Method and computing device for non-invasively estimating blood glucose levels, device for non-invasively measuring electrocardiogram signals, and non-transitory computer-readable recording medium

Abstract

To provide a method for non-invasively estimating a glucose level, suitable for correctly and non-invasively estimating a glucose level of a user by a computing device.SOLUTION: A method includes: receiving a plurality of electrocardiogram waveforms of a user; extracting at least two first electrocardiogram features from each of the plurality of electrocardiogram waveforms of the user; determining first feature peak positions respectively corresponding to the first electrocardiogram features; calculating at least one peak distance among the plurality of first feature peak positions; and estimating a glucose level of the user on the basis of the peak distance. The first electrocardiogram feature is selected from the group consisting of P wave, Q wave, R wave, S wave, T wave, and U wave. There are also provided a computing device for non-invasively estimating a glucose level, a device for non-invasively measuring an electrocardiogram signal, and a non-transient computer readable recording medium that are used in the method.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present disclosure relates to a method and a computing device for estimating, a device for measuring, and a non-transitory computer-readable recording medium, and in particular to a method and a computing device for non-invasively estimating blood glucose levels, a device for non-invasively measuring electrocardiogram (ECG) signals, and a non-transitory computer-readable recording medium. [Background technology]

[0002] According to the World Health Organization (WHO), the number of people with diabetes increased from 108 million in 1980 to 422 million in 2014. In the United States alone, 30 million adults suffer from diabetes, of which 7.2 million are unaware they have the condition. This makes measuring and monitoring blood glucose levels a significant challenge in the medical field.

[0003] Conventional blood glucose measurement, whether continuous or not, is invasive. For example, when measuring blood glucose levels using a blood glucose meter or continuous glucose monitor, users must insert a needle into the body. One of the problems with invasive measurement is that it causes physical pain and significant discomfort to users. Therefore, the users of blood glucose measurement devices are often limited to diabetic patients who need to manage their blood glucose levels. In addition to the cost of medication, the costs of test strips, measurement devices, and readers must also be added to the total cost. Therefore, developing methods and devices for non-invasively estimating blood glucose levels would be of great benefit to diabetic patients.

[0004] In conventional non-invasive blood glucose measurement, to obtain a single-lead electrocardiogram signal, the user's hands must contact the measurement device as electrodes to form the V1 lead. To obtain an electrocardiogram signal, the diner must put down their dishes every 15 minutes during the meal and spend one minute measuring the electrocardiogram signal with both hands. If the signal quality is not satisfactory during this period, the user must spend another minute measuring the signal until the signal quality is sufficiently improved. Although this is a non-invasive measurement, it does not achieve the goal of continuous measurement, and therefore does not provide detailed insight into the trends in blood glucose levels.

[0005] In addition, the method of estimating a user's blood glucose level using an electrocardiogram signal places very high requirements on the signal quality, and in order to obtain accurate results, it is further necessary to accurately detect and extract features of the P, Q, R, S, and T waves.

[0006] There are two main methods for measuring blood glucose levels based on ECG signals. One method extracts features, such as the QT interval, from the morphology of the ECG waveform. Statistical analysis and machine learning are commonly used analytical methods. Statistical analysis results show that the QT interval and ST interval are highly correlated with hypoglycemia, while the PR interval and ST interval are highly correlated with hyperglycemia. The other method uses the ECG signal itself as a feature. However, this method has been shown to vary significantly in performance between individuals. Therefore, using the morphology of the ECG signal as a feature provides more stable performance than using the ECG signal directly.

[0007] Although the method based on extracting features from the morphology of the ECG waveform is an excellent method, it only shows the relationship between hypoglycemia and the QT interval (and / or ST interval) and the relationship between hyperglycemia and the PR interval (and / or ST interval). In other words, this method cannot estimate the user's blood glucose level because the ST interval is highly correlated with both hypoglycemia and hyperglycemia.

[0008] Therefore, improved non-invasive blood glucose measurement methods are desired to estimate a user's blood glucose level based on electrocardiogram features extracted from the electrocardiogram waveform. Summary of the Invention [Problem to be solved by the invention]

[0009] An object of the present disclosure is to provide an advantageous technique for non-invasively estimating a user's blood glucose level using multiple electrocardiogram (ECG) waveforms, which improves the accuracy and non-invasive estimation of a user's blood glucose level using multiple ECG waveforms. [Means for solving the problem]

[0010] To achieve at least the above-mentioned objectives, the present disclosure provides a method for non-invasively estimating a blood glucose level of a user by a computing device, the method including: receiving a plurality of electrocardiogram (ECG) waveforms of the user; extracting at least two first electrocardiogram features from each of the plurality of electrocardiogram waveforms of the user; determining first feature peak positions corresponding to each of the first electrocardiogram features; calculating at least one peak distance between the first feature peak positions for each of the plurality of electrocardiogram waveforms; and estimating the blood glucose level of the user based on the at least one peak distance. The first electrocardiogram features are selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0011] In one embodiment, the method for non-invasively estimating blood glucose levels described above further comprises calculating a corrected blood glucose level of the user using an equation for correcting the blood glucose level based on the blood glucose level of the user.

[0012] In one embodiment, the method for non-invasively estimating blood glucose levels further includes calculating, for each of the plurality of electrocardiogram waveforms, at least one peak-to-peak slope between the first characteristic peak locations, and estimating the user's blood glucose level based on the at least one peak-to-peak slope.

[0013] In one embodiment, the method for non-invasively estimating blood glucose levels further includes extracting at least three second electrocardiographic features from each of a plurality of electrocardiographic waveforms of the user and calculating at least one incidence rate based on the second electrocardiographic features. The step of estimating the blood glucose level of the user is further performed based on the at least one incidence rate. The second electrocardiographic features are selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0014] In one embodiment, the method for non-invasively estimating blood glucose levels further includes extracting at least four third electrocardiogram features from each of a plurality of electrocardiogram waveforms of the user and calculating at least one amplitude ratio based on the third electrocardiogram features. The step of estimating the blood glucose level of the user is further based on the at least one amplitude ratio. The third electrocardiogram features are selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0015] In one embodiment, the method for non-invasively estimating blood glucose levels further includes extracting at least one fourth electrocardiogram feature from each of a plurality of electrocardiogram waveforms of the user and calculating at least one sharpness result based on the at least one fourth electrocardiogram feature. Estimating the blood glucose level of the user is further based on the at least one sharpness result. The at least one fourth electrocardiogram feature is selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0016] In one embodiment, the method for non-invasively estimating blood glucose levels further includes evaluating the quality of a plurality of electrocardiogram waveforms of a user, and outputting the user's real-time blood glucose levels when the user's plurality of electrocardiogram waveforms are evaluated as normal electrocardiogram signals, and outputting the user's historical blood glucose levels when the user's plurality of electrocardiogram waveforms are evaluated as noise signals. The user's real-time blood glucose levels are blood glucose levels estimated in real time. The user's historical blood glucose levels are blood glucose levels previously estimated.

[0017] In one embodiment, the method for non-invasively estimating blood glucose levels further includes determining whether the number of the user's electrocardiogram waveforms is less than a preset value, and re-receiving the user's electrocardiogram waveforms if the number of the user's electrocardiogram waveforms is less than the preset value.

[0018] The present disclosure also provides a computing device for non-invasively estimating blood glucose levels, the computing device being signally connected to an electrocardiogram (ECG) measurement device to receive a plurality of electrocardiogram waveforms of a user from the ECG measurement device. The computing device includes a storage module and a blood glucose level estimation module. The blood glucose level estimation module is configured to be electrically connected to the storage module. A plurality of program codes are stored in the storage module. After the blood glucose level estimation module executes the plurality of program codes stored in the storage module, the blood glucose level estimation module performs the steps of any one of the methods for non-invasively estimating blood glucose levels described above.

[0019] The present disclosure also provides a device for non-invasively measuring electrocardiogram (ECG) signals, electrically connected to a computing device and outputting a plurality of electrocardiogram waveforms of a user to the computing device. The device includes a measurement module and a signal transmission module. The measurement module has a plurality of electrodes electrically connected to the user. The signal transmission module is configured to be electrically connected to the measurement module. The measurement module is configured to measure the plurality of electrocardiogram waveforms of the user. The signal transmission module transmits the plurality of electrocardiogram waveforms of the user to the computing device, and the computing device is configured to perform the steps of any one of the methods for non-invasively estimating blood glucose levels described above.

[0020] The non-transitory computer-readable recording medium can realize any one of the above-mentioned methods for non-invasively estimating blood glucose levels after a plurality of program codes stored therein are loaded and executed by a computing device. [Effects of the Invention]

[0021] The present disclosure provides an improvement in the technical field of non-invasively estimating a user's blood glucose level using multiple electrocardiogram waveforms, thereby enabling accurate non-invasive estimation of a user's blood glucose level. [Brief explanation of the drawings]

[0022] [Figure 1] 1A-1C are waveform diagrams illustrating two examples of electrocardiogram waveforms received from a user, according to one embodiment of the present disclosure. [Figure 2] FIG. 2 is a waveform diagram illustrating an example of multiple electrocardiogram features extracted from each of multiple electrocardiogram waveforms, according to an embodiment of the present disclosure. [Figure 3] FIG. 10 is a waveform diagram illustrating calculating at least one peak distance between a plurality of first feature peak locations for each of a plurality of electrocardiogram waveforms according to an embodiment of the present disclosure. [Figure 4] FIG. 1 is a schematic diagram illustrating an example of a computing device electrically connected to an electrocardiogram measurement device and to at least one of a machine learning model, a neural network model, and a convolutional neural network model, in accordance with an embodiment of the present disclosure. [Figure 5] FIG. 2 is a schematic diagram illustrating an example of a calculation unit electrically connected to a determination unit and an estimation unit according to an embodiment of the present disclosure. [Figure 6] 1 is a flowchart illustrating a first example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 7A] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distances, according to one embodiment of the present disclosure. [Figure 7B] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distances, according to one embodiment of the present disclosure. [Figure 8A] FIG. 1 is a schematic diagram illustrating an example of a machine learning model according to an embodiment of the present disclosure. [Figure 8B] FIG. 1 is a schematic diagram illustrating an example of a neural network model according to an embodiment of the present disclosure. [Figure 8C] FIG. 1 is a schematic diagram illustrating an example of a convolutional neural network model, according to an embodiment of the present disclosure. [Figure 8D] FIG. 1 is a schematic diagram illustrating an example of a convolutional layer, according to an embodiment of the present disclosure. [Figure 9] 10 is a flowchart illustrating a second example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 10] 10 is a flowchart illustrating a third example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 11] 10 is a flowchart illustrating a fourth example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 12]10 is a detailed flowchart illustrating a method for calculating at least one incidence rate based on a second electrocardiogram feature, according to an embodiment of the present disclosure. [Figure 13] 10A-10C are waveform diagrams illustrating a method for calculating at least one incidence rate based on a second electrocardiogram feature according to an embodiment of the present disclosure. [Figure 14A] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and incidence rate, according to one embodiment of the present disclosure. [Figure 14B] 1 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and incidence rate, according to one embodiment of the present disclosure. [Figure 15] 10 is a flowchart illustrating a fifth example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 16] 10 is a detailed flowchart illustrating a method for calculating at least one amplitude ratio based on a third electrocardiogram feature, according to one embodiment of the present disclosure. [Figure 17A] FIG. 10 is a waveform diagram illustrating an example of calculating at least one amplitude ratio based on a third electrocardiogram feature according to an embodiment of the present disclosure. [Figure 17B] FIG. 10 is a waveform diagram illustrating another example of calculating at least one amplitude ratio based on a third electrocardiogram feature according to an embodiment of the present disclosure. [Figure 18A] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and amplitude ratio, according to one embodiment of the present disclosure. [Figure 18B] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and amplitude ratio, according to one embodiment of the present disclosure. [Figure 19] 10 is a flowchart illustrating a sixth example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 20] 10 is a detailed flowchart illustrating a method for calculating a sharpness result based on a fourth electrocardiogram feature, according to an embodiment of the present disclosure. [Figure 21] FIG. 10 is a waveform diagram illustrating a method for calculating a sharpness result based on a fourth electrocardiogram feature according to an embodiment of the present disclosure. [Figure 22A] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and sharpness results, according to one embodiment of the present disclosure. [Figure 22B] 10 is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and sharpness results, according to one embodiment of the present disclosure. [Figure 23] FIG. 10 is a schematic diagram illustrating another example of a computing device electrically connected to an electrocardiogram measurement device and to at least one of a machine learning model, a neural network model, and a convolutional neural network model, in accordance with an embodiment of the present disclosure. [Figure 24] 10 is a flowchart illustrating a seventh example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 25] 10 is a flowchart illustrating an eighth example of a method for non-invasively estimating blood glucose levels, according to an embodiment of the present disclosure. [Figure 26] FIG. 1 is a schematic diagram illustrating an example of a computing device for non-invasively estimating blood glucose levels, electrically connected to at least one of an electrocardiogram sensor, an electrocardiogram monitoring device, and a server, and electrically connected to at least one of a display device and a server, in accordance with an embodiment of the present disclosure. [Figure 27] FIG. 1 is a schematic diagram illustrating an example of a device for non-invasively measuring electrocardiogram signals electrically connected to at least one of a non-invasive blood glucose level estimation device, a computing device, and a server, according to an embodiment of the present disclosure. [Figure 28] FIG. 1 is a schematic diagram illustrating that a user's blood glucose level is estimated based on multiple electrocardiogram waveforms received from the user, according to an embodiment of the present disclosure. [Figure 29A] FIG. 10 is a schematic diagram showing the comparison results of blood glucose levels estimated by the present disclosure compared with other devices. [Figure 29B] FIG. 10 is a schematic diagram showing another comparison result of blood glucose levels estimated by the present disclosure compared with other devices. DETAILED DESCRIPTION OF THE INVENTION

[0023] The present disclosure will be described in detail with reference to the following embodiments and accompanying drawings so that those skilled in the art can understand the objects, features, and advantages of the present disclosure.

[0024] Before the present disclosure is described in detail, it should be noted that in the following description, identical components or steps may be represented by identical reference numerals.

[0025] It should also be noted that in the context of this disclosure, terms such as "first," "second," "third," "fourth," "fifth," "sixth," etc. are used to distinguish between components, but are not used to limit the components themselves or to indicate a particular order of the components.

[0026] It should also be noted that each step described herein may be performed in order, in reverse order, or with steps appropriately modified or skipped during control and processing.

[0027] It should also be noted that the expression "the first step may be executed subsequently to the second step being executed" may also be expressed as meaning that the first step may be executed directly after the second step is executed, or that the first step may be executed after another step (e.g., a third step) is executed first.

[0028] Since the human body itself generates many different physiological signals, and many medical devices and sensors can convert these signals into specific electrical signals, important clinical information can be obtained by processing the physiological signals digitally using electronic devices.

[0029] For example, the heart itself is composed of multiple muscle groups that beat spontaneously and contract regularly. When the multiple muscle groups of the heart beat or contract, a small amount of electric current is generated, and the electric current is reflected to the surface of the body through the conductive tissues and body fluids surrounding the heart. Therefore, the voltage changes caused by the multiple muscle groups of the heart can be converted into a corresponding electrocardiogram by an associated electronic device. More specifically, the electrical signals caused by the multiple muscle groups of the heart can be received and recorded by multiple electrodes in contact with the skin of the human body.

[0030] The method provided by the present disclosure is suitable for signal processing multiple electrocardiogram waveforms to estimate a blood glucose level based on the multiple electrocardiogram waveforms. Because the multiple electrocardiogram waveforms of a user can be received by a non-invasive measurement device, the method provided by the present disclosure can estimate the blood glucose level of the user in a non-invasive manner.

[0031] Please refer to Figure 1, which is a waveform diagram illustrating two examples of electrocardiogram waveforms received from a user. The horizontal axis of Figure 1 represents time (in seconds), and the vertical axis of Figure 1 represents the amplitude value of the electrocardiogram waveform (in millivolts).

[0032] As shown in Fig. 1, the first example indicates that the number of electrocardiogram waveforms received from the user over a 10-second period is 20. In other words, the first example shown in Fig. 1 indicates that the user's heart rate is 120 beats per minute. As shown in Fig. 1, the second example indicates that the number of electrocardiogram waveforms received from the user over a 10-second period is 30. In other words, the second example shown in Fig. 1 indicates that the user's heart rate is 180 beats per minute.

[0033] The number of electrocardiogram waveforms received from a user within a certain period of time may vary from user to user. For example, the first and second examples shown in FIG. 1 each show electrocardiogram waveforms received from two different users. Furthermore, the number of electrocardiogram waveforms received from a user within a certain period of time may also vary depending on different states of the same user. For example, the first example may show electrocardiogram waveforms received from a user in a normal state, and the second example may show electrocardiogram waveforms received from the same user in a state of heightened emotion, such as tension, excitement, or fear.

[0034] Please refer to Figure 2. Figure 2 is a waveform diagram illustrating an example of multiple electrocardiogram features extracted from each of multiple electrocardiogram waveforms. The horizontal axis of Figure 2 represents time (in seconds), and the vertical axis of Figure 2 represents amplitude values ​​(in millivolts) of the multiple electrocardiogram waveforms. As shown in Figure 2, each of the multiple received electrocardiogram waveforms may have multiple electrocardiogram features, including P waves, Q waves, R waves, S waves, T waves, and U waves. The multiple electrocardiogram features may be used to represent voltage changes due to multiple muscle groups of the heart.

[0035] Multiple cardiac muscle groups may contract due to atrial depolarization, resulting in a first deflection (i.e., first voltage change) on the ECG waveform, called the P wave. Multiple cardiac muscle groups may contract due to left and right ventricular depolarization, resulting in a second deflection (i.e., second voltage change) on the ECG waveform, called the QRS complex. The second deflection is significantly larger than the first deflection. Typically, a QRS complex has three deflections: a first downward deflection, a first upward deflection, and a second downward deflection. The first downward deflection is called the Q wave, the first upward deflection is called the R wave, and the second downward deflection is called the S wave. Multiple cardiac muscle groups may contract due to ventricular repolarization, resulting in a third deflection (i.e., third voltage change) on the ECG waveform, called the T wave. The deflection immediately following the T wave (i.e., fourth voltage change) is sometimes called the U wave.

[0036] Please refer to Fig. 3. Fig. 3 is a waveform diagram illustrating the calculation of at least one peak distance between a plurality of first feature peak positions for each of a plurality of electrocardiogram waveforms. Note that Fig. 3 shows, as an example, how the peak distance is calculated using only one electrocardiogram waveform, but the peak distances of each of the received plurality of electrocardiogram waveforms can all be calculated in the same manner as shown in Fig. 3.

[0037] 3, each of the received electrocardiogram waveforms has multiple electrocardiogram features, such as P wave, Q wave, R wave, S wave, T wave, and U wave, so that the peak distance can be calculated based on the multiple electrocardiogram features. To calculate the peak distance, it is necessary to define the specific position of each of the multiple electrocardiogram features.

[0038] In some embodiments, the specific position of the P wave may refer to the point where the amplitude value changes most significantly during the first voltage change, i.e., the maximum value of the P wave; the specific position of the Q wave may refer to the point where the amplitude value changes most significantly during the first downward deviation during the second voltage change, i.e., the minimum value of the Q wave; the specific position of the R wave may refer to the point where the amplitude value changes most significantly during the first upward deviation during the second voltage change, i.e., the maximum value of the R wave; the specific position of the S wave may refer to the point where the amplitude value changes most significantly during the second downward deviation during the second voltage change, i.e., the minimum value of the S wave; the specific position of the T wave may refer to the point where the amplitude value changes most significantly during the third voltage change, i.e., the maximum absolute value of the T wave; and the specific position of the U wave may refer to the point where the amplitude value changes most significantly during the fourth voltage change, i.e., the maximum value of the U wave.

[0039] From the above, the peak distance between the P wave and the Q wave may be the difference between the specific position of the P wave and the specific position of the Q wave, i.e., the time difference between the maximum value of the P wave and the minimum value of the Q wave; the peak distance between the Q wave and the S wave may be the difference between the specific position of the Q wave and the specific position of the S wave, i.e., the time difference between the minimum value of the Q wave and the minimum value of the S wave; and the peak distance between the S wave and the T wave may be the difference between the specific position of the S wave and the specific position of the T wave, i.e., the time difference between the minimum value of the S wave and the maximum absolute value of the T wave.

[0040] For example, as shown in FIG. 3, if the time when the maximum value of the R wave occurs is 13:44:20.234 and the time when the maximum absolute value of the T wave occurs is 13:44:20.385, the calculated peak distance between the R wave and the T wave is 0.151 seconds (i.e., the time difference between the maximum value of the R wave and the maximum absolute value of the T wave).

[0041] Therefore, the peak distance between each of the multiple electrocardiogram features can be calculated by calculating the time difference between each of the two electrocardiogram features, and the unit of the peak distance may be seconds or milliseconds.

[0042] Please refer to Figure 4. Figure 4 is a schematic diagram illustrating an example of a computing device 400A signally connected to an electrocardiogram measurement device 300 and to at least one of a machine learning model 500, a neural network model 510, and a convolutional neural network model 520, in accordance with one embodiment of the present disclosure.

[0043] Because the computing device 400A is electrically connected to the electrocardiogram measurement device 300, the computing device 400A may receive multiple electrocardiogram waveforms of a user from the electrocardiogram measurement device 300. In some embodiments, the computing device 400A may receive multiple electrocardiogram waveforms of a user from the electrocardiogram measurement device 300 via a physical signal line connection. For example, the physical signal line connection may be, but is not limited to, a network signal line connection compliant with the Internet Protocol (IP). Furthermore, in some embodiments, the computing device 400A may receive multiple electrocardiogram waveforms of a user from the electrocardiogram measurement device 300 via a virtual signal line connection. For example, the virtual signal line connection may be, but is not limited to, a Wi-Fi connection compliant with a wireless network protocol.

[0044] In some embodiments, the computing device 400A may be electrically connected to another device (e.g., a wearable measurement device) capable of providing multiple electrocardiogram waveforms and may receive multiple electrocardiogram waveforms of the user from the other device (not shown).

[0045] The computing device 400A includes a receiving unit 402, an extracting unit 404, a determining unit 406, a calculating unit 408, and an estimating unit 410, and the computing device 400A may perform a series of specific signal processing on the received electrocardiogram waveforms of the user. That is, after the computing device 400A receives the electrocardiogram waveforms of the user, the computing device 400A may estimate the blood glucose level of the user based on the electrocardiogram waveforms of the user.

[0046] The receiving unit 402 may be configured to receive a plurality of electrocardiogram waveforms of a user, i.e., the computing device 400A may receive, via the receiving unit 402, a plurality of electrocardiogram waveforms of a user from the electrocardiogram measurement device 300 or another device capable of providing a plurality of electrocardiogram waveforms via a wired or wireless signal transmission path.

[0047] The extracting unit 404 may be configured to extract multiple electrocardiographic features from each of the multiple electrocardiographic waveforms. That is, the extracting unit 404 may extract multiple electrocardiographic features from each of the multiple electrocardiographic waveforms in time series, respectively, to extract multiple electrocardiographic features from each of the multiple electrocardiographic waveforms. The extracted multiple electrocardiographic features may include, but are not limited to, P waves, Q waves, R waves, S waves, T waves, and U waves (as shown in FIG. 2 ). In some embodiments, multiple first electrocardiographic features, multiple second electrocardiographic features, and / or multiple third electrocardiographic features may be extracted from each of the multiple electrocardiographic waveforms by the extracting unit 404, respectively. Therefore, taking multiple first electrocardiographic features as an example, the extracting unit 404 may extract multiple first electrocardiographic features from each of the multiple electrocardiographic waveforms, and the multiple first electrocardiographic features may include P waves, Q waves, R waves, S waves, T waves, and U waves. In some embodiments, the extractor 404 may input a plurality of electrocardiogram waveforms of the user to the convolutional neural network model 520, and each of the plurality of electrocardiogram features may be extracted by the convolutional neural network model 520.

[0048] The determiner 406 may be configured to determine a feature peak location corresponding to each of a plurality of electrocardiographic features. In some embodiments, the determiner 406 may determine a first feature peak location corresponding to each of a plurality of first electrocardiographic features, which may include a P wave, a Q wave, an R wave, an S wave, a T wave, and a U wave. For example, the determiner 406 may determine a location of a maximum value of a P wave, a location of a minimum value of a Q wave, a location of a maximum value of an R wave, a location of a minimum value of an S wave, a location of a maximum absolute value of a T wave, and a location of a maximum value of a U wave, respectively.

[0049] The calculation unit 408 may be configured to calculate a peak distance between multiple feature peak positions. In some embodiments, the calculation unit 408 may calculate a peak distance between two first feature peak positions among the multiple first feature peak positions, for example, but not limited to, the peak distance between the position of the maximum value of the P wave and the position of the minimum value of the Q wave, the peak distance between the position of the minimum value of the Q wave and the position of the minimum value of the S wave, and / or the peak distance between the position of the minimum value of the S wave and the position of the maximum absolute value of the T wave. The calculation unit 408 may calculate the peak distance between any two first feature peak positions among the multiple first feature peak positions. For example, if the multiple first feature peak positions include the position of the maximum value of the P wave, the position of the minimum value of the Q wave, the position of the maximum value of the R wave, the position of the minimum value of the S wave, the position of the maximum absolute value of the T wave, and the position of the maximum absolute value of the U wave, the calculation unit 408 may select any two of them to calculate the peak distance.

[0050] The estimation unit 410 may be configured to estimate the user's blood glucose level based on at least one peak distance. That is, the estimation unit 410 may estimate the user's blood glucose level based on the calculation result of each peak distance. Specifically, the calculation result of each peak distance is input to an estimation model, and the estimation model estimates the user's blood glucose level. In some embodiments, the estimation unit 410 may input the calculation result of each peak distance to a machine learning model 500, and the machine learning model 500 may estimate the user's blood glucose level. In other embodiments, the estimation unit 410 may input the calculation result of each peak distance to a neural network model 510, and the neural network model 510 may estimate the user's blood glucose level.

[0051] In some embodiments, the estimator 410 may be configured to estimate the user's blood glucose level based on at least one peak distance and at least one incidence rate. In other embodiments, the estimator 410 may be configured to estimate the user's blood glucose level based on at least one peak distance and at least one amplitude ratio. In other embodiments, the estimator 410 may be configured to estimate the user's blood glucose level based on at least one peak distance, at least one incidence rate, and at least one amplitude ratio.

[0052] The machine learning model 500 may be configured to estimate the user's blood glucose level based on at least one peak distance. In some embodiments, the machine learning model 500 may be further configured to estimate the blood glucose level based on at least one occurrence rate and / or at least one amplitude ratio. The machine learning model 500 may include or utilize a machine learning algorithm. In some embodiments, the machine learning model 500 may be a supervised learning model, and the machine learning model 500 may be trained with a plurality of electrocardiogram data so that the machine learning model 500 can estimate the user's blood glucose level. In some embodiments, each of the plurality of electrocardiogram data may include, but is not limited to, at least one peak distance and a corresponding blood glucose level of each of the plurality of electrocardiogram waveforms. In some embodiments, the machine learning model 500 may be implemented by a decision tree algorithm, a random forest algorithm incorporating bagging, an extreme gradient boosting (XGBoost) algorithm, or a support vector machine algorithm so that the machine learning model 500 can estimate the user's blood glucose level based on the at least one peak distance.

[0053] In some embodiments, the machine learning model 500 may be a semi-supervised learning model or an unsupervised learning model. For example, the machine learning model 500 may be implemented by CycleGAN.

[0054] In some embodiments, the machine learning model 500 may be integrated into the computing device 400A, i.e., the computing device 400A may store algorithms associated with the machine learning model 500 and implement the functionality of the machine learning model 500 by executing the algorithms stored on the computing device 400A.

[0055] The neural network model 510 may be configured to estimate the user's blood glucose level based on at least one peak distance. In some embodiments, the neural network model 510 may be configured to estimate the blood glucose level further based on at least one occurrence rate and / or at least one amplitude ratio. The neural network model 510 may include an input layer, a hidden layer, and an output layer. The input layer of the neural network model 510 may be configured to receive multiple electrocardiogram features input from outside the neural network model 510, and the number of nodes in the input layer may depend on the extracted multiple electrocardiogram features. For example, if the multiple electrocardiogram features extracted from each of the multiple electrocardiogram waveforms include P waves, Q waves, R waves, S waves, T waves, and U waves, the number of nodes in the input layer may be, for example, six. The hidden layer of the neural network model 510 may be located between the input layer and the output layer and configured to perform a series of operations on the variables of the nodes in the input layer. In some embodiments, to increase the computational complexity of the neural network model 510, the number of hidden layers may be configured to be at least three. The number of nodes in each hidden layer may be greater than, less than, or the same as the number of nodes in the previous layer, depending on the computational complexity. For example, if the neural network model 510 is configured with an input layer, three hidden layers, and an output layer, the number of nodes in the first hidden layer may be greater than the number of nodes in the input layer, the number of nodes in the second hidden layer may be less than the number of nodes in the first hidden layer, and the number of nodes in the third hidden layer may be less than the number of nodes in the second hidden layer. More specifically, if the number of nodes in the input layer is set to six, the number of nodes in the first hidden layer may be greater than six. In some embodiments, the activation function of the hidden layer may be, but is not limited to, a ReLU function, a Tanh function, a sigmoid function, or the like. In some embodiments, the hidden layer may be standardized to ensure that values ​​input to the hidden layer are within the sensitive range of the activation function.The output layer of the neural network model 510 is configured to output a final calculation result, which may be obtained by performing a series of calculation processes by the hidden layer. That is, after receiving each of the multiple electrocardiogram features, the neural network model 510 performs a series of calculation processes by the hidden layer between the input layer and the output layer, so that the output layer of the neural network model 510 can be configured to output the user's blood glucose level (i.e., the final calculation result).

[0056] In some embodiments, neural network model 510 may be integrated into computing device 400A, i.e., computing device 400A may store algorithms associated with neural network model 510 and may implement the functionality of neural network model 510 by executing the algorithms stored on computing device 400A.

[0057] Convolutional neural network model 520 may be configured to extract multiple electrocardiogram features based on multiple electrocardiogram waveforms. In some embodiments, convolutional neural network model 520 may include a convolutional layer configured to extract multiple electrocardiogram features from the multiple electrocardiogram waveforms and a fully connected layer configured to have substantially the same configuration and purpose as neural network model 510. In some embodiments, the number of convolutional layers may be configured to be at least three. In some embodiments, the fully connected layer of convolutional neural network model 520 has substantially the same configuration as neural network model 510, so that the user's blood glucose level may also be estimated by convolutional neural network model 520 (i.e., each of the multiple electrocardiogram features is extracted by the convolutional layer, and the user's blood glucose level is estimated by the fully connected layer).

[0058] In some embodiments, convolutional neural network model 520 may be integrated into computing device 400A, i.e., computing device 400A may store algorithms related to convolutional neural network model 520 and may implement the functionality of convolutional neural network model 520 by executing the algorithms stored on computing device 400A.

[0059] In some embodiments, the computing device 400A may include an electrocardiogram waveform database 452 and a blood glucose level database 454. The electrocardiogram waveform database 452 may be configured to store a plurality of electrocardiogram waveforms of the user. That is, the computing device 400A may first store the plurality of electrocardiogram waveforms of the user received from the electrocardiogram measurement device 300 in the electrocardiogram waveform database 452, and then receive the plurality of electrocardiogram waveforms of the user from the electrocardiogram waveform database 452 via the receiving unit 402. The blood glucose level database 454 may be configured to store the blood glucose level of the user. That is, after estimating the user's blood glucose level and / or calculating the user's corrected blood glucose level, the computing device 400A may store the user's estimated blood glucose level and / or the user's corrected blood glucose level in the blood glucose level database 454.

[0060] As a result, the computing device 400A provided by the present disclosure can be used to implement the method for non-invasively estimating blood glucose levels described below, in which the computing device 400A can receive a plurality of electrocardiogram waveforms of a user, and then estimate the user's blood glucose level and / or calculate the user's corrected blood glucose level based on the user's plurality of electrocardiogram waveforms.

[0061] The computing device 400A shown in Fig. 4 can provide a user with a means for non-invasively estimating the user's blood glucose level, as well as more accurately estimating the user's blood glucose level. Furthermore, the computing device 400A shown in Fig. 4 can advance the technical field of non-invasively estimating a user's blood glucose level by utilizing multiple electrocardiogram waveforms.

[0062] See FIG. 5, which is a schematic diagram illustrating an example of a calculation unit 408 signally connected to the determination unit 406 and the estimation unit 410, according to one embodiment of the present disclosure. In some embodiments, the calculation unit 408 may be configured to calculate at least one peak distance between a plurality of feature peak locations. Furthermore, the calculation unit 408 may be configured to calculate, but is not limited to, at least one incidence rate, at least one amplitude ratio, a corrected blood glucose level, a peak-to-peak slope, and / or at least one sharpness result. Accordingly, the calculation unit 408 may include a peak spacing calculator 502, and may further include an incidence rate calculator 504, an amplitude ratio calculator 506, a corrected blood glucose level calculator 508, a peak-to-peak slope calculator, and / or a sharpness result calculator.

[0063] The peak distance calculator 502 may be configured to calculate at least one peak distance between the plurality of feature peak locations, i.e., the peak distance calculator 502 may be configured to calculate the peak distance between two of the plurality of feature peak locations, for example, but not limited to, the time difference between the location of the maximum value of the P wave and the location of the minimum value of the Q wave (as shown in FIG. 3 ).

[0064] The incidence calculator 504 may be configured to calculate a respective incidence of at least one of the electrocardiographic features in each of the plurality of electrocardiographic waveforms, i.e., the incidence calculator 504 may be configured to calculate a ratio of occurrence of the second electrocardiographic feature in each electrocardiographic waveform, such as, but not limited to, an incidence of P waves.

[0065] The amplitude ratio calculator 506 may be configured to calculate a ratio of amplitude values ​​between each of the plurality of electrocardiogram features, i.e., the amplitude ratio calculator 506 may be configured to calculate, but is not limited to, a peak-to-peak ratio, such as a P-wave amplitude ratio, between each of the plurality of third electrocardiogram features.

[0066] The corrected blood glucose value calculator 508 may be configured to calculate a corrected blood glucose value of the user. That is, after the estimation unit 410 estimates the blood glucose value of the user, the corrected blood glucose value calculator 508 of the calculation unit 408 may receive the estimated blood glucose value of the user from the estimation unit 410, and further perform a correction operation on the estimated blood glucose value of the user according to an equation for correcting blood glucose values ​​to calculate a corrected blood glucose value of the user.

[0067] Please refer to Figure 6. Figure 6 is a flowchart illustrating a first example of a method for non-invasively estimating blood glucose levels according to one embodiment of the present disclosure. The method shown in Figure 6 may include steps S610, S620, S630, S640, and S650.

[0068] In step S610, a plurality of electrocardiogram waveforms of the user are received. Step S610 may be performed by the receiving unit 402 of the computing device 400A shown in FIG. 4. In some embodiments, by performing step S610, the plurality of electrocardiogram waveforms of the user may be received in real time from, for example, but not limited to, the electrocardiogram measurement device 300, an electrocardiogram sensor (not shown), or a wearable device (not shown). In other embodiments, by performing step S610, the plurality of electrocardiogram waveforms of the user may be received from the electrocardiogram waveform database 452, thereby indirectly receiving the plurality of electrocardiogram waveforms of the user.

[0069] In some embodiments, the number of the plurality of electrocardiogram waveforms may depend on the duration of electrocardiogram reception, i.e., in step S610, the plurality of electrocardiogram waveforms of the user may be received during the duration of electrocardiogram reception, and the duration of electrocardiogram reception may be set to at least one minute to ensure that a sufficient number of the plurality of electrocardiogram waveforms are received to estimate the user's blood glucose level and to more accurately estimate and / or calculate the blood glucose level.

[0070] In step S620, at least two first electrocardiographic features are extracted from each of the user's multiple electrocardiographic waveforms. Step S620 may be performed by the extraction unit 404 of the computing device 400A shown in FIG. 4. In some embodiments, step S620 may be performed subsequently after step S610 is performed. By performing step S620, at least two first electrocardiographic features may be extracted from each of the multiple electrocardiographic waveforms, and the first electrocardiographic features may be selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0071] In some examples, by performing step S620, P waves and Q waves of each of the plurality of electrocardiographic waveforms may be extracted as the first electrocardiographic feature. In other embodiments, by performing step S620, S waves and T waves of each of the plurality of electrocardiographic waveforms may be extracted as the first electrocardiographic feature. In other embodiments, by performing step S620, P waves, Q waves, R waves, S waves, T waves, and U waves of each of the plurality of electrocardiographic waveforms may be extracted as the first electrocardiographic feature. That is, by performing step S620, at least two of P waves, Q waves, R waves, S waves, T waves, and U waves of each of the plurality of electrocardiographic waveforms may be extracted as the first electrocardiographic feature.

[0072] In some embodiments, the first electrocardiogram feature may be extracted by the convolutional neural network model 520. That is, after the user's multiple electrocardiogram waveforms are input to the convolutional neural network model 520, the first electrocardiogram feature may be extracted by a convolutional layer of the convolutional neural network model 520.

[0073] In step S630, a first feature peak position corresponding to each of the plurality of first electrocardiogram features is determined. Step S630 may be performed by the determination unit 406 of the computing device 400A shown in FIG. 4. In some embodiments, step S630 may be performed subsequently after step S620 is performed. The first feature peak positions corresponding to each of the plurality of first electrocardiogram features may be determined by performing step S630, and the first feature peak positions may be substantially the same as those described in FIG. 3.

[0074] In some examples, if the extracted first electrocardiographic features are P waves and Q waves, the position of the maximum value of the P wave and the position of the minimum value of the Q wave of each of the multiple electrocardiographic waveforms may be determined by performing step S630. In other embodiments, if the extracted first electrocardiographic features are S waves and T waves, the position of the minimum value of the S wave and the position of the maximum absolute value of the T wave of each of the multiple electrocardiographic waveforms may be determined by performing step S630. In other embodiments, if the extracted first electrocardiographic features are P waves, Q waves, R waves, S waves, T waves, and U waves, the position of the maximum value of the P wave, the position of the minimum value of the Q wave, the position of the maximum value of the R wave, the position of the minimum value of the S wave, the position of the maximum absolute value of the T wave, and the position of the maximum value of the U wave of each of the multiple electrocardiographic waveforms may be determined by performing step S630.

[0075] Since the first feature peak position is the position of the maximum or minimum value of the first electrocardiogram feature, in some embodiments, the method for finding the first feature peak position may be realized by successively comparing the numerical values ​​of two adjacent points, and in other embodiments, the method for finding the first feature peak position may be realized by respectively calculating the slope of the tangent to each point.

[0076] Compared to existing techniques, the first feature peak position can be more easily determined and utilized, which contributes to estimating a user's blood glucose level.

[0077] In step S640, at least one peak distance between a plurality of first feature peak positions is calculated for each of a plurality of electrocardiogram waveforms. Step S640 may be performed by peak distance calculator 502 of calculation unit 408 shown in FIG. 5. In some embodiments, step S640 may be performed subsequently after step S630 is performed. By performing step S640, the time difference between any two of the plurality of first feature peak positions is calculated as a peak distance, and the peak distance is substantially the same as that described in FIG. 3.

[0078] In some examples, when the extracted first electrocardiographic features are a P wave and a Q wave, the time difference between the position of the maximum value of the P wave and the position of the minimum value of the Q wave is calculated as the peak distance by performing step S640. In other embodiments, when the extracted first electrocardiographic features are an S wave and a T wave, the time difference between the position of the minimum value of the S wave and the position of the maximum absolute value of the T wave is calculated as the peak distance by performing step S640. In other embodiments, when the extracted first electrocardiographic features are a P wave, a Q wave, an R wave, an S wave, a T wave, and a U wave, the time difference between any two positions of the position of the maximum value of the P wave, the position of the minimum value of the Q wave, the position of the maximum value of the R wave, the position of the minimum value of the S wave, the position of the maximum absolute value of the T wave, and the position of the maximum value of the U wave is calculated as the peak distance by performing step S640. That is, the peak distance may include, but is not limited to, the time difference between the P and Q waves, the time difference between the P and R waves, the time difference between the P and S waves, the time difference between the P and T waves, the time difference between the P and U waves, the time difference between the Q and R waves, the time difference between the Q and S waves, the time difference between the Q and T waves, the time difference between the Q and U waves, the time difference between the R and S waves, the time difference between the R and T waves, the time difference between the R and U waves, the time difference between the S and T waves, the time difference between the S and U waves, and / or the time difference between the T and U waves.

[0079] In step S650, the user's blood glucose level is estimated based on the peak distance. Step S650 may be performed by the estimation unit 410 of the computing device 400A shown in FIG. 4. In some embodiments, step S650 may be performed subsequently after step S640 is performed. By performing step S650, the user's blood glucose level is estimated based on the calculation of the peak distance. More specifically, the calculation of the peak distance is input into an estimation model, and then the user's blood glucose level is estimated by the estimation model.

[0080] The method for non-invasively estimating blood glucose levels shown in Fig. 6 not only provides a user with a means for non-invasively estimating their blood glucose levels, but also allows the user's blood glucose levels to be more accurately estimated using the peak distances described above. Furthermore, the method for non-invasively estimating blood glucose levels shown in Fig. 6 can improve the technical field related to non-invasively estimating a user's blood glucose levels by utilizing multiple electrocardiogram waveforms.

[0081] Please refer to Fig. 7A. Fig. 7A is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distances according to one embodiment of the present disclosure. That is, step S650 shown in Fig. 6 includes steps S710A, S720A, S730A, and S740A, and may be completed by executing steps S710A, S720A, S730A, and S740A, and steps S710A, S720A, S730A, and S740A may be executed by the estimating unit 410 of the computing device 400A shown in Fig. 4.

[0082] In step S710A, the peak distances are normalized. In some embodiments, step S710A may be performed subsequently after step S640 is performed. By performing step S710A, the peak distance calculations may be normalized to corresponding values ​​between ranges (e.g., between 0 and 1).

[0083] In step S720A, the normalized peak distances are input to the machine learning model 500. In some embodiments, step S720A may be executed subsequently after step S710A is executed. By executing step S720A, the normalized peak distances may be input to the machine learning model 500, such as a decision tree algorithm, a random forest algorithm incorporating bagging, an extreme gradient boosting (XGBoost) algorithm, or a support vector machine algorithm.

[0084] In step S730A, the user's blood glucose level is estimated by the machine learning model 500. In some embodiments, step S730A may be executed subsequently after step S720A is executed. By executing step S730A, the normalized peak distances are input into the machine learning model 500, and then the user's blood glucose level may be estimated based on the normalized peak distances. Since the machine learning model 500 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by the machine learning model 500 with a prediction accuracy of 0.80 or higher, preferably 0.90 or higher.

[0085] In step S740A, the user's blood glucose level is output by the machine learning model 500. In some embodiments, step S740A may be executed subsequently after step S730A is executed. By executing step S740A, the user's estimated blood glucose level may be output by the machine learning model 500 to a display device and / or a database. In some examples, the user's estimated blood glucose level may be output to the blood glucose level database 454 of the computing device 400A and synchronously stored in the blood glucose level database 454.

[0086] Please refer to FIG. 8A . FIG. 8A is a schematic diagram illustrating an example of a machine learning model 500 according to one embodiment of the present disclosure. As shown in FIG. 8A , the machine learning model 500 may be implemented using an extreme gradient boosting (XGBoost) algorithm. The XGBoost algorithm is designed using multiple decision trees, and the results of a previous decision tree affect the results of the subsequent decision trees, so that the multiple decision trees are interrelated, thereby making the prediction results of the machine learning model 500 more accurate. In some embodiments, the machine learning model 500 may be implemented using other algorithms, such as a decision tree algorithm, a random forest algorithm incorporating bagging, an extreme gradient boosting (XGBoost) algorithm, or a support vector machine algorithm.

[0087] Please refer to Figure 7B, which is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distances according to one embodiment of the present disclosure. That is, step S650 shown in Figure 6 includes steps S710B, S720B, S730B, and S740B, and may be completed by executing steps S710B, S720B, S730B, and S740B, and steps S710B, S720B, S730B, and S740B may be executed by the estimation unit 410 of the computing device 400A shown in Figure 4.

[0088] In step S710B, the peak distances are normalized. In some embodiments, step S710B may be performed subsequently after step S640 is performed. In some embodiments, step S710B may be substantially the same as step S710A.

[0089] In step S720A, the normalized peak distance is input to the neural network model 510. In some embodiments, step S720B may be performed subsequently after step S710B is performed. By performing step S720A, the normalized peak distance is input to the neural network model 510.

[0090] In step S730B, the user's blood glucose level is estimated by the neural network model 510. In some embodiments, step S730B may be executed subsequently after step S720B is executed. By executing step S730B, the normalized peak distances are input into the neural network model 510, and then the user's blood glucose level may be estimated based on the normalized peak distances. Since the neural network model 510 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by the neural network model 510 with a prediction accuracy of 0.80 or higher, preferably 0.90 or higher.

[0091] In step S740B, the user's blood glucose level is output to the neural network model 510. In some embodiments, step S740B may be executed subsequently after step S730B is executed. By executing step S740B, the user's estimated blood glucose level may be output by the neural network model 510 to a display device and / or a database. In some examples, the user's estimated blood glucose level may be output to the blood glucose level database 454 of the computing device 400A and synchronously stored in the blood glucose level database 454.

[0092] See FIG. 8B. FIG. 8B is a schematic diagram illustrating an example of a neural network model 510 according to an embodiment of the present disclosure. The neural network model 510 may include an input layer IL, at least three hidden layers HL, and an output layer OL. In some embodiments, after receiving the peak distance, the neural network model 510 may perform a series of calculations through each node in the hidden layer HL to estimate the user's blood glucose level, and output the estimated blood glucose level of the user through the output layer OL. In some embodiments, the neural network model 510 may directly receive information of each of the multiple electrocardiogram features, and estimate the user's blood glucose level by performing a series of calculations on the information of each of the multiple electrocardiogram features through the hidden layer HL.

[0093] In some embodiments, electrocardiogram features of each of the plurality of electrocardiogram waveforms may be extracted by the convolutional neural network model 520. Furthermore, since the fully connected layer of the convolutional neural network model 520 has substantially the same configuration as the neural network model 510, the convolutional neural network model 520 may extract electrocardiogram features of each of the plurality of electrocardiogram waveforms, and then directly estimate the user's blood glucose level based on each of the plurality of electrocardiogram waveforms, in particular, based on the peak distances between the plurality of electrocardiogram features.

[0094] Please refer to FIG. 8C. FIG. 8C is a schematic diagram illustrating an example of a convolutional neural network model 520 according to an embodiment of the present disclosure. After receiving a plurality of electrocardiogram waveforms EW of a user, the convolutional neural network model 520 can extract each of a plurality of electrocardiogram features by a convolutional layer CL. Please also refer to FIG. 8D. FIG. 8D is a schematic diagram illustrating an example of a convolutional layer CL according to an embodiment of the present disclosure. As shown in FIG. 8D, the convolutional neural network model 520 may include at least three (e.g., three) convolutional layers CL.

[0095] Please refer to Figure 9. Figure 9 is a flowchart illustrating a second example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 9 may include steps S610, S620, S630, S640, S650, and S610, in which steps S610, S620, S630, S640, and S650 are substantially the same as the steps shown in Figure 6.

[0096] In step S910, the user's corrected blood glucose value is calculated based on the user's blood glucose level. Step S910 may be performed by the corrected blood glucose value calculator 508 of the calculation unit 408 shown in FIG. 5. In some embodiments, step S910 may be performed subsequently after step S650 is performed. By performing step S910, the user's corrected blood glucose value is calculated using an equation that uses the corrected blood glucose value based on the user's blood glucose level estimated by the estimation unit 410.

[0097] In some embodiments, the equation for the corrected blood glucose level is, for example,

[0098]

number

[0099] Each parameter (a, b, c, d) in the equation for the corrected blood glucose level may be determined based on the user's estimated blood glucose level (variable x) and the user's actual blood glucose level measured by the blood glucose meter (variable y).

[0100] As an example, if a user's estimated blood glucose levels are 90, 135, and 180, respectively, and the user's actual blood glucose levels measured by a blood glucose meter are 85, 120, and 190, respectively, the values ​​of each parameter (a, b, c, d) are known by solving the simultaneous equations.

[0101]

number

[0102] Therefore, the corrected blood glucose calculator 508 calculates the corrected blood glucose equation:

[0103]

number

[0104] The estimated blood glucose level of the user may be corrected by performing a calculation operation on the estimated blood glucose level of the user by the above formula.

[0105] In some embodiments, the equation for the corrected blood glucose level is, for example,

[0106]

number

[0107] Each parameter (a, b, c) in the equation may be determined based on the user's estimated blood glucose level (variable x) and the user's actual blood glucose level measured by a blood glucose meter (variable y).

[0108]

number

[0109] Therefore, the corrected blood glucose calculator 508 calculates the corrected blood glucose equation:

[0110]

number

[0111] The estimated blood glucose level of the user may be corrected by performing a calculation operation on the estimated blood glucose level of the user by the above formula.

[0112] 9, the method for non-invasively estimating a blood glucose level not only provides a user with a means for non-invasively estimating the user's blood glucose level, but also uses the equation for the corrected blood glucose level to calculate a blood glucose level (i.e., the user's corrected blood glucose level) that is the same as or close to the actual measurement result of the blood glucose meter. As a result, the method shown in FIG. 9 can compensate for the difference between the estimated blood glucose level and the actual measurement result of the blood glucose meter, and provide the user with a blood glucose level that is substantially the same as or close to the actual measurement result of the blood glucose meter.

[0113] In some embodiments, if the blood glucose level estimated by the estimation unit 410 is sufficiently accurate (for example, if the difference between the estimated blood glucose level and the actual measurement result by the blood glucose meter is within 0.05), step S910 may be omitted. That is, whether to perform step S910 may depend on the user's requirement for the accuracy of the blood glucose level estimation result. If the user desires a more accurate blood glucose level result, the method shown in FIG. 9 provides a more accurate blood glucose level result (i.e., the user's corrected blood glucose level).

[0114] Please refer to Figure 10. Figure 10 is a flowchart illustrating a third example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 10 may include steps S610, S620, S630, S640, S1010, and S1020, in which steps S610, S620, S630, and S640 are substantially the same as the steps shown in Figure 6.

[0115] In step S1010, at least one peak-to-peak slope between a plurality of first characteristic peak positions of each of the plurality of electrocardiogram waveforms is calculated. Step S1010 may be performed by the operation unit 408 of the computing device 400A shown in FIG. 4. In some embodiments, step S910 may be performed subsequent to step S630. Furthermore, step S1010 may be performed simultaneously with step S640.

[0116] In some examples, when the extracted first electrocardiographic features are a P wave and a Q wave, the slope between the position of the maximum value of the P wave and the position of the minimum value of the Q wave is calculated as the peak-to-peak slope by performing step S1010. In other embodiments, when the extracted first electrocardiographic features are an S wave and a T wave, the slope between the position of the minimum value of the S wave and the position of the maximum absolute value of the T wave is calculated as the peak-to-peak slope by performing step S640. In other embodiments, when the extracted first electrocardiographic features are a P wave, a Q wave, an R wave, an S wave, a T wave, or a U wave, the slope between any two positions among the position of the maximum value of the P wave, the position of the minimum value of the Q wave, the position of the maximum value of the R wave, the position of the minimum value of the S wave, the position of the maximum absolute value of the T wave, and the position of the maximum value of the U wave is calculated as the peak-to-peak slope by performing step S640. That is, the peak-to-peak slope may include, but is not limited to, the slope between the P and Q waves, the slope between the P and R waves, the slope between the P and S waves, the slope between the P and T waves, the slope between the P and U waves, the slope between the Q and R waves, the slope between the Q and S waves, the slope between the Q and T waves, the slope between the Q and U waves, the slope between the R and S waves, the slope between the R and T waves, the slope between the R and U waves, the slope between the S and T waves, the slope between the S and U waves, and / or the slope between the T and U waves.

[0117] In step S1020, the user's blood glucose level is estimated based on the peak distance and the peak-to-peak slope. Step S1020 may be performed by the estimation unit 410 of the computing device 400A shown in FIG. 4. In some embodiments, step S1130 may be performed subsequently after step S640 and step S1010 are performed. By performing step S1020, the user's blood glucose level is estimated based on the calculated peak distance and the calculated peak-to-peak slope. More specifically, the calculated peak distance and the calculated peak-to-peak slope are input into an estimation model, and then the user's blood glucose level is estimated by the estimation model.

[0118] The method for non-invasively estimating blood glucose levels shown in Figure 10 not only provides a user with a means for non-invasively estimating their blood glucose levels, but also allows the user's blood glucose levels to be more accurately estimated using the peak distance and peak-to-peak slope described above. Furthermore, the method for non-invasively estimating blood glucose levels shown in Figure 10 can improve the technical field related to non-invasively estimating a user's blood glucose levels by utilizing multiple electrocardiogram waveforms.

[0119] Also, in some embodiments, the method for non-invasively estimating blood glucose levels shown in FIG. 10 may further include step S910 shown in FIG. 9, and may calculate the user's corrected blood glucose level by performing step S910.

[0120] Please refer to Figure 11. Figure 11 is a flowchart illustrating a fourth example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 11 may include steps S610, S620, S630, S640, S1110, S1120, and S1130, in which steps S610, S620, S630, and S640 are substantially the same as the steps shown in Figure 6.

[0121] In step S1110, at least three second electrocardiographic features are extracted from each of the user's multiple electrocardiographic waveforms. Step S1110 may be performed by extraction unit 404 of computing device 400A shown in FIG. 4. In some embodiments, step S1110 may be performed subsequent to step S610. Furthermore, step S1110 may be performed simultaneously with step S620. By performing step S1110, at least three second electrocardiographic features may be extracted from each of the multiple electrocardiographic waveforms, and the second electrocardiographic features may be selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0122] In some examples, by performing step S1110, P waves, Q waves, and R waves of each of the multiple electrocardiogram waveforms are extracted as second electrocardiogram features. In other embodiments, by performing step S1110, S waves, T waves, and U waves of each of the multiple electrocardiogram waveforms may be extracted as second electrocardiogram features. In other embodiments, by performing step S1110, P waves, Q waves, R waves, S waves, T waves, and U waves of each of the multiple electrocardiogram waveforms may be extracted as second electrocardiogram features. That is, by performing step S1110, at least three of P waves, Q waves, R waves, S waves, T waves, and U waves of each of the multiple electrocardiogram waveforms may be extracted as second electrocardiogram features.

[0123] In some embodiments, the second electrocardiogram feature may be extracted by the convolutional neural network model 520. That is, after the user's multiple electrocardiogram waveforms are input to the convolutional neural network model 520, the second electrocardiogram feature may be extracted by a convolutional layer of the convolutional neural network model 520.

[0124] In step S1120, at least one incidence rate is calculated based on the second electrocardiogram feature. Step S1120 may be performed by the incidence rate calculator 504 of the calculation unit 408 shown in FIG. 5 . In some embodiments, step S1120 may be performed subsequently after step S1110 is performed. By performing step S1120, at least one incidence rate, for example, the incidence rate of P waves, may be calculated based on the plurality of second electrocardiogram features, but is not limited to this. Step S1120 may be a step of calculating at least one incidence rate based on amplitude values ​​(e.g., maximum value, minimum value, root-mean-square value, or peak-to-peak value) of the plurality of second electrocardiogram features. More specifically, the at least one incidence rate may be calculated based on comparing the amplitude values ​​of the plurality of second electrocardiogram features with a predetermined value to respectively determine whether each of the plurality of second electrocardiogram features occurs in each of the plurality of electrocardiogram waveforms. Furthermore, by comparing the amplitude values ​​of the plurality of second electrocardiogram features, it may be determined whether each of the plurality of second electrocardiogram features occurs in each of the plurality of electrocardiogram waveforms, and at least one occurrence rate may be calculated based on the determined results.

[0125] In some examples, if the extracted second electrocardiographic features are P waves, Q waves, and R waves, the incidence of P waves, the incidence of Q waves, and / or the incidence of R waves may be calculated by performing step S1120. In other embodiments, if the extracted second electrocardiographic features are S waves, T waves, and U waves, the incidence of S waves, the incidence of T waves, and / or the incidence of U waves may be calculated by performing step S1120. In other embodiments, if the extracted second electrocardiographic features are P waves, Q waves, R waves, S waves, T waves, and U waves, the incidence of P waves, the incidence of Q waves, the incidence of R waves, the incidence of S waves, the incidence of T waves, and / or the incidence of U waves may be calculated by performing step S1120.

[0126] In step S1130, the user's blood glucose level is estimated based on the peak distance and the incidence rate. Step S1130 may be executed by the estimation unit 410 of the computing device 400A shown in FIG. 4. In some embodiments, step S1130 may be executed subsequently after step S640 and step S1120 are executed. By executing step S1130, the user's blood glucose level may be estimated based on the calculated peak distance and the calculated incidence rate. More specifically, the calculated peak distance and the calculated incidence rate are input into an estimation model, and then the user's blood glucose level is estimated by the estimation model.

[0127] The method for non-invasively estimating blood glucose levels shown in Fig. 11 not only provides a user with a means for non-invasively estimating their blood glucose levels, but also allows the user's blood glucose levels to be more accurately estimated using the peak distances and occurrence rates described above. Furthermore, the method for non-invasively estimating blood glucose levels shown in Fig. 11 can improve the technical field related to non-invasively estimating a user's blood glucose levels by utilizing multiple electrocardiogram waveforms.

[0128] Also, in some embodiments, the method for non-invasively estimating blood glucose levels shown in FIG. 11 may further include step S910 shown in FIG. 9, and may calculate the user's corrected blood glucose level by performing step S910.

[0129] Please refer to Figure 12. Figure 12 is a detailed flowchart illustrating a method for calculating at least one incidence rate based on a second electrocardiogram feature according to an embodiment of the present disclosure. That is, step S1120 shown in Figure 11 includes steps S1210, S1220, S1230, S1240, S1250, and S1260, and may be completed by performing steps S1210, S1220, S1230, S1240, S1250, and S1260, among which steps S1210, S1220, S1230, S1240, S1250, and S1260 may be performed by the incidence rate calculator 504 of the calculation unit 408 shown in Figure 5.

[0130] In step S1210, an occurrence interval is determined. In some embodiments, step S1210 may be performed subsequently after step S1110 is performed. By performing step S1210, the occurrence interval may be determined based on one of the second electrocardiogram features.

[0131] Step S1210 may be a step of determining an occurrence interval by using one of the second electrocardiogram features as a reference point and moving forward or backward by a specific occurrence interval distance. In some embodiments, the occurrence interval distance may depend on the distance between two adjacent electrocardiogram waveforms, i.e., the R-R interval.

[0132] In some embodiments, by performing step S1210, the occurrence interval may be determined by taking the location of the maximum R wave as the reference point and going forward by 0.33 times the R-R interval. In other embodiments, by performing step S1210, the occurrence interval may be determined by taking the location of the maximum R wave as the reference point and going backward by 0.67 times the R-R interval. In other embodiments, the occurrence interval may be determined by taking the location of the maximum absolute value of the T wave as the reference point and going backward by a specific occurrence interval distance (i.e., until a new P wave of the electrocardiogram waveform is encountered).

[0133] In step S1220, a first peak-to-peak and a second peak-to-peak within the occurrence interval are calculated. In some embodiments, step S1220 may be performed subsequently after step S1210. By performing step S1220, a first peak-to-peak and a second peak-to-peak within the occurrence interval of each of the plurality of electrocardiogram waveforms may be calculated.

[0134] In some examples, when the extracted second electrocardiographic features are P waves, Q waves, and R waves, step S1220 may be executed to calculate, for each of the multiple electrocardiographic waveforms, the peak-to-peak between the maximum value of the P wave and the minimum value of the Q wave within an occurrence interval (i.e., 0.33 times the R-R interval forward from the position of the maximum value of the P wave) as the first peak-to-peak, and the peak-to-peak between the maximum value of the R wave and the minimum value of the Q wave as the second peak-to-peak. In other embodiments, when the extracted second electrocardiographic features are R waves, S waves, and T waves, step S1220 may be executed to calculate, for each of the multiple electrocardiographic waveforms, the peak-to-peak between the maximum absolute value of the T wave and the minimum value of the S wave within an occurrence interval (i.e., 0.67 times the R-R interval backward from the position of the maximum value of the R wave) as the first peak-to-peak, and the peak-to-peak between the maximum value of the R wave and the minimum value of the S wave as the second peak-to-peak. In another embodiment, when the extracted second electrocardiogram feature is an S wave, a T wave, or a U wave, step S1220 may be performed to calculate, for each of the multiple electrocardiogram waveforms, the peak-to-peak between the maximum value of the U wave and the minimum value of the S wave within an occurrence interval (i.e., a specific occurrence interval distance backward from the position of the maximum absolute value of the T wave until a new P wave is encountered in the electrocardiogram waveform) as the first peak-to-peak, and the peak-to-peak between the maximum absolute value of the T wave and the minimum value of the S wave as the second peak-to-peak.

[0135] In step S1230, a ratio of the first peak-to-peak to the second peak-to-peak is calculated. In some embodiments, step S1230 may be performed subsequently after step S1220 is performed. The ratio of the first peak-to-peak to the second peak-to-peak may be calculated as a peak-to-peak ratio for each of the plurality of electrocardiogram waveforms.

[0136] In some examples, when the extracted second electrocardiographic feature is a P wave, a Q wave, or an R wave, the ratio of the peak-to-peak between the P wave and the Q wave to the peak-to-peak between the R wave and the Q wave may be calculated as the respective peak-to-peak ratio (which may be referred to as the first peak-to-peak ratio) by performing step S1230. In other embodiments, when the extracted second electrocardiographic feature is an R wave, an S wave, or a T wave, the ratio of the peak-to-peak between the T wave and the S wave to the peak-to-peak between the R wave and the S wave may be calculated as the respective peak-to-peak ratio (which may be referred to as the second peak-to-peak ratio) by performing step S1230. In other embodiments, when the extracted second electrocardiographic feature is an S wave, a T wave, or a U wave, the ratio of the peak-to-peak between the U wave and the S wave to the peak-to-peak between the T wave and the S wave may be calculated as the respective peak-to-peak ratio (which may be referred to as the third peak-to-peak ratio) by performing step S1230.

[0137] In step S1240, the peak-to-peak ratios are each compared to a preset ratio. In some embodiments, step S1240 may be performed subsequently after step S1230. For each of the plurality of electrocardiogram waveforms, the peak-to-peak ratios may each be compared to a preset ratio (e.g., by comparing the two ratio values).

[0138] In some examples, when the extracted second electrocardiographic features are P waves, Q waves, and R waves, step S1240 may be performed to compare a first peak-to-peak ratio of each of the plurality of electrocardiographic waveforms with a predetermined ratio, where the predetermined ratio may be set to a ratio value between 0.05 and 0.15, specifically 0.1, to determine whether the peak-to-peak ratio between the P wave and the Q wave is greater than the peak-to-peak ratio between the R wave and the Q wave multiplied by the predetermined ratio. In other embodiments, when the extracted second electrocardiographic features are R waves, S waves, and T waves, step S1240 may be performed to compare a second peak-to-peak ratio of each of the plurality of electrocardiographic waveforms with a predetermined ratio, where the predetermined ratio may be set to a ratio value between 0.2 and 0.4, specifically 0.25, to determine whether the peak-to-peak ratio between the T wave and the S wave is greater than the peak-to-peak ratio between the R wave and the S wave multiplied by the predetermined ratio. In another embodiment, if the extracted second electrocardiogram features are S waves, T waves, and U waves, step S1240 may be performed to compare the third peak-to-peak ratios of each of the multiple electrocardiogram waveforms with a preset ratio, and the preset ratio may be set to a ratio value between 0.12 and 0.18, specifically 0.15, to determine whether the peak-to-peak between the U wave and the S wave is greater than the peak-to-peak between the T wave and the S wave multiplied by the preset ratio.

[0139] In step S1250, a comparison result is generated. In some embodiments, step S1250 may be performed subsequently after step S1240 is performed.

[0140] In some examples, when the extracted second electrocardiographic features are P waves, Q waves, and R waves, performing step S1250 may generate a comparison result for each of the plurality of electrocardiographic waveforms, where the comparison result may be a result in which the first peak-to-peak ratio is greater than a predetermined ratio or a result in which the first peak-to-peak ratio is less than or equal to the predetermined ratio. In other embodiments, when the extracted second electrocardiographic features are R waves, S waves, and T waves, performing step S1250 may generate a comparison result for each of the plurality of electrocardiographic waveforms, where the comparison result may be a result in which the second peak-to-peak ratio is greater than a predetermined ratio or a result in which the second peak-to-peak ratio is less than or equal to the predetermined ratio. In another embodiment, when the extracted second electrocardiogram feature is an S wave, a T wave, or a U wave, by performing step S1250, a comparison result for each of the multiple electrocardiogram waveforms may be generated, among which the comparison result may be a result that the third peak-to-peak ratio is greater than a predetermined ratio, or a result that the third peak-to-peak ratio is less than or equal to the predetermined ratio.

[0141] In step S1260, the incidence rate is calculated based on the results of the multiple comparisons. In some embodiments, step S1260 may be performed subsequently after step S1250. By performing step S1260, the incidence rate may be calculated based on the results of each comparison of the multiple electrocardiogram waveforms.

[0142] In some examples, when the extracted second electrocardiographic features are P waves, Q waves, and R waves, by performing step S1260, the occurrence rate of P waves (i.e., the ratio at which the first peak-to-peak ratio results in a result greater than the preset ratio) may be calculated based on the results when the first peak-to-peak ratio is greater than the preset ratio and the results when the first peak-to-peak ratio is equal to or less than the preset ratio. In other embodiments, when the extracted second electrocardiographic features are R waves, S waves, and T waves, by performing step S1260, the occurrence rate of T waves (i.e., the ratio at which the second peak-to-peak ratio results in a result greater than the preset ratio) may be calculated based on the results when the second peak-to-peak ratio is greater than the preset ratio and the results when the second peak-to-peak ratio is equal to or less than the preset ratio. In another embodiment, if the extracted second electrocardiogram features are S waves, T waves, and U waves, by performing step S1260, the occurrence rate of U waves (i.e., the ratio at which the third peak-to-peak ratio occurs greater than the preset ratio) may be calculated based on the results at which the third peak-to-peak ratio is greater than the preset ratio and the results at which the third peak-to-peak ratio is less than or equal to the preset ratio.

[0143] By performing the steps shown in FIG. 12 , multiple comparison results may be generated by comparing the peak-to-peak ratio with a preset ratio to more accurately calculate the incidence rate, which may include, but is not limited to, the incidence rate of P waves, the incidence rate of Q waves, the incidence rate of R waves, the incidence rate of S waves, the incidence rate of T waves, and / or the incidence rate of U waves.

[0144] Please refer to FIG. 13. FIG. 13 is a schematic diagram illustrating a method for calculating at least one incidence rate based on a second electrocardiogram feature according to an embodiment of the present disclosure. As shown in FIG. 13, taking an electrocardiogram waveform as an example, when the extracted second electrocardiogram features are P waves, Q waves, R waves, S waves, T waves, and U waves, by performing the steps shown in FIG. 12, the peak-to-peak ratios between the second electrocardiogram features and the peak-to-peak ratios between the peak-to-peak features are calculated, respectively, to calculate the incidence rates. The peak-to-peak ratios may be compared with a predetermined ratio, where the incidence rates may be, but are not limited to, the incidence rate of P waves, the incidence rate of Q waves, the incidence rate of R waves, the incidence rate of S waves, the incidence rate of T waves, and / or the incidence rate of U waves.

[0145] Please refer to Fig. 14A. Fig. 14A is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and incidence rate according to one embodiment of the present disclosure. That is, step S1130 shown in Fig. 11 includes steps S1410A, S1420A, S1430A, and S1440A, and may be completed by performing steps S1410A, S1420A, S1430A, and S1440A, among which steps S1410A, S1420A, S1430A, and S1440A may be performed by the estimation unit 410 of the computing device 400A shown in Fig. 4.

[0146] In step S1410A, the peak distances are normalized. In some embodiments, step S1410A may be performed subsequent to step S640 and step S1120. In some embodiments, step S1410A may be substantially the same as step S710A shown in FIG. 7A.

[0147] In S1420A, the normalized peak distance and the incidence rate are input to the machine learning model 500. In some embodiments, step S1420A may be executed subsequently after step S1410A is executed. By executing step S1420A, the normalized peak distance and the incidence rate may be input to the machine learning model 500, such as a decision tree algorithm, a random forest algorithm incorporating bagging, an extreme gradient boosting (XGBoost) algorithm, or a support vector machine algorithm.

[0148] In step S1430A, the user's blood glucose level is estimated by machine learning model 500. In some embodiments, step S1430A may be executed subsequently after step S1420A is executed. By executing step S1430A, the normalized peak distance and the incidence rate are input into machine learning model 500, and then the user's blood glucose level may be estimated based on the normalized peak distance and the incidence rate. Since machine learning model 500 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by machine learning model 500 with a prediction accuracy of 0.80 or more, preferably 0.90 or more.

[0149] In step S1440A, the user's blood glucose level is output by machine learning model 500. In some embodiments, step S1440A may be performed subsequent to step S1430A being performed. In some embodiments, step S1440A may be substantially the same as step S740A shown in FIG. 7A.

[0150] Please refer to Figure 14B, which is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and incidence rate according to one embodiment of the present disclosure. That is, step S1130 shown in Figure 11 includes steps S1410B, S1420B, S1430B, and S1440B, and may be completed by performing steps S1410B, S1420B, S1430B, and S1440B, among which steps S1410B, S1420B, S1430B, and S1440B may be performed by the estimation unit 410 of the computing device 400A shown in Figure 4.

[0151] In step S1410B, the peak distances are normalized. In some embodiments, step S1410B may be performed subsequently after steps S640 and S1120 have been performed. In some embodiments, step S1410B may be substantially the same as step S710A shown in FIG. 7A.

[0152] In step S1420B, the normalized peak distance and the incidence rate are input to the neural network model 510. In some embodiments, step S1420B may be executed subsequently after step S1410B is executed. By executing step S1420B, the normalized peak distance and the incidence rate may be input to the neural network model 510.

[0153] In step S1430B, the user's blood glucose level is estimated by neural network model 510. In some embodiments, step S1430B may be executed subsequently after step S1420B is executed. By executing step S1430B, the normalized peak distance and the incidence rate are input into neural network model 510, and then the user's blood glucose level may be estimated based on the normalized peak distance and the incidence rate. Since neural network model 510 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by neural network model 510 with a prediction accuracy of 0.80 or more, preferably 0.90 or more.

[0154] In step S1440B, the user's blood glucose level is output by neural network model 510. In some embodiments, step S1440B may be performed subsequently after step S1430B is performed. In some embodiments, step S1440B may be substantially the same as step S740B shown in FIG. 7B.

[0155] Please refer to Figure 15. Figure 15 is a flowchart illustrating a fifth example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 15 may include steps S610, S620, S630, S640, S1510, S1520, and S1530, in which steps S610, S620, S630, and S640 are substantially the same as the steps shown in Figure 6.

[0156] In step S1510, at least four third electrocardiographic features are extracted from each of the user's multiple electrocardiographic waveforms. Step S1510 may be performed by extraction unit 404 of computing device 400A shown in FIG. 4. In some embodiments, step S1510 may be performed subsequent to step S610. Furthermore, step S1510 may be performed simultaneously with step S620. By performing step S1510, at least four third electrocardiographic features may be extracted from each of the multiple electrocardiographic waveforms, and the third electrocardiographic features may be selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

[0157] In some examples, by performing step S1510, the P wave, Q wave, R wave, and S wave of each of the multiple electrocardiogram waveforms may be extracted as the third electrocardiogram feature. In other embodiments, by performing step S1510, the Q wave, R wave, S wave, and T wave of each of the multiple electrocardiogram waveforms may be extracted as the third electrocardiogram feature. In other embodiments, by performing step S1510, the Q wave, R wave, S wave, and U wave of each of the multiple electrocardiogram waveforms may be extracted as the third electrocardiogram feature. In other embodiments, by performing step S1510, the P wave, Q wave, R wave, S wave, T wave, and U wave of each of the multiple electrocardiogram waveforms may be extracted as the third electrocardiogram feature. That is, by performing step S1510, at least four of the P wave, Q wave, R wave, S wave, T wave, and U wave of each of the multiple electrocardiogram waveforms may be extracted as the third electrocardiogram feature.

[0158] In some embodiments, the third electrocardiogram feature may be extracted by the convolutional neural network model 520. That is, after the user's multiple electrocardiogram waveforms are input to the convolutional neural network model 520, the third electrocardiogram feature may be extracted by a convolutional layer of the convolutional neural network model 520.

[0159] In step S1520, at least one amplitude ratio is calculated based on a plurality of third electrocardiogram features. Step S1520 may be performed by the amplitude ratio calculator 506 of the calculation unit 408 shown in FIG. 5 . In some embodiments, step S1520 may be performed subsequently after step S1510 is performed. By performing step S1520, at least one amplitude ratio, for example, but not limited to, an amplitude ratio of P waves, may be calculated based on a plurality of third electrocardiogram features. Step S1520 may be a step of calculating at least one amplitude ratio based on amplitude values ​​(for example, but not limited to, maximum value, minimum value, root-mean-square value, or peak-to-peak value) of a plurality of third electrocardiogram features. More specifically, at least one amplitude ratio is calculated by performing a series of calculation operations on a plurality of third electrocardiogram features.

[0160] In some examples, when the extracted third electrocardiographic features are P waves, Q waves, R waves, and S waves, the amplitude ratio of the P waves may be calculated by performing step S1520. In other embodiments, when the extracted third electrocardiographic features are Q waves, R waves, S waves, and T waves, the amplitude ratio of the T waves may be calculated by performing step S1520. In other embodiments, when the extracted third electrocardiographic features are Q waves, R waves, S waves, and U waves, the amplitude ratio of the U waves may be calculated by performing step S1520. In other embodiments, when the extracted third electrocardiographic features are P waves, Q waves, R waves, S waves, T waves, and U waves, the amplitude ratio of the P waves, the amplitude ratio of the T waves, and / or the amplitude ratio of the U waves may be calculated by performing step S1520, but this is not limiting.

[0161] In step S1530, the user's blood glucose level is estimated based on the peak distance and the amplitude ratio. Step S1530 may be executed by the estimation unit 410 of the computing device 400A shown in FIG. 4. In some embodiments, step S1530 may be executed subsequently after steps S640 and S1520 are executed. By executing step S1530, the user's blood glucose level may be estimated based on the calculated peak distance and the calculated amplitude ratio. More specifically, the calculated peak distance and the calculated amplitude ratio are input into an estimation model, and then the user's blood glucose level is estimated by the estimation model.

[0162] The method for non-invasively estimating blood glucose levels shown in Fig. 15 not only provides a user with a means for non-invasively estimating their blood glucose levels, but also allows the user's blood glucose levels to be more accurately estimated using the peak distance and amplitude ratio as described above. Furthermore, the method for non-invasively estimating blood glucose levels shown in Fig. 15 can improve the technical field related to non-invasively estimating a user's blood glucose levels by utilizing multiple electrocardiogram waveforms.

[0163] Also, in some embodiments, the method for non-invasively estimating blood glucose levels shown in FIG. 15 may further include step S910 shown in FIG. 9, and may calculate the user's corrected blood glucose level by performing step S910.

[0164] Please refer to Fig. 16. Fig. 16 is a detailed flowchart illustrating a method for calculating at least one amplitude ratio based on a third electrocardiogram feature according to an embodiment of the present disclosure. That is, step S1520 shown in Fig. 15 includes steps S1610, S1620, S1630, and S1640, and may be completed by performing steps S1610, S1620, S1630, and S1640A, and steps S1610, S1620, S1630, and S1640 may be performed by the amplitude ratio calculator 506 of the calculation unit 408 shown in Fig. 5.

[0165] In step S1610, the third peak-to-peak, fourth peak-to-peak, fifth peak-to-peak, and sixth peak-to-peak are calculated. In some embodiments, step S1610 may be performed immediately after step S1510. By performing step S1610, the third peak-to-peak, fourth peak-to-peak, fifth peak-to-peak, and sixth peak-to-peak may be calculated for each of the plurality of electrocardiogram waveforms.

[0166] In some embodiments, when the extracted third electrocardiogram features are P waves, Q waves, R waves, and S waves, by performing step S1610, the peak-to-peak between the maximum value of the P wave and the minimum value of the Q wave may be calculated as the third peak-to-peak, the peak-to-peak between the maximum value of the P wave and the minimum value of the S wave as the fourth peak-to-peak, the peak-to-peak between the maximum value of the R wave and the minimum value of the Q wave as the fifth peak-to-peak, and the peak-to-peak between the maximum value of the R wave and the minimum value of the S wave as the sixth peak-to-peak for each of the multiple electrocardiogram waveforms. In another embodiment, when the extracted third electrocardiogram feature is a Q wave, an R wave, an S wave, or a T wave, by executing step S1610, for each of the multiple electrocardiogram waveforms, the peak-to-peak between the maximum absolute value of the T wave and the minimum value of the Q wave may be calculated as the third peak-to-peak, the peak-to-peak between the maximum absolute value of the T wave and the minimum value of the S wave as the fourth peak-to-peak, the peak-to-peak between the maximum R wave and the minimum value of the Q wave as the fifth peak-to-peak, and the peak-to-peak between the maximum R wave and the minimum value of the S wave as the sixth peak-to-peak. In another embodiment, when the extracted third electrocardiogram feature is a Q wave, an R wave, an S wave, or a U wave, by executing step S1610, the peak-to-peak between the maximum value of the U wave and the minimum value of the Q wave may be calculated as the third peak-to-peak, the peak-to-peak between the maximum value of the U wave and the minimum value of the S wave as the fourth peak-to-peak, the peak-to-peak between the maximum value of the R wave and the minimum value of the Q wave as the fifth peak-to-peak, and the peak-to-peak between the maximum value of the R wave and the minimum value of the S wave as the sixth peak-to-peak for each of the multiple electrocardiogram waveforms.

[0167] In some embodiments, the third peak-to-peaks of each of the multiple electrocardiogram waveforms may be further summed, and the summed result may be divided by the number of third peak-to-peaks to calculate an average third peak-to-peak value. Similarly, the fourth, fifth, and / or sixth peak-to-peaks may be calculated using the same calculation operation to calculate an average fourth, fifth, and / or sixth peak-to-peak value, respectively. Thus, steps S1620, S1630, and S1640 described below may be continuously performed based on the average third, fourth, fifth, and sixth peak-to-peak values ​​(i.e., steps S1620, S1630, and S1640 may avoid performing calculations for each of the multiple electrocardiogram waveforms) to calculate the amplitude ratio with fewer calculation operations.

[0168] In step S1620, a first average value of the third peak-to-peak and the fourth peak-to-peak is calculated. In some embodiments, step S1620 may be performed subsequently after step S1610 is performed. By performing step S1620, the third peak-to-peak and the fourth peak-to-peak may be summed, and the summed result may be divided by two to calculate the first average value. In some embodiments, by performing step S1620, the average value of the third peak-to-peak and the average value of the fourth peak-to-peak may be summed, and the summed result may be divided by two to calculate the first average value.

[0169] In some embodiments, when the extracted second electrocardiographic features are a P wave, a Q wave, an R wave, and an S wave, step S1620 may be performed to sum the peak-to-peak between the maximum value of the P wave and the minimum value of the Q wave (i.e., the third peak-to-peak) and the peak-to-peak between the maximum value of the P wave and the minimum value of the S wave (i.e., the fourth peak-to-peak), and the summed result may be divided by 2 to calculate the first average value. In one embodiment, when the extracted second electrocardiographic features are a Q wave, an R wave, an S wave, and a T wave, step S1620 may be performed to sum the peak-to-peak between the maximum value of the absolute value of the T wave and the minimum value of the Q wave (i.e., the third peak-to-peak) and the peak-to-peak between the maximum value of the absolute value of the T wave and the minimum value of the S wave (i.e., the fourth peak-to-peak), and the summed result may be divided by 2 to calculate the first average value. In one embodiment, if the extracted second electrocardiogram features are a Q wave, an R wave, an S wave, and a U wave, step S1620 may be performed to sum the peak-to-peak between the maximum value of the U wave and the minimum value of the Q wave (i.e., the third peak-to-peak) and the peak-to-peak between the maximum value of the U wave and the minimum value of the S wave (i.e., the fourth peak-to-peak), and the summed result may be divided by 2 to calculate a first average value.

[0170] In step S1630, a second average value of the fifth peak-to-peak and the sixth peak-to-peak is calculated. In some embodiments, step S1630 may be performed subsequently after step S1620 is performed. Furthermore, step S1630 may be performed simultaneously with step S1620. By performing step S1630, the fifth peak-to-peak and the sixth peak-to-peak may be summed and the summed result may be divided by two to calculate the second average value. In some embodiments, by performing step S1630, the fifth peak-to-peak average value and the sixth peak-to-peak average value may be summed and the summed result may be divided by two to calculate the second average value.

[0171] In some embodiments, when the extracted second electrocardiographic features are a P wave, a Q wave, an R wave, and an S wave, step S1630 may be performed to sum the peak-to-peak between the maximum of the R wave and the minimum of the Q wave (i.e., the fifth peak-to-peak) and the peak-to-peak between the maximum of the R wave and the minimum of the S wave (i.e., the sixth peak-to-peak), and the summed result may be divided by 2 to calculate the second average value. In one embodiment, when the extracted second electrocardiographic features are a Q wave, an R wave, an S wave, and a T wave, step S1630 may be performed to sum the peak-to-peak between the maximum of the R wave and the minimum of the Q wave (i.e., the fifth peak-to-peak) and the peak-to-peak between the maximum of the R wave and the minimum of the S wave (i.e., the sixth peak-to-peak), and the summed result may be divided by 2 to calculate the second average value. In one embodiment, if the extracted second electrocardiogram features are a Q wave, an R wave, an S wave, and a U wave, step S1630 may be performed to sum the peak-to-peak between the maximum value of the R wave and the minimum value of the Q wave (i.e., the fifth peak-to-peak) and the peak-to-peak between the maximum value of the R wave and the minimum value of the S wave (i.e., the sixth peak-to-peak), and the summed result may be divided by 2 to calculate a second average value.

[0172] In step S1640, the ratio of the first average value to the second average value is calculated as the amplitude ratio. In some embodiments, step S1640 may be executed subsequently after step S1630 is executed. By executing step S1640, the ratio of the first average value to the second average value is calculated as the amplitude ratio.

[0173] By performing the steps shown in FIG. 16 , at least one amplitude ratio can be more accurately calculated based on the ratio of the first average value to the second average value by converting the peak-to-peak between the plurality of third electrocardiogram features into a first average value and a second average value, and the amplitude ratio may include, but is not limited to, the amplitude ratio of the P wave, the amplitude ratio of the T wave, and / or the amplitude ratio of the U wave.

[0174] 17A and 17B simultaneously. Fig. 17A is a waveform diagram illustrating an example of calculating at least one amplitude ratio based on a third electrocardiogram feature according to an embodiment of the present disclosure, and Fig. 17B is a waveform diagram illustrating another example of calculating at least one amplitude ratio based on a third electrocardiogram feature according to an embodiment of the present disclosure. As shown in Figs. 17A and 17B, for example, taking one electrocardiogram waveform as an example, if the extracted third electrocardiogram features are P waves, Q waves, R waves, S waves, T waves, and U waves, peak-to-peaks between multiple third electrocardiogram features may be calculated, and first and second average values ​​may be calculated based on the peak-to-peaks to calculate at least one amplitude ratio by performing the steps shown in Fig. 16.

[0175] Please refer to Fig. 18A. Fig. 18A is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on a peak distance and an amplitude ratio according to one embodiment of the present disclosure. That is, step S1530 shown in Fig. 15 includes steps S1810A, S1820A, S1830A, and S1840A, and may be completed by performing steps S1810A, S1820A, S1830A, and S1840A, among which steps S1810A, S1820A, S1830A, and S1840A may be performed by the estimation unit 410 of the computing device 400A shown in Fig. 4.

[0176] In step S1810A, the peak distance and the amplitude ratio are each normalized. In some embodiments, step S1810A may be performed subsequently after step S640 and step S1520 are performed. By performing step S1810A, the calculated peak distance may be normalized to a corresponding value between ranges (e.g., between 0 and 1), and the calculated amplitude ratio may be normalized to a value between ranges (e.g., between 0 and 1).

[0177] In step S1820A, the normalized peak distance and the normalized amplitude ratio are input to the machine learning model 500. In some embodiments, step S1820A may be executed subsequently after step S1810A is executed. By executing step S1820A, the normalized peak distance and the normalized amplitude ratio may be input to the machine learning model 500, such as a decision tree algorithm, a random forest algorithm incorporating bagging, an extreme gradient boosting (XGBoost) algorithm, or a support vector machine algorithm.

[0178] In step S1830A, the user's blood glucose level is estimated by machine learning model 500. In some embodiments, step S1830A may be executed subsequently after step S1820A is executed. By executing step S1830A, the normalized peak distance and the normalized amplitude ratio are input into machine learning model 500, and then the user's blood glucose level may be estimated based on the normalized peak distance and the normalized amplitude ratio. Since machine learning model 500 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by machine learning model 500 with a prediction accuracy of 0.80 or more, preferably 0.90 or more.

[0179] In step S1840A, the user's blood glucose level is output by machine learning model 500. In some embodiments, step S1840A may be performed subsequently after step S1830A is performed. In some embodiments, step S1840A may be substantially the same as step S740A shown in FIG. 7A.

[0180] Please refer to Fig. 18B, which is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on a peak distance and an amplitude ratio according to one embodiment of the present disclosure. That is, step S1530 shown in Fig. 15 includes steps S1810B, S1820B, S1830B, and S1840B, and may be completed by executing steps S1810B, S1820B, S1830B, and S1840B, among which steps S1810B, S1820B, S1830B, and S1840B may be executed by the estimation unit 410 of the computing device 400A shown in Fig. 4.

[0181] In step S1810B, the peak distance and amplitude ratio are each normalized. In some embodiments, step S1810B may be performed subsequently after steps S640 and S1520 have been performed. In some embodiments, step S1810B may be substantially the same as step S1810A shown in FIG. 18A.

[0182] In step S1420B, the normalized peak distance and the normalized amplitude ratio are input to the neural network model 510. In some embodiments, step S1820B may be executed subsequently after step S1810B is executed. By executing step S1820B, the normalized peak distance and the normalized amplitude ratio may be input to the neural network model 510.

[0183] In step S1830B, the user's blood glucose level is estimated by neural network model 510. In some embodiments, step S1830B may be executed subsequently after step S1820B is executed. By executing step S1430B, the normalized peak distance and the normalized amplitude ratio are input into neural network model 510, and then the user's blood glucose level may be estimated based on the normalized peak distance and the normalized amplitude ratio. Since neural network model 510 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by neural network model 510 with a prediction accuracy of 0.80 or more, preferably 0.90 or more.

[0184] In step S1840B, the user's blood glucose level is output by neural network model 510. In some embodiments, step S1840B may be performed subsequently after step S1830B is performed. In some embodiments, step S1840B may be substantially the same as step S740B shown in FIG. 7B.

[0185] Please refer to Figure 19. Figure 19 is a flowchart illustrating a sixth example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 19 may include steps S610, S620, S630, S640, S1910, S1920, and S1930, where steps S610, S620, S630, and S640 are substantially the same as the steps shown in Figure 6.

[0186] In step S1910, at least one fourth electrocardiogram feature is extracted from each of the user's multiple electrocardiogram waveforms. Step S1910 may be performed by extraction unit 404 of computing device 400A shown in FIG. 4. In some embodiments, step S1910 may be performed subsequent to step S610. Furthermore, step S1910 may be performed simultaneously with step S620. The fourth electrocardiogram feature may be selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves. In some embodiments, step S1910 may be substantially similar to step S620 shown in FIG. 6.

[0187] In some embodiments, the fourth electrocardiogram feature may be extracted by the convolutional neural network model 520. That is, after the user's multiple electrocardiogram waveforms are input to the convolutional neural network model 520, the fourth electrocardiogram feature may be extracted by a convolutional layer of the convolutional neural network model 520.

[0188] In step S1920, at least one sharpness result is calculated based on the fourth electrocardiogram feature. Step S1920 may be performed by the automatic calculation unit 408 of the computing device 400A shown in FIG. 4. In some embodiments, step S1920 may be performed subsequently after step S1910 is performed. Step S1920 may be a step of calculating a sharpness result based on an amplitude value (e.g., but not limited to, a maximum or minimum value) of the fourth electrocardiogram feature. More specifically, the sharpness result is calculated by performing a series of calculation operations on the fourth electrocardiogram feature.

[0189] In step S1930, the user's blood glucose level is estimated based on the peak distance and the sharpness result. Step S1930 may be executed by the estimation unit 410 of the computing device 400A shown in FIG. 4. In some embodiments, step S1930 may be executed subsequently after step S640 and step S1920 are executed. By executing step S1930, the user's blood glucose level may be estimated based on the peak distance calculation result and the sharpness calculation result. More specifically, the peak distance calculation result and the sharpness calculation result are input into an estimation model, and then the user's blood glucose level is estimated by the estimation model.

[0190] The method for non-invasively estimating blood glucose levels shown in Fig. 19 not only provides a user with a means for non-invasively estimating their blood glucose levels, but also allows the user's blood glucose levels to be more accurately estimated using the peak distance and sharpness results described above. Furthermore, the method for non-invasively estimating blood glucose levels shown in Fig. 19 can improve the technical field related to non-invasively estimating a user's blood glucose levels by utilizing multiple electrocardiogram waveforms.

[0191] Furthermore, according to the method for non-invasively estimating blood glucose levels shown in FIG. 19, when multiple electrocardiogram waveforms having inverted T waves are received, the sharpness results can be used to more accurately estimate the user's blood glucose level.

[0192] Also, in some embodiments, the method for non-invasively estimating blood glucose levels shown in FIG. 19 may further include step S910 shown in FIG. 9, and may calculate the user's corrected blood glucose level by performing step S910.

[0193] Please refer to Fig. 20. Fig. 20 is a detailed flowchart illustrating a method for calculating a sharpness result based on a fourth electrocardiogram feature according to an embodiment of the present disclosure. That is, step S1920 shown in Fig. 19 includes steps S2010, S2020, S2030, and S2040, and may be completed by performing steps S2010, S2020, S2030, and S2040, among which steps S2010, S2020, S2030, and S2040 may be performed by the calculation unit 408 of the computing device 400A shown in Fig. 4.

[0194] In step S2010, a fourth feature peak position corresponding to each fourth electrocardiographic feature is determined. In some embodiments, step S2010 may be executed subsequently after step S1910 is executed. In some embodiments, step S2010 may be substantially similar to step S630 shown in FIG. 6. In some embodiments, the fourth feature peak position may refer to the position of the maximum value of the P wave, the position of the minimum value of the Q wave, the position of the maximum value of the R wave, the position of the minimum value of the S wave, the position of the maximum absolute value of the T wave, and / or the position of the maximum absolute value of the U wave. In some embodiments, by executing step S2010, if the extracted fourth electrocardiographic feature is a T wave, the position of the maximum absolute value of the T wave of each of the multiple electrocardiographic waveforms (i.e., "T" shown in FIG. 21) may be determined. P ") may be determined respectively.

[0195] In step S2020, the first slope and the second slope are each calculated based on the fourth feature peak position. In some embodiments, step S2020 may be performed subsequently after step S2010 is performed.

[0196] The first slope is calculated based on the fourth feature peak position. More specifically, the first slope is calculated by the equation: In some embodiments, when the extracted fourth electrocardiographic feature is a T wave, the above equation is:

[0197]

number

[0198] where "SL1" is the first slope, "T0" is the amplitude value of the T wave position, and "N" is the number of sampling points. For example, "N" is 10.

[0199] The second slope is calculated based on the fourth feature peak position. More specifically, the second slope is calculated by the equation: In some embodiments, when the extracted fourth electrocardiographic feature is a T wave, the above equation is:

[0200]

number

[0201] where "SL2" is the second slope, "T0" is the amplitude value of the fourth characteristic peak position (i.e., "T P " amplitude value), and "N" is the number of sampling points. For example, "N" is 10.

[0202] In step S2030, a first sharpness is calculated based on the first slope. In some embodiments, step S2030 may be performed subsequently after step S2020 is performed. The first sharpness is calculated by an equation. In some embodiments, when the extracted fourth electrocardiographic feature is a T wave, the above equation is:

[0203]

number

[0204] where "SP1" is the first sharpness, "T0" is the amplitude value of the fourth feature peak position (i.e., "T P " is the amplitude value of "), "SL1" is the first slope, and "N" is the number of sampling points. For example, "N" is 10.

[0205] In step S2040, a second sharpness is calculated based on the second slope. In some embodiments, step S2040 may be performed subsequently after step S2030 is performed. Furthermore, step S2040 may be performed simultaneously with step S2030. The second sharpness is calculated by an equation. In some embodiments, when the extracted fourth electrocardiographic feature is a T wave, the above equation is:

[0206]

number

[0207] where "SP2" is the second sharpness, "T0" is the amplitude value of the fourth feature peak position (i.e., "T P " is the amplitude value of "), "SL2" is the second slope, and "N" is the number of sampling points. For example, "N" is 10.

[0208] The first sharpness and the second sharpness are regarded as the sharpness result, and therefore the sharpness result can be calculated by performing steps S2010, S2020, S2030, and S2040.

[0209] See Figure 21, which is a waveform diagram illustrating a method for calculating a sharpness result based on a fourth electrocardiogram feature, according to one embodiment of the present disclosure. In some embodiments, the sharpness result may be calculated by a derivative calculation and an integral calculation.

[0210] The first slope is calculated based on the position of the maximum absolute value of the T wave. More specifically, the first slope is calculated using a differential equation. The above-mentioned differential equation is

[0211]

number

[0212] where "SL1" is the first slope, and "T P" is the amplitude value of the fourth characteristic peak position (i.e., "T P " amplitude value), "T S " is the amplitude value of the starting point of the T wave.

[0213] The second slope is calculated based on the position of the maximum absolute value of the T wave. More specifically, the second slope is calculated using a differential equation. The above-mentioned differential equation is

[0214]

number

[0215] where "SL2" is the second slope, and "T P " is the amplitude value of the fourth characteristic peak position (i.e., "T P " amplitude value), "T e " is the amplitude value at the end of the T wave.

[0216] The first sharpness is calculated based on the position of the maximum absolute value of the T wave. More specifically, the first sharpness is calculated using an integral equation. The integral equation is as follows:

[0217]

number

[0218] where "SP1" is the first sharpness, "T P " is the amplitude value of the fourth characteristic peak position (i.e., "T P " amplitude value), "T S " is the amplitude value of the starting point of the T wave, and "ECG(x)" is multiple electrocardiogram waveforms.

[0219] The second sharpness is calculated based on the position of the maximum absolute value of the T wave. More specifically, the second sharpness is calculated using an integral equation. The integral equation is as follows:

[0220]

number

[0221] where "SP2" is the second sharpness, "T P " is the amplitude value of the fourth characteristic peak position (i.e., "T P " amplitude value), "T e " is the amplitude value at the end of the T wave, and "ECG(x)" is multiple electrocardiogram waveforms.

[0222] Please refer to Fig. 22A, which is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and sharpness results, according to one embodiment of the present disclosure.

[0223] In step S2210A, the peak distance and sharpness result are each normalized. In some embodiments, step S2210A may be performed subsequently after steps S640 and S1920 have been performed. By performing step S2210A, the calculated peak distance may be normalized to a corresponding value between a range (e.g., between 0 and 1), and the calculated sharpness result may be normalized to a corresponding value between a range (e.g., between -2,000 and 2,000).

[0224] In step S2220A, the normalized peak distance and the normalized sharpness result are input to the machine learning model 500. In some embodiments, step S2220A may be executed subsequently after step S2210A is executed. By executing step S2220A, the normalized peak distance and the normalized sharpness result may be input to the machine learning model 500, such as a decision tree algorithm, a random forest algorithm incorporating bagging, an extreme gradient boosting (XGBoost) algorithm, or a support vector machine algorithm.

[0225] In step S2230A, the user's blood glucose level is estimated by machine learning model 500. In some embodiments, step S2230A may be executed subsequently after step S2220A is executed. By executing step S2230A, the normalized peak distance and the normalized sharpness result are input into machine learning model 500, and then the user's blood glucose level may be estimated based on the normalized peak distance and the normalized sharpness result ratio. Since machine learning model 500 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by machine learning model 500 with a prediction accuracy of 0.80 or more, preferably 0.90 or more.

[0226] In step S2240A, the user's blood glucose level is output by machine learning model 500. In some embodiments, step S2240A may be performed subsequently after step S2230A is performed. In some embodiments, step S2240A is substantially the same as step S740A shown in FIG. 7A.

[0227] Please refer to Figure 22B, which is a detailed flowchart illustrating a method for estimating a user's blood glucose level based on peak distance and sharpness results, according to one embodiment of the present disclosure.

[0228] In step S2210B, the peak distance and sharpness results are each normalized. In some embodiments, step S2210B may be performed subsequently after steps S640 and S1920 have been performed. In some embodiments, step S2210B is substantially the same as step S2210A shown in FIG. 22A.

[0229] In step S2220B, the normalized peak distance and the normalized sharpness result are input to the neural network model 510. In some embodiments, step S2220B may be performed subsequently after step S2210B is performed. By performing step S2220B, the normalized peak distance and the normalized sharpness result may be input to the neural network model 510.

[0230] In step S2230B, the user's blood glucose level is estimated by neural network model 510. In some embodiments, step S2230B may be executed subsequently after step S2220B is executed. By executing step S2230B, the normalized peak distance and the normalized sharpness result are input into neural network model 510, and then the user's blood glucose level may be estimated based on the normalized peak distance and the normalized sharpness result. Since neural network model 510 is trained with a plurality of data and validated with prediction results, the user's blood glucose level can be estimated by neural network model 510 with a prediction accuracy of 0.80 or more, preferably 0.90 or more.

[0231] In S1840B, the user's blood glucose level is output by neural network model 510. In some embodiments, step S1840B may be performed subsequently after step S1830B is performed. In some embodiments, step S1840B is substantially the same as step S740B shown in FIG. 7B.

[0232] Please refer to Figure 23. Figure 23 is a schematic diagram illustrating another example of a computing device 400B electrically connected to the electrocardiogram measurement device 300 and electrically connected to at least one of the machine learning model 500, the neural network model 510, and the convolutional neural network model 520, according to an embodiment of the present disclosure. The computing device 400B has the same configuration and function as the computing device 400A shown in Figure 4 and may further include an evaluation unit 412, a counting unit 414, and an output unit 416.

[0233] The evaluation unit 412 may be configured to evaluate the quality of the plurality of electrocardiogram waveforms. That is, the computing device 400B may evaluate the quality of the received plurality of electrocardiogram waveforms by the evaluation unit 412. More specifically, the evaluation unit 412 may evaluate whether the plurality of electrocardiogram waveforms contain noise. If the result of the evaluation of the plurality of electrocardiogram waveforms for noise is less than a preset noise threshold, the quality of the plurality of electrocardiogram waveforms may be determined to be good electrocardiogram waveforms (i.e., normal electrocardiogram signals). Conversely, if the result of the evaluation of the plurality of electrocardiogram waveforms for noise is equal to or greater than the preset noise threshold, the quality of the plurality of electrocardiogram waveforms may be determined to be poor electrocardiogram waveforms (i.e., noisy). In some embodiments, the evaluation unit 412 may evaluate the quality of the received plurality of electrocardiogram waveforms by calculating a distribution of R-R interval values ​​between adjacent electrocardiogram waveforms, a duration of each of the plurality of electrocardiogram waveforms, a user's heart rate, a distribution of each of the plurality of electrocardiogram features, an amplitude value of each of the plurality of electrocardiogram features, and / or a similarity between the plurality of electrocardiogram waveforms.

[0234] In some embodiments, the quality of the multiple electrocardiogram waveforms may be determined to be good if the values ​​of the R-R intervals between adjacent electrocardiogram waveforms are slightly different from each other (e.g., within a 0.1 error rate) and the percentage of outliers is less than a preset value (e.g., 0.1). In some embodiments, the quality of the multiple electrocardiogram waveforms may be determined to be good if the durations of each of the multiple electrocardiogram waveforms are slightly different from each other (e.g., within a 0.1 error rate) and the percentage of outliers is less than a preset value (e.g., 0.1). In some embodiments, the quality of the multiple electrocardiogram waveforms may be determined to be good if the user's heart rate is within a normal range. In some embodiments, the quality of the multiple electrocardiogram waveforms may be determined to be good if the distributions of each of the multiple electrocardiogram features are slightly different from each other (e.g., within a 0.1 error rate of the time difference between the P wave and the R wave of each of the multiple electrocardiogram waveforms) and the percentage of outliers is less than a preset value (e.g., 0.1). In some examples, if the amplitude values ​​of each of the plurality of electrocardiogram features are slightly different from each other (e.g., within an error rate of 0.1) and the percentage of outliers is less than a preset value (e.g., 0.1), the quality of the plurality of electrocardiogram waveforms may be determined to be good. In some embodiments, if the similarity between the plurality of electrocardiogram waveforms is large (e.g., the similarity is greater than 0.9) and the percentage of outliers is less than a preset value (e.g., 0.1), the quality of the plurality of electrocardiogram waveforms may be determined to be good.

[0235] The counting unit 414 may be configured to count the number of received electrocardiogram waveforms of the user. That is, when the computing device 400B starts receiving electrocardiogram waveforms, the computing device 400B may start receiving amplitude values ​​of each point of the electrocardiogram waveforms, convert the amplitude values ​​of each point into corresponding waveforms, and calculate the number of corresponding waveforms.

[0236] In some embodiments, the counting unit 414 may be further configured to compare the number of corresponding waveforms (i.e., the received electrocardiogram waveforms) with a preset value to determine whether the computing device 400B has received a sufficient number of electrocardiogram waveforms. In some embodiments, the preset value may be set to, but is not limited to, 60 or more.

[0237] The output unit 416 may be configured to output the blood glucose level of the user. That is, the output unit 416 may output the blood glucose level of the user estimated by the estimation unit 410 and / or the corrected blood glucose level of the user calculated by the corrected blood glucose level calculator 508. In some embodiments, the blood glucose level of the user estimated by the estimation unit 410 and / or the corrected blood glucose level of the user calculated by the corrected blood glucose level calculator 508 may be synchronously stored in the blood glucose level database 454, so that the output unit 416 can output the blood glucose level stored in the blood glucose level database 454.

[0238] As a result, the computing device 400B provided by the present disclosure can be used to implement a method for non-invasively estimating blood glucose levels, so that the computing device 400B can receive a plurality of electrocardiogram waveforms of a user, and then estimate the user's blood glucose level and / or calculate the user's corrected blood glucose level based on the user's plurality of electrocardiogram waveforms.

[0239] Furthermore, the computing device 400B provided by the present disclosure can further evaluate the quality and quantity of the received electrocardiogram waveforms and ensure that the method for non-invasively estimating blood glucose levels is implemented with a certain number and / or quality of the electrocardiogram waveforms, thereby more accurately estimating the user's blood glucose level and / or calculating the user's corrected blood glucose level.

[0240] The computing device 400B shown in Fig. 23 can provide a user with a means for non-invasively estimating the user's blood glucose level, as well as more accurately estimating the user's blood glucose level. Furthermore, the computing device 400B shown in Fig. 23 can advance the technical field related to non-invasively estimating the user's blood glucose level by utilizing multiple electrocardiogram waveforms.

[0241] Please refer to Figure 24. Figure 24 is a flowchart illustrating a seventh example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 24 may include steps S610, S2410, S2420, and S2430, where step S610 is substantially the same as the step shown in Figure 6.

[0242] In step S2410, the quality of the plurality of electrocardiogram waveforms is evaluated. Step S2410 is performed by the evaluation unit 412 of the computing device 400B shown in FIG. 23. In some embodiments, step S2410 may be performed subsequently after step S610 is performed. By performing step S2410, the quality of the plurality of electrocardiogram waveforms received by the receiving unit 402 is evaluated.

[0243] If the plurality of electrocardiogram waveforms are evaluated as noise NS (i.e., the evaluation result for noise in the plurality of electrocardiogram waveforms is equal to or greater than a preset noise threshold), the user's past blood glucose level may be output (i.e., step S2420 is executed). If the plurality of electrocardiogram waveforms are evaluated as normal electrocardiogram signals NM (i.e., the evaluation result for noise in the plurality of electrocardiogram waveforms is less than a preset noise threshold), the user's real-time blood glucose level may be output (i.e., step S2430 is executed).

[0244] In step S2420, the user's past blood glucose levels are output. Step S2420 may be performed by the output unit 416 of the computing device 400B shown in Figure 23. In some embodiments, step S2420 may be performed subsequently after step S2410 is performed. By performing step S2420, the user's past blood glucose levels may be output.

[0245] In some embodiments, the method provided by the present disclosure estimates the user's blood glucose level over a certain period of time (e.g., every second), and when multiple electrocardiogram waveforms are evaluated as noise NS at time point A and step S2420 is performed, the user's past blood glucose level may be output, and the past blood glucose level may refer to the user's blood glucose level obtained before time point A (e.g., the blood glucose level estimated at time point A-1 or the blood glucose level estimated at the time when the electrocardiogram signal was last evaluated as normal NM). In some embodiments, the past blood glucose level may refer to the blood glucose level stored in the blood glucose level database 454, and the past blood glucose level may refer to the user's blood glucose level estimated by the estimator 410 and / or the user's corrected blood glucose level calculated by the corrected blood glucose level calculator 508.

[0246] In step S2430, the user's real-time blood glucose level is output. Step S2430 is performed by the output unit 416 of the computing device 400B shown in FIG. 23 . In some embodiments, step S2430 may be performed subsequently after step S2410 is performed. In some embodiments, the user's blood glucose level estimated by the estimation unit 410 and / or the user's corrected blood glucose level calculated by the corrected blood glucose level calculator 508 may be output in real time. In other embodiments, the user's blood glucose level synchronously stored in the blood glucose level database 454 may be output in real time.

[0247] The method for non-invasively estimating blood glucose levels shown in Figure 24 not only provides a user with a means for non-invasively estimating their blood glucose level, but also allows for more accurate estimation of their blood glucose level using multiple electrocardiogram waveforms of specific quality. That is, by implementing the method for non-invasively estimating blood glucose levels shown in Figure 24, it is possible to ensure that the method estimates a user's blood glucose level based on multiple electrocardiogram waveforms of specific quality (i.e., the estimated result of the user's blood glucose level is not significantly biased by multiple electrocardiogram waveforms with noise). Furthermore, the method for non-invasively estimating blood glucose levels shown in Figure 24 can advance the technical field related to non-invasively estimating a user's blood glucose level by utilizing multiple electrocardiogram waveforms.

[0248] Please refer to Figure 25. Figure 25 is a flowchart illustrating an eighth example of a method for non-invasively estimating blood glucose levels, according to one embodiment of the present disclosure. The method shown in Figure 25 may include steps S610, S620, S630, S640, S650, and S6510, where steps S610, S620, S630, S640, and S650 are substantially the same as the steps shown in Figure 6.

[0249] In step S2510, it is determined whether the number of the user's electrocardiogram waveforms is less than a preset value. Step S2510 may be performed by the counting unit 414 of the computing device 400B shown in FIG. 23. In some embodiments, step S2510 may be performed subsequently after step S610 is performed. Step S2510 may be performed to compare the number of received electrocardiogram waveforms with a preset value to determine whether the number of the user's electrocardiogram waveforms is less than the preset value.

[0250] If the number of received electrocardiogram waveforms is greater than or equal to the preset value (i.e., "NO" shown in FIG. 25), step S620 is executed to complete the next step in the method for non-invasively estimating blood glucose levels. If the number of received electrocardiogram waveforms is less than the preset value (i.e., "YES" shown in FIG. 25), multiple electrocardiogram waveforms of the user may be received again (i.e., return to step S610) until the number of received electrocardiogram waveforms exceeds the preset value.

[0251] The method for non-invasively estimating blood glucose levels shown in Figure 25 not only provides a user with a means for non-invasively estimating the user's blood glucose level, but also allows the user's blood glucose level to be more accurately estimated using a sufficient number of multiple electrocardiogram waveforms. That is, by implementing the method for non-invasively estimating blood glucose levels shown in Figure 25, it is possible to ensure that the method estimates the user's blood glucose level based on a sufficient number of multiple electrocardiogram waveforms (i.e., the estimation result of the user's blood glucose level will not be significantly biased due to a lack of multiple electrocardiogram waveforms). Furthermore, the method for non-invasively estimating blood glucose levels shown in Figure 25 can advance the technical field related to non-invasively estimating a user's blood glucose level by utilizing multiple electrocardiogram waveforms.

[0252] In some embodiments, the method for non-invasively estimating blood glucose levels estimates a user's blood glucose level based on at least one peak distance, at least one occurrence rate, at least one amplitude ratio, and a sharpness result, thereby not only providing a user with a means for non-invasively estimating the user's blood glucose level, but also enabling the user to more accurately estimate the user's blood glucose level by utilizing the peak distance, occurrence rate, amplitude ratio, and sharpness result. Furthermore, the method for non-invasively estimating blood glucose levels described above can advance the technical field related to non-invasively estimating a user's blood glucose level by utilizing multiple electrocardiogram waveforms.

[0253] In addition, in some embodiments, the above-mentioned method for non-invasively estimating blood glucose levels may further include step S910 shown in FIG. 9, and may calculate the user's corrected blood glucose level by performing step S910.

[0254] Please refer to Figure 26. Figure 26 is a schematic diagram illustrating an example of a computing device 2600 for non-invasively estimating blood glucose levels, electrically connected to at least one of an electrocardiogram sensor 310, an electrocardiogram monitoring device 320, and a server 330, and electrically connected to at least one of a display device 380 and a server 390, in accordance with an embodiment of the present disclosure.

[0255] Because the computing device 2600 is electrically connected to at least one of the electrocardiogram sensor 310, the electrocardiogram monitoring device 320, and the server 330, the computing device 2600 may receive the user's electrocardiogram waveforms from at least one of the electrocardiogram sensor 310, the electrocardiogram monitoring device 320, and the server 330. In some embodiments, the computing device 2600 may receive the user's electrocardiogram waveforms from at least one of the electrocardiogram sensor 310, the electrocardiogram monitoring device 320, and the server 330 via a physical signal line connection. For example, but not limited to, the physical signal line connection may be a network signal line connection compliant with the Internet Protocol (IP). Furthermore, in some embodiments, the computing device 2600 may receive the user's electrocardiogram waveforms from at least one of the electrocardiogram sensor 310, the electrocardiogram monitoring device 320, and the server 330 via a virtual signal line connection. For example, but not limited to, the virtual signal line connection may be a Wi-Fi connection compliant with a wireless network protocol.

[0256] In some embodiments, the computing device 2600 may be electrically connected to another device (e.g., a wearable measurement device) that can provide multiple electrocardiogram waveforms and receive multiple electrocardiogram waveforms of the user from the other device (not shown).

[0257] The computing device 2600 may include a storage module 2610 and a blood glucose level estimation module 2620 so as to perform specific signal processing on the received plurality of electrocardiogram waveforms. That is, after the computing device 2600 receives the plurality of electrocardiogram waveforms of a user, the computing device 2600 may estimate the user's blood glucose level and / or calculate the user's corrected blood glucose level based on the user's plurality of electrocardiogram waveforms, and output the user's estimated blood glucose level and / or the user's corrected blood glucose level.

[0258] Since the computing device 2600 is electrically connected to at least one of the display device 380 and the server 390, the computing device 2600 may output the user's estimated blood glucose value and / or the user's corrected blood glucose value to at least one of the display device 380 and the server 390.

[0259] In some embodiments, the computing device 2600 may be electrically connected to other devices that can receive, utilize, store and / or display the user's estimated blood glucose level and / or the user's corrected blood glucose level for subsequent use by the other devices (not shown).

[0260] The storage module 2610 may be configured to store a program, a set of program code, and / or a set of instructions. In some embodiments, the storage module 2610 may include, but is not limited to, one or more non-volatile memories and one or more volatile memories. The non-volatile memory may be, for example, but is not limited to, a read-only memory, a flash memory, or a non-volatile random access memory. The volatile memory may be, for example, but is not limited to, a dynamic random access memory or a static random access memory. In some embodiments, the storage module 2610, particularly the non-volatile memory, may store a program, a set of program code, and / or a set of instructions relating to a method for non-invasively estimating blood glucose levels as described above.

[0261] The blood glucose level estimation module 2620 may be configured to be electrically connected to the storage module 2610 and may be configured to estimate the user's blood glucose level based on the user's multiple electrocardiogram waveforms. That is, after the user's multiple electrocardiogram waveforms are received, the blood glucose level estimation module 2620 may perform the steps of any one of the methods for non-invasively estimating blood glucose levels as described above, so that the blood glucose level estimation module 2620 can estimate the user's blood glucose level based on the user's multiple electrocardiogram waveforms. In some embodiments, the blood glucose level estimation module 2620 may also calculate a corrected blood glucose level for the user.

[0262] In some embodiments, the blood glucose level estimation module 2620 may include, but is not limited to, one or more processors, which may be, for example, but are not limited to, a central processing unit.

[0263] The memory module 2610 stores a program, a set of program codes and / or a set of instructions relating to the method for non-invasively estimating blood glucose levels as described above, so that the blood glucose level estimation module 2620 can load and execute the program, code and / or instruction set, and then use the blood glucose level estimation module, particularly the processor, to realize any one of the methods for non-invasively estimating blood glucose levels as described above.

[0264] In some embodiments, the computing device 2600 may further include a signal output module (not shown), which may be configured to be electrically connected to the blood glucose level estimation module 2620 and configured to output the user's blood glucose level estimated by the blood glucose level estimation module 2620 and / or the user's corrected blood glucose level calculated by the blood glucose level estimation module 2620.

[0265] As a result, the computing device 2600 provided by the present disclosure can be used to perform any of the steps of the method for non-invasively estimating blood glucose levels as described above, and the computing device 2600 can estimate the user's blood glucose level based on the user's multiple electrocardiogram waveforms after the user's multiple electrocardiogram waveforms are received.

[0266] 26 , the computing device 2600 can provide a user with a means for non-invasively estimating the user's blood glucose level, as well as more accurately estimating the user's blood glucose level. Furthermore, the computing device 2600 can advance the technical field of non-invasively estimating a user's blood glucose level by utilizing multiple electrocardiogram waveforms.

[0267] See Figure 27, which is a schematic diagram illustrating an example of a device 2700 for non-invasively measuring electrocardiogram signals electrically connected to at least one of computing device 2200, computing device 400A, computing device 400B, and server 350, in accordance with an embodiment of the present disclosure.

[0268] Device 2700 may be signally connected to at least one of computing device 2600, computing device 400A, computing device 400B, and server 350 such that device 2700 may provide the user's electrocardiogram waveforms to at least one of computing device 2600, computing device 400A, computing device 400B, and server 350. In some embodiments, device 2700 may provide the user's electrocardiogram waveforms to at least one of computing device 2600, computing device 400A, computing device 400B, and server 350 via a physical signal line connection. For example, but not limited to, the physical signal line connection may be an Internet Protocol (IP)-compliant network signal line connection. Furthermore, in some embodiments, device 2700 may provide the user's electrocardiogram waveforms to at least one of computing device 2600, computing device 400A, computing device 400B, and server 350. For example, the virtual signal line connection may be, but is not limited to, a Wi-Fi connection that complies with a wireless network protocol.

[0269] The device 2700 includes a measurement module 2710 and a signal transmission module 2720, and the device 2700 can provide a plurality of electrocardiogram waveforms of the user to at least one of the computing device 2600, the computing device 400A, the computing device 400B, and the server 350, thereby causing at least one of the computing device 2600, the computing device 400A, the computing device 400B, and the server 350 to perform any step of the above-described method for non-invasively estimating blood glucose levels to estimate the user's blood glucose level based on the plurality of electrocardiogram waveforms measured by the device 2700.

[0270] The measurement module 2710 may be configured with a plurality of electrodes electrically connected to a user and configured to measure a plurality of electrocardiogram waveforms of the user. Because the cardiac depolarization process causes voltage changes on the body surface, the voltage changes on the body surface can be measured by a plurality of electrodes electrically connected to the skin surface of the user to obtain a plurality of electrocardiogram waveforms of the user.

[0271] The signal transmission module 2720 may be configured to be electrically connected to the measurement module 2710 and to transmit a plurality of electrocardiogram waveforms of the user to at least one of the computing device 2600, the computing device 400A, the computing device 400B, and the server 350, and to cause at least one of the computing device 2600, the computing device 400A, the computing device 400B, and the server 350 to perform the steps of any one of the methods for non-invasively estimating blood glucose levels as described above.

[0272] As a result, the device 2700 provided by the present disclosure is used to transmit the measured electrocardiogram waveforms to at least one of the computing device 2600, the computing device 400A, the computing device 400B, and the server 350, and at least one of the computing device 2600, the computing device 400A, the computing device 400B, and the server 350 can estimate the user's blood glucose level based on the measured electrocardiogram waveforms by the device 2700.

[0273] Please refer to Figure 28. Figure 28 is a schematic diagram illustrating that a user's blood glucose level is estimated based on multiple electrocardiogram waveforms received from the user, according to one embodiment of the present disclosure.

[0274] The method for non-invasively estimating a blood glucose level may be realized by the computing device described above. Specifically, the computing device described above can estimate a user's blood glucose level based on the user's multiple electrocardiogram waveforms, particularly the peak distance. For example, the peak distance may be calculated based on the multiple electrocardiogram waveforms shown in FIG. 28, and the user's blood glucose level may be estimated based on the peak distance. As shown in FIG. 28, the peak distance may be input into a machine learning model, i.e., an extreme gradient boosting (XGBoost) algorithm, and the user's blood glucose level may be estimated by the machine learning model. As shown in FIG. 28, the estimated user's blood glucose level is 124 mg / dL, and in this case, the user's actual blood glucose level measured by the blood glucose meter is 128 mg / dL. Since the estimated blood glucose level is close to the actual blood glucose level, the present disclosure not only provides a user with a means for non-invasively estimating the user's blood glucose level, but also more accurately estimates the user's blood glucose level by utilizing multiple electrocardiogram waveforms, particularly the peak distance described above.

[0275] Please refer to Figure 29A, which is a schematic diagram showing the comparison results of blood glucose levels estimated according to the present disclosure compared with other devices.

[0276] As shown in FIG. 29A, "UUID:48" is the subject's identification (ID), "CGM MARD" is the error between the true blood glucose level and the blood glucose level monitored by another device, and "XGB MARD" is the error between the true blood glucose level and the blood glucose level estimated by the device provided by the present disclosure. The comparison results show that the estimated results of "XGB MARD" are more accurate than the monitored results of "CGM MARD". In some embodiments, "MARD" is:

[0277]

number

[0278] where "BGi" is the estimated blood glucose level, "Compi" is the true blood glucose level, and "N" is the number of sampling points. For example, if the estimated blood glucose level is {105,95,85} and the true blood glucose level is {100,90,90}, then "MARD" is calculated, and the calculated result of "MARD" is 5.37%.

[0279]

number

[0280] See Figure 29B, which is a schematic diagram showing another comparison result of blood glucose levels estimated according to the present disclosure compared with other devices.

[0281] 29B, "UUID:9" is the identity (ID) of the subject (i.e., another subject), "CGM MARD" is another device for monitoring the user's blood glucose level, and "XGB MARD" is a device provided by the present disclosure. The comparison results also show that the estimated results of "XGB MARD" are more accurate than the monitored results of "CGM MARD."

[0282] In some embodiments, the steps of the method for non-invasively estimating blood glucose levels described above may be stored as a series of specific program codes or a series of specific instruction sets in a non-transitory computer-readable recording medium, which may be, but is not limited to, a hard disk, a compact disk (CD-ROM), a magnetic disk, or a flash drive (USB), and the non-transitory computer-readable recording medium enables the method for non-invasively estimating blood glucose levels described above when the program code or instruction sets stored in the non-transitory computer-readable recording medium are loaded and executed by a computer.

[0283] Although the present disclosure has been described by means of specific embodiments, modifications and changes may be made by those skilled in the art to which the present disclosure pertains without departing from the scope and spirit of the present disclosure as set forth in the claims. Therefore, the scope of protection of the present application should be limited by the claims, rather than by the content disclosed in the specification. [Explanation of symbols]

[0284] 300 Electrocardiogram measuring device 310 Electrocardiogram Sensor 320 Electrocardiogram Monitoring Device 330 Server 350 servers 380 display devices 390 Server 400A, 400B Computing Devices 402 Receiver 404 Extraction part 406 Decision Section 408 Calculation Unit 410 Estimation Department 412 Evaluation Department 414 Counting Unit 416 Output section 452 Electrocardiogram Waveform Database 454 Blood Glucose Level Database 500 machine learning models 502 Peak Distance Calculator 504 Incidence Calculator 506 Amplitude ratio calculator 508 Corrected Blood Glucose Calculator 510 Neural Network Model 520 Convolutional Neural Network Model 2600 Computing Devices 2610 Storage Module 2620 Blood Glucose Level Estimation Module 2700 devices 2710 Measurement Module 2720 ​​Signal Transmitting Module BGV blood sugar level CL convolutional layer EF ECG characteristics EW electrocardiogram waveform HL hidden layer IL input layer NM Normal ECG signal NS Noise OL output layer S610, S620, S630 Step S640,S650 Step S710A, S710B Step S720A, S720B Step S730A, S730B Step S740A, S740B Step S910 Step S1010, S1020 steps S1110, S1120, S1130 steps S1210, S1220, S1230 steps S1240, S1250, S1260 Step S1410A, S1410B Step S1420A, S1420B Step S1430A, S1430B Step S1440A, S1440B Step S1510, S1520, S1530 Step S1610, S1620 Step S1630, S1640 Step S1810A, S1810B Step S1820A, S1820B Step S1830A, S1830B Step S1840A, S1840B Step S1910, S1920, S1930 steps S2010, S2020 Step S2030, S2040 steps S2210A, S2210B Step S2220A, S2220B Step S2230A, S2230B Step S2240A, S2240B Step S2410, S2420, S2430 Step S2510 Step

Claims

1. 1. A method for non-invasively estimating blood glucose levels, estimating a blood glucose level of a user by a computing device, comprising: receiving a plurality of electrocardiogram (ECG) waveforms of the user; extracting at least two first electrocardiographic features and at least one fourth electrocardiographic feature from each of a plurality of electrocardiographic waveforms of the user; determining a first feature peak location corresponding to each of the first electrocardiogram features; calculating at least one peak distance between first feature peak positions for each of the plurality of electrocardiogram waveforms; calculating at least one sharpness result based on the at least one fourth electrocardiographic feature; estimating a blood glucose level of a user based on the at least one peak distance and the at least one sharpness result; Including, The method for non-invasively estimating blood glucose levels, wherein the first electrocardiogram features are selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves, and the at least one fourth electrocardiogram feature is selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves, and the at least one peak distance is a time difference.

2. estimating the blood glucose level of the user based on the at least one peak distance includes: normalizing the at least one peak distance; inputting the normalized at least one peak distance into a machine learning model; outputting the blood glucose level of the user by the machine learning model; Including, The method for non-invasively estimating blood glucose levels of claim 1 , wherein the blood glucose level of the user is estimated by the machine learning model based on a normalized distance of the at least one peak.

3. estimating the blood glucose level of the user based on the at least one peak distance includes: normalizing the at least one peak distance; inputting the normalized at least one peak distance into a neural network model; outputting the blood glucose level of the user by the neural network model; Including, The method for non-invasively estimating blood glucose levels of claim 1 , wherein the blood glucose level of the user is estimated by the neural network model based on a normalized distance of the at least one peak.

4. Calculating a corrected blood glucose level of the user using an equation for correcting blood glucose levels based on the blood glucose level of the user; 10. The method for non-invasively estimating blood glucose levels of claim 1, further comprising:

5. calculating at least one peak-to-peak slope between the first characteristic peak positions for each of the plurality of electrocardiogram waveforms; further comprising The method for non-invasively estimating blood glucose levels of claim 1 , wherein the estimation of the user's blood glucose level is further based on the at least one peak-to-peak slope.

6. extracting at least three second electrocardiographic features from each of the plurality of electrocardiographic waveforms of the user; calculating at least one incidence rate based on these second electrocardiographic features; further comprising the estimation of the user's blood glucose level is further based on the at least one incidence rate; and 2. The method for non-invasively estimating blood glucose levels of claim 1, wherein these second electrocardiographic features are selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

7. Calculating the at least one incidence rate based on these second electrocardiogram features includes: determining an interval between occurrences of each of a plurality of electrocardiogram waveforms of the user; calculating a first peak-to-peak and a second peak-to-peak within the occurrence interval for each of the plurality of electrocardiogram waveforms based on the second electrocardiogram features; calculating a peak-to-peak ratio of the first peak-to-peak to the second peak-to-peak for each of the plurality of electrocardiogram waveforms; comparing the peak-to-peak ratio with a predetermined ratio for each of the plurality of electrocardiogram waveforms to generate a comparison result; calculating the at least one incidence rate based on the results of these comparisons; and 7. The method for non-invasively estimating blood glucose levels according to claim 6, comprising:

8. estimating the blood glucose level of the user normalizing the at least one peak distance; inputting the normalized at least one peak distance and the at least one incidence rate into a machine learning model; outputting the blood glucose level of the user by the machine learning model; Including, 7. The method for non-invasively estimating blood glucose levels of claim 6, wherein the blood glucose level of the user is estimated by the machine learning model based on a normalized distance of the at least one peak and the at least one incidence rate.

9. estimating the blood glucose level of the user normalizing the at least one peak distance; inputting the normalized at least one peak distance and the at least one incidence rate into a neural network model; outputting the blood glucose level of the user by the neural network model; Including, 7. The method for non-invasively estimating blood glucose levels of claim 6, wherein the blood glucose level of the user is estimated by the neural network model based on the normalized at least one peak distance and the at least one occurrence rate.

10. extracting at least four third electrocardiographic features from each of a plurality of electrocardiographic waveforms of the user; calculating at least one amplitude ratio based on these third electrocardiographic features; further comprising the estimation of the user's blood glucose level is further based on the at least one amplitude ratio; and 2. The method for non-invasively estimating blood glucose levels according to claim 1, wherein the third electrocardiographic features are selected from the group consisting of P waves, Q waves, R waves, S waves, T waves, and U waves.

11. Calculating the at least one amplitude ratio based on these third electrocardiographic features includes: calculating a third peak-to-peak, a fourth peak-to-peak, a fifth peak-to-peak, and a sixth peak-to-peak for each of the plurality of electrocardiogram waveforms based on the third electrocardiogram features; calculating a first average value of the third peak-to-peak and the fourth peak-to-peak; calculating a second average value of the fifth peak-to-peak and the sixth peak-to-peak; calculating the at least one amplitude ratio of the first average value to the second average value; 11. The method for non-invasively estimating blood glucose levels according to claim 10, comprising:

12. estimating the blood glucose level of the user normalizing the at least one peak distance and the at least one amplitude ratio, respectively; inputting the normalized at least one peak distance and the normalized at least one amplitude ratio into a machine learning model; outputting the blood glucose level of the user by the machine learning model; Including, 11. The method for non-invasively estimating blood glucose levels of claim 10, wherein the blood glucose level of the user is estimated by the machine learning model based on a normalized peak distance and a normalized amplitude ratio.

13. estimating the blood glucose level of the user normalizing the at least one peak distance and the at least one amplitude ratio, respectively; inputting the normalized at least one peak distance and the normalized at least one amplitude ratio into a neural network model; outputting the blood glucose level of the user by the neural network model; Including, 11. The method for non-invasively estimating blood glucose levels of claim 10, wherein the blood glucose level of the user is estimated by the neural network model based on a normalized peak distance and a normalized amplitude ratio.

14. Calculating the at least one sharpness result based on the at least one fourth electrocardiographic feature includes: determining a fourth feature peak location corresponding to each of the at least one fourth electrocardiographic feature; calculating a first slope and a second slope based on the fourth characteristic peak position; calculating a first sharpness based on the first slope; calculating a second sharpness based on the second slope; 2. The method for non-invasively estimating blood glucose levels according to claim 1, comprising:

15. estimating the blood glucose level of the user normalizing the at least one peak distance and the at least one sharpness result, respectively; inputting the normalized at least one peak distance and the normalized at least one sharpness result into a machine learning model; outputting the blood glucose level of the user by the machine learning model; Including, 2. The method for non-invasively estimating blood glucose levels of claim 1, wherein the blood glucose level of the user is estimated by the machine learning model based on a normalized peak distance and a normalized sharpness result.

16. estimating the blood glucose level of the user normalizing the at least one peak distance and the at least one sharpness result, respectively; inputting the normalized at least one peak distance and the normalized at least one sharpness result into a neural network model; outputting the blood glucose level of the user by the neural network model; Including, 2. The method for non-invasively estimating blood glucose levels of claim 1, wherein the blood glucose level of the user is estimated by the neural network model based on a normalized peak distance and a normalized sharpness result.

17. assessing the quality of a plurality of electrocardiogram waveforms of the user; outputting a real-time blood glucose level of the user when the plurality of electrocardiogram waveforms of the user are evaluated as normal electrocardiogram signals; outputting the user's past blood glucose levels when the user's electrocardiogram waveforms are evaluated as noise signals; further comprising the user's real-time blood glucose level is a blood glucose level estimated in real time; and The method for non-invasively estimating blood glucose levels according to claim 1 , wherein the user's past blood glucose levels are previously estimated blood glucose levels.

18. determining whether the number of the plurality of electrocardiogram waveforms of the user is less than a predetermined value; re-receiving the plurality of electrocardiogram waveforms of the user if the number of the plurality of electrocardiogram waveforms of the user is less than a predetermined value; 10. The method for non-invasively estimating blood glucose levels of claim 1, further comprising:

19. 1. A computing device for non-invasively estimating blood glucose levels, the computing device being signally connected to an electrocardiogram (ECG) measurement device to receive a plurality of electrocardiogram waveforms of a user from the ECG measurement device, the computing device comprising: a storage module; a blood glucose level estimation module configured to be signally connected to the storage module; Including, A plurality of program codes are stored in the storage module; and 20. A computing device for non-invasively estimating blood glucose levels, wherein after the blood glucose level estimation module executes a plurality of program codes stored in the storage module, the blood glucose level estimation module performs the steps of the method for non-invasively estimating blood glucose levels described in any one of claims 1 to 18.

20. 1. A device for non-invasively measuring electrocardiogram signals, electrically connected to a computing device and outputting a plurality of electrocardiogram waveforms of a user to the computing device, comprising: a measurement module having a plurality of electrodes electrically connected to the user; a signal transmission module configured to be electrically connected to the measurement module; Including, the measurement module is configured to measure a plurality of electrocardiogram waveforms of the user; and 19. A device for non-invasively measuring electrocardiogram (ECG) signals, wherein the signal transmission module is configured to transmit a plurality of electrocardiogram waveforms of the user to the computing device, and the computing device is configured to perform the steps of the method for non-invasively estimating blood glucose levels according to any one of claims 1 to 18.

21. 20. A non-transitory computer readable recording medium having program code recorded thereon, the program code being executed by a computing device to cause the device to perform the method for non-invasively estimating blood glucose according to any one of claims 1 to 18.

Images