WO2025003717A1

WO2025003717A1 - Non-invasive blood glucose monitor

Info

Publication number: WO2025003717A1
Application number: PCT/HU2024/050048
Authority: WO
Inventors: Dávid SZABÓ; János MINÁROVITS; Ákos Lajos LÉSTYÁN; Sándor Imre HEGEDŰS
Original assignee: Nibsm Hungary Korlátolt Felelősségű Társaság
Priority date: 2023-06-30
Filing date: 2024-06-28
Publication date: 2025-01-02

Abstract

The subject of the invention is a non-invasive blood glucose monitoring device and a method for measuring blood glucose levels. The solution consists of the following elements: an infrared light source, three high-sensitivity detectors, a signal conditioning unit, an analogue-to-digital converter with a minimum resolution of 12 bits, a regulator controlling the intensity and excitation of the light source (1), the detectors, the signal conditioning unit, and the analogue- to-digital converter, and a signal processing algorithm that estimates blood glucose levels using mathematical methods based on the measured signal values. The detectors are placed around the light source, rotated by 120 +/- 10° along a circle, at a distance of 3-4.5 mm from the light source. The signal conditioning unit consists of one or more analogue and/or digital frequency filters and a circuit that amplifies the signals measured by the detectors. During the applied measurement procedure, the determination of blood glucose levels is done using the averaged signal from the three detectors.

Description

Title of Invention:

Non-invasive blood glucose monitor

The subject of the invention is a non-invasive blood glucose monitoring device and a method for measuring blood glucose levels.

In the case of diabetes, blood glucose monitoring is a tool for therapy and also an indicator of the success of the therapy, as it provides a picture of the blood glucose level and its changes, helping to compile and check insulin, diet, medication, and exercise therapies, and also provides information about the occurrence of hypo- and hyperglycemia. Therefore, depending on the severity of the disease, diabetics are required to regularly self-monitor their blood glucose levels, even several times a day. Currently, invasive devices are typically used for blood glucose monitoring, alongside an increasing number of semi-invasive solutions. For both types of devices, it is characteristic that the measurement cannot be carried out without damaging the skin surface, which, due to the discomfort of the sampling procedure, represents a continuous burden for patients. In recent years, therefore, numerous solutions have been developed aimed at non-invasive blood glucose measurement. The non-invasive design of the device means that the patient can measure their blood glucose levels without physical sampling and thus without damaging the skin surface, thereby offering a much simpler, more comfortable, and cheaper alternative for monitoring blood glucose levels.

Most of the non-invasive blood glucose monitors available in the state of the art use an optical method employing infrared light. During their operation, they examine the radiation that is scattered, transmitted, or reflected from the surface illuminated with infrared light. From the characteristics of the absorbed, scattered, or reflected radiation, which is modified by the molecular properties of the sample, material characteristics can be inferred. Electromagnetic waves in the infrared spectrum easily reach the dermis layer of the skin and show correlation with blood glucose levels, thus the method is suitable for estimating blood glucose levels.

The optical method using infrared light and the non-invasive blood glucose monitoring device are described, for example, in the patent document with publication number US 10687739 B2. The subject of the description is a device and method that, after calibration, is suitable for performing non-invasive blood glucose measurements. The device enables non-invasive measurement of tissue glucose concentration and invasive measurement of capillary blood glucose concentration. A mathematical correlation system embedded in a special program code is used to establish a relationship between the actual blood glucose concentration measured invasively and the values measured non-invasively. The design includes four touch buttons to assist control, a display, a speaker, and a microcontroller unit (MCU), which is responsible for the user interface, managing processes, internal data storage, and device power management. The blood glucose monitor optionally includes a finger positioning unit, as well as four monochromatic light sources, which emit light in the visible light and infrared wavelength range (600 nm-1000 nm). Additional components include one or more colour image sensors, a signal processor (DSP), and an invasive blood glucose monitor module used for calibration.

The patent document with publication number EP 1518494 Al describes a blood glucose monitoring device that can also non-invasively determine blood glucose concentration and oxygen saturation. The aim of the invention is to provide a method and device for highly accurate determination of blood glucose concentration based on the test subject's temperature data and measured optical characteristics, without blood sampling. According to their approach, blood glucose levels can be accurately determined based on the surface temperature values of the body, blood oxygen concentration, and measurement parameters related to blood flow volume. An appropriate measurement surface could be the skin of the fingertip, where heat transfer is measured with a thermometer, and the content of oxy/deoxyhemoglobin in the blood can be calculated based on spectroscopic measurement. The parameter related to the amount of blood flow can be determined by measuring the rate of heat transfer from the skin. Several preferable implementation designs are described, in which the measuring device includes up to three light sources of different wavelength and their corresponding photodetectors, as well as a thermometer device suitable for measuring convective heat transfer. The light sources alternate in emitting light, and the photodetectors detect light reflected from the tissues. The solution resulted in a desktop measuring device that includes four push buttons, an LCD display, and a measuring and positioning unit protected by a cover. The measured data are stored in the machine memory with a date.

The patent document with publication number WO 2017/168432 Al describes a device that is also a non-invasive blood glucose monitor, consisting of the following elements. It includes a near-infrared LED light source at 980 nm and a photodetector at the wavelength of 1100 nm. The light source and photodetector are supplemented with a 970-1000 nm bandpass filter to ensure the precise 980 nm wavelength. The light source and photodetector are integrated into a device housing with a microprocessor, LCD display, power supply, battery, and keyboard. During operation, it emits light onto body tissues and then detects the reflected radiation. The device can be placed on a finger or any other body part, such as the palm, earlobe, face, finger, or wrist. The instrument converts the recorded signal from analogue to digital and displays the calculated blood glucose value on the display.

The patent document with publication number US 5601079 A describes a small and portable sensor capable of non-invasively measuring blood glucose concentration. The solution uses an optical light source to excite glucose molecules, then detects the characteristic emission of the molecules as they return to their ground state. The solution primarily uses light with a wavelength of 270-370 nm, as it can pass through the epidermal layer without significant attenuation. Various optical sources, such as semiconductor LEDs and semiconductor laser diodes, may also be suitable. The solution uses three infrared sensors. Each sensor is placed in a separate compartment to prevent crosstalk or light leak between them. Each sensor has its own narrow-band interference filter, which only allows radiation at 378 nm, 9.02 microns, and 3.80 microns to pass through. The sensors are housed in a sensor case, which is temperature- controlled at 37°C.

The patent document with publication number CN 104990038 A describes a multi -wavelength light source used for measuring blood characteristics on an optical principle. The solution includes four LED light sources of different wavelength, three of which are placed along an arc and the fourth in the centre of a circle. During operation, the four LEDs emit light of different wavelengths.

In the article titled "Non-invasively accuracy enhanced blood glucose sensor using shallow dense neural networks with NIR monitoring and medical features” published by Chevis Srichan et al. in 2022

, a non-invasive blood glucose monitor is described, which incorporates three different wavelength light sources and one detector. The measurement takes place at near-infrared wavelengths, and neural networks are used for data processing. These solutions also indicate that numerous solutions are known in the state of the art aimed at non-invasive measurement of blood glucose levels using infrared light/radiation. In addition to solutions using infrared light, for example, solutions using Raman spectroscopy and ultrasound can also be found. The main disadvantage of the known solutions is their limited accuracy, and the developed methods struggle to handle variations among users (e.g., skin colour, skin dryness, etc.) and measurement conditions (e.g., temperature, light, etc.). Because of these, no known solution achieves the measurement accuracy and reliability that would allow for regulatory approval, medical application, and thus the replacement of invasive and semi- invasive devices. The task to be solved by the invention is therefore to improve the accuracy of non-invasive blood glucose measurement by providing a new non-invasive blood glucose monitoring device and a method for measuring blood glucose levels. The solution must ensure the tracking of significant variations in blood glucose levels, thereby potentially replacing traditional blood glucose monitors that use blood samples from the fingertip based on the measurement results. It is crucial that the blood glucose monitor be portable and easy to handle, allowing patients to perform non-invasive blood glucose measurements at home without medical assistance.

The blood glucose monitoring device according to the invention is based on optical technology. In terms of structural composition, its main parts are the optical elements, the signal conditioning unit, the AD converter, and the control responsible for driving the individual elements. The optical elements include at least one high-power infrared light source and three high-sensitivity detectors placed around the light source, which convert the infrared light reflected from the skin/tissue into voltage. We recognize that for successful measurement, it is crucial that the three detectors measure the reflected light from the infrared light source from three different positions. The three independent measured signals ensure that certain measurement errors can be reduced by averaging. The detectors are placed around the infrared light source along a circle, spaced 120+/-10 degrees apart, with a radius between 3 and 4.5 mm. The measurement wavelength range of the detectors is between 550-1600 nm. The infrared light source is preferably an infrared LED, with a wavelength between 550 and 1600 nm. The detectors are preferably photodiodes. Preferably, light deflectors and/or optical filters can also be placed around the detectors. The light deflectors reduce crosstalk between the detectors.

The signal conditioning unit is a photovoltaic voltage conversion unit. Its task is to filter, amplify, and thus provide a suitable input signal for the AD converter (analogue-to-digital converter). The purpose of the filtering performed by the signal conditioning unit is to filter out high-frequency noise from the raw signal. Preferably, each detector's signal passes through a 4- 5 kHz cut-off frequency low-pass filter before and after amplification, which performs 20 decib el s/decade of filtering. The signal conditioning unit also performs an amplification function to properly amplify the signals measured by the detectors and bring them to the appropriate level for input to the next processing step, the AD converter. The degree of amplification depends on the current output of the used detector and the reference voltage of the AD converter.

The filtered and amplified signals are sampled, quantized, and encoded in the analogue/digital converter at least at 12 bits. A control responsible for driving the optical elements is connected to the optical elements. To achieve the measurement state, the device's electronics are heated to a temperature suitable for the application area between 30-37 C° and then controlled by a microcontroller to manage the light source state. For the heating to be feasible, the blood glucose monitor preferably contains heating resistors, which ensure that the device can be heated to 30-37 C°. This achieves a constant device temperature corresponding to the skin temperature, thus the device's temperature fluctuations do not affect the measurement.

The control can preferably also be supplemented with excitation electronics, which can produce periodic signals in addition to constant signals. For example, generating a sinusoidal or square wave signal might be advantageous, but the LED can also emit a constant signal according to its own profile. During the measurement, the infrared LED preferably emits a periodic or constant signal between 550 and 1600 nm. The state of the three detectors is preferably measured with 16,000Hz ADC sampling. If we previously used periodic excitation (sinusoidal or square/PWM), then, from the sampled channel signals, we separate the carrier frequency with a LOCK-in algorithm. Proper use of the LOCK-in algorithm requires the resolution of the excitation waveform, which can be 16 or 32 bits. The resulting data stream will be IKhz, which we then average every 10 samples to obtain a data stream arriving at 80Hz or faster, on which further processing is performed. This speed is fully sufficient to adequately detect rapid changes in the peripheral vascular network (changes in the volume of blood pumped by the heart).

The blood glucose monitor preferably includes sensors suitable for measuring temperature, e.g., skin temperature. The skin temperature measuring sensor is used to examine body temperature. Its purpose is to provide extra measurement data for more accurate diagnosis. Preferably, skin temperature data are measured with a resolution and accuracy of 0.1 °C using a 12-bit thermometer.

The device's power supply can be provided by a battery or another power source. Preferably, the device is powered through a battery. The device may optionally include one or more buttons for turning the device on/off and for controlling the device. For the success of the measurement, it is crucial that only the infrared radiation returning from the skin or from within the skin reaches the detectors, thus minimizing the noise level. Therefore, the design of the device and its housing is crucial. The device must be provided with a housing that stabilizes the connection between the sensors and the skin surface, as the position and the distance between the skin surface and the sensors must not change during measurement, and the possibility of external light entering must also be excluded. During the measurement, the pressure exerted by the device on the skin surface must be kept constant and care must be taken not to compress the blood vessels at the measurement point.

The measured, filtered, amplified, and digitized data can be processed and the blood glucose level calculated either on the blood glucose monitoring device or remotely from the device. If the calculation of blood glucose measurement takes place on the device, the blood glucose monitor includes a processing unit that performs the calculation, as well as a display that shows the measurement results. If the processing of the measurement is done remotely, or if telemedicine-related considerations justify data transmission, the device includes a communication module. The function of the communication module, located within the blood glucose monitor, is to transmit the measured voltage values and also to enable remote programming and control of the device. The preferred communication channel is Bluetooth. (The communication module can also be used to transfer data to a mobile device even if the calculation is done on the device.) The remote calculation of blood glucose levels can be implemented in several ways. The communication module can directly connect to an external processing unit. Alternatively, the communication module can connect to a smart device, which is in communication with a central server. The smart device's role is to collect the voltage signals recorded by the blood glucose monitor and transmit them to the central server for processing. Additionally, it displays the blood glucose value determined by the central server to the user and assists in performing and recording calibration measurements. This is facilitated by a mobile application on the smart device. The mobile application transmits all data via an internet connection to the central server, where they are recorded in a database and then evaluated. After evaluation, the blood glucose value is recorded in the database, from which the mobile application reads and then displays it to the user. The smart device, based on the incoming data, can check the measured values and inform the user in case of inappropriate values, and can initiate or interrupt the actual measurement process.

During the blood glucose measurement procedure implemented with the blood glucose monitoring device, the determination of blood glucose levels is done from the uniquely averaged signal of the three photodiodes. This is because we recognize that averaging the signals can filter out minor movements, various noises and interfering factors. It is important to note that the reflected infrared signal during measurement does not come from a homogeneous sample. For example, reflections can occur in the blood or in interstitial fluid. These reflections are inseparable from the measured voltage signal, so it must be ensured that the measured values correlate with the blood glucose concentration. According to our findings, the shape of the measured signal follows the cardiac cycle, and characteristic points such as systolic peaks and, in some cases, diastolic peaks can be identified. During the measurement, we identify the local maxima (systolic peaks) on the measured voltage curve, supplemented with other characteristics of the curve (diastolic peak if identifiable, locations of the peaks, heights, bases, distances, parameters of at least one Gaussian curve fitted to the curve, area under the curve) which correlate with the blood glucose level.

The local maxima play an important role during measurement, corresponding to the systolic peaks of the heart rhythm curve. These are the points where the pressure and volume of blood in the vascular network are highest. This is the point where the measured values are most accurate. Furthermore, local maxima also assume similar conditions, where comparable values can be identified. In accordance with this, during the measurement, it is necessary to find the local maximum peaks and the related characteristics of the curve within each cardiac cycle, and then take their average. This average is the average voltage value, which, supplemented with other identifiable characteristics on the heart rhythm curve (diastolic peak if identifiable, locations of the peaks, heights, bases, distances, parameters of at least one Gaussian curve fitted to the curve, area under the curve) shows correlation with the blood glucose level.

Considering our insights regarding the procedure for measuring blood glucose levels, the measurement procedure is completed as follows. In the first step, the blood glucose monitoring device must be placed on the skin surface. When placing the blood glucose monitor, it is crucial that the light source and the detectors are in close contact with the skin/tissue. After placing the device, it is heated to the appropriate skin temperature for the application area, and then the IR light source is turned on. The measurement procedure can be performed with various signal shapes. The light/radiation emitted by the light source is periodic, preferably sinusoidal, square, or constant. The light/radiation emitted by the light source penetrates the skin and tissue, and part of the light/radiation is absorbed by the skin and tissue, while another part of the light/radiation is reflected from the skin and tissue, and enters the detectors. The reflected light is detected from three positions with the three detectors, and the detected light is converted into electrical signals. The signals from the sensor unit are filtered and amplified with the signal conditioning unit, and then filtered again after amplification. The filter used in both cases is a 4-5 kHz low-pass filter, with amplification preferably 80, 000-120, OOOx. Subsequently, the analogue-to-digital converter converts the signal into a digital format, with a resolution of at least 12 bits. On the measured signals, we determine the locations of the local maximum peaks, their heights, bases, identifiable diastolic peaks, and the distances between the peaks. Then, using Gaussian fitting, at least one Gaussian curve is fitted to the waveforms of the curve. The fitting occurs at the peak location of the waveform. If additional peaks are identifiable - for example, at the location of the diastolic peak in the measured signal - further curve fitting is performed. The parameters determined for the fitted Gaussian curves are the peak heights, the X-axis locations associated with the peaks, and the parameters defining the widths of the curves, which essentially specify the normal distribution of the curves to be fitted. The three signals are averaged point by point, and then the local maximum peaks, their heights, bases, diastolic peaks, distances between the peaks, and the average voltage (which is the average of the peak heights) are re-determined from the averaged signal. Gaussian fitting can also be performed on the waveforms of the averaged signal. Thus, using Gaussian fitting, at least one Gaussian curve is fitted to the waveforms of the curve. The fitting occurs at the peak location of the waveform. If additional peaks are identifiable - for example, at the location of the diastolic peak in the measured signal - further curve fitting is performed. Again, the parameters determined for the fitted Gaussian curves are the peak heights, the X-axis locations associated with the peaks, and the parameters defining the widths of the curves, which essentially specify the normal distribution of the curves to be fitted. These values are calculated and stored. Additionally, the area under the curve is calculated and stored. The blood glucose value is determined based on these parameters using polynomial regression and/or machine learning techniques. The applied machine learning solution preferably involves a Feed Forward type neural network with supervised learning using a Backpropagation learning algorithm. The reason for using polynomial regression and machine learning techniques is that the relationship between the blood glucose value and the average voltage is not entirely linear, making the correlation non-trivial to find. When using polynomial regression, the regression first examines the relationships in the training dataset between the average voltage-skin temperature pair and the blood glucose values measured with an invasive device. It examines the newly obtained average voltage values, the data from the curves, and estimates the blood glucose values these might correspond to. It then returns the predicted blood glucose value to the process.

When using neural networks, supervised learning methods are applied using the parameters determined from the processed signal. In this supervised learning method, one or more neural networks are configured so that based on known input and desired output blood glucose data, our method learns the relationship between input and output. In this case, the input data appear in the form of time series, which contain sequences of data sets (features) filtered from data recorded at different times. The combination of input data and their assigned desired outputs forms the training set. The applied Feed Forward neural network has a multi-layer architecture, which includes an input layer, one or more hidden layers, and an output layer. The input layer receives the input data as time series, which are then transmitted through the network. Neurons in the hidden layers transform the inputs with weights and activation functions and forward them to the output layer. During the learning process, the neural network optimizes the weights to minimize errors between the input measurement data and the desired blood glucose values. This is achieved using a backpropagation algorithm, which calculates the error in the output layer of the network and backpropagates this error to earlier layers of the network, modifying the weights to reduce the error. After training, the Feed Forward neural network is capable of approximating the desired output blood glucose values based on individual input data. The neural network thus predicts the blood glucose value from the data generated during measurements and adapts to new environments and conditions when calibration is performed.

Naturally, a training dataset is necessary for the measurement. This is produced through calibration measurements. The user also measures their blood glucose level with a traditional, invasive device. After the user has recorded the measured value, the blood glucose level is also measured with the non-invasive blood glucose monitoring device. The data pairs measured with invasive and non-invasive devices are entered into a database, which forms the training database. At least 3 calibration measurements are necessary for the system to function. Calibration is crucial for the accuracy of the measurement, according to our understanding. The accuracy of the measurement is regulated by a large number of unique, user-dependent parameters. For example, factors such as body fat percentage and unique skin characteristics all influence the relationship between the measured voltage signal and the blood glucose level. We have recognized that because of this, the relationship between blood glucose level and voltage cannot be described with a universally valid fixed correlation. A dynamic, user-dependent relationship must be determined, which can also respond to changes. This is ensured by the blood glucose value collected from the user with an invasive device. We accept that the exact blood glucose level can be determined using the blood-based invasive device. By recording these values and associating them with the values measured by the blood glucose meter, the characteristic data relationship between the voltage values and blood glucose levels can be determined for each user. The device is operational with just 3 calibration measurements, but accuracy can be further improved with each new calibration measurement.

In a preferable implementation design of the measurement procedure, the temperature sensors, including the skin temperature sensor, are also considered during the determination of the blood glucose level. The skin temperature at the measurement point is related to blood circulation, thus it can positively influence the measurement accuracy. The data from the temperature sensors must be recorded both in the reference measurements stored in the training dataset and in the actual blood glucose measurements.

Before averaging the three signals during the measurement, several preconditioning examinations may be necessary to determine the final result, which is more accurate. The preconditioning examinations preferably include the following steps. The measurement procedure is based on the average of the peak heights found, so after determining the heights, positions, and widths of the peaks in the three measurement signals, it must be examined whether enough peaks have been found for processing. This is a multi -iteration examination, as initially, we only search for peaks with a given division, then dynamically search for peaks again considering the average distance between peaks. The peaks thus found will be considered our initial set of peaks. Initially, all peaks are qualified as processable, but during subsequent examinations, some peaks that do not pass a particular test are discarded. To determine whether we have a sufficient number of peaks, conclusions are drawn based on the ideal number of cardiac cycles. For example, in a 30-second measurement, the ideal number of cardiac cycles is around 30, so by comparing the number of identified peaks to this, we can infer the accuracy of the measurement. Deviation of the number of peaks from the ideal number of cycles according to the length of the measurement may indicate a faulty measurement. During the operation of the device, the permissible tolerance is +/- 10-15%. In the second step, we check whether the peak heights are approximately the same. This is important because significant fluctuations may indicate hardware faults, patient movement, and similar phenomena, but this does not change the fact that if there are many peaks of varying heights, the signal cannot be processed. When examining the peak heights, the threshold is +/-10-15% of the average peak height. At each peak, we check whether the peak height is within these threshold values; if not, the peaks that fall outside these thresholds during the examination are classified as unprocessable. The next examination performed on the peaks examined so far involves fitting a Gaussian curve to the peaks. If the peak is supplemented by additional features, such as a diastolic peak, Gaussian curve fitting is also performed on these. If fitting is successfully performed on 20-60% of the peaks, then the measurement is considered evaluable.

The objectives of the invention are achievable with the non -invasive blood glucose monitor described in claim I, and the blood glucose measuring procedure described in claim 14, whose preferable implementations are included in the sub-claims.

Our invention is described in detail with reference to the attached drawings, where Figure 1 illustrates the operation of the blood glucose monitoring device, Figure 2 shows the units of the blood glucose monitoring device and their connections, Figure 3 shows the positions of the detectors and the light source, Figure 4 illustrates the operation of blood glucose measurement, Figure 5 shows the process of blood glucose measurement, Figure 6 shows the process of calibration measurement, Figure 7 outlines the steps of the preconditioning procedure, Figure 8 shows the results of the measurement.

Figure 1 shows the operating principle of the blood glucose monitor. The infrared radiation emitted (10) by the light source (1) reaches the skin surface (6) and penetrates the epidermis (7), dermis (8), and hypodermis (9) layers of the skin. During the measurement, the infrared radiation (11) reflected from glucose molecules in the micro-arterial, capillary, and (circulatory) micro-venous blood already used by the tissues is detected by detectors (2, 3, 4) in the dermis layer (8). To ensure minimal crosstalk between detectors (2, 3, 4), light deflectors (5) are placed around them. Optical filters (not shown) can also be placed on detectors (2, 3, 4.)

Figure 2 shows the units of the blood glucose monitor and their connections. The figure identifies the optical elements of the blood glucose monitor, which include at least one high- powered infrared light source (1) and three high-sensitivity detectors (2, 3, 4) placed around light source (1). The radiation reflected from the skin surface (6) and the skin layers (7, 8, 9) (not shown) is converted into voltage by detectors (2, 3, 4). The voltage signal produced by detectors (2, 3, 4) enters the signal conditioning unit (13). The signal conditioning unit (13) contains two filters and an amplification unit. The filters are responsible for filtering the raw signal from high-frequency noise. Each detector's signal passes through a low-pass filter with a cut-off frequency of 4-5 kHz before and after amplification, which performs 20 decib el s/decade of filtering. The function of the amplifying unit of the signal conditioning unit

(13) is to properly amplify the signals measured by the detectors and bring them to the appropriate level for input to the next processing step, the AD converter. The degree of amplification applied depends on the current output of the used detector and the reference voltage of the AD converter. The filtered and amplified signals are digitized by the AD converter

(14). In the AD converter (14), the measured signal is sampled, quantized, and encoded at least at 12 bits. The detectors (2, 3, 4) are connected to the control (15), responsible for their operation, via a data bus (19) with the help of the AD converter (14) and the radiation source (1). The signal conditioning unit (13) operates automatically without control. Temperature measuring sensors (12) also connect to the control (15), which records the skin temperature and device temperature at the time of measurements. Additional elements of the solution include a communication unit (16) that transmits the measured, filtered, amplified, and digitized data for processing, a USB connector (17) that assists in charging the blood glucose monitor, and a battery (18) responsible for the power supply. The control (15) can also be supplemented with excitation electronics (not shown), which can generate signals such as sine and square waves, or emit a constant signal according to the LED's own profile. Heating resistors (not shown) can also be connected to the control (15), ensuring that the device can be heated to 30-37°C.

Figure 3 shows the positions of the detectors and the light source. The number and position of optical elements placed in the blood glucose monitor are crucial for successful and reliable measurement. In the blood glucose monitor, the infrared radiation reflected off the skin from the infrared source (1) is measured by detectors (2, 3, 4). The three detectors produce three independent measured signals, which allow averaging of certain measurement errors. The three independent signals also provide the opportunity to filter out faulty sensor placements. Establishing the correct sensor-skin contact is crucial for successful measurement. Comparing the measured signals from each sensor and significant deviations in the signal shapes can determine the faulty positions of the sensors. The detectors (2, 3, 4) are arranged around the radiation source (1). The detectors (2, 3, 4) are arranged around the radiation source (1) in a circle, rotated by 120 +/-10 degrees. The distance between the detectors (2, 3, 4) and the radiation source (1) is between 3 and 4.5 mm.

Figure 4 illustrates the operation of blood glucose measurement. The diagram shows the voltage signal (20) measured and averaged by the detectors (2, 3, 4). The diagram shows that the measured voltage signal (20) is periodic and follows the cardiac cycle. It displays observable local maximum peaks / systolic peaks (21) and diastolic peaks (22). The local maximum peaks play an important role during measurement. These correspond to the points where, as a result of the heart's pumping function, the pressure and volume of blood in the vascular network are greatest. At this point in time, the measurement site contains the highest volume of blood, thereby yielding the most precise and comparable results. In accordance with this, during the measurement, we search for local maximum peaks within each cardiac cycle, and based on the voltage peak and other parameters of the curve (diastolic peak, locations of the peaks, heights, bases, distances, parameters of at least one Gaussian curve fitted to the curve, area under the curve), we determine the blood glucose level.

Figure 5 shows the process of blood glucose measurement, with the following steps. After initiating the measurement process (SI), the next step is to align the blood glucose monitor with the measurement point (S2). The device should be positioned so that the radiation source (1) and the detectors (2, 3, 4) are in close contact with the skin and/or tissue at the measurement site. During the measurement, the blood glucose monitor must not move relative to the measurement point and no external interfering light should enter, which can preferably be ensured by the design of the blood glucose monitor housing. After placing the device, it is heated to the appropriate skin temperature for the measurement site, then the next step is turning on the light source (S3). In the next step (S4), the reflected light is detected from three positions using the three detectors (2, 3, 4), preferably with three photodiodes. The detected light is converted into electrical signals. In the next step (S5), the signal conditioning unit (13) filters the signal coming from the sensor unit using a low-pass filter between 4-5 kHz. Then, in step S6, the signal from the detectors is amplified. Following this, in step S7, as part of the filtering and digitization of signals from the detectors, the signal is filtered with a 4-5 kHz low-pass filter, and then converted into digital format by the analogue-to-digital converter (14). The resolution of the signal is at least 12 bits. Subsequently, if there was periodic excitation, we remove the carrier signal from the real-life signal data using a LOCK-IN algorithm. Then we convert the measured data into a data stream of 80Hz or higher frequency by averaging, and then we send the signal for processing. In the next step (S8), the algorithm determines the location, height, bases, diastolic peaks, and the distances between the peaks from the three measured signals. Then, using Gaussian fitting, at least one Gaussian curve is fitted to the waveforms of the curve. The fitting occurs at the peak location of the waveform. If additional peaks are identifiable - for example, at the location of the diastolic peak in the measured signal - further curve fitting is performed. The parameters determined for the fitted Gaussian curves are the peak heights, the X-axis locations associated with the peaks, and the parameters defining the widths of the curves, which essentially specify the normal distribution of the curves to be fitted. Subsequently, in step S9, the algorithm averages the signals from the three channels point by point. In step S10, we also search for processable peaks in the averaged signal, determining the locations, heights, bases, diastolic peaks, and distances between the peaks. The necessary data, such as average voltage and preferably average temperature are determined. Subsequently, at least one Gaussian curve is also fitted to the waveforms of the averaged curve using Gaussian fitting. The fitting occurs at the peak of the waveform, and additional curves are fitted at other potential peaks - for example, if a diastolic peak is identifiable at the measured location. Again, the parameters determined for the fitted Gaussian curves are the peak heights, the X-axis locations associated with the peaks, and the parameters defining the widths of the curves, which essentially specify the normal distribution of the curves to be fitted. These values are calculated and stored. Additionally, the areas under the two curves are calculated and stored. Using the newly calculated data, the blood glucose level is determined using polynomial regression or neural network techniques. If the procedure uses polynomial regression, the first step (Sil) involves reading the calibration data. Then, in step SI 2, the measured average voltage, parameters related to the Gaussian curve, skin temperature values, and calibration data are passed to the regression. As part of the next step (SI 3), the regression first examines the relationships between the data in training dataset data and the invasive blood glucose values. It examines the newly obtained values and estimates the blood glucose values these might correspond to. As part of step S14, the polynomial regression returns the predicted blood glucose value to the process and displays it to the user.

Figure 6 shows the process of calibration measurement. After initiating the calibration measurement process in step SI 5, the next step is SI 6, which involves a reference measurement using a traditional blood glucose monitor. As part of this, the user measures their blood glucose level with a traditional, invasive device, then records the measured value. Subsequently, blood glucose level is measured with the non-invasive blood glucose monitoring device according to the process steps S2-S10 already described. After calculating the average voltage, the algorithm records the data (invasive blood glucose value, the corresponding average voltage value, Gaussian curve parameters, temperature) as part of step S17 . With this, the calibration measurement is completed.

Figure 7 shows the steps of the preconditioning procedure, which can preferably supplement the measurement procedure and contribute to determining a more accurate result. The preconditioning examination should be performed before averaging the signals from the three channels. It preferably includes the following steps. The first step (SI 8) is checking the number of peaks, where we check whether we have enough peaks for signal examination. In a 30-second measurement, the ideal number of cardiac cycles is around 30, so our checkpoint is to compare the measurement results and the number of identified peaks to the ideal number of cardiac cycles. During the measurement, considering the length of the measurement, the permissible tolerance is +/- 10-15%. The second step (S19) relates to examining the peak heights. Ideally, the peaks should have approximately the same heights, but there are times when this may not be the case for some reason. These cases are attempted to be handled in some way. For the examination, we need the average of the processable peak heights, and we allow deviations within a certain tolerance threshold from this average. When examining the peak heights, a deviation of +/-10-15% from the average peak height is the threshold. In the next step (S20) , we fit a Gaussian curve to the peaks. If individual peaks are supplemented by additional features, such as diastolic peaks, we also perform Gaussian curve fitting on these. If fitting is successfully performed on 20-60% of the peaks, then the measurement is considered evaluable.

Figure 8 shows the results and accuracy of the measurement. We conducted a series of measurements. During the measurement, the patient's blood glucose level was measured with a non-invasive blood glucose monitor. Concurrently, several invasive measurements were also performed for comparison between the two methods.

The non-invasive blood glucose monitor and the procedure for measuring blood glucose level according to the invention are capable of determining the blood glucose level with sufficient accuracy and stability. Numerous measurement errors can be eliminated through averaging across the three detector signals. By determining the local maxima, averaging them, and calculating the blood glucose level from the determined average voltage, accurate and reliable results are obtained.

Reference marks - light source - detector - detector - detector - light deflector - skin surface - epidermis - dermis - hypodermis - emitted infrared radiation - reflected infrared radiation - temperature measuring sensor - signal conditioning unit - AD converter - control - communication module - USB connector - battery - data bus - voltage signal - local maximum peaks - diastolic peaks - Initiation of measurement process - Fitting the blood glucose monitor to the measurement point - Excitation of the light source - Detection of reflected radiation from 3 positions - Filtering of signals from the detectors - Amplification of signals from the detectors - Filtering and digitizing signals from the detectors - Identification of parameters of local maximum peaks - Determination of average signal from 3 signals point by point 510 - Determination of parameters necessary for blood glucose measurement from the averaged signal

511 - Reading calibration data

512 - Transfer of measured average voltage, skin temperature value, and calibration data to the regression

513 - Prediction of blood glucose value by regression

514 - Display of blood glucose value to the user

515 - Initiation of calibration measurement process

516 - Reference measurement with a traditional blood glucose monitor

517 - Recording of data pairs (average voltage and reference value measured with traditional device)

518 - Checking the number of peaks

519 - Checking the heights of the peaks

520 - Fitting of Gaussian curves

Claims

1. Non-invasive blood glucose monitor, comprising: an infrared light source (1); three high-sensitivity detectors (2, 3, 4) for converting the voltage changes of infrared radiation reflected from the skin and/or tissue, wherein said detectors (2, 3, 4) are placed around the light source (1), rotated by 120 +/- 10° along a circle, at a distance of 3-4.5 mm from the light source; a signal conditioning unit (13) for signal filtering and amplification, consisting of one or more analogue and/or digital frequency filters and a circuit that amplifies the signals measured by the detectors; an analogue-to-digital converter with a minimum resolution of 12 bits (14); regulator (15) controlling the intensity and excitation of the light source (1), the detectors (2, 3, 4), the signal conditioning unit (13), and the analogue-to-digital converter (14); and a signal processing algorithm that estimates blood glucose levels using mathematical methods based on the measured signal values.

2. The non-invasive blood glucose monitor according to claim 1, characterized in that the light source (1) is an infrared LED with a wavelength between 550 and 1600 nm,

3. The non-invasive blood glucose monitor according to claim 1, characterized in that the detectors (2, 3, 4) are photodiodes, with detected wavelength ranges between 550 - 1600 nm.

4. A non-invasive blood glucose monitor according to any of the claims 1-3, characterized in that a skin temperature measuring sensor is connected to the control (15).

5. A non-invasive blood glucose monitor according to any of the claims 1-4, characterized in that a device temperature measuring sensor is connected to the control (15).

6. A non-invasive blood glucose monitor according to any of the claims 1-5, characterized in that heater resistors are connected to the control (15).

7. A non-invasive blood glucose monitor according to any of the claims 1-6, characterized in that it includes a signal processing unit and a display for calculating and displaying blood glucose levels.

8. A non-invasive blood glucose monitor according to any of the claims 1-6, characterized in that it includes a signal processing unit positioned remotely from the blood glucose monitor.

9. The non-invasive blood glucose monitor according to claim 7, characterized in that it has a communication module (16) and a smart device for establishing a communication link between the signal processing unit and the blood glucose monitor.

10. A non-invasive blood glucose monitor according to any of the claims 1-9, characterized in that it has a housing to be attached to the body, preferably at the measurement point on the finger.

11. A non-invasive blood glucose monitor according to any of the claims 1-10, characterized in that there are light deflectors (5) around the detectors (2, 3, 4).

12. A non-invasive blood glucose monitor according to any of the claims 1-11, characterized in that optical filters are present on the detectors (2, 3, 4).

13. A non-invasive blood glucose monitor according to any of the claims 1-12, characterized in that there are low-pass filters between 4-5 kHz in the signal conditioning unit (13).

14. Procedure for measuring non-invasive blood glucose using a non-invasive blood glucose monitor according to any of the claims 1-13, with the following procedural steps: a) fitting the blood glucose monitoring device to the point of measurement on the body so that the light source (1) and the detectors (2, 3, 4) are in close contact with the skin and/or tissue at the point of measurement; b) heating the blood glucose monitoring device to the appropriate skin temperature at the point of measurement; c) exciting the light source (1), d) detecting the reflected light with the detectors (2, 3, 4) from three positions. e) filtering and amplifying the signals from the sensor unit with the signal conditioning unit (13); f) converting the signals into a digital format with an analogue-to-digital converter, g) determining the local maximum peaks, the height of the peaks, their bases, and the distances between the peaks from the three measured signals; h) fitting at least one Gaussian curve to the waveforms of the three signals at the peak locations and any additional peaks; i) determining the peak height, full width at half maximum, the sigma parameter, the width of the bell curve, and the area under the curves from the Gaussian curves; j) an averaged signal is generated by averaging the three measured voltage signals point by point; k) determining the local maximum peaks, the height of the peaks, their bases, and the distances between the peaks, as well as average voltage from the averaged signal; l) fitting at least one Gaussian curve to the waveforms of the averaged signal at the peak locations and any additional peaks; m) determining the peak height, full width at half maximum, the sigma parameter, the width of the bell curve, and the area under the curves from the Gaussian curve; n) reading the calibration data comprising pairs of blood glucose values measured by the invasive device and the non-invasive blood glucose monitor; o) determining the blood glucose level using polynomial regression and/or machine learning techniques with the retrieved data; p) displaying the measured blood glucose level.

15. The procedure according to claims 13, characterized in that the calibration data are the voltage values measured with the non-invasive blood glucose monitor according to claim 1, as well as blood glucose values measured with an invasive device.

16. The procedure according to any of the claims 14-15, characterized in that skin temperature data are also recorded during the production of calibration data.

17. The procedure according to any of the claims 14-16, characterized in that the excitation of the light source is done with sinusoidal, square, or constant signals.

18. A procedure according to any of the claims 14-17, characterized in that we use periodic excitation and remove the carrier signal from the detected signal using a LOCK-IN algorithm.

19. Procedure according to any of the claims 14-18, characterized in that we also determine the locations of the diastolic peaks, the heights of the peaks, their bases, and the distances between the peaks from the three measured signals and the averaged signal, and use these as input data in determining the blood glucose level.

20. Procedure according to any of the claims 14-16, characterized in that the machine learning technique used for determining the blood glucose level is a Feed Forward type neural network with supervised learning using a Backpropagation learning algorithm.

21. Procedure according to any of the claims 14-20, characterized in that we supplement the measurement procedure with a preconditioning examination, which includes the following steps: a) checking the number of peaks, by comparing the number of identified peaks and the ideal number of cardiac cycles during the duration of the measurement, where the permissible deviation between the ideal number of cycles and the number of measured peaks is +/- 10-15%; b) checking the similarity of peak heights by excluding peaks of heights different from the average height, where the permissible deviation from the average peak height is +/- 10- 15%; c) fitting Gaussian curves to the identified peaks, with the expected ratio of successful fits being 20-60% of the peaks.