WIDE BAND FIELD RESPONSE MEASUREMENT FOR GLUCOSE DETERMINATION

[0001] The invention relates to a device and method for determining the glucose level of living tissue. In particular, it relates to technologies that derive the glucose level from the response of the tissue to an applied electrical field.

[0002] WO 02/069791 describes a device for determining the glucose level of living tissue. The device comprises an electrode arrangement mounted to a substrate, as well as a control unit adapted to determine the response of the tissue to the electric field generated by the electrodes. This type of device exploits the fact that the glucose level affects the dielectric and electric response of the same.

[0003] Similarly, WO 2005/053523 and WO2005/053526 describe hardware and measurement methodology to be used in such a device.

[0004] Since glucose level is not the only state variable of the tissue that affects its electric and dielectric response, accuracy with such device can only be achieved by measuring and combining appropriate measured parameters.

[0005] The problem to be solved by the present invention is therefore to provide an improved device and method of this type that allows a more accurate glucose level determination.

[0006] This problem is solved by the device and method of the independent claims.

[0007] Accordingly, AC signal voltages of at least a first, a second and a third frequency are applied, consecutively or simultaneously, to the electrode arrangement in order to measure a first, a second and a third electric parameter, respectively. The first frequency is between 1 kHz and 200 kHz, the second frequency is between 0.2 MHz and 100 MHz, and the third frequency is at least 1 GHz.

[0008] In addition, a temperature parameter indicative of the temperature of the tissue is measured.

[0009] The first, second and third electric parameter as well as the temperature parameter are then combined in order to derive the glucose level. For example, a sum of linear and/or quadratic terms of the first, second and third electric parameter and the temperature signal can be used, with coefficients obtained from calibration measurements.

[0010] This is based on the understanding that measurements in the three specific frequency ranges are advantageous in order to eliminate the influence of tissue state parameters other than glucose on the measured signals. For example, glucose mainly modulates the beta-dispersion in the frequency range 0.2 MHz to 100 MHz, and the effect of sweat can be eliminated because it primarily affects the signal in the frequency range between 1 kHz and 200 kHz, while it has a lesser effect on the signals at the other frequency ranges. Furthermore, the temperature of the tissue affects all the electrically measured parameters, therefore a direct temperature measurement allows to improve the accuracy further.

[0011] Advantageously, in order to compensate for changes in the water content in the skin and underlying tissue, the third frequency should be between 1 GHz and 30 GHz because the dielectric response of water primarily affects the signal in this frequency range, while it has a lesser effect on the signals at the other frequency ranges.

[0012] In a further advantageous embodiment, the first electric parameter is measured by means of a first pair of electrodes designed as interdigital electrodes because the separation gap needs to be very small to allow measuring only surface effects, and the interdigital arrangement allows for a suitable large measurement surface.

[0013] The second electric parameter is best measured by means of a second pair of electrodes separated by a gap of at least 2 mm. This allows to generate a field reaching sufficiently far into the tissue in order to measure the beta-dispersion of blood in the tissue. Advantageously, at least two pairs of electrodes with different gap widths are provided for measurements at the second frequency range. They are used for measuring at least two different electric parameters, which allows to obtain depth-resolved information. The two pairs of electrodes may have one electrode (e.g. the ground electrode) in common.

[0014] The third electric parameter is also best measured by means of a dedicated third pair of electrodes, which can be optimized for measurements at the Gigahertz range, e.g. be being designed as coplanar waveguides. Advantageously, at least two pairs of electrodes with different gap widths are provided for measurements at the third frequency range. They are used for measuring at least two different electric parameters, which again allows to obtain depth-resolved information. The two pairs of electrodes may also have one electrode (e.g. the ground electrode) in common.

[0015] The following parameters are found to be particularly indicative for a glucose measurement:

[0016] p1—a value depending on the phases of the impedances measured at several frequencies in the second frequency range at frequencies larger than 30 MHz, measured by means of an electrode having a separation gap of at least 2 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the logarithms of the phases of the impedances are used.

[0017] p2—a value depending on the phases of the impedances measured at several frequencies in the second frequency range at frequencies larger than 30 MHz, measured by means of an electrode pair having a separation gap of less than 2 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, logarithms of the phase of the phases of the impedances are used.

[0018] p3—a value depending on the phases of the impedances measured at several frequencies in the first frequency range at frequencies larger than 100 kHz. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, logarithms of the phase of the phases of the impedances are used.

[0019] p4—a value depending on the magnitudes of the impedances measured at several frequencies in the first frequency range at frequencies larger than 100 kHz. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, logarithms of the phase of the phases of the impedances are used.

[0020] p5—a value depending on the phases of the transmission coefficient measured at least one frequency f1 in the third frequency range, measured by means of an electrode having a separation gap of at least 1 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the square of the phase of the transmission is used.

[0021] p6—a value depending on the phases of the transmission coefficient measured at least one frequency f2 in the third frequency range, measured by means of an electrode having a separation gap of less than 1 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the square of the phase of the transmission is used.

[0022] p7—a value depending on the temperature.

[0023] In addition to this, the measurement can further be refined by also taking into account at least one of the following parameters

[0024] p8—a value depending on the phases of the impedances measured at several frequencies in the second frequency range between 10 and 30 MHz, measured by means of an electrode having a separation gap of at least 2 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the logarithms of the phases of the impedances are used.

[0025] p9—a value depending on the phases of the impedances measured at several frequencies in the second frequency range between 10 and 30 MHz, measured by means of an electrode having a separation gap of less than 2 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the logarithms of the phases of the impedances are used.

[0026] p10—a value depending on the magnitude of the impedances measured at several frequencies in the second frequency range at frequencies smaller than 10 MHz, measured by means of an electrode having a separation gap of at least 2 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the logarithms of the magnitudes of the impedances are used.

[0027] p11—a value depending on the magnitudes of the impedances measured at several frequencies in the second frequency range at frequencies smaller than 10 MHz, measured by means of an electrode having a separation gap of less than 2 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the logarithms of the magnitudes of the impedances are used.

[0028] p12—a value depending on the magnitudes of the transmission coefficient measured at least one frequency f1′<f1 in the third frequency range, measured by means of an electrode having a separation gap of at least 1 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the square of the magnitude of the transmission is used.

[0029] p13—a value depending on the magnitudes of the transmission coefficient measured at least one frequency f2′<f2 in the third frequency range, measured by means of an electrode having a separation gap of less than 1 mm. Advantageously, if the parameters are subsequently combined using linear or quadratic combinations, the square of the magnitude of the transmission is used.

[0030] Advantageously, when calculating the glucose level, the parameters are weighted with weights, with part of said weights being “global weights”, which are defined as weights that are common for a series of devices, i.e. which do not have to be adapted to the individual user. Another part of the weights are “user weights”, which are defined as weights that have to be calibrated for individual users. At least the following parameters are advantageously weighted by means of global weights:

[0031] p1, p2, p3, p4, p7, and, where applicable, p10, p11

[0032] In addition to this, or alternatively, at least the following parameters are advantageously weighed by means of local weights:

[0033] p5, p6, and, where applicable, p8, p9, p12, p13.

[0034] This is due to the fact that the parameters p5, p6 and, where applicable, p12, p13 mainly account for the changes in water content in the biological tissue, these are proportional to its thickness and may change from case to case. Furthermore, where applicable, the parameters p8 and p9 mainly account for the electrical losses in the biological tissue, which are again proportional to tissue thickness.

[0035] For example, the glucose level can be calculated using a linear, weighted sum and an offset, i.e. using a term such as

$c+\sum _{i=1}^Nk_i·p_i,$

wherein c is an offset, k_i are weights, p_i are the measured parameters and N is the number of measured parameters.

[0036] At least some of the weights k_i can be determined in calibration measurements. Advantageously, such calibration measurements comprise a calibration step where the parameters and a series of reference glucose values are measured over an extended period of time. In this context, a “reference glucose value” is a glucose level determined by other means than by means of the present invention, e.g. by analyzing blood samples, by optical measurements, etc.

[0037] During the calibration step, terms of the type

$c_j+\sum _{i=1}^Nk_i·p_i$

are calculated, with c₁, c₂, c_J being additive values. For a first series of consecutive measurements c₁ is used, for a second series of consecutive measurements c₂ is used, etc. In other words, the series of measurements is divided into sub-series, with each sub-series j having its own additive value c_j. Then, all the additive values c_j and at least part of the parameters k_i are fitted to the reference glucose values. The use of several additive values allows to compensate for a drift in the measurement by minimizing the effect of this latter in the estimation of the parameters k_i.

[0038] After this calibration step, the determination of the glucose level in “normal” operation can e.g. comprise the following steps:

[0039] measurement of a reference glucose value, i.e. in the morning,

[0040] determination of the offset c from the reference glucose value and the weights obtained in the calibration step, and

[0041] for subsequent measurements, e.g. during the day, determination of the glucose level from the measured parameters, the weights obtained in the calibration step, and the offset c.

[0053] General Hardware Design

[0054] The general design of an advantageous embodiment of the device is shown in FIGS. 1 and 2. The device comprises a housing 1, e.g. of plastics. One side of the device, in the following called the “bottom side”, is closed by a substrate 2. Substrate 2 is rectangular and elongate and has a longitudinal axis 3. A band or wristband 4 is attached to housing 1, extends perpendicular to longitudinal axis 3 and allows to mount the device e.g. to an arm or leg of a person, with longitudinal axis 3 extending parallel to the longitudinal axis of the arm or leg.

[0055] At a first side (bottom side) substrate 2 carries an electrode arrangement formed by a structured metal layer 5. Metal layer 5 is shown in gray in FIG. 2. On its second side (top side) electronic components 6 are mounted to substrate 2 and connected to metal leads on the surface or within substrate 2.

[0056] Housing 1 encloses at least one printed circuit 7 in addition to substrate 2, which carries further electronic components 8. Electrical connectors 9 are provided for connecting printed circuit board 7 to substrate 2.

[0057] In addition, the device typically also contains a battery as well as interface and/or display components (not shown).

[0058] As mentioned, substrate 2 has a first (bottom) side and a second (top) side, with the first side being applied to a person's skin during operation of the device.

[0059] Metal layer 5 is structured to form a ground electrode 10 having openings. Signal electrodes 12a, 12b, 13a, 13b, 13c and 14 are placed within these openings, such that gaps 15 are formed between the inner edges of the openings and the outer edges of the signal electrodes. Each signal electrode 12a, 12b, 13a, 13b, 13c and 14 is completely surrounded by such a gap 15.

[0060] The electrode arrangement can comprise an optional dielectric layer covering metal layer 5 for mechanically and/or chemically protecting the electrodes.

[0061] During a measurement, the device is worn with the electrode arrangement applied against the tissue, i.e. the skin of the user.

[0062] Electrodes

[0063] There are three sets of signal electrodes:

• • [0064] A first set of signal electrodes 14 are comprised of interdigital electrodes 22 and are operated at the first frequency range between 1 kHz and 200 kHz.
  • [0065] A second set of signal electrodes 13a, 13b, 13c are operated as capacitive sensors in the second frequency range, i.e. between 0.2 MHz and 100 MHz.
  • [0066] A third set of signal electrodes 12a, 12b are operated as “coplanar waveguides” (CPW) at frequencies in the third frequency range of at least 1 GHz.

[0067] Each signal electrode 12a, 12b, 13a, 13b, 13c, 14 forms an electrode pair together with ground electrode 10.

[0068] The primary purpose of the first set of signal electrodes 14 is the measurement of sweat and moisture, as described in section 2.2 of WO 2007/053963. The signal from the signal generator is fed to a contact point in the center of signal electrode 14, and a signal depending on the impedance Z between the signal electrode and the ground electrode is measured, as described in WO 2007/053963.

[0069] The width of the fingers of the interdigital electrodes as well their mutual distance is advantageously in the order of 0.15 mm.

[0070] The electrodes 13a, 13b, 13c of the second set are strip-shaped. The signal from the signal generator is fed to a contact point in the center of the signal electrode, and a signal depending on the impedance Z between the signal electrode and the ground electrode is measured, as e.g. described in WO 2007/053963 or WO 2005/053523.

[0071] Differing electrode geometries are used in order to generate electric fields reaching into different depths of the tissue. Hence, the widths of the gaps 15 around the signal electrodes 13a, 13b, 13c differ. Advantageously: for signal electrode 13a, the width of the gap as well as the width of the signal electrode are typically 4 mm, for signal electrode 13b the width of the gap as well as the width of the signal electrode are typically 1.5 mm, and for signal electrode 13b the width of the gap as well as the width of the signal electrode are typically 0.3 mm. The length of the signal electrodes 13a, 13b, 13c should be as large as possible in order to have a large measured volume of skin and underlying tissue.

[0072] The signal electrodes 13a, 13b, 13c extend parallel to each other and parallel to the longitudinal axis 3, which increases the interaction length between the electrodes and the tissue within the wearer's arm or leg.

[0073] Each signal electrodes 12a, 12b of the third set forms a conductor-backed coplanar waveguide together with the surrounding part of ground electrode 10 and a shield electrode embedded within or arranged on the opposite side of substrate 2. The signal from the signal generator is fed to a first end of the signal electrode and the signal at the second, opposite end is fed to a signal detector, which will be described below.

[0074] Again, as described in WO 2005/120332 and WO 2007/053963, it is advantageous to generate electric fields reaching into different depths of the tissue, for which purpose the width of the gaps around the signal electrodes 12a and 12b differ. Advantageously: for signal electrode 12a, the width of the gap is typically up to 4 mm, for signal electrode 12b the width of the gap is typically up to 0.15 mm. Both signal electrodes 12a, 12b have a width of 0.2 mm or less and a length of 20-23 mm.

[0075] The signal electrodes 12a, 12b of the third set are parallel to each other and extend perpendicularly to longitudinal axis 3. It has been found that for high-frequency sensors of this type, an arrangement perpendicular to the arm/leg of the wearer provides more robust measurements that are less prone to signal errors due to mechanical shifts related to sensor contact with the skin. However, the signal electrodes 12a, 12b may also extend parallel to longitudinal axis 3.

[0076] Optical Sensors

[0077] As described in WO 2007/053963, it is advantageous to combine the electrical measurements with optical measurements, in particular optical reflectance measurements. For this purpose, the device can be equipped with at least one optical reflection sensor. Such a sensor allows to obtain a measure of the perfusion of the tissue.

[0078] In the embodiment of FIG. 2, the device comprises two such optical reflection sensors 23a and 23b. Each optical reflection sensor 23a, 23b is arranged in the gap around signal electrode 13a.

[0079] Each optical reflection sensor 23a, 23b advantageously comprises at least one light source and at least one light detector. In the embodiment shown in FIG. 2, each optical reflection sensor comprises three light sources 27a, 27b, 27c arranged in a row that extends perpendicularly to the longitudinal axis of the sensor. The light sources 27a, 27b, 27c advantageously emit light in the visible or near-infrared spectral range. Furthermore, in the embodiment of FIG. 2, each optical reflection sensor comprises two light detectors 28a, 28b, with the light sources 27a, 27b, 27c located between the light detectors 28a, 28b, such that the light detectors 28a, 28b are able to sense light scattered in forward as well as backward direction respectively to the Iongitudinal axis of the sensor but having different separations to the light sources 27a, 27b and 27c allowing for perfusion at different depths in the tissue to be measured.

[0080] Using at least two light sources with different optical emission spectra has the advantage that differing tissue processes giving rise to a spectrally differing reflectance changes can be distinguished.

[0081] In particular, if two light sources are used, one should generate a wavelength below 600 nm and the other a wavelength above 700 nm, advantageously at approximately 568 nm and 800 nm. If three light sources are used, the first one should advantageously generate light at a wavelength below 600 mm, the second one light of a wavelength above 700 nm, and the third one light at a wavelength between 600 and 700 nm. Advantageous values were found to be 568 nm, 800 nm and 660 nm, respectively. It must be noted that 568 nm and 800 nm are “isosbestic” points where the haemoglobin absorption does not depend on the level of oxygenation. In order to correct for skin perfusion changes, the haemoglobin related signal can be calculated by the ratio of the absorption at 568 nm and 800 nm. The oxygen signal can be calculated at 660 nm, a wavelength where the difference between the absorbance of oxygenated and deoxygenated haemoglobin is at its largest, as a ratio to the 800 nm.

[0082] Electronics:

[0083] FIG. 3 shows a block diagram of an embodiment of the device. It comprises a control unit 30, e.g. a microprocessor with program and data memory as known to the skilled person, which controls the operation of the device. It is connected to various sensors, in particular:

[0084] a) A low-frequency sensor 31 operated by a signal generator and signal detector in the first frequency range, which uses the interdigital electrodes 22 for its measurements.

[0085] b) A medium-frequency sensor 32 operated by a signal generator and signal detector in the second frequency range, which uses the second set of signal electrodes 13a, 13b, 13c for its measurements.

[0086] c) A high-frequency sensor 33 operated by a signal generator and signal detector in the third frequency range, which uses the third set of signal electrodes 12a, 12b for its measurements.

[0087] d) An optical detector 34 measuring optical reflection by means of the optical reflection sensors 23a, 23b, 23c.

[0088] e) A temperature sensor 35 measuring a temperature of the surface of the tissue as well as, optionally, the temperature within housing 1, e.g. by means of a first temperature sensing device in direct thermal contact with substrate 2, as well as by means of a second temperature sensing device arranged inside housing 1.

[0089] Note: In the present embodiment, the signal generators of the low-frequency sensor 31, medium-frequency sensor 32 and high-frequency sensor 33 form the “signal generation circuit” as referred to in the claims.

[0090] In addition to the sensors, control unit 30 controls an interface 3 for exchanging data with an external device, which is used for analyzing and displaying the data measured by the present device. It must be noted, though, that this type of functionality can also be incorporated into the present device itself.

[0091] FIG. 4 shows a more detailed diagram of the high-frequency sensor 33. It comprises a voltage-controlled oscillator 40 with two identical outputs. One of the outputs is connected to a first switch 41, from where it is selectively sent to the input end Tx1, Tx2 of one of the signal electrodes 12a, 12b. The appropriate one of the signals Rx1, Rx2 from the output ends of the signal electrodes 12a, 12b is selected with a second switch 42 and fed to a first input of a magnitude/phase detector 43. The other output of oscillator 40 is routed through two static switches 44, 45 of the same type as the switches 41, 42 and then to the second input of magnitude/phase detector 43. The purpose of the static switches 44, 45 is to increase the symmetry of the two signal paths from oscillator 40 to magnitude/phase detector 43 in terms of temperature and technological variations.

[0092] Magnitude/phase detector 43 measures the relative magnitude and phase of the signals at its two inputs, which correspond to the complex transmission T of the coplanar waveguide, and feeds the corresponding value to an A/D converter 46.

[0093] Glucose Determination

[0094] Control unit 30 is structured and adapted by means of software and the above hardware to determine the glucose level g from the measured parameters.

[0095] The basic principles of operation of this type of device are described in WO 2007/053963. In most general terms, glucose level g can be determined from the measured parameters p_i with i=1, 2, . . . N using a suitable function ƒ as

g=(ƒ(p₁, . . . p_N).

[0096] In many cases, and in particular when suitably choosing the parameters p_i, function ƒ can be a linear or polynomial function in the parameters p_i. In a most simple approach, the glucose level can be determined from the measurements of the device described above by means of the following linear combination

c+k₁·p₁+k₂·p₂+k₃·p₃+k₄·p₄+k₅·p₅+k₆·p₆+k₇·p₇,

where c is an offset constant, to be determined via a reference measurement of glucose performed in the morning and kept constant during the day. The weights coefficient k_i are determined via calibration to reference measurements as described below, and the measured parameters p_i are as follows.

[0097] p1—the average or a weighted sum of the logarithms of the phase of the impedance as measured by the electrode 13a (“large”) of FIG. 1 at the frequencies: 35, 40, and 45 MHz, namely

$p_1=\frac{1}{3}\sum _{i=1}^3{\log }_{10}\left(\angle \left(Z_{\mathrm{el}2}\left(f_i\right)\right)\right)$ $f_i=\left\{35,40,45\right\}\mathrm{MHz},$

where ∠(Z) denotes the phase of the complex variable Z,

[0098] p2—the average of the logarithms of the phase of the impedance as measured by the electrode el3 (“middle”) of FIG. 1 at the frequencies: 35, 40, and 45 MHz, namely

$p_2=\frac{1}{3}\sum _{i=1}^3{\log }_{10}\left(\angle \left(Z_{\mathrm{el}3}\left(f_i\right)\right)\right)$ $f_i=\left\{35,40,45\right\}\mathrm{MHz},$

[0099] p3—the average of the logarithms of the phase of the impedance as measured by electrode 14 of FIG. 1 at the frequencies: 100, 150, and 200 kHz, namely

$p_3=\frac{1}{3}\sum _{i=1}^3{\log }_{10}\left(\angle \left(Z_{\mathrm{el}1}\left(f_i\right)\right)\right)$ $f_i=\left\{100,150,200\right\}\mathrm{kHz},$

[0100] p4—the average of the logarithms of the magnitude of the impedance as measured by electrode 14 of FIG. 1 at the frequencies: 100, 150, and 200 kHz, namely

$p_4=\frac{1}{3}\sum _{i=1}^3{\log }_{10}\left(Z_{\mathrm{el}1}\left(f_i\right)\right)$ $f_i=\left\{100,150,200\right\}\mathrm{kHz},$

[0101] p5—the square of the phase of the transmission coefficient as measured by the electrode 12a of FIG. 1 at the 2.02 GHz, namely

p₅=(∠/(T_el4(2.02 GHz)))²,

with f1 as defined above being 2.02 GHz;

[0102] p6—the square of the phase of the transmission coefficient as measured by the electrode 12b of FIG. 1 at the 2.02 GHz, namely

p₆=(∠(T_el5(2.02 GHz)))²,

with f2 as defined above being 2.02 GHz;

[0103] p7—the temperature measured in close proximity of the tissue by temperature sensor 35.

[0104] The first two terms are directly proportional to the changes in the beta dispersion induced by the changes in glucose concentration in the tissue (cf. FIG. 5). The two separation gaps allow addressing different penetration depths, where the beta dispersion is differently modulated because of different blood vessel concentration along the skin profile (cf. FIG. 6).

[0105] The third and fourth terms are necessary for accounting and compensating changes in impedance due to sweat events but unrelated to changes in glucose (cf. FIG. 7 for p₃, similar dependencies are observed for p₄).

[0106] The fifth and sixth terms are necessary for accounting and compensating changes in impedance due to changes in the water content of the skin but unrelated to changes in glucose (cf. FIG. 8 for p₅, similar dependencies are observed for p₆). Again, the two separation gaps allow addressing the water content distribution which is not constant along the skin profile (cf. FIG. 9).

[0107] The seventh term is necessary for accounting and compensating changes in impedance due to changes in temperature of the skin but unrelated to changes in glucose (cf. FIG. 10).

[0108] Advantageously, the glucose estimation can be further refined by considering also the following terms, namely by expanding the combination as follows

c+k₁·p₁+k₂·p₂+k₃·p₃+k₄·p₄+k₅·p₅+k₆·p₆+k₇·p₇++k₈·p₈+k₉·p₉+k₁₀·p₁₀+k₁₁·p₁₁+k₁₂·p₁₂+k₁₃·p₁₃,

where the first eight terms (from c to k₇·p₇) are the same as described above. The weights coefficient k_i (i=8 . . . 13) are similarly determined via calibration to reference measurements as described below, and the measured parameters p_i (i=8 . . . 13) are as followings.

[0109] p8—the average of the logarithms of the phase of the impedance as measured by electrode 13a of FIG. 1 at the frequencies: 15, 20, and 25 MHz, namely

$p_8=\frac{1}{3}\sum _{i=1}^3{\log }_{10}\left(\angle \left(Z_{\mathrm{el}2}\left(f_i\right)\right)\right)$ $f_i=\left\{15,20,25\right\}\mathrm{MHz},$

[0110] p9—the average of the logarithms of the phase of the impedance as measured by the electrode el3 (“middle”) of FIG. 1 at the frequencies: 15, 20, and 25 MHz, namely

$p_9=\frac{1}{3}\sum _{i=1}^3{\log }_{10}\left(\angle \left(Z_{\mathrm{el}3}\left(f_i\right)\right)\right)$ $f_i=\left\{15,20,25\right\}\mathrm{MHz},$

[0111] p10—the average of the logarithms of the magnitude of the impedance as measured by the electrode 13a of FIG. 1 at the frequencies: 0.5, 1, 2, and 5 MHz, namely

$p_{10}=\frac{1}{4}\sum _{i=1}^4{\log }_{10}\left(Z_{\mathrm{el}2}\left(f_i\right)\right)$ $f_i=\left\{0.5,1,2,5\right\}\mathrm{MHz},$

[0112] p11—the average of the logarithms of the magnitude of the impedance as measured by the electrode el3 (“middle”) of FIG. 1 at the frequencies: 0.5, 1, 2, and 5 MHz, namely

$p_{11}=\frac{1}{4}\sum _{i=1}^4{\log }_{10}\left(Z_{\mathrm{el}3}\left(f_i\right)\right)$ $f_i=\left\{0.5,1,2,5\right\}\mathrm{MHz},$

[0113] p12—the magnitude of the transmission coefficient as measured by electrode 12a of FIG. 1 at the 1.10 GHz, namely

p₁₂=|T_el4(1.10 GHz)|,

with f1′ as defined above being 1.10 GHz;

[0114] p13—the magnitude of the transmission coefficient as measured by electrode 12b of FIG. 1 at the 1.10 GHz, namely

p₁₃=|T_el4(1.10 GHz)|,

with f2′ as defined above being 1.10 GHz;

[0115] p8 and p9 allow better characterizing the modulation of the beta dispersion due to glucose concentration changes, whilst all other terms allow accounting and compensating for changes in the electrical dispersive properties of the tissue which are unrelated to glucose changes.

[0116] Experimental Results

[0117] The above comments on the properties of the various parameters refer to enclosed FIGS. 5-10. These figures show the measured values of the respective parameters as well as of a invasively measured glucose level as a function of time.

[0118] Study Procedure FIG. 5/FIG. 6/FIG. 8/FIG. 10:

[0119] The patients arrived in the clinical study unit in the morning. An intravenous (i.v.) insulin infusion was established and the Multisensor attached to the upper arm by an expandable band (proximal/distal location). After a run-in period of 75 minutes during which the glucose level was stabilized at an euglycaemic level, glucose was administered orally to induce one hyperglycaemic excursion. Euglycaemia was re-established by an i.v. insulin infusion.

[0120] In order to assure close blood glucose (BG) monitoring, changes in the BG were measured with a reference standard technique (HemoCue Glucose Analyzer) from intravenous blood samples.

[0121] Study Procedure FIG. 7/FIG. 9:

[0122] The patients have worn the Multisensor on the upper arm during the day time at home or work. No glucose excursions were induced, but normal daily glucose variations were present.

[0123] The patients have performed frequent self monitoring of blood glucose (SMBG) of at least 10 finger pricks per day with a standard blood glucose meter (BGM).

[0124] Events FIG. 5/FIG. 6:

[0125] The patient was lying in bed in hospital.

[0126] 13:00 h: The Multisensor was detached and reattached to the skin.

[0127] Events FIG. 7/FIG. 9:

[0128] The patient is at home and possibly active.

[0129] 10:00 h: The patient is sweating. The sweat on the skin surface and moisture balance in the skin are both affected. No physical exercise during that time.

[0130] Events FIG. 8/FIG. 10

[0131] The patient was lying in bed in hospital.

[0132] 12:10 h: The patient is changing his position. The water/tissue balance in the skin is affected.

[0133] 14:30 h: The room temperature and therefore also the skin temperature drops.

[0134] Calibration

[0135] Here follows a description of the calibration using “run3” (which is now called c_j in the claims.

[0136] The calibration of the coefficients ki (i=1 . . . 7 or i=1 . . . 13) can be performed as following.

[0137] Measurement data (p1 to p7 or p1 to p13) and the corresponding reference glucose values are collected across several days at a regular interval of one measurement every hour. The data are first divided into sub-series of 4 hours each, then, terms of the type

$c_j+\sum _{i=1}^Nk_i·p_i$

[0138] are fitted, with respect to the coefficient ki, in least square sense to the reference glucose values, where a different additive constant cj is allowed for each sub-series. The use of several additive values allows compensating for a drift in the measurement by minimizing the effect of this latter in the estimation of the parameters k_i.

[0139] Afterwards, the determination of the glucose level in “normal” operation is performed as following.

[0140] First, a measurement of a reference glucose value is collected in the morning concurrently with a measurement set from the device, i.e. values of the parameters p1 to p7 or p1 to p13. With these measurements the offset c is determined by subtracting from the reference glucose value the weighted sum of the parameters weighted with the weights obtained in the calibration step described above.

[0141] After computation of the offset, for subsequent measurements during the rest of the day, the glucose level is computed from the measured parameters, the weights obtained in the calibration step, and the offset c.

[0142] Notes

[0143] The above description as well as the claims mention that a voltage is generated by the signal generation circuit. This formulation is not meant to restrict the signal generation circuit to a voltage source controlled to provide a given voltage, but it may also be a current source controlled to provide a given current, or it may be any other circuit generating a voltage giving rise to a current in the tissue.

[0144] As mentioned above, the device can also be equipped with optical sensors. The signals, or the logarithms of the signals, measured by these sensors can also be incorporated as further parameters p14, p15 in the linear combination of parameters for determining the glucose level.

[0145] Various of the parameters described above are determined from an average of measurements at different frequencies. Alternatively, a weighted sum can be used for the same purpose, wherein the weights can e.g. be obtained in a fitting process varying the weights in order to fit the calculated glucose level to a sufficiently large number of experimentally measured reference glucose levels.

[0146] Alternatively to the magnitude and phase parameterization of the electrical properties, the real and imaginary parts can be considered instead.

[0147] Advantageously, fitting criteria alternative to the least square can be considered, e.g. least absolute deviation, max absolute deviation, robust least squares, and regularized least squares.

[0148] While there are shown and described presently preferred embodiments of the invention, it is to be distinctly understood that the invention is not limited thereto but may be otherwise variously embodied and practiced within the scope of the following claims.