Non-invasive continuous-time glucose monitoring system using a chipless printable sensor based on split ring microwave resonators

This paper reports a highly sensitive, non-invasive sensor for real-time glucose monitoring from interstitial fluid. The structure is comprised of a chip-less tag sensor which may be taped over the patient’s skin and a reader, that can be embedded in a smartwatch. The tag sensor is energized through the established electromagnetic coupling between the tag and the reader and its frequency response is reflected on the spectrum of the reader in the same manner. The tag sensor consumes zero power as there is no requirement for any active readout or communication circuitry on the tag side. When measuring changes in glucose concentrations within saline replicating interstitial fluid, the sensor was able to detect glucose with an accuracy of ~ 1 mM/l over a physiological range of glucose concentrations with 38 kHz of the resonance frequency shift. This high sensitivity is attained as a result of the proposed new design and extended field concentration on the tag. The impact of some of the possible interferences on the response of the sensor’s performance was also investigated. Variations in electrolyte concentrations within the test samples have a negligible effect on the response of the sensor unless these variations are supra-physiologically large.

non-invasive continuous-time glucose monitoring system using a chipless printable sensor based on split ring microwave resonators Masoud Baghelani 1,2,3* , Zahra Abbasi 1 , Mojgan Daneshmand 1,4,5 & peter e. Light 2,4* This paper reports a highly sensitive, non-invasive sensor for real-time glucose monitoring from interstitial fluid. The structure is comprised of a chip-less tag sensor which may be taped over the patient's skin and a reader, that can be embedded in a smartwatch. The tag sensor is energized through the established electromagnetic coupling between the tag and the reader and its frequency response is reflected on the spectrum of the reader in the same manner. The tag sensor consumes zero power as there is no requirement for any active readout or communication circuitry on the tag side. When measuring changes in glucose concentrations within saline replicating interstitial fluid, the sensor was able to detect glucose with an accuracy of ~ 1 mM/l over a physiological range of glucose concentrations with 38 kHz of the resonance frequency shift. This high sensitivity is attained as a result of the proposed new design and extended field concentration on the tag. The impact of some of the possible interferences on the response of the sensor's performance was also investigated. Variations in electrolyte concentrations within the test samples have a negligible effect on the response of the sensor unless these variations are supra-physiologically large.
The World Health Organization estimates that there are > 500 M people worldwide who have diabetes 1 . Diabetes is primarily characterized by poorly controlled blood glucose concentrations that, if allowed to remain chronically high (hyperglycemia), result in the development of serious and life-threatening diseases such as stroke, heart attack, heart failure, kidney failure, adult blindness and amputation 2 . Moreover, many patients also experience episodes of very low blood glucose (hypoglycemia) that can rapidly lead to coma and death 3 . The most common glucose-sensing technology in use today are finger-prick based glucose strips, although this requires sampling many times per day and the continual purchase of the consumable once-use strips. As such, patient compliance to regular glucose monitoring is often not possible. Moreover, real-time continuous glucose monitoring offers a much more accurate assessment of the large fluctuations in blood glucose that can occur in patients with diabetes 4,5 . In this regard, newer glucose-sensing technologies include the placement of a thin needle-like sensor under the skin to measure glucose levels within interstitial fluid that closely tracks with blood glucose levels 6 . Although this technology removes the need for finger-prick blood sampling and can sample glucose levels every few minutes, the sensor is consumable, is expensive, and requires the insertion of a new sensor every 10-14 days 7 . Therefore, many patients cannot take advantage of this technology. What is really required is a non-invasive, reliable and cost-effective technology that measures glucose in real-time 8 .
To date, there has been much interest in developing novel glucose-sensors and a variety of sensor technologies have been tested. One of the most promising therapeutic systems for diabetes patients is the artificial pancreas, whereby an automated insulin delivery system is coupled to a real-time glucose sensing devise 9,10 . Such technology promises much tighter control of blood glucose levels. However, the main drawback associated with the artificial pancreas is the real-time and continuous glucose monitoring system. Indeed, a great deal of research has focused on the development of optimal real-time non-invasive glucose-sensing [11][12][13][14][15] , that optical, transdermal and

Results and discussion
In this section the proposed glucose monitoring sensor is presented alongside with the schematics, design approach, featured parameters, characteristics, analysis, and various accomplished experiments for glucose concentration measurement in different conditions as well as an intense discussion including analysis of different parameters effects on the measurement. chipless tag resonator sensor design. Figure 2 presents the perspective view of the field concentrations of the chipless microwave sensor for glucose sensing applications. The sensor is a ring-shaped copper trace designed to work around 4 GHz, as shown in Fig. 2. This frequency is selected because there is a considerable difference between water, as the main material in interstitial fluid, and saturated glucose solution permittivity while their loss factors are still small, and therefore measuring at this frequency will result in a significant frequency shift and hence the device sensitivity 46 . Also, since loss factor at this frequency is still low for water, the quality factor of the resonator will remain high which is of high significance for high precision measurements. Since the sensor is constructed of two resonators, there are two peaks and notches in the spectrum. In this measurement, only the notch related to the tag will be considered. As shown in Fig. 2, the sensor contemplates the variations in the medium introduced to the tag which is skin and its underneath including interstitial fluid and blood depending on the sensor mounting location. Variations in the materials permittivity in the regions subjected to higher concentration fields has more contribution to frequency shift. For quantifying this fact, in Fig. 2, an MUT with different layers stacked above the sensor is presented. All the layers have the same dielectric permittivity of 1 and the same thickness of 1 mm and only permittivity of one of them is changed to 2 at each step. Results illustrated in Fig. 2c verifies our justifications. Based on this observation, it seems glucose concentration variations in ISF has much more impact on frequency shift of the sensor than its variations in blood. Therefore, in the succeeding subsections only fluids and components of ISF are modeled.
Detection mechanism. In this part, different parameters utilized as the outputs or detection mechanisms of the presented sensor for glucose monitoring are described. Also, some high frequency simulations and analysis will be provided verifying supremacy of the performance of the proposed sensor. www.nature.com/scientificreports/ When MUT is introduced to a resonator, the overall effective permittivity of the system is changed and therefore the resonance frequency of the resonator. This shift in the resonance frequency is therefore a measure for determining of the introduced material for a constant volume. Frequency shift measurement is a robust parameter against additive noise and also is easy to measure. Readout circuitry have been developed with the detection limits in the range of 100 ppb (parts per billion) easily which makes high resolution frequency shift measurement both precise and straightforward 48 .
Amplitude variation. Another output of microwave resonator which could be invaluable for attaining an insight into MUT is amplitude variation. Amplitude variation is mostly occurred as the result of variations in conductivity of MUT 49 . This usually happens when concentration of electrolytes changes inside ISF. Since the conductivity spectrum of materials differs in trend (if not completely orthogonal) from their permittivity, studying amplitude variations could be very useful. (1) The field concentration of a traditional microwave ring resonator; note that the field is concentrated between the resonator and its ground-plane and therefore the substrate plays the most important role in determining of the resonance frequency. Also, the field concentration at the region above the sensor, which usually considered as the sensing area is smaller which limits the sensitivity of microwave resonator-based sensors. (b) Perspective view and field concentration of the resonator, it could be seen that the fields around the tag is almost constantly distributed in both on top (where material-under the test is placed) and underneath (where could be considered as its substrate). Therefore, one could expect higher frequency dependency of the tag to the MUT which could be translated to higher sensitivity in comparison with traditional microwave resonator sensors. (c) The simulation setup for studying the effect of the distant of the MUT from the tag on the sensitivity of the sensor. In this simulations, ten layers with the same thickness of 1 mm above the tag have been tested with the same permittivity as the air and only the permittivity of one of them has been changed to 2 at each step determined by layer number in parts (d) and (e) [images in parts (a), (b), and (c) are obtained from HFSS]. (d) S21 spectrum of some of the simulation results. (e) frequency variations versus the layer number (i.e. distance), it could be seen that, as the distance becomes higher, the sensitivity decreases in an exponential manner and therefore variations in the permittivity of MUT at higher distances could negligible in comparison with the same variations at lower distances, therefore we expect the sensor to measure glucose concentration variation in ISF rather than blood, so all the experiments are designed based on ISF measurement and the variations of its components.
Scientific RepoRtS | (2020) 10:12980 | https://doi.org/10.1038/s41598-020-69547-1 www.nature.com/scientificreports/ Sensitivity analysis. Considering frequency shift as the main output parameter for the sensor, sensitivity could be defined as the frequency shift versus permittivity variations of MUT for a certain volume. Since, each research uses arbitrary container volume and shape, for having a meaningful understanding of sensitivity improvement in the proposed sensor, a comparison between traditional microwave resonators and the current introduced sensor designed at the same frequency is presented here. As illustrated in Fig. 3, a superficial material with specific volume and shape covering the whole area of both resonators with ε r = 4 is introduced as MUT. The frequency shift resulted from relative permittivity variation to 10 for the proposed sensor is 700 MHz which is more than 3.5 times higher than the frequency shift for the traditional resonator. Limited sensitivity of the traditional resonator is as the result of confined electromagnetic fields between the resonator and its ground plane (see Fig. 2a). In traditional resonators, because of this phenomenon, substrate has a more important role in defining the resonance frequency rather than MUT. Because of the removing of the substrate for the tag in the presented work, the main variable parameter defining the resonance frequency of the tag is the MUT permittivity. For studying this concept, another simulation has been accomplished for both conventional and presented resonators. As depicted in Fig. 4, different substrate permittivity has been used with different permittivity for MUT for both traditional and the proposed sensors. It could be seen that, for traditional resonator sensors, substrate permittivity is the dominant parameter in determining the resonant frequency of the structure while the impact of substrate permittivity variations on the proposed sensor is very small and even negligible. For the remaining of this paper, we define sensitivity as the frequency variation resulted from 1 mM/l of glucose concentration change for a specific test setup.

Distant measurement analysis.
Another notable feature of the presented work is the distant sensing capability. This characteristic is especially important for wearable electronic applications. In addition to capability of www.nature.com/scientificreports/ embedding the reader in a smart watch, phone or a gadget, this remarkable feature brings up new paramount benefits such as zero power consumption, extremely low cost, and small size for the sensing tag. For having a better insight into this characteristic, another simulation has been accomplished by placing MUT with specific relative permittivity on top of the tag and increasing the distance between the reader and the tag. It could be seen in Fig. 5 that tag continue to communicate with the reader for almost 11 mm with absolutely zero power which is completely enough for our application.
Experiments. Various measurements have been accomplished verifying the performance of the proposed non-invasive glucose measurement sensor. First of all, glucose concentration measurement in deionized (DI) water is carried out. For studying consistency and stability of the sensor as well as setup a return-to zero test is accomplished with as high concentrations of glucose as 200 mM/l (Fig. 6). Although this value is unrealistically high, but it will provide invaluable insight through consistency of the sensor performance by introducing DI water with zero glucose concentration and DI water with 200 mM/l glucose concentration alternatively to the sensor. Figure 6d sketches the resonance frequency notch amplitude of S21 response of the sensor. It could be seen that the sensor response is both stable and repeatable. Also, high sensitivity characteristic of the sensor is noticeable. To the best of our knowledge, the achieved sensitivity of this work, 60 kHz/1 mM/l of glucose concentration which is superior to the best results reported in literature regardless of shape and volume of MUT. This means, the response of the sensor is less susceptible to environmental noises than its conventional counterparts. For the next step, samples are prepared with 10 volumetric percent of horse serum for modelling ISF. Both return-to zero and small variations of glucose concentration samples have been tested with promising results achieved as sketched in Fig. 7. For achieving a better idea on the performance of the sensor, it is common to address the glucose concentration versus frequency shift as the measured data. An interpolation curve fitting process then accomplished based on the resulting data. These results are presented in Fig. 7d.
To further mimic a more physiological condition, we performed glucose sensing experiments through a layer of mouse skin. In these experiments, saline is included in the sample with electrolytes and ionic concentrations described in "Results and discussion" section. According to conductivity increasing of the samples, the amplitude of the notch frequency is increased. For this experiment, a shaved mice skin with about 300 µm thickness wrapped inside a sealed plastic bag is used between the sensor and the liquid. Hence, the sample is located in further distance from the sensor. As illustrated in Fig. 8, the sensitivity of the sensor is decreased with the same justification as Fig. 2 as the result of increasing the distance between ISF sample and the sensor. However, the sensitivity of the system to changes in glucose concentration is still superior to other non-invasive technologies published to date.

Discussion.
Although microwave resonators possess impressive characteristics, there is still a very challenging issue remained. Since any variation in the permittivity of MUT is reflected in frequency shift of the resonator, there is a concern about the uncertainty of the actual source of frequency shift. For addressing this issue, an extensive discussion part including some experiments is provided.
The presented sensor aims to measure glucose concentration in ISF which is a fluid contains around 40% of human body's water surrounding the cells acting as the nutrient transporting from blood capillaries and waste collecting medium for the cells. Beside water and plasma, ISF also contains glucose, fatty acids and salts. So far, glucose variation effects have been tested. Here, we provide some experiments for studying the effects of mineral www.nature.com/scientificreports/ variations on the frequency shift of the sensor. The main ions in ISF are, sodium, potassium, chloride, calcium, magnesium, bicarbonate and phosphate. Since sodium and chloride ions have one or more orders of magnitude higher variation range in comparison with the other ions, for the sake of simplicity, they are considered as the only variable ions the experiments. It could be seen from Fig. 9 that since ions mostly affect the conductivity of the MUT, it won't change the frequency of the sensor. Hence, since frequency change is considered as the main output of the sensor, ionic concentration variations is unlikely to not interfere with results from glucose related frequency shift. In addition, fatty acid concentration variation inside ISF is in the range of < 1 mM/l and therefore its effects are minimal on the frequency shift in comparison with effect of glucose variation. Another important parameter to consider is ionic concentration changes that manifest as a result of hydration levels. For example, mild dehydration often occurs regularly in humans. Dehydration directly affects the water content in ISF and therefore could change its permittivity and consequently affects the performance and precision of the sensor. Sample preparation method is presented in the next section. Figure 10 presents the frequency shift versus dehydration percentage with all the other variables remaining constant. Our results demonstrate that low to moderate dehydration has a minor effect on the frequency shift even less than the effect of 1 mM/l variation in glucose concentration. However, severe dehydration has the potential to interfere with the frequency shift resulting from glucose variations and therefore compromise the glucose sensitivity of the sensor. Therefore, further development of this sensor technology will have to consider the impact of severe dehydration on sensor accuracy. The real-time applicability of the sensor is achieved because of instant variation in glucose concentration in the MUT results in its dielectric permittivity which changes the effective permittivity of the sensor's environment and consequently results in frequency shift (see Eq. 1).
A comprehensive comparison between the presented structure and some of the state-of-the art works using methods other than microwave is outlined in Table 1. Another quantitative comparison between different microwave techniques-based glucose sensors and the current paper is presented in Table 2. Although, some of the summarized works seems to have higher sensitivity than the proposed work, but those are mostly as the result of lower distances between their resonators and sample due to using of extra-thin microfluidic channels. This www.nature.com/scientificreports/ justification is completely in agreement with the concept presented in Fig. 2. It could be seen from Fig. 2e that the frequency shift (i.e. sensitivity) is drastically reduced with increasing the distance of the sample from the sensor in an exponential manner. We present the design and testing of non-invasive glucose sensor with a very high sensitivity despite the considerable distance between the sensor and the testing medium that would be expected in real-life biosensing applications.

Methods
Sensor fabrication. The utilized sensor in this work includes two parts; sensing tag and reader, both fabricated with almost the same process. Top face of the resonator is first printed on a glossy paper using a highresolution printer. Printed pattern then transferred on a substrate by applying of high temperature and uniform pressure. For the reader part, ROGERS 5880 PTFE composite substrate with εr = 2.2, tan(δ) = 0.0004, thickness = 0.787 mm. Although it mentioned "substrate-less", for simplifying the lab-based fabrication process of tag, it is printed on a ROGERS RO4450F with εr = 3.52, tan(δ) = 0.0004, thickness = 101 µm.
Sample preparation. The saline composition is designed to mimic ISF as close as possible. A high concentration 10× stock solution is subsequently diluted with DI water to achieve actual concentration in ISF. The pH of the resulting saline is then buffered to pH 7.4 with NaOH and/or HCl, a value that mimics the pH of ISF. The final concentrations of the prepared solution are summarized in Table 3. DI water used for all the experiments was double distilled water with 18 MΩ/cm resistance. For dehydration testing, the saline compartments were the same and the percentage of diluting DI water was reduced according to the dehydration percentage. All the test saline contained 10% horse serum (SIGMAALDRICH). www.nature.com/scientificreports/ Simulation in Fig. 1c are accomplished using the tissue dielectric properties of humans as summarized in Table 4 58 . Skin samples were obtained from mice that were sacrificed for other research projects. The skin then shaved and sealed inside a plastic bag with a very small amount of saline surrounding to prevent excessive drying of the skin. The dielectric constant of the mouse skin is about 35 59 which is in a good agreement with human skin dielectric constant used in simulations.
instruments and setups. All the microwave measurements have been accomplished using S5085 2-port vector network analyzer (VNA) from COPPERMOUNTAIN TECHNOLOGIES INC. The liquid samples were also tested inside a borosilicate tube with εr = 4.3 and tan(δ) = 0.0047 and wall thickness of 1 mm. The total exposed volume of liquid was 200 µl. Simulations have also been carried out using High Frequency Spectrum Simulator (HFSS). Figure 11 presents the fabricated sensor including the reader and the sensing tag as well as the experimental setup. It could be seen that, according to introducing of the skin between the sensor and the sample, the overall sensitivity is reduced to 38 kHz/1 mM/l of glucose concentration variation. Figure 9. Effect of saline variations on the response of the sensor; here only Na and Cl concentrations have been changed as the major electrolytes in ISF from 0 to 150 mM/l. Although the maximum variation happens in human body is limited from 136-150 mM/l, an exaggerate variation is tested here to presents the proof of concept. It could be seen that saline concentration has in important impact on the amplitude of the response but its resulting frequency shift is less than 20 KHz which is completely negligible. The case would be even more negligible in real life case, because of less variations in the electrolytes.
Scientific RepoRtS | (2020) 10:12980 | https://doi.org/10.1038/s41598-020-69547-1 www.nature.com/scientificreports/ ethical approval. We confirm that all methods were carried out in accordance with relevant guidelines and regulations. We also confirm that all experimental protocols were approved by Research Ethics Office (REO) University of Alberta.  www.nature.com/scientificreports/ conclusion Herein, we report the design and testing of a novel non-invasive wearable glucose monitoring sensor with zero power consumption and high sensitivity that is based on microwave planar resonator technology. The sensor is  www.nature.com/scientificreports/ actually a metallic trace which could be taped over the skin. The impressive performance of the sensor, which removes many barriers against utilization of microwave resonator sensors for biomedical applications and especially wearable electronics, have been attained as the results of its improved design. Electromagnetic coupling between the reader and the tag, provides the distinct possibility of distant sensing and eliminating the requirement of a built-in power supply as the readout and communication circuitries can be integrated in the sensor's reader unit. Moreover, substrate removal of the tag results in enhanced sensitivity of the sensor. This important improvement has been achieved as the result of making the permittivity of the MUT as the main defining parameter of the resonance frequency of the sensor. For testing the robustness of the sensor against variations of possible interferers in ISF, impacts of electrolytes variations as well as dehydration have been tested. Any electrolyte variation effects, over a physiological range of concentrations, resulted in a negligible frequency shift of the sensor. Also, the confounding effects of low to moderate dehydration was very low but became significant for severe dehydration. In summary, our results present a novel approach to the biosensing of physiological parameters such as glucose that represents a marked improvement over existing non-invasive sensor technologies.