture both EEG and EMG signal components. Subsequently, signals were detrended using a median filter and notch filtered at 50 Hz to remove power fine interference. EEG signals were further bandpass filtered between 13 Hz and 30 Hz to isolate the beta band, which is most strongly associated with voluntary motor activity. EMG signals were bandpass filtered between 5 Hz and 400 Hz and rectified. To reduce high-frequency amplitude fluctuations and produce smoother signal envelopes, a moving average filter was applied to the rectified EMG signals. As a result of this smoothing, the EMG signal amplitude was limited to a maximum value of approximately 100 pV.

[0135] Prior to each movement, the participant was instructed to rest the hand on a table: palm facing upward for hand flexion trials and downward for wrist extension trials. Each movement was performed for a duration of two seconds, followed by a two-second rest period and repeated 100 times per movement condition.

[0136] As expected, during the motor planning phase immediately preceding movement onset betaband EEG activity was elevated, while EMG activity remained at baseline. Notably, beta activity exhibited greater magnitude during hand flexion trials, which may be attributed to inter-trial variability and external factors influencing cortical activation. The spatial distribution of cortical activation was largely similar across both movements, involving widespread brain regions with the exception of the superior right quadrant. Upon movement execution, EMG activity increased sharply, reaching the peak of the rectified and smoothed envelope.

[0137] The pattern of muscle activation differed between the two movements: wrist extension elicited a more uniform and intense response across the electrode array, particularly in the central region, whereas hand flexion was associated with more localised activation, primarily in the bottom right portion of the array, and exhibited overall lower contraction intensity. Consistent with established neurophysiological findings, beta-band EEG activity decreased during movement execution for both conditions. The relative reduction in beta activity was comparable between movements; however, due to the higher baseline beta amplitude observed during hand flexion planning, its post-movement signal retained a higher overall magnitude compared to wrist extension, for which beta activity was nearly absent.

[0138] The simultaneous recording of cortical and muscular activity enables the extraction of complex physiological metrics, such as reaction time, the interval between the initiation of a motor command in the premotor cortex and the resulting muscular contraction. As shown in Figure 16(a), reaction time was computed for motor commands propagating from the motor cortex to the forearm musculature. A single electrode from the EEG array, positioned near the premotor cortex, was used to estimate the onset of motor planning. This onset was defined by the peak amplitude in beta-band activity (13 Hz to 30 Hz), which occurred consistently approximately every four seconds, corresponding to the task cycle of two seconds of contraction followed by two seconds of rest. These peaks typically preceded the onset of EMG activity by approximately 300 ms. Reaction time was quantified by measuring the temporal delay between the EEG-derived onset of motor planning and the initiation of muscle contraction, as detected by the EMG array. For each EMG electrode, the contraction onset was defined as the point at which the rectified and smoothed EMG envelope crossed a predefined threshold. This threshold was set sufficiently above baseline to exclude noise, yet low enough to ensure consistent detection across all channels.

[0139] By applying this method across all forearm EMG electrodes, spatial variations in reaction time were assessed for both hand flexion and wrist extension, as illustrated in Figure 16(b). Average reaction times for both movements ranged between 275 ms and 350 ms, indicating minimal difference in overall motor execution delay between the two tasks. The variability across EMG electrodes remained below 6 ms on average, suggesting high spatial synchrony in muscle activation within each movement condition. The spatial distribution of reaction times provides novel insights into the temporal sequence of muscle fibre recruitment required to execute dexterous movements.

[0140] Small inter-electrode delays reflect the orderly, partially sequential activation of distinct muscle regions, enabling smooth and controlled motion. This hypothesis is supported by the observed patterns in the reaction time maps for each movement. Particularly, hand flexion exhibited longer delays predominantly in the left, right and lower regions of the electrode array, while wrist extension showed maximum delays in the top right and central regions. These opposing spatial delay patterns suggest that antagonistic movements are characterised by inverse sequences of muscle activation, consistent with the physiological requirements for generating coordinated and opposing motor outputs.

[0141] Figure 16(c) presents histograms of reaction time values computed across all trials for each movement type separately. Reaction times range from approximately 100 ms to 600 ms. Although the expected physiological range typically lies between 200 ms and 300 ms, the wider distributions in the recorded data likely reflect the influence of measurement noise and other experimental confounds. Gaussian fits to the histograms reveal mean reaction times centred around 300 ms for both hand flexion and wrist extension, consistent with values reported in electrophysiological studies of corticomuscular conduction.

[0142] Hand flexion exhibits a slightly higher average reaction time and a broader distribution compared to wrist extension. This difference is likely attributable to two main factors. First, as shown in Figure 14(c), the EMG signals during hand flexion are of lower amplitude, resulting in a reduced SNR and greater variability in reaction time estimates. Second, hand flexion may require the recruitment of a more complex set of forearm muscles, involving longer or more sequential activation patterns. These factors likely account for the rightward shift and increased spread in the reaction time distribution associated with hand flexion.

[0143] When forearm EMG BSP maps are viewed as outcomes of cortical activity, the simultaneous recording of brain and muscle signals can be reframed as a regression problem, with muscular maps serving as target outputs. This interpretation enables the development of machine learning algorithms capable of predicting high-density spatio-temporal muscle activation patterns from corresponding high-density cortical signals. As illustrated in Figure 17(a), brain BSPMs associated with specific movements can be used as input to a regression model, which outputs predicted EMG activity maps. Because ground truth muscular activity is simultaneously recorded using the EMG array, prediction accuracy can be quantitatively and visually assessed by comparing predicted and actual contraction patterns.

[0144] Unlike prior experiments described above, where BSPM data were used for discrete classification tasks such as posture recognition, object identification or sensory decoding, this approach represents a more advanced application of machine learning: the continuous -valued prediction of high-resolution muscular activity. This shift broadens the utility of corticomuscular mapping and demonstrates the potential of spatio-temporal regression models in decoding motor intent with high fidelity.

[0145] The model employed is based on the partial least squares (PLS) regression method, a statistical technique that identifies the fundamental relationship between an input matrix X (cortical maps) and an output matrix Y (muscular maps). PLS iteratively determines multidimensional projection directions w and c for X and Y, respectively, that maximise the covariance between the projected data. Specifically, it finds the directions in the input space that capture variance most predictive of the output space by maximising the covariance between X w and Ye. Unlike traditional linear regression, PLS reduces the dimensionality of both input and output data simultaneously by extracting latent components that represent the most relevant patterns linking cortical activity to muscle activation. This approach is robust to multicollinearity in X and effective in situations where the number of features exceeds the number of observations.

[0146] Two PLS regression models with identical hyperparameters were implemented separately for hand flexion and wrist extension. Each model was trained using 100 samples from the same trial. The input features consisted of cortical spectral power maps across the delta, theta, alpha, beta and gamma EEG frequency bands, while the target outputs were the EMG potential maps recorded at the peak of the EMG envelope. The dataset was split into training and testing sets using an 80:20 ratio. To assess the performance of the models in predicting muscle activity, Pearson correlation, Spearman rank correlation and cosine similarity were computed for each prediction in the test set. These metrics were chosen to evaluate different aspects of prediction accuracy, Pearson captures linear correspondence in activation magnitude, Spearman assesses consistency in activation ranking across electrodes and cosine similarity quantifies the spatial pattern similarity independent of scale. Together, they provide a complete and robust evaluation of both the strength and spatial distribution of predicted muscle activity maps.

[0147] Example predictions for both movements, which are shown alongside the corresponding ground truth muscle activity maps in Figure 17(b), demonstrate that the models accurately reproduced the spatial activation patterns of the muscles.

[0148] Although the predicted maps consistently showed slightly lower amplitude than the actual recordings, the overall distribution of activation was correctly preserved. For wrist extension, the average Pearson correlation, Spearman rank correlation and cosine similarity were 0.80 + 0.21, 0.73 + 0.21, and 0.99 + 0.01, respectively. For hand flexion, these values were 0.83 + 0.20, 0.80 + 0.19 and 0.95 + 0.29, respectively.

[0149] The similarity in performance across both movements suggests that the models are robust and capable of reliably predicting high-density muscular activation patterns from cortical spectral features, distinguishing between different motor commands consistently. Large-scale wireless BSPMfor arm electrophysiology

[0150] As analytically demonstrated above, the multi-layer lamination e-textile fabrication method according to embodiments of the present invention is inherently scalable, allowing for the integration of a large number of electrode channels without compromising wearability, comfort or compatibility with external electronics.

[0151] Increasing the number of recording channels enables high-resolution acquisition of BSP maps across larger anatomical regions. This capability opens new avenues for exploring systemic interactions between electrically active organs and systems in the human body. As a proof of concept, a 126-channel array according to an embodiment of the present invention was developed, as shown in Figure 18(a). This array was specifically designed for muscle activity recordings, featuring electrodes with a 1 mm diameter and 3 mm IED. Connectivity to external electronics is achieved via two 70-channel board-to-board connectors.

[0152] The EMG signals from the 126-channel array were acquired using a custom wireless recording unit, compact in size (comparable to two business cards) and powered by a rechargeable battery with a runtime of approximately 2 to 3 hours. The system performs onboard signal amplification and analogue band-pass filtering between 5 Hz and 1000 Hz, transmitting data wirelessly to a standard laptop via Wi-Fi at a sampling rate of 1954 Hz.

[0153] Once digitised, signals were further processed with a digital band-pass filter between 5 Hz and 400 Hz, detrended using a median filter, rectified and smoothed with a 500-sample moving average filter.

[0154] A reference electrode was placed over a bony landmark at the elbow. As illustrated in Figure 16(b), the array was initially positioned longitudinally over the forearm musculature. The participant performed three repetitions of voluntary hand flexion (i.e. finger flexor engagement) consisting of 5 -second contractions interleaved with 5 -second rest periods, preceded by a 10-second baseline recording at rest.

[0155] The array was then repositioned around the upper arm, covering approximately 80% of the arm’s circumference, to record muscle activity during three additional isolated movements. The first was elbow flexion, primarily activating the biceps brachii through a reduction of the elbow joint angle. The second was elbow extension, driven predominantly by triceps brachii contraction to increase the elbow joint angle. The third was shoulder abduction, engaging the middle deltoid by elevating the arm laterally within the frontal plane. Each movement followed the same recording protocol; three 5-second contractions with 5-second rest periods, following a 10-second resting baseline.

[0156] As shown in Figure 18(b), the recorded amplitude of EMG activity from the forearm is approximately four times lower than that of the muscles in the upper arm, which is consistent with anatomical expectations given the smaller size and mass of forearm muscles.

[0157] The maps represent the EMG envelope peaks during each contraction task. In the case of hand flexion, the forearm BSP map displays a heterogeneous potential distribution, indicative of the coordinated activation of multiple muscle groups responsible for finger and digit flexion. Two distinct bands of elevated activity are observed crossing the array diagonally and running roughly parallel to each other. Based on their spatial orientation, the upper diagonal trend likely corresponds to the brachioradialis, while the lower trend aligns with the combined activity of the flexor carpi radialis, palmaris longus and flexor carpi ulnaris. The highest amplitude is located at the centre-left region of the array along the upper diagonal. Anatomically, this location corresponds to the midsection of the brachioradialis, the largest muscle in the forearm, which supports its identification as the primary contributor to this activity pattern.

[0158] Upper arm activity during isolated biceps contraction exhibited the highest amplitude and broadest spatial distribution among the three tested movements, suggesting stronger engagement and likely greater muscle tone or mass relative to the triceps. In contrast, deltoid activity was weakest and lacked a distinct spatial pattern, which aligns with expectations given that the textile array was positioned around the mid-arm region at some distance from the belly of the middle deltoid, where its surface signals would be most prominent.

[0159] Across all three movements, a consistent baseline spatial trend is noticeable, which may stem from variable electrode-skin impedance or from low-level muscle activation common to all tasks. This underlying activity could reflect tonic engagement necessary to slightly abduct the arm and maintain spacing between the arm and torso to accommodate device cabling and ensure wireless connectivity. Biceps contraction was primarily localised to the lower-middle portion of the array, corresponding to the muscle’s cross-sectional centre along the anterior arm. Triceps activity appeared spatially close to the biceps signal, with its centre of activation only marginally shifted towards the central array region. This proximity may be due to volume conduction or overlapping innervation zones, as well as partial activation of synergistic or stabilising muscles. Triceps activation was also of lower magnitude than the biceps, which may reflect its relatively reduced engagement or muscle tone in the participant.

[0160] Overall, these findings highlight the ability of high-density textile EMG arrays to resolve fine spatial differences in muscle engagement across the upper limb, even in closely spaced or anatomically adjacent muscle groups.

[0161] The forgoing description is exemplary in nature only, and the skilled person will understand that changes and variations on the disclosed embodiments are possible within the scope of the claims. The claims define the invention.

Claims

CLAIMS1. A wearable device for placement on the skin of a wearer and configured to receive and / or transmit signals from or to the wearer, the device having a plurality of layers including: a first, textile layer having a first surface and a second surface opposite the first surface and having a plurality of conductive electrodes formed on the first surface and electrically connected through the first layer to the second surface; and a second layer, formed on the first layer, wherein the second layer has a plurality of conductive vias formed therethrough, the conductive vias connecting the electrodes of the first layer to a plurality of conductive tracks, wherein the plurality of conductive vias define a first array, wherein each of the plurality of conductive tracks has a contact point distal from the via to which it is connected, the plurality of contact points defining a second array; and wherein the area covered by the second array is smaller than the area covered by the first array.

2. The device of claim 1 wherein the first layer is flexible.

3. The device of claim 2 wherein the device is configured to conform to the skin of the wearer.

4. The device of any preceding claim wherein the second layer is flexible.

5. The device of any preceding claim further including a printed circuit board electrically connected to a plurality of said contact points.

6. The device of any preceding claim, further including an antenna configured to transmit signals from the plurality of electrodes to a remote receiver.

7. The device of any preceding claim further including a third layer formed on the second layer, the conductive tracks and contact points being encapsulated by the third layer such that the conductive tracks are not accessible through the third layer, but the contact points are accessible.

8. The device of any preceding claim wherein the vias, tracks and contact points are formed of silver or a silver-based substance, graphite, copper, gold, a conductive polymer or composites of these materials.

9. The device of any preceding claim wherein the area covered by the first array is at least two times the area covered by the second array.

10. A method of fabricating a wearable device, the method including the steps of: forming a plurality of conductive electrodes on a textile layer, wherein the electrodes extend from a first side of the textile layer to a second, opposite, side; forming a second, insulating layer on the second side of the textile layer, overlaying the electrodes; forming through-holes in the second layer at the locations of the electrodes; applying conductive material to the through-holes to form conductive vias connected to the electrodes, the plurality of conductive vias defining a first array; and forming a plurality of tracks on the second layer, each track connecting one of the conductive vias to a contact point distal from the conductive via, the plurality of contact points forming a second array, wherein the area covered by the second array is smaller than the area covered by the first array.

11. The method of claim 10, further including the step of forming a third layer over the tracks such that the conductive tracks are encapsulated between the second and third layers, but the contact points are accessible through the third layer.

12. The method of any of claims 10-11 further including the step of connecting a printed circuit board to the contact points.

13. The method of any of claims 10-12 wherein the step of forming the electrodes uses blade-coating, screen-printing or other additive manufacturing techniques.

14. The method of any of claims 10-13 further including the step of forming a coating on the electrodes on the first side of the textile layer.

Images