Deployment of an electrocorticography system with a soft robotic actuator

The eversion mechanism, strain sensor, and neural electrodes were combined into a single deployable ECoG system using a soft micromachining process (Fig. 1A). Two planar elastomeric membranes made of polydimethylsiloxane (PDMS) were stacked to form the implantable interface. The inner surface of the PDMS-made implant was treated to be hydrophilic (fig. S1, C and D), enabling lubrication with biocompatible aqueous fluids. The first membrane hosted the microelectrode arrays and strain sensors (24) prepared with the silicone-on-silicon process (31) and displayed natural dura mater–like compliance. Metallic tracks (Fig. 1Biii) were thermally evaporated (35-nm-thick gold thin film) on top of the elastomer surface to form microcracked elastic interconnects that sustain stretch over 30% strain without electrical failure (24, 32). Electrode sites were coated with a platinum (Pt)–PDMS composite that established an intimate contact with the nervous tissue to support low electrical impedance and mechanical compliance (Fig. 1Bii) (33). Resistive strain sensors were prepared on the elastomeric substrate with similar microcracked gold thin film (34). The second membrane of plain silicone was bonded at the edge of the first membrane using a soluble sacrificial layer as a spacer and completed the leg-shaped tubular stack. A thin flexible printed circuit board was integrated with the interconnects to form a low geometrical profile connector. The detailed fabrication process is available in Materials and Methods and the Supplementary Materials (fig. S1).

Figure 1B shows a photograph of a soft ECoG system containing six actuated legs (6 mm wide and 0.4 mm thick) (Fig. 1Bi). Each leg contains four interconnects, associated microelectrodes (red in Fig. 1Bi) surrounded by a strain sensor patterned on the outer contour of the deployable strip (green in Fig. 1Bi). In preparation for surgical implantation, the ECoG system was secured to a loader connector using a thin layer of vinylsiloxane silicone elastomer to provide a temporary seal between the array and the loader (fig. S2). Biocompatible aqueous solution [such as saline, phosphate-buffered saline (PBS) solution, or dextran solution] was used as a lubricant to reduce layer friction inside the soft ECoG system during folding and deployment.

Next, the ECoG legs were folded into the 20-mm-diameter loader (Fig. 1, C to E) that matched typical 20-mm burr holes. The rigid loader was positioned on the cranium immediately above the preliminary drilled circular burr hole. Figure 1Ei shows the three-step deployment process within a hydrogel phantom human brain [agarose, 0.5 weight % (wt %)] and a transparent plastic (polyethylene terephthalate) skull. A positive fluidic pressure differential was applied inside the loader, increasing the fluidic force F at the everting tip (Fig. 1Ci). The array stayed folded until the fluidic force overcame friction f with respect to the increase in the positive fluidic pressure differential. When the fluidic force became greater than the friction, the folded array was gently pushed out of the tip via eversion by flipping inside out and extended under the stiff skull (Fig. 1Cii). As the eversion created new volume to lengthen the array from the tip, the fluidic pressure differential dropped. The tip growth stopped as the fluidic pressure became smaller than the friction and until the pressure differential increased sufficiently to reiterate the above cycle. Note that the highest-pressure differential for deployment remained constant for a given leg cross-section (mm² range) independently of its length, which is one of the main advantages of the pressure-driven eversion.

A mechanical support made of 50-μm-thick polyimide film prevented the circular PDMS diaphragm at the center of the soft ECoG system from inflating and compressing the brain by bulging out with the positive pressure differential. Figure 1Dii shows a 40-mm-diameter, fully deployed implantable system. During deployment, which typically lasted 30 to 40 s per leg (movie S1), the inflated legs could slightly push against the brain. The resultant compressive strains are discussed in the next section. Once the system was fully deployed and depressurized, the high compliance (Young’s modulus, E, of about 700 kPa) and low profile (thickness, h, of about 0.4 mm) of the deployable ECoG grids allowed intimate contact with the surface of the brain with improved conformability to existing clinical ECoGs (fig. S3). After the ECoG monitoring, the deployed grids could be easily pulled out from the brain, similarly to nondeployable implants (31).

Implementing pressure-driven eversion within the subdural cortical space calls for minimal compression of the underlying brain tissue. To do so, we first investigated the shape and geometrical parameters of deployable legs, namely, the leg width w and thickness h on straight single legs (Fig. 2Ai), the taper angle a on tapered legs (Fig. 2Aii), and the radius of curvature r_c on curved legs (Fig. 2Aiii). Deployment characterization was performed using compressed air to readily monitor the air pressure differential P_a inside structural prototypes as the pressure gradually increased and then suddenly dropped at the moment of deployment. We defined the metric P_d, for deployment pressure, as the maximum air pressure differential to quantitatively evaluate the deployability of each design (Fig. 2B). The highest P_d indicates the most difficult design to deploy. Figure 2C (i and ii) and movie S2 display the deployment of two distinct leg geometries with the tapered and curved configuration. The width and thickness, w and h, of straight legs notably affected P_d during deployment (Fig. 2Di): P_d increased as the leg narrowed and thickened. In the present design, P_d increased by 2.8 times regardless of the width w if the thickness h was doubled. On the other hand, the taper angle (curvature) minimally affected P_d and the leg deployability (Fig. 2D, ii and iii). Increasing the curvature of the legs was of particular interest because larger surface coverage may be reached (Fig. 2Aiii and movie S3). P_d stayed nearly constant, at 10.4 kPa on average, regardless of r_c ranging from 6 to 30 mm, showing that r_c did not have any substantial effects on P_d (Fig. 2Diii).

Upon pressurization, each leg inflated with oval cross-sections owing to its rectangular relaxed shape. Such inflation may locally and momentarily compress the underlying brain. To quantify the resulting displacements, we measured the local strain maps in phantom brain models using a digital image correlation technique and computed them using finite element modeling (Fig. 3). The cross-sectional shape of the inflated leg evolved from oval to circular with increasing P_d (fig. S4A). The circumferential elongation of the leg envelope remained negligible until the circular shape appeared with high enough P_d (fig. S4A, iii and iv). Such large radial expansion was, however, not observed for legs with h < 0.4 mm and 4 mm < w < 8 mm.

To quantify eventual brain compression upon pressurization of the leg, we designed a mock system of the brain and cranium using a soft hydrogel brain model [agarose; 0.5 wt %, E ∼ 5.3 kPa (35)] and a stiff plastic frame, respectively. The brain and skull models are 1 mm apart, accounting for the epidural space (side views in Fig. 3A). The gel was coated with randomly dispersed graphene nanoplatelets to allow for displacement tracking. A straight leg prototype (h = 0.4 mm and a = 0°) was inserted in the gap and pressurized to P_d defined in Fig. 2Di (movie S4). Upon pressurization, the cross-section of the soft leg changed its shape from flat to oval, thereby compressing the hydrogel brain model (front views in Fig. 3A). The resulting density of the graphene nanoplatelets increased in the vicinity of the inflated leg (front view in Fig. 3Aii). Figure 3B displays maps of vertical displacement of the hydrogel brain model computed from the tracked displacement of the carbon suspension (36). The resulting indentation d_i on the surface of the brain model was defined as the vertical displacement of the graphene nanoplatelets referenced to their initial configuration (solid and dashed lines, respectively, in front views in Fig. 3A, i and ii). The center of the coordinate system (or the origin) of the indentation map was defined at the bottom center of the inflated leg as shown in Fig. 3 (Aii and B).

Figure 3C displays d_i along the z axis (x = y = 0) as a function of the hydrogel depth z for different leg widths (5 mm < w < 8 mm). The gel displacement d_i was maximal d_i = d_max immediately below the central axis of the leg (x = y = z = 0) and plateaued at about 0.6 mm at a depth of about 6 mm, independent of the leg width. Wider legs induced larger overall displacement of the underlying gel (d_max = 1.7 to 2.5 mm for w = 5 to 8 mm wide, respectively), although its slope became smaller with the increase in leg width as predicted by the finite element models (FEMs) (Fig. 3D). This may be explained as wider legs being able to deploy with a smaller P_d, thereby inducing lower compression on the brain phantom. The experiments and simulations were in good agreement (Fig. 3D), with an overall deviation of only 0.06 mm from the average of the measured d_max.

With the verified model, we estimated the effect of reduction in device thickness on the indentation of the cortex. We numerically calculated d_max of leg strips with half the thickness (h = 0.2 mm) compared with the experimentally tested legs (h = 0.4 mm). The 0.2-mm-thick leg strip in the FEM was inflated virtually with its corresponding P_d measured in Fig. 2Di (Fig. 3E). d_max for the 0.2-mm-thick leg strip (dotted line) showed about 0.6 mm less indentation on average than d_max for the 0.4-mm-thick leg strip (dashed line). This decrease in d_max was attributed to the prompt eversion associated with 2.8 times smaller P_d for the thinner legs. Note that this FEM computes d_max on the basis of P_d when the implantable system is deployed in an unconstrained environment. The computed d_max values are therefore underestimated. This is because of factors, such as poor lubrication, a reduced gap between the brain and the skull, adhesion between the skull and the dura mater in case of epidural implantation, or high curvature of the skull, that can increase P_d (see the effects of the increased P_d on d_max as shown by the colored solid lines in Fig. 3E).

Next, we investigated the potential of the deployable ECoG system in vivo to record cortical brain activity acutely in a minipig model. Geometrical features of the minipig’s brain (such as curvature of the brain) are close to that of the human brain (37). Design considerations were made to limit the indentation on the cortex to as small as 2 mm. This dimension proved to be a safe limit in human brain surgeries (38). The system also needs to be deployed with minimal fluidic pressure to permit slow and less invasive eversion. We developed an implantable soft ECoG prototype with a straight single leg with 6-mm width, 0.4-mm thickness, and 15-mm length that satisfies the above design criteria. The estimated indentation depth was about 2 mm based on the data in Fig. 3 with the lowest deployment pressure among possible design options. The 15-mm leg strip was long enough to target the rostrum somatosensory cortex when deployed from the top of the brain.

Photographs of the folded (preimplantation) and deployed soft ECoG are shown in Fig. 4A. The deployable system supports a three-by-four electrode array (0.3-mm diameter, 1.5-mm center-to-center pitch) with metallic interconnects and a strain sensor (Fig. 4Aiii) that are made of microcracked gold thin film (Fig. 1Biii) (24). The strain sensor surrounded the array and allowed real-time monitoring of the status of the leg deployment. First, the implant was folded within the holder using a small plastic rod (movie S5). Here, ethanol lubricated the contact surface between the ECoG leg strip and the plastic rod. The lubricant evaporated quickly from the leg surface and disappeared shortly after the plastic rod was removed before the in vivo use. The initial resistance R_o of the strain sensor (in its flat form) was about 10 kilohms and increased up to several megohms upon folding (Fig. 4B). The resistance after folding R_f remained in the megohm range during deployment with an increasing P_a and recovered its initial low value once the array laid flat (Fig. 4C, R_d < 15 kilohms at t > 27 s; movie S5). Although there were two orders of magnitude difference in electrical track resistance between R_f and R_d, the resistance after deployment R_d remained constant after 50 cycles of repetitive folding and deployment (Fig. 4D), indicating that the microcracked gold film sustained such demanding mechanical loading. Figure 4E shows that folding and deployment also had minimal effect on electrochemical impedance of the electrode sites, confirming their robustness to extreme deformations. Instead of using compressed air, deployment could also be induced by incompressible fluid medium (15 wt % dextran in this case), which resulted in slower deployment speed (Fig. 4F). The incompressible aqueous solution filling the entire deployment system prevented buildup of the pneumatic energy compared with the compressed air, which led to fast and uncontrolled deployment (fig. S5 and movie S6).

We demonstrated the in vivo deployment and recording capability of the soft ECoG system via eversion in an anesthetized Göttingen minipig model. The model has a large gyrencephalic brain with a vascular density and skull thickness similar to those of humans (39). We recorded somatosensory evoked potentials (SSEPs) at the surface of the brain in response to the electrical stimulation of the snout. A single-legged, straight ECoG strip (Fig. 4A) was deployed subdurally at the surface of the cortex targeting the frontal lobe and the somatosensory region. The dimensions of the soft implant supported a deployment from the superior part of the skull and thus did not require a craniotomy above the central sinus vein or displacement of the temporal muscle laterally.

The deployable ECoG system was first filled from the inside with PBS and was folded using a plastic rod with ethanol as shown in movie S5. After complete ethanol evaporation, the folded array was thoroughly rinsed again with PBS to make sure that there was no ethanol residue in contact with the brain. After durotomy (Fig. 5Ai; see Materials and Methods for details), the loader was placed at the surface of the brain. The base of the device was placed in contact with the cortex, at the frontal edge of the craniotomy with the folded leg facing the anterior section (Fig. 5Aii). A syringe pump was used to inject PBS into the loader to initiate and control the subdural deployment. The strain sensor output indicated when the soft leg was fully deployed (R_d = 21 kilohms at t = 42 s; Fig. 5B). Next, the loader was removed, and the implant was electrically connected to the electrophysiology recording system.

We recorded evoked potentials from the somatosensory cortex induced by electrical stimulation of the snout delivered at three distinct locations of the snout and two stimulation amplitudes (Fig. 5Ci). Averaged signals from selected channels presented depolarization and polarization peaks (Fig. 5C, ii and iii) and a timing similar to that of a previous report for SSEPs (40). The SSEPs with amplitudes reaching >30-μV peak amplitudes displayed location-specific characteristics of amplitude and waveform that were consistent with a spatial map of the snout representation (41). The position of the deployed array on the somatosensory area was confirmed postmortem (Fig. 5D) (37). The soft implant did not present any folded part and conformed well with the cortical surface (Fig. 5E, i to iii). No visible damage was evident under or in the vicinity of the deployment location after whole-brain extraction. In particular, morphological change or ruptured tissues due to the deployment of the everting strip were not visible (Fig. 5Eiv and fig. S6). Postmortem histochemistry was also performed to further analyze inflammation and neuronal health. Coronal sections were stained using three markers—Iba1, GFAP, and NeuN—for microglial cells, astrocytes, and neurons, respectively, at three different locations under the implanted zone corresponding to base, stem, and tip of the deployed ECoG (fig. S7A). Cross-sectional images confirmed that there was no visible damage outside and under the implanted zone and no indentation of the brain, suggesting that the eversion-based deployment was soft and gentle to the brain tissue as designed (fig. S7, B and F). At the microscopic level, sporadic microglial activation was noticed in the vicinity of the base and the stem (zones i and ii, fig. S7E, i and ii) but not in the tip (zone iii, fig. S7Eiii). No notable glial cell reaction (fig. S7E) or limited astrocyte proliferation was observed (fig. S7C). In addition, neuron imaging showed intact cortical layers (fig. S7D), except for the stem that displayed some loss of neurons on a limited part under the implanted zone (fig. S7Bii). Note that no glial cell activation, astrocyte proliferation, or neuronal loss was observed in the tip where the cortical tissue sustains lower compression than that at the implant base (lower form factor of the implant rounded tip). Although multiple factors (such as indentation depth and deployment speed) can contribute to the damage mentioned above, this result suggests that minor modifications to the current design, such as thinner or narrower legs, may be sufficient to ensure the safety of the implanted brain at a microscopic level. These are initial observations from an acute test (N = 1) that will need consolidation across a larger number of animals and longer time points.

The main objective of this study was to combine soft robotic sensing, an actuation mechanism called eversion, and implantable soft microelectrode arrays for minimally invasive ECoG. Experiments in this study were designed to characterize the effect of various design parameters on the system performance and to validate the proposed functionalities of the implant as follows. The deployment pressure was measured five times (N = 5) for each deployable ECoG strip design corresponding to a different combination of design parameters in leg width, thickness, taper angle, and radius of curvature. The indentation of the hydrogel blocks was measured three times (N = 3) for each straight ECoG strip (h = 0.4 mm) with four different leg widths (w = 5, 6, 7, and 8 mm). The electrical track resistance of the strain sensor was measured by folding and unfolding a single-legged ECoG array 50 times. The electrochemical impedance spectroscopy of electrode sites and gold interconnects was performed on 10 working electrode sites (N = 10). Last, the measurements of deployment speed were performed five times (N = 5) on a straight ECoG strip (h = 0.4 mm, a = 6 mm, and w = 6 mm) filled with two different fluid medium conditions at three different syringe pump speeds (40, 120, and 240 ml/hour). All experiments were performed more than three times (N > 3) to estimate mean value and SD of each measurement. Also, there was no specific treatment of outliers, and all measured data were included. Detailed information about measurement techniques can be found in the Supplementary Materials.

Both functional and structural samples of deployable ECoG systems were made of medical-grade PDMS elastomer (LSR M130, Elkem Silicones) based on the silicone-on-silicon process established in our previous work. Samples prepared for structural characterization were fabricated without the interconnects and Pt-PDMS electrode coatings. According to our experiments, PBS (pH 7.4; Gibco) washes out the dextran coating inside the deployable ECoG strip deposited in the process (fig. S1E) over multiple foldings and deployments. The PDMS surface inside the deployable ECoG tube changed from hydrophilic to hydrophobic, which meant it could not be lubricated with the aqueous solution anymore. Therefore, those structural samples were filled with 15 wt % dextran solution serving as a lubricant for long-term and repetitive eversion. In the case of the functional deployable ECoG strip equipped with electrodes and interconnects for in vivo testing, on the other hand, the sample was filled with PBS to achieve the maximum biocompatibility with the brain tissue for a few instances of eversion within a short period. Please see the “Fabrication methods” and “Experimental methods” sections in the Supplementary Materials for a complete description.

Surgical procedures were approved by the local ethical committee in accordance with the guidelines for the Care and Use of Laboratory Animals and approved by local (Canton of Geneva) and federal (Swiss) veterinary authorities with authorization number GE11120A. The single-legged deployable soft robotic ECoG array shown in Fig. 4Ai was used for the in vivo experiment. A Göttingen minipig of 8.5 kg (3 months old) was anesthetized with a premedication mix of azaperone (0.4 mg/kg), midazolam (0.75 mg/kg), and atropine (0.2 mg/ml) and induced with 6% sevoflurane. After intubation, an intravenous line was placed on the ear for continuous administration of propofol (10 ml/hour) at 2 wt % and along a bolus of fentanyl (2 μg/kg) during the heavy surgical procedures. The surgery was performed as follows: A large frontal-to-posterior incision was made over the midline. The muscle was retracted, and the periosteum was removed using an elevator. The midline and bregma were identified. The craniotomy was planned 2 mm laterally from the midline to 20 mm and −10 mm to +20 mm anterior to frontal. The skull was drilled with a neurosurgery drill (B. Braun ELAN 4 using a 2.5-mm neuro cutter drill bit) until reaching the dura. Then, the boneflap was broken free using a spatula. The dura mater was incised with a microscalpel (stab scalpel knife, Fine Science Tools) in a U shape away from the midline. The dura mater was placed toward the midline and kept hydrated. The loader system with the soft robotic ECoG array, filled with 15% dextran, was placed at the frontal edge of the craniotomy and kept in place using a bench stand. The soft robotic ECoG array was deployed using the linear syringe pump, and electrical resistance of the strain sensor was monitored with the microcontroller using the customized LABVIEW code. After the deployment, the sensory evoked potentials were elicited using bipolar needles inserted in the snout (2 mm inside the skin) and stimulated with an isolated pulse generator (isolated pulse generator model 2100, AM Systems) at 5 and 10 mA, symmetric, cathodic-first, biphasic pulses with a 300-μs pulse width at a 2-Hz repetition rate. The brain signals were acquired by connecting the FPCB on the implant to a custom-printed circuit board connected to a wireless head stage amplifier system (Multichannel Systems Wireless W2100). A ground wire was inserted epidurally on the contralateral side. The acquisition was performed at a sampling frequency of 2 kHz with a digital Butterworth bandpass filter of 1 to 200 Hz and a notch filter at 50 Hz. The recorded signals were averaged over each individual stimulation pulse for a total of 30 s (roughly 60 epochs per condition). The experiment was performed for different stimulation pulse amplitudes (baseline, 5 and 10 mA) for the three snout stimulation positions. After the recordings, two catheters were inserted in the right jugular aorta vein, and the animal was sacrificed using 5 ml of pentobarbital injected intravenously. Then, 1 liter of 0.1 M PBS with heparin (100,000 U for 1 liter) was injected at 200 ml/min using a peristaltic pump (Shenchen LabS3/UD15) followed by 5 liters of a PBS solution containing 4% paraformaldehyde (PFA) at 200 ml/min. After the perfusion, the animal was decapitated using a scalpel knife to incise the skin and muscle between the first and second vertebrae. The head was then postfixed in 4% PFA solution for a few days. The brain with the deployed soft robotic ECoG array was then extracted by removing first the skin and muscle using a scalpel knife and then the bone using bone rongeurs. The dura mater was carefully incised to reveal the deployment status of the implant. Last, after removal of the device, the whole brain was extracted from the skull cavity.

The explanted brain was postfixed in 4% PFA solution for 48 hours before the hemispheres were separated using a scalpel knife. Then, parallel cuts were made at the front and back of the implantation area to isolate a block for further sectioning. The block was immersed in 15 wt % and then 30 wt % sucrose solution for 3 days each. The brain block was then frozen in isopentane at −55°C in a tissue-freezing system (Excilone Snapfrost). The frozen brain was then sliced in 40-μm sections using a cryostat (Leica CM1950). Immunostaining was then performed on several sections covering the entire zone of implantation. After incubation in 0.3% Triton and 3% bovine serum albumin, tissues were incubated with primary antibodies for 48 hours at room temperature (Anti-GFAP, Rat, Thermo Fisher Scientific, #13-0300; Anti-Iba1, Rabbit, FujiFilm, 019-19741; Anti-NeuN, GuineaPig, Sigma-Aldrich, ABN90) followed by secondary antibody incubation for 2 hours at room temperature (Alexa Fluor 488, Thermo Fisher Scientific, #A-11006; Alexa Fluor 647, Thermo Fisher Scientific, #A-21245; Alexa Fluor 555, Thermo Fisher Scientific, #A-21435). After mounting, slides were imaged at ×10 magnification using a Slide Scanner microscope (Olympus VS120).

Mean and SD were used for statistical analysis in quantifying the effect of different design variables on the system performance, whereas the minimum and the maximum were used as a shaded area in visualizing the range of system response to a given input (such as compression on the brain in Fig. 3C or electrical impedance in Fig. 4E). One-way analysis of variance (ANOVA) with post hoc Tukey’s honestly significant difference test was performed using Origin 2022 (OriginLab Corp.) when estimating a statistical significance depending on more than two groups of a parameter. The statistical significance was determined using P values, and a value higher than 0.05 (P > 0.05) was considered statistically significant (*P < 0.05 and **P < 0.01). The ANOVA results were reported in figure legends using the following format “F(dfB, dfW) = F-statistic, P = P value”, where dfB and dfW are degrees of freedom between groups and degrees of freedom within groups, respectively.