A noninvasive flexible conformal sensor for accurate real-time monitoring of local cerebral edema based on electromagnetic induction

Detection principle

Magnetic Induction Phase Shift (MIPS) monitoring is a non-invasive, real-time, continuous method to measure the conductivity change of an object. This method is based on the magnetic induction between the excitation-sensing coils and an object in its vicinity. Figure 1 sketched its basic physical principle. The signal source generates a sinusoidal signal and splits into two equivalent signals, one as reference signal and the other one as excitation signal sending to the excitation coil. The current-driven excitation coil sends out a primary alternating magnetic field (B) where this field penetrates the biological tissue in its vicinity and induces an eddy current. The secondary magnetic field (ΔB) generated by the eddy current is collected by the sensing coil, together with the primary magnetic field. There is a phase difference between the reference signal and detection signal. According to Griffiths, Stewart & Gough (1999), ΔB is related to B as follows: (1)${\displaystyle \Delta B∕B=Q\omega {\mu }_0\left[\omega {\epsilon }_0\left({\epsilon }_r-1\right)-i\sigma \right]+R\left({\mu }_r-1\right)}$

where ω is the signal frequency, σ, ε_r and μ_r are the electrical conductivity, relative permittivity and relative permeability of the sample, separately. ε₀ and μ₀ are the permittivity and permeability of free space. Q and R are geometrical constants. Also, the magnetic field can be characterized by the voltage (V, ΔV) in the excitation and sensing coil (Griffiths, 2001; Griffiths, Stewart & Gough, 1999): (2)${\displaystyle \Delta V∕V=\Delta B∕B.}$

Thus, the total signal (V + ΔV) collected by the sensing coil lags the primary signal by an angle θ. The change of phase angle θ is MIPS. The magnitude of ε_r in the MHz range is much lower than σ. Thus, there is a relationship between θ and the signal frequency ω and σ (Griffiths, 2001; Griffiths, Stewart & Gough, 1999): (3)${\displaystyle \theta \propto \omega \sigma .}$

Finite element method simulations

In target to investigate the sensitivity map of the conformal two-coil and the single-coil structure, through which selecting a suitable coil designment for CE monitoring, simulation using the FEM was conducted using commercial software (HFSS, ANSYS Electronics Desktop 2018.0, USA). Two spiral coils were used to resemble a conformal two-coil structure, where the inner ring is excitation coil and the outer ring is sensing coil. Conformal refers specifically to the structure in which multiple coils are located on the same geometric plane, which comes from the planar compact Wireless Power Transfer (WPT) systems (Jonah, Geogakopoulos & Tentzeris, 2013; Hu & Georgakopoulos, 2017). The inner radius of excitation coil and sensing coil is 9 mm and 19 mm, separately (Fig. 2A). Each coil has 15 turns and with wire diameter (w) of 0.15 mm and line spacing (d) of 0.25 mm. Considering the skin effect of metal, the coil model in the simulation uses a plane model without thickness instead of a volume model. The boundary condition of coil was set as PEC. The sensor was developed on a 0.1-mm-thick (t) polyamide substrate, whose relative permittivity (ε_r) and loss tangent are 3.2 and 0.003. Two coils were driven by lumped port excitation at 55 MHz. When cerebral edema occurs, the CSF will first be discharged from the cranial cavity due to compensation and followed by a certain range of brain blood compensation (Wykes & Vindlacheruvu, 2015). In the 1–100 MHz frequency band, the conductivity of intracranial cerebrospinal fluid (CSF) is the highest, followed by blood. Choosing the frequency of the system in this frequency band can better reflect the pathological process of CE. At the same time, the selection of this frequency is close to our previous work, in which we’ve conducted experiments using 65.5 MHz (Pan et al., 2015), 64.1 MHz (Zhao et al., 2019), 67.1 MHz (Yang et al., 2019a; Yang et al., 2019b). The two-coil sensor at the same lateral is a kind of traditional structure whereas it overcomes the disadvantages of using rigid materials to control the relative distance between excitation coil and sensing coil in practical applications. The conformal structure enables direct textile or FPC integration without affecting the relative position of coils. Subsequently, the outer coil was deleted and thus constructed a single-coil sensor (Fig. 2B), consistent with the structure used by Teichmann et al. (2015) and Teichmann et al. (2014). In both sensors, a 10 ml cylinder (radius r = 15 mm, height h ≈ 26.31 mm) was placed in the front center of the sensors. The initial distance between the center of beaker and center of substrates was D = d + r = 20 mm, as shown in Fig. 3. The boundary condition of the boundary of the analyzed domain was set as radiation. This cylinder was used to simulate the 1% saline solution with conductivity of 1.71 S/m.

Simulation investigation was carried out by displacing the 10 ml saline solution cylinder with a step of 1 mm from (0, 20) along the x axis and z axis. Through subtracting the phase of S21 at 55 MHz in free-space condition from those at all positions, MIPS sensitivity of the two-coil sensor at each position can be obtained. Likewise, through subtracting the phase of S11 at 55 MHz in free-space condition from those at all positions, MIPS sensitivity of the single-coil sensor can be obtained.

Physical experiments

To investigate the effect of sensor bending on MIPS, flexible MIPS sensor was fabricated using FPC, as shown in Fig. 4. In this paper, a RF vector network analyzer (VNA, Agilent E5061B, USA) was utilized to realize MIPS detection. The output power of the VNA was 10dBm. The frequency band was set from 1 MHz to 100 MHz, with 1601 sample points and intermediate frequency bandwidth of 30 kHz. MIPS data were obtained from the phase of S21 at its characteristic frequency (53.7 MHz), where the transmission coefficient reach its maximum. The selection method of characteristic frequency comes from previous researches (Pan et al., 2015; Li et al., 2017a; Li et al., 2017b; Li et al., 2017c; Oziel et al., 2016). Physical and subsequent animal experiments were both carried out under those parameter settings. The characteristic frequency of animal experiments appeared at 57.7 MHz. In the three experiments, the system measurement frequency was slightly different. The detailed discussion on this phenomenon was arranged in the discussion part.

The physical experiments were shown in Fig. 5. In this scenario, four 3D-printed substrates with different bending radius (flat, 100 mm, 75 mm, 50 mm) made of photosensitive resin were used for sensor fixing. Saline solution (1%) was used as the measured object. A beaker was fixed in front of the sensor to load the solution. Measurements utilizing each of those substrates was successively done while the distance between the outer ring of beaker and center point of substrates was kept at 5 mm (There was 4 mm between the mouth of the beaker and its inner wall), consistent with the settings of simulation experiments. Saline solution was injected into the beaker via injection pump (Longer, LSP01-1A, UK) from 0 ml to 10 ml with an increment of 1 ml, during which the MIPS data could be obtained. In each bending condition, the injection procedure was repeated for 20 times to get 20 sets of measurement data.

Subsequently, to verify the relationship between detection distance and sensor’s sensitivity, beaker was placed at another two positions for same measurements where the outer ring of the beaker is 10 mm and 15 mm from the center of the sensor, while the sensor was kept flat (using substrate 1).

Animal experiments

In order to verify that the conformal MIPS sensor can monitor the development of CE, seventeen rabbits (available from Daping Hospital, 2.0–3.0 kg) were enrolled in the animal experiments. Rabbits were arrived one day before experiments and housed under ambient conditions (22 °C, 50% relative humidity, and a 12-h light/dark cycle), with free access to water and chow. These experiments were conducted under the guidance of the Administration of Animal Experiments for Medical Research Purpose issued by the Ministry of Health of China. The protocol was reviewed and approved by the Animal Experiments and Ethical Committee of Army Medical University (AMU, Chongqing, China), approval number AMUWEC2019237. All procedures were conducted while minimizing the suffering of rabbits. The rabbits were randomly divided into experimental group (n = 14) and control group (n = 3).

Although it is of great practical significance to conduct research and data analysis on clinical CE patients, the complex pathological conditions and possible observation and treatment procedure shadows the data of clinical scenario. Thus, this experiment conduct animal models under controlled pathological conditions to better simulate the pathological progression of CE in a repeatable manner. Our previous researches had proved the validity of the CE model established by epidural freezing method (Kawai et al., 2003; Li et al., 2017a; Li et al., 2017b; Li et al., 2017c; Li et al., 2019). Rabbits were first anesthetized with pentobarbital (3%, 1 ml/kg) via ear vein. During surgery and monitoring, rabbits were given inhalation anesthesia (isoflurane, 1.5%) to maintain adequate anesthesia depth. After sedation, head and neck hair was removed. Then, rabbits underwent tracheal intubation for long-term monitoring purpose. After intubation, rabbits’ scalp was cut to expose the skull. Next, a bone window was drilled to expose the dura matter at 2 mm on the right side of sagittal suture and 2 mm on the posterior side of coronal suture. During drilling, the cryo-pencil was immersed into liquid nitrogen (−196 °C). After sufficient cold, the cryo-pencil were vertically put on the bone window for 120s. Then, the bone window was sealed using dental cement. In contrast, rabbits in control group experienced the same procedure but without freezing. After the establishment of CE and control model, rabbits were fit on the board and monitored for 24 h while the conformal MIPS sensor was placed close to the freezing point of the head. The characteristic frequency appeared at 57.7 MHz. The initial sampling rate was 12 times per hour. Rabbits in both groups were euthanized via IV pentobarbital overdose at the end of monitoring. Experimental arrangement was shown in Fig. 6.

Signal processing and statistical analysis

Sensitivity results in simulation experiments was normalized to 0-1 for mapping and comparsion purpose. S_t Represented the normalized sensitivity of two-coil structure and S_s represented the normalized sensitivity of single-coil structure. The calculation formula of normalized sensitivity was given by:

(4)${\displaystyle S_t\left(x,y\right)=\frac{S_t^{\ast }\left(x,y\right)}{max\left(S_t^{\ast }\right)-min\left(S_t^{\ast }\right)}\left|x=0,1,\dots ,20,y=0,1,\dots ,20\right.}$ (5)${\displaystyle S_s\left(x,y\right)=\frac{S_s^{\ast }\left(x,y\right)}{max\left(S_s^{\ast }\right)-min\left(S_s^{\ast }\right)}=\left|x=0,1,\dots ,20,y=0,1,\dots ,20\right.}$

where $S_t^{\ast }$ and $S_s^{\ast }$ represents sensitivity values of two-coil structure and single-coil structure, respectively.

In order to quantitatively analyze the change trend of sensitivity, R_t and R_s were used to represent the ratio of the coordinate points of the normalized sensitivity in a specific interval to all the measured coordinate points. The value of R_t and R_s were given by:

(6)${\displaystyle R_t=\frac{n}{21\ast 21}\ast 100\text{\%},i=1,2,\dots ,n\left|S_t\left(x,y\right)\ge a\right.}$ (7)${\displaystyle R_s=\frac{n}{21\ast 21}\ast 100\text{\%},i=1,2,\dots ,n\left|S_s\left(x,y\right)\ge a\right.}$

where n represents the total number higher than the threshold a.

In animal experiments, the initial sampling rate of the MIPS signal was 12 times/h. Considering that the original signal was mixed with heart and lung activities plus external interference, a wavelet transform was used to reduce noise. Data processing was performed by MATLAB R2015a (MathWorks, Inc., USA). Moreover, The physical and animal experimental data were statistically analyzed utilizing nonparametric multi-independent/independent sample test and two-sample independent t-test, seperately. Statistical analyses were performed by SPSS 19.0 (SPSS Inc., Chicago, IL, USA). The significance level was set at p < 0.05.

Results of simulation experiments

Figure 7 shows the normalized sensitivity map of the conformal two-coil sensor structure (S_t ∈ [0, 1]). It can be found that this structure had the characteristic of local focus, where the sensitivity got higher when close to the center. The highest sensitivity (S_t = 1) appeared at (x, z) = (3, 20). S_t gradually attenuated form the center outward. Figure 8 shows the normalized sensitivity map of the single-coil sensor structure (S_s ∈ [0, 1]). The highest sensitivity (S_s = 1) also appeared at (x, z) = (3, 20). Both sensitivity map of the two sensor structures showed a manner of elliptical shape, where their sensitivity attenuated sharper axially. Especially, S_s attenuated faster both axially and radially than that of two-coil structure. Moreover, as shown in Fig. 7, S_t is better distributed while there are many burrs in Fig. 8. Table 1 lists the sensitivity attenuation trend in both two sensor structures, which further confirmed the fact that the S_s was more concentrated near the center and attenuated quickly. In comparison, R_t is gentler, with more proportion up to ≥0.3 level. Due to the burrs in the low-sensitivity area of Fig. 8, R_s in ≥0.1 interval was higher. It can be concluded that single-coil structure also has the characteristic of local focus whereas its flatness and uniformity of sensitivity map is poorer. Based on this conclusion, MIPS sensor with the two-coil structure was selected and fabricated in physical experiments.

Results of physical experiments

Figure 9 demonstrates the MIPS at 53.7 MHz as a function of injection volume (mean ± SD), in which four color lines indicated four bending radius conditions, respectively. Obviously, the MIPS data showed a downward trend with increasing volume, indicating that MIPS can reflect the conductivity change of the measured object.

To quantitatively study whether bending is related to the sensor’s performance, the MIPS data under three volumes (3 ml, 6 ml, 9 ml) were extracted for statistical analysis. First, those data were enrolled for nonparametric multi-independent sample test (Kruskal Wallis test), as shown in Table S1. It can be found that MIPS under different bending radius in all of the three volumes were statistically significant (p < 0.05). This shows that at any volume, there was a significant difference in those data. But the relationship between the data within each group remains unclear. Thus, Non-parametric independent sample test was conducted between each bending radius (Table S2).

Statistical data shows that, in all of the three volume, there was no significant difference between different bending radius (2-3/2-4/3-4) (p > 0.05), which means that the performance of the sensor is independent of its bending degree. Besides, MIPS in each of the three bending cases was significantly different from the flat one (1-2/1-3/1-4) (p < 0.05). That can also be found in Fig. 9 that the MIPS data under those bending conditions was slightly larger. This may be caused by the reduced gap between the sensor and beaker, as a result of bending. Smaller gaps were provided with higher S_t. Figure 10 shows the MIPS changes under three detection distance (using substrate 1). As shown below, the sensitivity of the MIPS did decay with increasing distance, consistent with the simulation results where S_t attenuated quickly along its axial direction.

Results of animal experiments

Figure 11 shows the MIPS data at 57.7 MHz of the 9th rabbit in the experimental group. There was a clear upward trend over time. Also, it can be found that the raw data was doped with breathing, head movement and external interference, which influenced the actual trend of MIPS-characterized CE. Therefore, the original data is processed through wavelet transform. Figure 12 shows the MIPS change at 57.7 MHz in 24 h monitoring of rabbits in experimental group and control group. For mapping needs, the initial MIPS value of all data was normalized to 0. Rabbits in experimental group all showed a significant upward trend with maximum change at 24th hour (15.66 ± 2.41°). After the same surgical procedures, the development of CE seemed to have individual differences. In contrast, rabbits in control group merely had a slight increase in MIPS and kept stable with insignificant variation in the whole 24 h measurement (0.78 ± 0.47° at 24th hour). Our previous finding showed a 24 h downward trend of the MIPS with largest change of −13.1121 ± 2.3953° at the 24th hour in CE group and no obvious trend in the control group (−0.87795 ± 1.5146° at 24th hour). This conformal MIPS sensor is consistent with previous studies in monitoring and distinguishing CE (Li et al., 2017a; Li et al., 2017b; Li et al., 2017c; Zhao et al., 2019; Li et al., 2019), whereas the change trend of MIPS in the experimental group was opposite.

Then, the differences between the experimental group and the control group were statistically analyzed (Table 2). Through two-sample independent t-test, it can be found that those two groups showed significant difference from the 1st hour. The early diagnostic ability of the conformal MIPS sensor in this study was the same as that in previous study by Li et al. (2017a), Li et al. (2017b) and Li et al. (2017c).