WO2025003415A1

WO2025003415A1 - Continuous wave thz spectroscopy to quantify glucose in liquids

Info

Publication number: WO2025003415A1
Application number: PCT/EP2024/068269
Authority: WO
Inventors: Miriam GALBIATI; Oscar Garcia GARCIA; Henrik Bo LARSEN; Bjørn Hübschmann MØLVIG; Simon Jappe LANGE
Original assignee: Danmarks Tekniske Universitet
Priority date: 2023-06-30
Filing date: 2024-06-28
Publication date: 2025-01-02

Abstract

A method for quantitative detection of glucose in a liquid sample using continuous wave (CW) Terahertz (THz) spectroscopy such as THz cross-correlation spectroscopy (CCS) or THz frequency domain spectroscopy (FDS) in an attenuated total reflection (ATR) setup is disclosed. A continuous-wave (CW) light signal is used to generate a THz signal reflecting off an internal face of a sample surface of an ATR element. An electric detection signal resulting from an interference between THz radiation from the sample and the CW light signal can be recorded while a phase or a frequency domain parameter of the CW light signal is adjusted. A spectrum of the electric detection signal as a function of an adjustment of a phase or a frequency domain parameter is recorded while an external face of the sample surface is in contact with a sample comprising a liquid solution of glucose. The recorded spectrum is analysed using a supervised machine learning algorithm trained on similarly recorded spectra from similar liquid solutions with known glucose levels to estimate a glucose level in the liquid solution of the sample. The method can be used for quantitative, non-invasive, in-vivo detection of glucose in living subjects and other active metabolic systems.

Description

CONTINUOUS WAVE THZ SPECTROSCOPY TO QUANTIFY GLUCOSE IN LIQUIDS

FIELD

The invention relates to a device and a method for quantifying glucose in liquids and living tissue using continuous wave THz radiation.

BACKGROUND

Quantitative analysis of glucose is of interest in the food industry where glucose forms as a product of fermentation processes and in the medical industry where glucose monitoring is the main diagnostic method used by diabetic patients. In both areas, glucose is dissolved in a liquid consisting largely of water.

Optical methods for quantitative analysis of glucose can be non-invasive, which is an advantage for glucose monitoring in diabetes patients and continuous monitoring of fermentation processes. The strongest interaction between electromagnetic (EM) radiation and glucose in water is found in the infrared (IR) and THz ranges. Whereas interactions in IR range probe primarily the intramolecular vibrations of glucose molecules in the solution. These vibrations involve the oscillation and movement of atoms within a molecule, such as stretching, bending, and twisting of chemical bonds. Interactions in the THz range probe primarily intermolecular forces such as hydrogen bonds between water and glucose molecules, see Figs. 1A and B.

IR spectroscopy is the most studied optical techniques for non-invasive glucose analysis since an IR spectrometer can be built using available off-the-shelf components. The main drawback of IR spectroscopy is that IR radiation also interacts strongly with many other substances in liquid solution, in particular other blood analytes so that a spectrum over a considerable range of wavelengths and with high spectral resolution is needed. This, in part, imposes requirements on the laser system such as high beam quality, high spectral repeatability, high tuning speed, and good power across the tuning spectrum to enable consistent and accurate measurements. Such laser systems, although readily available, are bulky and have considerable power consumption.

Terahertz (THz) time domain spectroscopy (TDS) is an established method for material characterization. Some preliminary result of glucose detection using TDS have been reported, see e.g.:

- Cherkasova et al., Noninvasive blood glucose monitoring in the terahertz frequency range, Opt Quant Electron (2016) 48:217, DOI: 10.1007/s11082-016-0490-5

- Cherkasova et al., Studying Human and Animal Skin Optical Properties via Terahertz Pulsed Spectroscopy, Bulletin of the Russian academy of sciences. Physics Vol. 80 No. 42016, DOI: 10.3103/S1062873816040067 - Wang et al., Critical Factors for In Vivo Measurements of Human Skin by Terahertz Attenuated Total Reflection Spectroscopy, Sensors 2020, 20, 4256; DOI: 10.3390/s20154256.

Huang et al., Attenuated Total Reflection for Terahertz Modulation, Sensing, Spectroscopy and Imaging Applications: A Review, Appl. Sci. 2020, 10, 4688; D0l:10.3390/app10144688.

These papers report in vivo measurements on human and rat skin using THz-TDS measurements. The measurements are made before and after intake of glucose and measured variations in skin optical properties showed some correlation with changes in independently measured blood glucose concentration. No quantitative detection of glucose is indicated therein. In TDS, the THz generation requires the use of an expensive and bulky femtosecond (fs) laser systems which then also imposes limitations on the optical components of the system. Devices for quantitative detection of glucose using the spectroscopic techniques reported above are disadvantageous since scaling down the physical size of the device is extremely challenging.

Other THz based approaches for glucose detection have been reported, including:

- Takuro et al., Double-beam CW THz system with photonic phase modulator for sub-THz glucose hydration sensing, 2016 IEEE MTT-S International Microwave Symposium, 2016, DOI:

10.1109/MWSYM.2016.7540168. It relates to transmission-type CW-THz spectroscopy using homodyne detection and demonstrates characterization of the hydration of glucose in water, and not a quantitative detection of glucose.

- Suhandy et al., A quantitative study for determination of sugar concentration using ATR-THz spectroscopy, Sensing For Agriculture And Food Quality And Safety III, 2011 , DOI: 10.1117/12.886183. It relates to recording spectra of glucose in water using a mercury lamp THz source in ATR setup and applying chemometrics to the determined spectra to determine a glucose concentration.

These approaches relate to pure solutions of glucose in water, and there is no indication that they could apply to more complicated systems such as metabolic systems.

SUMMARY

Accordingly, there is a need for devices and methods for quantitative detection of glucose in a sample, which may mitigate, alleviate or address the shortcomings existing and may provide a small and portable solution to determining glucose levels.

A method for quantitative detection of glucose in a sample using continuous wave (CW) Terahertz (THz) spectroscopy in an attenuated total reflection (ATR) setup is disclosed. The method is performed using a THz CCS setup comprising: a light source for outputting a continuous-wave (CW) light signal; a THz transmitter optically coupled to the light source via a first optical path, the THz transmitter configured to emit THz radiation when modulated by the CW light signal; an ATR element positioned for THz radiation from the THz transmitter to reflect off an internal face of a sample surface of the ATR element; a THz receiver optically coupled to the light source via a second optical path, the THz receiver being positioned to receive THz radiation reflected off the internal face of the sample surface and configured to detect received THz radiation by generating an electric detection signal that is a result of an interference between the THz radiation and the CW light signal; and means for adjusting a phase or a frequency domain parameter of the CW light signal.

The method comprises at least the steps of recording a CW THz spectroscopy spectrum of the electric detection signal as a function of an adjustment of a phase or a frequency domain parameter of the CW light signal while an external face of the sample surface is in contact with a sample comprising a liquid solution of glucose; and analysing the recorded CW THz spectroscopy spectrum using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples with liquid solutions having known glucose levels to estimate a glucose level in the liquid solution of the sample.

Further, a device for recording a continuous wave (CW) Terahertz (THz) spectrum from a sample comprising a liquid is disclosed. The devise comprises: a THz CCS setup comprising: a light source (2, 52) for outputting a continuous-wave (CW) light signal; a THz transmitter (4) optically coupled to the light source via a first optical path (11), the THz transmitter configured to emit THz radiation (5) when modulated by the CW light signal; an attenuated total reflection (ATR) element (7) positioned for THz radiation from the THz transmitter to reflect off an internal face of a sample surface (8) of the ATR element; a THz receiver (6) optically coupled to the light source via a second optical path (12), the THz receiver being positioned to receive THz radiation (5) reflected off the internal face of the sample surface and configured to detect received THz radiation by generating an electric detection signal that is a result of an interference between the THz radiation and the CW light signal; and means for adjusting a phase or a frequency domain parameter of the CW light signal.

The device may also comprise a control unit comprising an electronic processor configured to control the means for adjusting to adjust a phase or a frequency domain parameter of the CW light signal and record a THz CCS spectrum by recording the electric detection signal from the THz receiver as a function of the adjustment of a phase or a frequency domain parameter of the CW light signal.

Also, a system for quantitative detection of glucose in a sample is disclosed. The system comprises the device for recording a CW THz spectroscopy spectrum disclosed above and a data analysis unit comprising an electronic processor being configured to access the recorded CW THz spectroscopy spectrum and analyse the recorded CW THz spectroscopy spectrum using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples comprising liquid solutions with known glucose levels to estimate a glucose level in the liquid solution of the sample.

It is an advantage of the present disclosure that the device can be made much smaller than prior art devices, whereby small, portable or wearable devices with low power consumption are made possible.

It is an advantage that the THz generation in CW THz spectroscopy relies on an off-the-shelf light source, such as an Amplified Spontaneous Emission (ASE) source or a Superluminescent Diode (SLED). The CW light source overcomes the cost and footprint-related issues with the lasers required in IR and TDS. In comparison to TDS, the invention can be used solely with an optical waveguide and thus does not require free space optical paths. Thereby, the optical system can be made smaller and less sensible to shocks and environmental parameters like temperature and humidity.

The disclosed system offers specific advantages over the disclosures of Cherkasova et al. and Takuro et al. discussed above. Firstly, the use of CW THz radiation in an ATR setup greatly simplifies the physical setup and reduces sensitivity to environmental factors in comparison to the double-beam and time-domain methods described in these references. Secondly, the adjusting of phase or frequency domain parameters of the CW light signal enhances the ability to fine-tune the spectroscopic analysis, which is not disclosed in in these references. Thirdly, it allows the use of more accessible light sources like Amplified Spontaneous Emission (ASE) sources or Superluminescent Diodes (SLEDs), which are less costly and have lower power consumption compared to the more complex systems used in these references. In summary, these effects of using THz-CCS are that the device can be made cheaper, smaller, and more robust, thereby allowing for a device for quantitative glucose measurements in active metabolic systems that can be purchased and operated by normal users from diabetes patients to homebrewers.

The use of analysing the recorded spectrum using a supervised machine learning algorithm trained on similarly recorded spectra is advantageous since it allows for weaker spectral features to be evaluated and compared. In one or more exemplary embodiments, the sample and the training samples comprise or are part of an active metabolic system, such as comprises living cells and/or a liquid solution containing glucose and other metabolites. In such embodiments, the disclosure relates to quantitative detection of glucose in samples that contain many organic and inorganic compounds in addition to glucose. The presence of such additional compounds related to metabolism extensively complicates the quantification of glucose relative to prior art quantitative detection of pure solutions of glucose in water. It is an advantage of the present disclosure that the glucose level estimation uses a supervised machine learning algorithm trained on similar samples that also comprise or are also comprised by an active metabolic system. This approach makes it possible to include parts of the spectral data which would otherwise not be considered due to its complexity, resulting in an enhanced accuracy and efficiency of interpreting intricate patterns correlating with different glucose levels.

In an exemplary embodiment, the CW THz spectroscopy may be THz cross-correlation spectroscopy (CCS). CCS differentiates itself in comparison to other types of THz spectroscopy systems in that the CCS system utilizes a broadband CW incoherent source that generates a broadband THz signal.

In another exemplary embodiment, the CW THz spectroscopy may be the Frequency-Domain spectroscopy (FDS). FDS differentiates itself in comparison to other types of THz spectroscopy systems in that it utilizes two or more narrowband lasers, from which at least one is swept in wavelength, that generate a single frequency THz signal for every wavelength of the swept laser.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and other features and advantages of the present disclosure will become readily apparent to those skilled in the art by the following detailed description of examples thereof with reference to the attached drawings, in which:

Figs. 1A and B illustrate the interactions between glucose in water and EM radiation in the IR range (1A) and in the THz range (1B).

Fig. 2 illustrates an exemplary continuous broadband spectrum according to this disclosure.

Fig. 3 is a flow-chart illustrating an example method for quantitative detection of glucose in a sample.

Figs. 4 and 5 are diagrams illustrating exemplary devices according to this disclosure.

Fig. 6 is a block diagram illustrating an example control unit, an example data analysis unit, and an example glucose level monitoring unit according to this disclosure.

Fig. 7 is a diagram illustrating an exemplary device according to this disclosure.

Fig. 8 is a diagram illustrating an exemplary system according to this disclosure.

Figs. 9-11 are graphs showing results from demonstrations of exemplary embodiments according to this disclosure.

DETAILED DESCRIPTION

The invention provides a method and a system for quantitative detection of glucose in a sample using continuous wave (CW) Terahertz (THz) radiation in in an attenuated total reflection (ATR) setup. The system comprises, and the method utilizes, a CW THz spectroscopy setup comprising at least the following parts: a light source; a beam splitter, a THz transmitter, an ATR element, and a THz receiver. The system further comprises a control unit for recording a CW THz spectrum and a data analysis unit for analysing the recorded CW THz spectrum and estimating a glucose level. The CW THz setup and the control unit form a device for recording a CW THz spectrum from a sample comprising a liquid. Terahertz radiation lies in the electromagnetic spectrum between microwave and infrared wavelengths, typically ranging from 0.1 to 10 THz (or 100 GHz to 10 THz). In CW THz spectroscopy, a continuous wave terahertz source emits a coherent or incoherent continuous beam of terahertz radiation. This radiation is then directed towards the sample of interest. The THz radiation is measured after it has interacted with the sample to obtain information about the sample's properties.

The interaction between terahertz radiation and matter is based on the absorption, reflection, transmission, and scattering of the radiation by the sample's constituents. Different materials exhibit characteristic absorption and refractive properties in the terahertz frequency range, allowing for the identification and characterization of compounds, detection of impurities, and analysis of molecular structures.

It's worth noting that CW THz spectroscopy is just one approach to terahertz spectroscopy utilizing a CW THz signal. Other techniques, such as time-domain spectroscopy (TDS) and Fourier-transform spectroscopy (FTS), are also commonly used for terahertz analysis, each with its own advantages and limitations. In CW systems, it is easier to trim the spectral properties of the source, and therefore the spectral properties of the THz wave, to maximize the interaction with analytes in water solutions.

In the present disclosure, at least two types of CW THz spectroscopy may be applied, terahertz crosscorrelation spectroscopy (CCS) and Terahertz Frequency-Domain spectroscopy (FDS).

In an exemplary embodiment, the CW THz spectroscopy is Terahertz Cross-Correlation Spectroscopy (THz CCS). In THz CCS, a time-domain signal in the form of the cross-correlation between a CW pump source and the THz radiation generated by the CW pump source in a THz emitter is measured. When transformed to frequency domain, its spectral composition can be used to extract information from the sample. CCS enables phase-sensitive measurements with a simple physical setup but complex data analysis. The bandwidth of CCS systems is generally smaller than TDS systems, but the dynamic range at low frequencies is larger. This is beneficial for liquid samples, since high frequencies are absorbed. Background descriptions of terahertz cross-correlation spectroscopy may be found in e.g. “Two decades of terahertz cross-correlation spectroscopy”, Appl. Phys. Rev. 8, 021311 (2021 ); doi: 10.1063/5.0037395, or "Terahertz cross-correlation spectroscopy driven by incoherent light from a superluminescent diode”, Opt. Express 27, 12659-12665 (2019); doi 10.1364/OE.27.012659.

In an exemplary embodiment, the CW THz spectroscopy is Terahertz Frequency-Domain spectroscopy (FDS). Here, at least two CW light sources, of which at least one is a wavelength or frequency sweeping laser, are coupled to form a combined optical signal which is divided into two optical paths. One path is employed to generate a CW THz wave in a THz transmitter. The other path is employed in coherent/phase-sensitive detection with a THz receiver. Compared to CCS, FDS does not require scanning an optical delay between the paths, resulting in a CW terahertz spectrometer with simplified architecture and increased recording speed. Background descriptions of FDS may be found in e.g. Preu, S., Dbhler, G. H., Malzer, S., Wang, L. J. & Gossard, A. C. Tunable, continuous- wave Terahertz photomixer sources and applications. J. Appl. Phys. 109, 061301 (201 1 ) or Deninger, A. J., Roggenbuck, A., Schindler, S., & Preu, S. (2015). 2.75 THz tuning with a triple-DFB laser system at 1550 nm and InGaAs photomixers. Journal of Infrared, Millimeter, and Terahertz Waves, 36, 269-277.

Although other types of CW THz spectroscopy may be applied, in the present description CW THz spectroscopy preferably refers to THz CCS and/or THz FDS. Also, CW THz spectroscopy spectrum preferably refers to CCS spectrum and/or FDS spectrum, and CW THz spectroscopy device preferably refers to CCS device and/or FDS device. In one or more exemplary embodiments, there is no active adjustment of a phase domain parameter of the CW light signal, and only a frequency domain parameter of the CW light signal is adjusted, whereby the electric detection signal is recorded as a function of only a frequency domain parameter of the CW light signal to obtain the CW THz spectroscopy spectrum.

CCS and FDS have so far been used in characterization of solid materials, in particular to determine porosity or thickness of individual material layers. The quantification of individual substances in liquid solution has so far not been considered feasible or, at best, been speculative.

The visibility and exact position of spectral peaks of glucose in aqueous in the THz range can be highly sensitive to experimental conditions (temperature, concentration, etc.) and the specific spectroscopic setup. Changes in these parameters can shift or alter the appearance of the peaks. Further, as described in more detail later herein, the picture gets even more complex when the liquid of the sample comprises or is comprised by an active metabolic system. To reliably detect and characterize these peaks, especially at low concentrations typical of physiological levels, high- resolution and highly sensitive methods are needed. In one or more exemplary embodiments, the.CW THz spectroscopy setup is configured to emit and detect THz radiation having a bandwidth in the range 100 to 2000 GHz, such as 500 to 1500 GHz or 100 to 1000 GHz.

CCS and FDS setups involve many of the same components, a main difference being that a CCS setup comprises an optical delay setup whereas the CW light source in a FDS setup is a combination of at least two light sources, at least one of which can be frequency swept. The following provides a description of the components that are common in a CCS and a FDS setup, followed by descriptions of the components that differ.

The CW THz spectroscopy setup, such as a CCS or FDS setup, comprises a light source generating a continuous wave light signal, such as a light source with wavelengths in the visible or near infrared (NIR). The term ‘light’ is used herein to distinguish the light signal from the THz signals, both of which are technically both 'optical signals’, i.e. electromagnetic wave signals. However, in this description, the light signal and the components used for this, are sometimes also referred to using the adjective ‘optical’. The CW light signal is later used to generate THz signals interacting with the sample. But since THz radiation is more difficult to manipulate than light signals, the detailed tailoring of the optical properties of the THz signals are prepared in the visible or near-infrared. The light source generating the CW light signal may comprise several individual light sources with the output from the different light sources being combined to one CW light signal. It is a major advantage that the CW THz spectroscopy setup utilizes a CW light signal, as opposed to a pulsed light signal which may have very high peak power and intensity. This means that there is no free-space-propagation requirement for the optical paths so that waveguides and cheaper components can be used instead. Therefore, in an exemplary embodiment, the CW THz spectroscopy setup comprises optical waveguides providing the first and second optical paths. In the present specification, an optical waveguide is a system or material designed to confine and direct electromagnetic waves in a direction determined by the physical boundaries of the waveguide. Typical waveguide types are optical fibres, channel waveguides and planar waveguides. In the CW THz spectroscopy setup utilized by the invention, optical paths for the CW signal from the light source to the optical delay component, the THz transmitter, and the THz receiver are preferably provided by optical waveguides, thus ensuring a full optical waveguide path for the device. Thereby, the device does not rely on lenses and mirrors to control the direction and vergence of the optical signal along the optical path. In a free-space setup, lenses and mirrors are bulky components and require a precise and stable alignment. A full optical waveguide path is thus advantageous since it is less sensitive to variations in humidity and temperature, mechanical vibrations, and shocks. In addition, the use of optical waveguides allows for reducing the overall size of the device. In an alternative formulation, it is preferred that the CW optical signal does not propagate in free space (i.e. through the air of the environment) at any point during its path to the THz antennae. In one or more exemplary embodiments, the CW THz spectroscopy setup of the disclosed method and system comprises a light source comprising a distributed feedback laser (DFB) and/or a Fabry-Perot laser. These are advantageous since they are small, low power, and robust and can be incorporated in devices for personal and/or handheld use.

The CW THz spectroscopy setup comprises a beam-splitter for receiving the CW light signal from the light source and defining a first arm providing a first optical path coupling the beam splitter and the THz transmitter, and a second arm providing a second optical path coupling the beam splitter and the THz receiver. Such beam splitter will divide a light signal from the light source into two separate light signals. In one example, the beam splitter is a 50/50 splitter providing two light signals of similar power. In another example, the beam splitter provides two signals with one having significantly higher power than the other, such as a 60/40, 70/30 or 80/20 split, where the higher power signal will typically be for the THz transmitter.

The CW THz spectroscopy setup comprises a THz transmitter optically coupled to the light source via a first optical path, the THz transmitter configured to emit THz radiation when modulated by the CW light signal. The setup comprises an ATR element positioned for THz radiation from the THz transmitter to reflect off an internal face of a sample surface of the ATR element. The setup comprises a THz receiver optically coupled to the light source via a second optical path, the THz receiver being positioned to receive THz radiation reflected off the internal face of the sample surface and configured to detect received THz radiation by generating an electric detection signal that is a result of an interference between the THz radiation and the CW light signal. INV The THz transmitter and receiver are generally THz antennas. One exemplary type is a ‘CW photomixer’ with a semiconductor structure (which may in itself be composed of many different layers of semiconductor) and a metallic antenna (typically bowtie or dipole antenna). Here, the optical signal excites the semiconductor while a voltage bias is applied to the antenna poles (for the transmitter) or the current generated in the antenna is measured (for the receiver). Other applicable THz antennas exist, and new ones may be developed that are equally applicable in the devise of the invention. Antennas optimized for narrow-band frequency mixing or for broadband, pulsed mixing exists. In one or more exemplary embodiments, the THz transmitter and the THz receiver comprise antennas configured for narrow-band frequency mixing.

The shape of the ATR element and the position of the THz transmitter are preferably configured so that an angle of incidence of the THz radiation from the THz transmitter onto the internal face of the sample surface is greater than the critical angle so that total internal reflectance occurs. The ATR element is selected to have good transmitting properties in the THz range. The ATR element may be a crystal such as a Silicon crystal, sapphire or quartz. The ATR element may also be formed in noncrystalline material such as an amorphous solid, such as fused silica. The ATR element may be a prism cut with precise angles and plane faces configured to reverse the direction of THz radiation by internal reflection.

It is noted that the sample surface of the ATR element may be extended by components having the same or similar refractive index in the THz range as the material of the ATR element. Thus, a component, such as a window or a protective layer, separate from the material composition of the ATR element but abutting the sample surface and fulfilling the geometric requirements for total internal reflectance at an external surface, can be used to extend the ATR element. Thereby, the external surface of the component becomes the new sample surface of the ATR element. This is advantageous if it is desired to attach replaceable components to the sample surface for protective or hygienic reasons. It is also advantageous if the CCS setup is to be brought into contact with liquid samples contained in containers or conveyers, in which case the component can be a window in the container or conveyer.

Whereas the above description is common to CCS and FDS setups, the CCS setup comprises a light source emitting a broadband, continuous spectrum (without longitudinal modes). In an exemplary embodiment, the spectrum of the optical signal from the light source is a broadband spectrum having an optical bandwidth of at least 8 nm at a center wavelength of 1550 nm such as at least 10 nm. With present THz antennas, the bandwidth of the THz signal is, to some extent, determined by the bandwidth of the CW optical signal. In this context, THz means electromagnetic radiation in a spectral range [0,1 ; 10 THz]. Therefore, in another exemplary embodiment, the spectrum from the light source is a broadband spectrum that, when received by the THz transmitter, results in the generation of a THz signal with a bandwidth of at least 0,1 THz, such as at least 0,5 THz or at least 1 THz. As a note, at 1550nm, an optical bandwidth of 8 nm corresponds to a THz bandwidth of approximately 1 THz. This conversion is only true for 1550 nm, at shorter wavelengths, smaller optical bandwidth is needed to obtain the same THz bandwidth. In the present specification, a broadband spectrum of the light source means a spectrum having a -3dB bandwidth A and a centre wavelength , with cAX/X² > 0,1 THz, such as > 0,2 THz, such as > 0,5 THz. The broader the broadband spectrum of the light source is, the more spectral information can be retrieved from the sample, thus increasing the functionality of the device for the user. In an exemplary embodiment, the light source is a superluminescent light emitting diode (SLED).

The center frequency of the broadband spectrum can be selected depending on the light source or the type of antenna substrate in the THz transmitter and receiver. In one embodiment, the center frequency is around 1550nm. This is advantageous since commercial light sources and fibre optics from telecom can be used. In another embodiment, the centre wavelength is shorter - such as 1064nm or 960nm - which is advantageous since a smaller bandwidth of the light source is needed to meet the 0,1 THz requirement. A further advantage is that cheaper and/or better semiconductor materials for the antennae are available for these shorter wavelengths.

Fig. 2 shows an exemplary continuous broadband spectrum of a light source according to exemplary embodiments. The spectrum is centred around 1550 nm, has a bandwidth of 40 nm, and is void of any mode peaks or other discontinuities.

As will be described in more detail later, the light source of the CCS setup may comprise an amplified spontaneous emission (ASE) source, an SLED, a light emitting diode (LED), an erbium-doped fibre amplifier (EDFA), and any combination of those.

The CCS setup comprises an optical delay setup configured to adjust a synchronization of the THz receiver to the THz transmitter by the CW optical signal. The beam splitter is positioned between the optical source and the optical delay setup. In exemplary embodiments, the optical delay setup is based on optical fibres and/or solid-state waveguides and does not involve mechanically moving parts or separate mirrors and lenses that must be kept in precise alignment with other optical components. These embodiments are advantageous since they are more robust and less sensitive to mechanical vibrations and shocks. They are further advantageous since the optical signal propagates in solid matter instead of free space and is therefore much less sensitive to environmental parameters such as pressure, temperature, humidity, gases, aerosols etc. Exemplary optical delay setups are described later herein.

To achieve high-precision terahertz cross-correlation measurements, the optical signals driving both THz antennae are preferably in phase. On the other hand, using a broadband spectrum optical source will typically result in a coherence length of the order of centimetres. Therefore, in an exemplary embodiment, an optical path length of the first arm, Li, and an optical pathlength of the second arm, L2 are equal within a range corresponding to a coherence length of the optical signal. In another exemplary embodiment, an optical path length of the first arm, Li, and an optical pathlength of the second arm, L2 are preferably equal within half the stroke of the optical delay setup, where the stoke is the maximum difference in optical path length between the two arms achievable by the optical delay setup. In yet another exemplary embodiment, an optical path length of the first arm, Li, and an optical pathlength of the second arm, L2 are preferably equal within a length of 5 cm. The scanning range of the optical delay setup may be in the order of tenths to thousands of picoseconds, with time-steps typically in the order of 10-100 femtoseconds. For example, for a system with 3 THz bandwidth, Nyquist theorem states that the time-step should be 166 fs.

Features that are common to CCS and FDS setups and features that are specific to CCS have been described above. The FDS setup comprises at least two narrowband light sources, such as two single mode (single frequency) lasers, out of which at least one can be swept in wavelength. In an exemplary embodiment, these light sources are Distributed-Feedback (DFB) lasers with a linewidth in the range of hundreds of kHz, such as 600 kHz, and a center wavelength of 1550 nm. These lasers typically have thermal tuning coefficients in the range of 10 GHz/K. Therefore, a change in temperature via a built-in thermoelectric cooler on the laser of, for example. 50K will lead to a change in the center wavelength of 500 GHz. Therefore, sweeping the two lasers allows for generation of a THz signal on the range from 0 THz to 1 THz. Cascading more than two laser sources expands the bandwidth.

In another exemplary embodiment, the light sources are centred at 800 nm. These have the advantage of typically larger thermal tuning coefficients, such as 20 GHz/K. This allows for increased bandwidth.

The light sources above are combined in a single waveguide such as an optical fibre using a coupler. The setup might or might not include an optical amplifier, such as an erbium doped fibre amplifier, to boost the optical power from the lasers into the required power for the antennas. The setup might or might not include optical isolators to prevent reflections into the light sources.

The THz spectrum is recorded as a function of the wavelength tuning of the sources, such as the temperature tuning. A control unit is used to control and adjust the wavelength, such as by adjusting the temperature, while recoding the photocurrent in the receiver.

In one or more exemplary embodiments, the method and system relate to quantitative detection of glucose in a sample that comprises or is part of an active metabolic system such as a human or animal body (in-vivo), human or animal cells in vitro, and microorganisms in liquid suspension such as fermentation tanks used for everything from beer brewing to fertilizer production. Such systems generally comprise many organic and inorganic compounds in addition to glucose. The presence of such additional compounds related to metabolism, including metabolic compounds and metabolites, means that obtainable spectra are “messier” in that contributions from the other compounds may overlap or skew spectral information from glucose that would typically be relied upon to perform a quantitative detection. In the present disclosure, an active metabolic system designates a biological system with living organisms that undergo metabolism.

In addition, such metabolic systems pose a challenge when it comes to providing similar reference samples with known levels of glucose. Several prior art references report quantitative detection of pure glucose in distilled water where reference samples will be indistinguishable from measured samples with the same concentration. Under these conditions, standard chemometric techniques can be applied to the measured spectra and the reference spectra to make a quantitative detection. The present disclosure utilizes spectra from similar samples with liquid solutions having known glucose levels. For metabolic systems, however, such similar or reference sample will often be distinguishable from a measured sample with the same glucose concentration, since the concentration of several other metabolic compounds will differ. As a result, direct comparison of spectra may yield differences that are not correlated to differences in glucose concentration, rendering standard chemometric techniques unsuitable.

A further challenge with quantitative detection in samples that contains or are part of an active metabolic system is, that the relative concentration of glucose to other metabolites in the sample continuously change, whereby the recorded spectra will represent some average sample composition over the duration of the recording of the spectrum. For in-vivo measurements on the skin, additional temporal variations related to the skin occlusion effect occurs. The skin occlusion effect occurs when the skin's surface is covered in a way that prevents moisture on the skin from evaporating normally, which increases skin hydration. This may impact the measured spectra because changes in hydration and other factors such as sweat could affect the optical properties of the skin. As such effects may occur at similar timescales as the recording of a spectrum utilizing the disclosed method and system, they may complicate the interpretation of the recorded spectra.

All the factors discussed above can complicate the quantification of glucose in active metabolic systems extensively in comparison to prior art quantitative detection of glucose in water. For this purpose, the disclosed method and system estimates a glucose level in the sample using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples with known glucose concentrations, hereinafter also referred to as training samples. In one or more exemplary embodiments, the training samples also comprise or are part of an active metabolic system.

In one or more exemplary embodiments, the liquid of the sample and the training samples comprise living biological cells such as living tissue. Exemplary samples comprise bacteria and eukaryotes, living cells from a human or animal in vivo or in vitro, as well as other microorganisms such as yeast cells. In one or more exemplary embodiments, the liquid of the sample and the training samples comprise a liquid solution containing glucose and other metabolites. In one or more exemplary embodiments, the liquid of the sample and the training samples comprise interstitial fluids and/or liquids containing products of a fermentation process.

According to embodiments of the invention, the method and system relates to quantitative detection of glucose in a sample, such as a sample comprising a liquid solution of glucose or glucose and other analytes. The invention uses CW THz spectroscopy in an attenuated total reflection setup which can record a CW THz spectroscopy spectrum of a sample simply by contacting the sample “from the outside”, i.e. without having to extract or isolate the sample from its arrangement and insert it in a detection apparatus such as a spectrometer. This is advantageous in arrangements where the glucose level in the sample changes continuously so that repeatedly extracting a sample from the arrangement to monitor the glucose level becomes laborious or inconvenient. In real-life situations, glucose is found in liquid solutions also containing numerous other compounds, typically liquids consisting of a large degree of water. The invention is further advantageous in that it uses CW THz spectroscopy which can record a spectrum indicative of a glucose level in a sample comprising a liquid solution of glucose.

In exemplary embodiments, the sample is a liquid solution containing glucose, and the liquid solution can be brought into contact with the external face of the sample surface of the ATR element. Here, the external face of the sample surface may form a window in a container or conveyer containing the liquid solution. In a preferred embodiment, the liquid solution is a product of a fermentation process.

In exemplary embodiments, the method and system relate to quantitative, non-invasive, in-vivo detection of glucose in a living subject, such as a human or animal. Here, the sample is an epidermis segment of the subject and the liquid solution is interstitial fluid in the epidermis segment. The interstitial fluid surrounds every cell of the body and continuously exchanges substances with the blood plasma across capillary walls. Hence, the glucose concentration in the interstitial fluid correlates with the glucose concentration in the blood. The CW THz spectroscopy spectrum is recorded while an external face of the sample surface of the ATR element is in contact with the stratum corneum, the outermost layer of the epidermis segment. Since THz radiation is non-ionizing, it is completely safe for use in-vivo.

The method and system for quantitative detection of glucose, as well as the device for recording a CW THz spectroscopy spectrum, involves controlling the recording of a CW THz spectroscopy spectrum, such as a CCS spectrum or a FDS spectrum. For this purpose, the device and the system comprise a control unit with memory circuitry, processor circuitry, and an interface utilizing e.g. USB or Wi-Fi communication to an external consumer. The control unit is connected to components of the spectrometer components and the control unit, or the processor circuitry thereof, is configured to perform operations to record a CW THz spectrum. The operations of the control unit may be considered method steps that the control unit is configured to carry out. In exemplary embodiments utilizing a CCS setup, these operations may comprise:

- control the optical delay setup to adjust the difference in optical path length between the first and second optical paths; and record a THz CCS spectrum by recording the electric detection signal from the THz receiver as a function of the difference in optical path length.

In exemplary embodiments utilizing an FDS setup, the operations may comprise:

- sweeping the frequencies of the CW light sources for THz generation and detection; and record a THz FDS spectrum by recording the electric detection signal from the THz receiver as a function of the frequency difference between the CW light sources.

In exemplary embodiments, operations of the control unit may comprise one or more of: powering up the CW THz spectroscopy setup, including providing power to the light source and the THz transmitter and the THz receiver. providing an electric potential to the optical delay setup (for CCS), such as to put it in a particular starting position or initiating frequency sweeping in the optical source (for FDR).

- Applying an electronic modulation signal to the THz transmitter.

- Applying the same modulation signal as input to a lock-in amplifier, which additionally takes in the read-out voltage from a transimpedance amplifier, which is takes an electrical current input directly from the THz receiver unit.

Performs numeric operations on a microcontroller of the output from the lock-in amplifier. The result of these numeric operations is a set of numbers that are transferred to a peripheral computer. in exemplary embodiments, a protocol or instruction manual for a user or an operator performing in vivo or in vitro measurements is provided.

Protocol for in vivo measurements:

- To minimise drift due to e.g. temperature sensitivity of many light sources, turn on the control unit and await stabilisation of the different components.

- Select the site on the body to perform the measurements. Areas with no hair and thin skin (such as, but not limited to, forearm, fingertips, or palm) are best suited to perform the measurements. Dry the area of the skin chosen as measurements site to remove sweat that could affect the refractive properties of the skin.

- The external face of the sample surface of the ATR element should be clean and free of impurities. Thus, the ATR surface is cleaned with alcohol, such as ethanol, before starting the measurements.

- The contact between the skin and the ATR prism surface leads to mechanical deformation of the skin’s surface. This means that the applied pressure affects the THz response of skin due to skin hydration. Therefore, a constant contact pressure should preferably be maintained over the duration of the measurement. To this aim, the setup can be equipped with a pressure sensor to maintain a constant pressure in the optimal range to get the best THz contrast. By controlling the contact pressure, the disturbances caused by variations in the water concentration of the skin are reduced or eliminated. This, in turn, improves the stability and accuracy of the measurements.

- When the skin is in contact with the prism, the contact pressure may cause water to accumulate on the surface of the skin, leading to a change in the water concentration of the skin. Thus, the measurements are performed when the water content in the skin is stabilized.

- After every measurement, clean the surface of the ATR prism with alcohol such as ethanol.

Protocol for in-vitro measurements:

- To minimise drift due to e.g. temperature sensitivity of many light sources, turn on the device and await stabilisation of the different components.

Let the liquid samples warm up to room temperature. - The external face of the sample surface of the ATR element should be clean and free of impurities. Thus, the ATR surface is cleaned with alcohol, such as ethanol, before starting the measurements.

Use a pipette to take the liquid sample from the liquid container and transfer it onto the ATR element. To dispense all the liquid in the pipette and avoid air bubbles, operate the plunger slowly and smoothly.

- The liquid sample of glucose to investigate is placed directly in contact with the external face of the sample surface of ATR element. The droplet of glucose solution is centred with respect to the position of the beam on the sample surface of the ATR prism. For each sample, the same volume of glucose solution is used for acquisition of THz signal.

If it is a series of measurements, randomize the order of the different glucose solutions to investigate.

- The THz signal of water is recorded for each glucose sample measurement and used as reference signal to detect small differences in frequency response.

Device settings such as integration time and number of time traces are the same for each measurement, for both reference and glucose sample.

- After each measurement, the liquid sample is aspirated with a pipette. The external face of the sample surface of the ATR element is wiped and cleaned with ethanol.

While the described functions and operations may be implemented in software, such functionality may as well be carried out via dedicated hardware or firmware, or some combination of hardware, firmware and/or software.

Based on the recorded spectrum, a glucose concentration in a liquid in the sample can be inferred by a nonlinear regression model based on supervised machine learning, such as but not limited to decision trees and neural networks. The model is trained on a labelled dataset comprising pairs of input features and target values. Specifically, the input features consist of pre-processed sample and reference measurements, where sample and reference measurements always include CW THz spectroscopy spectra of the liquid solutions under investigation and may include temperature, humidity, force and pressure measurements. Reference measurements are CW THz spectroscopy spectra measured on a known reference sample such as a measurement with no sample present. The target values are the known glucose concentrations of the samples.

The nonlinear regression model is trained by minimizing a suitable loss function such as the mean squared error (MSE) or the mean absolute relative difference (MARD), which measures the difference between the estimated glucose concentration and the true glucose concentration. The minimization process uses a suitable optimization algorithm for the given model such as backpropagation for neural networks or gradient boosting for decision trees. The various hyperparameters of the regression model, such as the learning rate, width, and depth in the case of neural networks or the learning rate or maximum depth in the case of decision trees, are tuned through a suitable hyperparameter optimization scheme such as random search or Bayesian optimization. Once optimized, the model is saved and used for inference on new, unknown samples.

In an exemplary embodiment, the applied force to the ATR prism at the time of recording is used as an input feature in the training of the nonlinear regression model. Hence, training spectra are recorded while systematically varying an applied force. The device for recording a CW THz spectrum may include a force sensor for determining the applied while the spectrum is being recorded. Such determined force, or a value indicative thereof, may be provided to a nonlinear regression model together with the spectrum.

The method and system for quantitative detection of glucose involves analysing a recorded CW THz spectroscopy spectrum to estimate a glucose level in the liquid solution of the sample. For this purpose, the system comprises a data analysis unit comprising memory circuitry, processor circuitry, and an interface such as a wireless interface. The data analysis unit is configured to analyse the recorded CW THz spectroscopy spectrum using the supervised machine learning algorithm described above and estimate a glucose level in the liquid solution of the sample. The operations of the data analysis unit may be considered method steps that the data analysis unit is configured to carry out. In exemplary embodiments, these operations may comprise:

- Accessing the recorded sample and reference CW THz spectroscopy spectra

- Calibration of the CW THz spectroscopy spectra

- Averaging of several spectra to reduce noise.

Fourier transformation of sample and reference signals.

Extraction and preprocessing of features from spectra.

Nonlinear regression model inference.

Postprocessing of the regression model output to determine final glucose concentration estimate.

In exemplary embodiments, the method and system for quantitative detection of glucose involves initiating an estimation of a glucose level in a sample and providing the estimated glucose level. For this purpose, the system comprises a glucose level monitoring unit comprising memory circuitry, processor circuitry, and an interface such as a wireless interface. The glucose level monitoring unit initiate a quantitative detection of glucose in a sample and to provide an estimated glucose level in the sample. The operations of the glucose level monitoring unit may be considered method steps that the glucose level monitoring unit is configured to carry out. In exemplary embodiments, these operations may comprise: instructing the control unit to record a CW THz spectroscopy spectrum; instructing the data analysis unit to estimate a glucose level based on the recorded THz CCS spectrum; and retrieving the estimated glucose level from the data analysis unit.

Various examples and details are described hereinafter, with reference to the figures when relevant. It should be noted that the figures may or may not be drawn to scale and that elements of similar structures or functions are represented by like reference numerals throughout the figures. It should also be noted that the figures are only intended to facilitate the description of the examples. They are not intended as an exhaustive description of the disclosure or as a limitation on the scope of the disclosure. In addition, an illustrated example needs not have all the aspects or advantages shown. An aspect or an advantage described in conjunction with a particular example is not necessarily limited to that example and can be practiced in any other examples even if not so illustrated, or if not so explicitly described.

The figures are schematic and simplified for clarity, and they merely show details which aid understanding the disclosure, while other details have been left out. Throughout, the same reference numerals are used for identical or corresponding parts.

The invention relates to a method for quantitative detection of glucose in a sample using CW THz spectroscopy in an ATR setup. The method may be performed by the system according to this disclosure. In an exemplary embodiment illustrated in the flowchart of Fig. 3, the method being performed using a CW THz spectroscopy setup with an ATR setup as described in this disclosure and comprises:

- adjusting (S104) a phase or a frequency domain parameter of the optical signal, such as adjusting a frequency content in the optical signal or a difference in optical path length between the first and second optical paths; recording (S106) a CW THz spectroscopy spectrum of the electric detection signal as a function of the adjusted parameter while an external face of the sample surface of the ATR element is in contact with a sample comprising a liquid solution of glucose; and

- analysing (S110) the recorded CW THz spectroscopy spectrum using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar liquid solutions with known glucose levels to estimate a glucose level in the liquid solution of the sample.

In exemplary embodiments, the method may comprise one or more of the following steps:

Instructing a user to prepare the sample and bring it into contact with the external face of the sample surface. For in-vivo detection, this may comprise instructing a user to bring an external face of the sample surface of the ATR element in contact with an epidermis segment. Instructing (S102) a control unit to record the CW THz spectroscopy spectrum. This may comprise establishing a connection between a glucose level monitoring unit giving the instruction and a control unit recording the spectrum.

Instructing a data analysis unit to estimate a glucose level based on the recorded CW THz spectroscopy spectrum. This may comprise establishing a connection between a glucose monitoring unit and a data analysis unit estimating the glucose level.

- Accessing (S108) a recorded CW THz spectroscopy spectrum. This may comprise establishing a connection between a control unit recording the spectrum and a data analysis unit estimating the glucose level.

Retrieving (S112) an estimated glucose level.

Displaying the estimated glucose level on a display.

- Generating a signal indicative of the estimated glucose level.

Providing an estimated glucose level in a sample at predetermined times, such as at regular intervals. This preferably comprises performing at least steps S104, S106 and S110.

Fig. 4 is a diagram illustrating an exemplary device 1 for recording a Terahertz cross-correlation spectrum according to an exemplary embodiment. The device 1 comprises a light source 2 for outputting a CW optical signal and a THz transmitter 4 optically coupled to the light source 2 via a first optical path 11 , the THz transmitter configured to emit THz radiation 5 when modulated by the CW light signal. The device 1 further comprises an ATR element 7 positioned for THz radiation 5 from the THz transmitter 4 to reflect off an internal face of a sample surface 8 of the ATR element. The device 1 further comprises a THz receiver 6 optically coupled to the light source 2 via a second optical path 12, the THz receiver being positioned to receive THz radiation 5 reflected off the internal face of the sample surface 8 and configured to detect received THz radiation 5 by generating an electric detection signal that is a result of an interference between the THz radiation 5 and the CW light signal. The device 1 further comprises an optical delay setup 3 configured to adjust a difference in optical path length between the first and second optical paths 11 and 12.

Optical paths 11 and 12 for the CW signal from the optical source to the optical delay component, the THz transmitter, and the THz receiver are preferably provided by optical waveguides, such as optical fibre, channel waveguides and planar waveguides. In an exemplary embodiment, the optical path is a full optical waveguide path meaning that no part of the optical path of the CW optical signal is situated outside of an optical waveguide. This has the effect that the CW optical signal at no point propagates in free space, i.e. in the atmosphere of the environment where the device 1 is situated.

The optical delay setup 3 may have one or more optical delay components in both the arm to the THz transmitter 4 (first optical path 11) and in the arm to the THz receiver 6 (second optical path 12), or only in only one of the arms (as in e.g. Fig. 4). Exemplary embodiments of the one or more optical delay components of the optical delay setup 3 will be described later.

The sample 9 to be measured upon is brought into contact with the sample surface 8. The sample 9 comprises a liquid but need not be a liquid as illustrated in Fig. 4 ATR uses a property of total internal reflection resulting in an evanescent wave 10. An evanescent wave is an oscillating electric and/or magnetic field that does not propagate as an electromagnetic wave but whose energy is spatially concentrated in the vicinity of the source (oscillating charges). When a beam of light passes through the ATR crystal, it is reflected at least once off the internal surface of the crystal. An evanescent wave is formed in the sample by this reflection. Variations in the incidence angle affect the amount of reflection. A detector then collects the beam at the exit of the crystal. When the incidence angle exceeds the “critical” angle, internal reflection occurs. In this case, the angle of refraction is determined by the real parts of the refractive indices of the ATR crystal and the sample. The evanescent effect only occurs when the crystal is made from an optical material with a higher refractive index than the sample being examined. Otherwise, light is lost in the sample.

In the THz region, the penetration depth of the evanescent waves ranges from several micrometres to hundreds of micrometres. Typical materials for ATR crystals include germanium, KRS-5 and zinc selenide, while silicon is ideal for use in the THz region of the electromagnetic spectrum. The excellent mechanical properties of diamond make it an ideal material for ATR.

Satisfaction of the boundary conditions of electromagnetic waves at the interface between the sample surface of the ATR prism and the sample itself require that part of the electromagnetic field “leaks” into the sample, known as evanescent wave. This evanescent wave strongly interacts with the sample and is used to probe its properties. The amplitude of this evanescent wave decays exponentially with the distance from the boundary and the exponential decay constant can be tuned by adjusting parameters such as the refractive index of the ATR prism, the incidence angle, etc. By adjusting this, one can obtain the penetration depth that best fits the sample under analysis. Contact gel might or might not be used to trim the penetration depth. ATR measurements benefit from a strong interaction between radiation and sample but without decreasing the signal strength when the sample is very absorptive, such as when the sample contains water. Therefore, ATR is preferrable to transmission. If even stronger sample interaction is desired, multiple-pass ATR prisms can be used.

For in-vivo detection of glucose in a subject, the sample is an epidermis segment of the subject and the liquid solution of glucose is interstitial fluid in the epidermis segment. In order to ensure proper contact between the external face of the sample surface 8 and the epidermis segment, and in order to ensure a sufficient penetration depth of the evanescent field 10 into the epidermis, it is preferable that the sample surface and the epidermis segment are held closely together. In an exemplary embodiment, a pressure between the sample surface and the epidermis segment during recording of a spectrum is between 5 and 50 kN/m². In an exemplary embodiment, the sample surface and the epidermis segment are pressed together with a force between 1 and 10 N, such as between 1 and 5 N, such as between 2 and 4 N during the recording of the spectrum,

In an exemplary embodiment, the device for recording a CW THz spectrum comprises a pressure gauge for determining a pressure or a force between the sample surface and the epidermis segment. In an exemplary embodiment, the control system 13 further comprises output means for generating an output indicative of the determined pressure or force and for presenting the generated output to a user. The presented output may be a value indicative of the presently applied force or an indication of whether the presently applied force lies within a predetermined range around an empirically determined optimal force for the given device. Such optimal force typically depends on the size of a contact area of the external face of the sample surface and the epidermis segment.

In an exemplary embodiment, the control unit is configured to automatically initiate the recording of a spectrum when the determined force lies within or exceeds a predetermined range, such as a range around an empirically determined optimal force for the given device. This may be combined with an indication to the user to maintain the applied force for a length of time corresponding to the recording of a spectrum.

The components above describe the THz CCS setup utilized by exemplary embodiments of the method, device, and system according to the invention. The device 1 then further comprises a control unit 13 for recording a THz CCS spectrum using the THz CCS setup. The control unit 13 is connected to at least the optical delay setup 3 and the THz receiver 6. The control unit 13 may also be connected to the light source 2.

Fig. 5 is a diagram illustrating an exemplary device 50 for recording a FDS spectrum according to an exemplary embodiment. The device 50 comprises a light source 52 for outputting a CW light signal and a THz transmitter 4 optically coupled to the light source 52 via a first optical path 11 , the THz transmitter configured to emit THz radiation 5 when modulated by the CW light signal. The device 50 further comprises an ATR element 7 positioned for THz radiation 5 from the THz transmitter 4 to reflect off an internal face of a sample surface 8 of the ATR element. The device 50 further comprises a THz receiver 6 optically coupled to the light source 52 via a second optical path 12, the THz receiver being positioned to receive THz radiation 5 reflected off the internal face of the sample surface 8 and configured to detect received THz radiation 5 by generating an electric detection signal that is a result of an interference between the THz radiation 5 and the CW light signal.

Optical paths 11 and 12 for the CW signal from the optical source to the optical delay component, the THz transmitter, and the THz receiver are preferably provided by optical waveguides, such as optical fibre, channel waveguides and planar waveguides. In an exemplary embodiment, the optical path is a full optical waveguide path meaning that no part of the optical path of the CW optical signal is situated outside of an optical waveguide. This has the effect that the CW optical signal at no point propagates in free space, i.e. in the atmosphere of the environment where the device 50 is situated.

The light source 52 comprises at least two narrowband light sources 53 and 54, at least one of which is a frequency sweeping laser 54. The control unit 13 controls the temperature of lasers 53 and 54 and sweeps the wavelength of at least one of them as it records the signal from THz receiver 6.

The sample 9 to be measured upon is brought into contact with the sample surface 8. The sample 9 comprises a liquid but need not be a liquid as illustrated in Fig. 5. Details relating to the ATR setup are described above in relation to Figure 4. The components above describe the THz FDS setup utilized by exemplary embodiments of the method, device, and system according to the invention. The device 50 then further comprises a control unit 13 for recording a THz FDS spectrum using the THz FDS setup. The control unit 13 is connected to at least the light source 2 and the THz receiver 6.

Fig. 6 is a block diagram illustrating an example control unit 13 according to this disclosure. The control unit 13 comprises memory circuitry 201 , processor circuitry 202, and a communication interface 203. The control unit 13 may further comprise a power source 204 such as battery. The control unit 13 is configured to record a CW THz spectroscopy spectrum by performing one or more of the method steps disclosed in relation to Fig. 3, such as:

- control (S104) a phase or a frequency domain parameter of the optical signal, such as control the optical delay setup 3 to adjust the difference in optical path length between the first and second optical paths, 11 and 12 or control the frequency swept laser 53 and/or 54 to adjust a frequency content in the optical signal. record (S106) a CW THz spectroscopy spectrum by recording the electric detection signal from the THz receiver 6 as a function of the adjusted parameter.

The operations of the control unit 13 may be embodied in the form of executable logic routines (e.g., lines of code, software programs, etc.) that are stored on a non-transitory computer readable medium (e.g., the memory circuitry 201 ) and are executed by the processor circuitry 202).

Memory circuitry 201 may be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, a random-access memory (RAM), or another suitable device. In a typical arrangement, memory circuitry 201 may include a non-volatile memory for long term data storage and a volatile memory that functions as system memory for processor circuitry 202. Memory circuitry 201 may exchange data with processor circuitry 202 over a data bus. Control lines and an address bus between memory circuitry 201 and processor circuitry 202 also may be present (not shown). Memory circuitry 201 is considered a non-transitory computer readable medium. The memory circuitry 201 stores instructions that when executed by the processor circuitry 202 causes the device 1 to perform the operations.

The communication interface 203 can establish a data connection with one or more other units for sending and receiving data and control signals. The communication interface 203 is preferably a wireless interface capable of establishing a wireless connection such as via Bluetooth, WiFi, 3GPP wireless communication, etc.

Application-specific embodiments of CW THs spectroscopy devices

A continuous blood glucose monitor (CGM) is disclosed herein. The continuous blood glucose monitor comprises a device for recording a CW THz spectroscopy spectrum from interstitial fluid in an epidermis segment of a subject, and a data analysis unit using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar subjects with known glucose levels. There are two different embodiments of the CGM hardware:

For vivo measurements in a clinical setting, an exemplary embodiment of a CW THs spectroscopy device may be implemented as a countertop device version that utilizes an ASE light source, fibre stretchers for optical delay and THz antennae activated by 1550 nm light. The data analysis unit may be connected directly to the device or be configured to access the recorded spectrum via a network connection.

For in vivo measurements by a diabetic user, an exemplary embodiment of a CW THs spectroscopy device may be implemented as a wearable-sized version that utilized a SLED light source, solid-state delay lines for delay and THz antennae activated by either 1550 nm or 1060 nm light, or even shorter wavelengths. Here, the data analysis unit is preferably a cloud-based solution accessing or receiving the recorded spectrum and returning a determined blood glucose level.

Fig. 7 is a diagram illustrating an exemplary THz CCS setup 50 according to the disclosure, with reference numbers indicating the components already described in relation to Fig. 4. In this exemplary THz CCS setup 50, the optical waveguide, and thus the first and second optical paths 11 and 12, is provided by a channel waveguide or a planar waveguide formed on or integrated in a substrate 14 of a semiconductor device. In this exemplary embodiment, the components of the CCS setup may be integrated on-chip or on a photonic circuit whereby the setup can be made more compact. At least two different options may be considered:

- Surface mounted components on a chip, put together as a lumped system. This approach is, in simple terms, to take already available components, take them out of their respective housings and mount the active parts directly on a chip surface.

Monolithic chip integration. Here, all optical circuitry components will be built directly on a chip using cleanroom processing. This option is likely to be implementable as a consumer-aimed, wearable product.

The control unit and potentially the data analysis unit may be integrated on the same or an adjacent chip to provide a wearable patch for quantitative detection of glucose levels. A further advantage of these embodiments is that fabrication of the CCS setup or the device or the system can be scaled-up by using already existing methods in semiconductor processing. Besides the components already described in relation to Fig. 4, the CCS setup 50 may comprise electronic circuitry 15 for delivering power and control signals and conveying data signals, such as one or more electronic processor implementing the control unit 13 described in relation to Fig. 5. The electronic circuitry 15 may comprise the connections to at least parts 2, 3, and 6 illustrated by lines. The setup may also comprise an interface 16, such as connector to a control unit 15 formed on a different platform.

In general, power consumption is a concern for on-chip systems. The optical-to-THz conversion efficiency is very low (typically 10^-4 in power), so the light source has to have a relatively large current injection compared to other on-chip systems found in wearables. This can be solved either by creating a dedicated wearable device where the entire power consumption can be taken up by our chip, or by increasing the optical-to-THz conversion efficiency in the THz antennae. The latter would be done by more careful material engineering in the cleanroom, and this approach has the largest potential to solve the power consumption issue.

The light source for CCS is preferably a single broadband CW source.

The optical delay can be obtained in various ways, some embodiments of which will be described later herein.

Similarly, an exemplary embodiment of the THz FDS setup 50 according to the disclosure and described in relation to Fig. 5 may have the components of the FDS setup integrated on-chip or on a photonic circuit whereby the setup can be made more compact.

Fig. 8 is a diagram illustrating an exemplary system 60 for quantitative detection of glucose in a sample according to the disclosure. The system 60 comprises a device 1 for recording a CW THz spectroscopy spectrum as described in relation to Figures 4, 5, and 7, the device comprising the control unit 13. The system 60 further comprises a data analysis unit 17 comprising an electronic processor capable of establishing a connection to the control unit 13 and being configured to analyse a recorded CW THz spectroscopy spectrum using a supervised machine learning algorithm. In an exemplary embodiment, the data analysis is a cloud-based solution, and the data analysis unit 17 may be a server accessible via e.g. an internet connection.

The block diagram of Fig. 6 also illustrates an exemplary embodiment of the data analysis unit 17. The data analysis unit 17 comprises memory circuitry 301 , processor circuitry 302, and a communication interface 303. The data analysis unit 17 is configured to perform one or more of the method steps disclosed in relation to Fig. 3, such as:

- accessing (S108) a recorded CW THz spectroscopy spectrum;

The operations of the data analysis unit 17 may be embodied in the form of executable logic routines (e.g., lines of code, software programs, etc.) that are stored on a non-transitory computer readable medium (e.g., the memory circuitry 301) and are executed by the processor circuitry 302).

Memory circuitry 301 may be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, a random-access memory (RAM), or another suitable device. In a typical arrangement, memory circuitry 301 may include a non-volatile memory for long term data storage and a volatile memory that functions as system memory for processor circuitry 302. Memory circuitry 301 may exchange data with processor circuitry 302 over a data bus. Control lines and an address bus between memory circuitry 301 and processor circuitry 302 also may be present (not shown). Memory circuitry 301 is considered a non-transitory computer readable medium. The memory circuitry 301 stores instructions that when executed by the processor circuitry 302 causes the system 60 to perform the operations.

The communication interface 303 can establish a data connection with one or more other units for sending and receiving data and control signals. The communication interface 303 is preferably a wireless interface capable of establishing a wireless connection such as via Bluetooth, WiFi, 3GPP wireless communication, etc.

In an exemplary embodiment, the system 60 in Fig. 8 also comprises a glucose level monitoring unit 18 capable of establishing a connection to the control unit 13 and the data analysis unit 17, and comprising an electronic processor configured to provide an estimated glucose level in a sample. In an exemplary embodiment, the glucose level monitoring unit 18 is comprised in a user equipment 19, such as a smartphone or a dedicated remote control for the device 1.

The block diagram of Fig. 6 also illustrates an exemplary embodiment of the glucose level monitoring unit 18. The glucose level monitoring unit 18 comprises memory circuitry 401 , processor circuitry 402, and a communication interface 403. The glucose level monitoring unit 18 is configured to perform one or more of the method steps disclosed in relation to Fig. 3, such as: instructing (S102) the control unit 13 to record a CW THz spectroscopy spectrum; instructing the data analysis unit 17 to estimate a glucose level based on the recorded CW THz spectroscopy spectrum; and retrieving (S112) the estimated glucose level from the data analysis unit 17.

In exemplary embodiments, operations of the glucose level monitoring unit 18 may comprise one or more of:

Instructing a user to prepare the sample. For in-vivo detection, this may comprise instructing a user to bring an external face of the sample surface of the ATR element in contact with an epidermis segment, potentially including selecting and preparing the segment and applying a given force when contacting the segment with the sample source.

Providing an estimated glucose level in a sample at predetermined times, such as at regular intervals.

Displaying the estimated glucose level on a display.

- Generating a signal indicative of the estimated glucose level.

The operations of the glucose level monitoring unit 18 may be embodied in the form of executable logic routines (e.g., lines of code, software programs, etc.) that are stored on a non-transitory computer readable medium (e.g., the memory circuitry 401) and are executed by the processor circuitry 402).

Memory circuitry 401 may be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, a random-access memory (RAM), or another suitable device. In a typical arrangement, memory circuitry 401 may include a non-volatile memory for long term data storage and a volatile memory that functions as system memory for processor circuitry 402. Memory circuitry 401 may exchange data with processor circuitry 402 over a data bus. Control lines and an address bus between memory circuitry 401 and processor circuitry 302 also may be present (not shown). Memory circuitry 401 is considered a non-transitory computer readable medium. The memory circuitry 401 stores instructions that when executed by the processor circuitry 402 causes the system 60 to perform the operations.

The communication interface 403 can establish a data connection with one or more other units for sending and receiving data and control signals. The communication interface 403 is preferably a wireless interface capable of establishing a wireless connection such as via Bluetooth, WiFi, 3GPP wireless communication, etc.

In an exemplary embodiment of the system 60, both the control unit 13 and the data analysis unit 17 are part of the device 1 , such as formed on the same platform. In an exemplary embodiment, the control unit 13 and the data analysis unit 17 are implemented using the same memory circuitry 201 , 301 , processor circuitry 202, 302, and a communication interface 203, 303. In an exemplary embodiment of the system 60, the glucose level monitoring unit 18 is also implemented using the same memory circuitry 201 , 301 , 401 , processor circuitry 202, 302, 402, and a communication interface 203, 303, 403.

In an exemplary embodiment, the optical delay setup 3 described in relation to the CCS device of Figure 4 comprises:

• a double-pass polarization-conserving fibre stretcher comprising a circulator, a fibre stretcher, and a Faraday mirror arranged for the circulator and the fibre stretcher to receive the CW optical signal propagating in a first direction and in a second, opposite direction, the change in direction being due to a reflection in the Faraday mirror; and/or

• a variable solid state optical delay (SSOD) comprising two or more optical ports, one or more optical waveguide sections for connecting two optical ports, and actuation means for varying an optical path length between the two connected optical ports.

In an exemplary embodiment, the optical delay setup comprises a double-pass polarizationconserving fibre stretcher in each of the first and second arms. These double-pass polarizationconserving fibre stretchers are preferably identical or identical except for a small difference in the optical path length of the fibre stretcher, this difference being of the order of the desired scanning range. This setup is advantageous since it provides identical or almost identical optical paths - with respect to both the optical path-length and any distorting effects - for the CW optical signal in the first and second arms. This again ensures that the CW optical signals applied to the THz antennae are in phase over the stroke of the optical delay setup.

The double-pass polarization-conserving fibre stretcher is an optical delay component comprising an optical circulator, a fibre stretcher, and a Faraday mirror arranged for the optical circulator and the fibre stretcher to receive the CW optical signal propagating in a first direction and in a second, opposite direction, the change in direction being due to a reflection in the Faraday mirror.

The optical circulator is an optical device with three or more ports designed such that light entering any port exits from the next. This means that if light enters port 1 it is emitted from port 2, but if some of the emitted light is reflected back to the circulator, it does not come out of port 1 but instead exits from port 3. Optical circulators are typically used to separate optical signals that travel in opposite directions, for example to achieve bi-directional transmission over a single fibre.

The double-pass polarization-conserving fibre stretcher also comprises a Faraday mirror which is a combination of a 45 degrees Faraday rotator and a mirror. Since the Faraday rotator rotates the polarization of the light in the same direction with respect to the direction of propagation on both passes, the optical signal reflected by the Faraday mirror will return with the polarization rotated by 90 degrees.

In the double-pass polarization-conserving fibre stretcher, the delay is produced by physically stretching an optical fibre of the fibre stretcher to extend the optical path. In an exemplary embodiment, the fibre stretcher is a section of optical fibre, e.g. 50-100 meters, tightly wound around a piezo crystal or another electrostrictive material that can be strained by applying an electric voltage. The longer the fibre and the more it is stretched, the larger is the resulting optical delay. Numerous fibre stretchers are commercially available that may be used in this setup.

One end of the fibre stretcher is optically coupled to the optical source via a waveguide and the optical circulator, and the other end of the fibre stretcher is equipped with a Faraday mirror. The optical signal will be received by the circulator, acquire a delay in the fibre stretcher, be reflected by the faraday mirror and then pass through the fibre stretcher once more to acquire an additional delay. Arriving at the optical circulator from the fibre stretcher, the optical signal will be coupled to a different optical waveguide propagating towards the THz transmitter or receiver. The Faraday mirror will rotate the polarization of the optical signal 90 degrees so that any polarization changes caused by the stretching of the optical fibre in the fibre stretcher during the first pass is reciprocated during the second pass. The 90 degrees rotation is advantageous as the THz transmitter and receiver are sensitive to the polarization of the optical signal. The double-pass polarization-conserving fibre stretcher is also advantageous because the optical signal passes the fibre stretcher twice in opposite directions - thus ‘double-pass’. As explained above, this has the effect of reciprocating the birefringence existing in the fibre which is the effect leading to the polarisation rotation. Further, it has the effect that the additional optical pathlength resulting from stretching the fibre, and thereby also the optical delay, is doubled, which means that the component can be made smaller. In addition, fibre stretchers are advantageous since they provide a continuous adjustment of the optical delay and can therefore be used to provide time-steps at any desired length.

The variable solid state optical delay (SSOD) is an optical delay component comprising two or more optical ports, such as ports for coupling an input signal and an output signal to an optical waveguide, one or more optical waveguide sections for connecting two optical ports, and actuation means for varying an optical path length between the two connected optical ports. Such solid-state optical delay is also referred to as Non-Mechanical Variable Optical Time Delay Line or solid-state delay line (SSDL),

In an exemplary embodiment, the variable solid state optical delay comprises two or more optical waveguide sections of different lengths for connecting two optical ports, and wherein the actuation means are configured to select one or a combination of the two or more optical waveguide sections to connect the two or more optical ports. Hence, in this embodiment, the optical path length between the two ports is varied by the selection of different routes of different length. In this embodiment, the two or more different optical waveguide sections in the variable solid state optical delay may be fibre loops or channel/planar waveguides sections of varying lengths formed on a substrate. The actuation means can select one of the optical waveguide sections or connect two or more of the optical waveguide sections in series to provide a set of selectable discrete optical delays. Some SSODs may, in addition, provide some continuous adjustment of the optical pathlength around or between one or more discrete optical delays. The actuation means may for example be an opto-mechanical fibre switch or a MEMS (Micro Electro Mechanical System) switch which are advantageous since they do not involve macroscopic moving parts. In an exemplary embodiment, the variable solid state optical delay is a non-mechanical optical delay, where the actuation means may comprise one or more of thermo-optic switch, electro-optic switch, acousto-optic switch, magneto-optic switch. In addition to the increased robustness and lack of free-space propagation, a variable solid state optical delay is advantageous since it is stripped of any moving parts, mirrors, lenses, and can be minoritized to fit on a single chip. In an exemplary embodiment, the SSOD comprises durable and high reliability optical switches capable of at least 10⁹ switching cycles.

In another exemplary embodiment of the variable solid state optical delay, the optical path length between the two ports is varied by modulating a material property (preferably excluding stretching) of an optical waveguide section connecting the ports. In one embodiment, this could be a modulation of a refractive index or a birefringence of the waveguide material. One example of such variable solid state optical delay could be an acousto-optic delay module where the actuation means transmits an acoustic signal through a birefringent crystal, leading to a change of grating positions and thereby a change in the diffraction of the optical signal and ultimately a different optical pathlength.

In an exemplary embodiment of the variable solid state optical delay, the SSOD comprises temperature stabilisation to stabilise the temperature of the SSOD. This is advantageous since changes in temperature will change the refractive index of the waveguide media, resulting in a different path length, which will affect the interference pattern. In another exemplary embodiment, the SSOD comprises low-loss waveguide media, such that propagation through the different paths in the SSOD does not change the optical power at the output substantially.

In an exemplary embodiment, the optical delay setup of the CCS setup comprises at least a first optical delay component in the first arm and a second optical delay components in the second arm, i.e. each arm comprises at least one optical delay component. Preferably, each of the first and second delay components comprises:

• a double-pass polarization-conserving fibre stretcher; or

• a variable solid state optical delay.

In one example, the optical delay setup of the CCS setup comprises both a double-pass polarizationconserving fibre stretcher and a variable solid state optical delay. These may be provided in parallel (in different arms) or in series (in the same arm). The variable solid state optical delay preferably provides a set of discrete optical delays that can be selected by the actuation means and where a largest difference between two following optical delays is D. In one example, the double-pass polarization-conserving fibre stretcher is adapted to provide a continuous optical delay adjustment equal to or larger than D. This combination is advantageous since it allows for a large range of optical delays to be scanned continuously.

CCS Results from In-vitro study

Different solutions of glucose in deionized water were prepared within the concentration range 0.5-5.0 mg/mL (2.8-27.8 mmol/L). Reference measurements of air and pure water are recorded right before starting the measurement series. A droplet of circa 200 pL for each solution is placed on top of the ATR surface. For each droplet, circa 200 consecutive time traces are acquired with integration time of 6 seconds. The recorded time traces are aligned on the time (x-axis) and the corresponding Fast Fourier Transformations (FFT) are used to perform the analyses. The collected measurements are randomly divided in train and test datasets. 80% of the datapoints are used as training set for the model. The remaining 20% of datapoints represents the test dataset. The gradient boosting model is trained on the training data and the test set is used to evaluate the performance of the machine learning model. However, statistical results often do not express the clinical relevance of the measurements as diabetes places individuals at different levels of risk depending on the level and duration of glucose values. Low levels for any length of time are considered acutely dangerous, while high levels have chronic impact over days or years. Different levels of hazard are assigned to errors of different kinds. A common way of expressing this is the use of the Clarke Error Grid as presented in Figs. 9A and 9B. It has been widely adopted for use in the evaluation of blood glucose monitoring systems.

The error grid divides the possible errors into five error groups of varying severity based on the difference between the reference and the estimated blood glucose levels. The different error groups are:

Region A: predicted values fall within 20% of the reference values.

Region B: predicted values deviate more than 20% from the reference but they would not lead to inappropriate treatment. Region C: predicted values leading to unnecessary treatment.

Region D: predicted values indicate a potentially dangerous failure to detect hypoglycemia or hyperglycemia.

Region E: predicted values would confuse treatment of hypoglycemia for hyperglycemia and vice versa.

This shows that recording a CW THz spectroscopy spectrum, in accordance with this disclosure, of an in vitro sample comprising a liquid solution with an unknown glucose concentration, allows for a glucose concentration in the liquid solution to be estimated using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples with known glucose concentrations.

CCS Results from Ex-vivo study

The following describes Ex vivo experiments on porcine skin samples. Porcine skin has been found to be a good model for human skin from a spectroscopy point of view, in particular for transdermal diffusion dynamics studies. The glucose dependent contrast in the recorded spectra comes from a change in dielectric properties, which is characterised for the specific skin sample. Therefore, different reference spectra are recorded prior to recording a spectrum_on the treated skin sample. Figure 10 shows a series of recordings of the electric detection signal as a function of a difference in optical path length between the first and second optical paths for different references and for a sample treated with a 1 ,00 mg/mL glucose concentration.

Fresh porcine skin is cut into samples of 2x2 cm² and the subcutaneous fat is removed. To delay trans epidermal water loss effects, the samples are kept in cling film until the measurement takes place. First, a baseline measurement is recorded when no sample is placed on the ATR surface, this is reference 101 in Figure 10. Then, a reference measurement of the bare skin sample is recorded, this is reference 102 in Figure 10. The skin sample is placed with the stratum corneum facing the surface of the prism. Micro-holes are pierced into the skin sample from the dermis side. Micro holes are pierced into the samples from the hypodermis side, to create small pathways to aid the diffusion of the solution past the hypodermis and into the dermis and epidermis layers of the skin. A cotton pad soaked with either pure water or a glucose solution is applied to the exposed hypodermis and the skin is left to absorb the selected solution for 30 minutes. A baseline measurement is recorded with no sample on top of the ATR surface to monitor the signal stability, this is reference 103 in Figure 10. Both air reference signals 101 and 104 overlap with each other for the entire time window, as shown in Figure 10.

The measurement on the treated skin sample is recorded, this is reference 104 in Figure 10. The skin sample is placed with the stratum corneum facing the surface of the prism. To ensure good contact between the sample and the surface of the prism and the reproducibility of the measurements, the same constant and uniform contact pressure is applied to all skin samples while performing the measurements. This is because the contact pressure greatly affects the THz response of skin due to skin hydration. After every measurement on skin samples, the surface of the prism is cleaned by ethanol. The time traces collected are aligned with respect to the position of the maximum peaks and normalized by their reference measurements.

To reduce variance induced by using different porcine samples, feature sets for each concentration have been generated by averaging spectra from different porcine samples. Additionally, to artificially increase the train and test set sizes, bootstrap sampling is used to generate the aforementioned averages. The Uniform Manifold Approximation and Projection (UMAP) algorithm, a non-linear dimensionality reduction technique, which seeks to learn the inherent structure of the data and find a low dimensional embedding that preserves the notion of similarity between samples, has been used to visualize a 2-dimensional representation of the samples as shown in Figure 11 . While the specific values of the new features “Spectrum feature 1” and “Spectrum feature 2” are not directly interpretable, the formation of distinct clusters show that samples treated with different solutions have distinct, measurable signals.

This shows that recording a CW THz spectroscopy spectrum, in accordance with this disclosure, of an ex-vivo sample comprising a liquid solution with an unknown glucose concentration, allows for a glucose concentration in the liquid solution to be estimated using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples with known glucose concentrations.

Claims

1. A method for quantitative detection of glucose in a sample using continuous wave (CW) Terahertz (THz) spectroscopy in an attenuated total reflection (ATR) setup, the method being performed using a THz CCS setup comprising: a light source for outputting a continuous-wave (CW) light signal; a THz transmitter optically coupled to the light source via a first optical path, the THz transmitter configured to emit THz radiation when modulated by the CW light signal; an ATR element positioned for THz radiation from the THz transmitter to reflect off an internal face of a sample surface of the ATR element; a THz receiver optically coupled to the light source via a second optical path, the THz receiver being positioned to receive THz radiation reflected off the internal face of the sample surface and configured to detect received THz radiation by generating an electric detection signal that is a result of an interference between the THz radiation and the CW light signal; means for adjusting a phase or a frequency domain parameter of the CW light signal; the method comprising: recording a CW THz spectroscopy spectrum of the electric detection signal as a function of an adjustment of a phase or a frequency domain parameter of the CW light signal while an external face of the sample surface is in contact with a sample comprising a liquid solution of glucose;

- analysing the recorded CW THz spectroscopy spectrum using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples with liquid solutions having known glucose levels to estimate a glucose level in the liquid solution of the sample.

2. The method according to claim 1 , wherein the CW THz spectroscopy is THz cross-correlation spectroscopy (CCS), wherein the CW light source has a continuous broadband spectrum; wherein the means for adjusting a phase or a frequency domain parameter of the CW light signal comprises an optical delay setup configured to adjust a difference in optical path length between the first and second optical paths; wherein recording a CW THz spectroscopy spectrum of the electric detection signal as a function of an adjustment of a phase or a frequency domain parameter of the CW light signal comprises recording a CCS spectrum of the electric detection signal as a function of a difference in optical path length between the first and second optical paths; and wherein the recorded CW THz spectroscopy spectra are recorded CCS spectra.

3. The method according to claim 1 , wherein the CW THz spectroscopy is THz frequency domain spectroscopy (FDS), wherein the CW light source comprises at least two narrowband light sources of which at least one is a frequency sweeping laser; wherein the means for adjusting a phase or a frequency domain parameter of the CW light signal comprises controlling the frequency sweeping of the at least one frequency sweeping laser; wherein recording a CW THz spectroscopy spectrum of the electric detection signal as a function of an adjustment of a phase or a frequency domain parameter of the CW light signal comprises recording a FDS spectrum of the electric detection signal as a function of the frequency sweep of the frequency swept laser; and wherein the recorded CW THz spectroscopy spectra are recorded FDS spectra.

4. The method according to any of the preceding claims, wherein the liquid solution of glucose comprises or is part of an active metabolic system.

5. The method according to any of the preceding claims, wherein the liquid solution of glucose is a product of a fermentation process.

6. The method according to any of claims 1-4, wherein the method is a method for quantitative, non- invasive, in-vivo detection of glucose in a subject, wherein the sample is an epidermis segment of the subject and the liquid solution of glucose is interstitial fluid in the epidermis segment, and wherein the THz CCS spectrum is recorded while an external face of the sample surface is in contact with the epidermis segment.

7. The method according to claim 6, wherein the external face of the sample surface and the epidermis segment are pressed together with a force between 1 - 10 N during recording a CW THz spectroscopy spectrum

8. The method according to claim 6, wherein a pressure between the external face of the sample surface and the epidermis segment during recording a CW THz spectroscopy spectrum is between 5 - 50 kN/m².

9. A device (1 , 50) for recording a continuous wave (CW) Terahertz (THz) spectrum from a sample comprising a liquid, comprising: a THz CCS setup comprising: a light source (2, 52) for outputting a continuous-wave (CW) light signal; a THz transmitter (4) optically coupled to the light source via a first optical path (11), the THz transmitter configured to emit THz radiation (5) when modulated by the CW light signal; an attenuated total reflection (ATR) element (7) positioned for THz radiation from the THz transmitter to reflect off an internal face of a sample surface (8) of the ATR element; a THz receiver (6) optically coupled to the light source via a second optical path (12), the THz receiver being positioned to receive THz radiation (5) reflected off the internal face of the sample surface and configured to detect received THz radiation by generating an electric detection signal that is a result of an interference between the THz radiation and the CW light signal; means for adjusting a phase or a frequency domain parameter of the CW light signal; a control unit (13) comprising an electronic processor configured to

- control the means for adjusting to adjust a phase or a frequency domain parameter of the CW light signal; record a THz CCS spectrum by recording the electric detection signal from the THz receiver as a function of the adjustment of a phase or a frequency domain parameter of the CW light signal.

10. The device according to claim 9, wherein at least the light source, the THz transmitter, the THz receiver, the means for adjusting a phase or a frequency domain parameter are monolithically integrated on a chip.

11. The device according to any of claims 9 - 10, further comprising a pressure or force sensor for detecting a pressure or force applied to the external face of the sample surface.

12. The device according to any of claims 9 - 11 , wherein the device is for recording a THz crosscorrelation spectroscopy (CCS) spectrum; wherein the CW light source has a continuous broadband spectrum; wherein the means for adjusting a phase or a frequency domain parameter of the CW light signal comprises an optical delay setup configured to adjust a difference in optical path length between the first and second optical paths; and wherein the electronic processor of the control unit is configured to

- control the optical delay setup to adjust the difference in optical path length between the first and second optical paths; and record a THz CCS spectrum by recording the electric detection signal from the THz receiver as a function of a difference in optical path length between the first and second optical paths.

13. The device according to any of claims 9 - 12, wherein the light source is a superluminescent light emitting diode (SLED) and t.

14. A system for quantitative detection of glucose in a sample, comprising the device for recording a CW THz spectroscopy spectrum according to any of claims 7 - 9; a data analysis unit comprising an electronic processor being configured to access the recorded CW THz spectroscopy spectrum and analyse the recorded CW THz spectroscopy spectrum using a supervised machine learning algorithm trained on similarly recorded CW THZ spectroscopy spectra from similar samples comprising liquid solutions with known glucose levels to estimate a glucose level in the liquid solution of the sample.

15. The system according to claim 14, further comprising a glucose level monitoring unit capable of establishing a connection to the control unit and the data analysis unit, and comprising an electronic processor configured to provide an estimated glucose level in a sample by: instructing the control unit to record a CW THz spectroscopy spectrum; instructing the data analysis unit to estimate a glucose level based on the recorded CW THz spectroscopy spectrum; and retrieving the estimated glucose level from the data analysis unit.