Method and apparatus for predicting blood compound concentration of a target

Abstract

The present invention relates to a method and apparatus for predicting the blood compound concentration of a target, comprising the steps of: receiving spectral data including data instances collected at multiple time intervals related to a target region using near-infrared spectroscopy by a blood compound monitoring device; classifying each data instance of the spectral data into one of a plurality of labeled classes predetermined using historical spectral data related to a plurality of learning targets; obtaining one or more optimal models from a plurality of prediction models corresponding to the plurality of learning targets based on the classification; determining a blood compound concentration value corresponding to each of the plurality of optimal models; and predicting the blood compound concentration of the target using a weighted average of the blood compound concentration values.

Description

Technology Field

[0001] The invention relates to health monitoring, and more specifically, to a method and apparatus for predicting the concentration of a target compound in the blood. Background Technology

[0002] Using spectroscopy, in particular, as a non-invasive technique for measuring parameters in the bodies of animals or humans is known to be a highly desirable method.

[0003] Current non-invasive methods for monitoring or measuring metabolic parameters include polarimetry, mid-infrared spectroscopy, Raman spectroscopy, chromoscopy, fluorescence spectroscopy, nuclear magnetic resonance spectroscopy, radio-frequency spectroscopy, ultrasound, transdermal measurements, and photoacoustic spectroscopy. In particular, near-infrared (NIR) spectroscopy is currently a widely studied and preferred method for the non-invasive monitoring of glucose levels.

[0004] Conventional methods transmit electromagnetic radiation of a specific wavelength through the skin to a target measurement area, collect and analyze the light, and compare it with a stored reference calibration value or curve. However, since the calibration value or curve varies from person to person or animal to animal, the reference calibration value or curve used to predict glucose concentration may not be realistic, and the metabolic parameter values ​​obtained or predicted in this way may not be accurate. Therefore, there is an urgent need for a method and device that can accurately predict blood glucose concentration by selecting the best prediction model. Prior art literature

[65535] US 2015 / 0220702 (Published: 2015.08.06) The problem to be solved

[0005] The purpose is to provide a method and apparatus for predicting the blood compound concentration of a target for health monitoring. means of solving the problem

[0006] A blood compound monitoring device according to one embodiment for achieving the above objective, in a method for predicting a blood compound concentration, comprises: receiving spectral data including a plurality of data instances collected at a plurality of time intervals related to a target region using near-infrared spectroscopy; classifying each data instance of the spectral data into one of a plurality of predetermined labeled classes using historical spectral data related to one of a plurality of training subjects; obtaining one or more optimal models from a plurality of prediction models corresponding to the plurality of training subjects based on the classification; determining a blood compound concentration value corresponding to each of the plurality of optimal models; and predicting the blood compound concentration of the target using a weighted average of the blood compound concentration values.

[0007] At this time, a plurality of prediction models are obtained as machine learning regression models trained using reference blood compound concentration values ​​for the historical spectrum data related to the learning target.

[0008] At this time, each of the plurality of labeled classes represents one of the plurality of learning targets.

[0009] At this time, the step of obtaining one or more optimal models comprises: determining, for each labeled class, the number of data instances of the spectrum data to which the corresponding labeled class is assigned; identifying one or more labeled classes having the largest number of data instances; and obtaining a prediction model corresponding to the one or more labeled classes as one or more optimal models.

[0010] At this time, the above prediction model corresponds to a model that establishes a relationship or association between the above historical spectrum data and the above blood compound concentration value.

[0011] At this time, the spectrum data is obtained by near-infrared spectroscopy by removing abnormal data from the spectrum data using an anomaly detection technique.

[0012] At this time, the target corresponds to a human or animal in which the concentration value of the compound in the blood is predicted.

[0013] In addition, a blood compound monitoring device for predicting the blood compound concentration of a target according to one embodiment comprises a processor; a memory connected to communicate with the processor and storing processor-executable instructions, wherein when executed, the processor receives spectrum data including a plurality of data instances collected at a plurality of time intervals related to a target region using near-infrared spectroscopy, classifies each of the plurality of data instances of the spectrum data into one of a plurality of predetermined labeled classes using historical spectrum data related to one of a plurality of learning targets, obtains one or more optimal models from a plurality of prediction models corresponding to the plurality of learning targets based on the classification, determines a blood compound concentration value corresponding to each of the plurality of optimal models, and predicts the blood compound concentration of the target using the weighted average of the blood compound concentration values.

[0014] At this time, the processor obtains each of the plurality of prediction models as machine learning regression models learned using reference blood compound concentration values ​​for the historical spectrum data related to the learning target.

[0015] At this time, each of the plurality of labeled classes represents one of the plurality of learning targets.

[0016] At this time, for each labeled class, the processor determines the number of data instances of spectrum data to which the corresponding labeled class is assigned, identifies one or more labeled classes having the largest number of data instances, and obtains one or more optimal models by obtaining a prediction model corresponding to the one or more labeled classes as one or more optimal models.

[0017] At this time, the above prediction model corresponds to a model that establishes a relationship or association between the above historical spectrum data and the above blood compound concentration value.

[0018] At this time, the spectrum data is obtained by near-infrared spectroscopy by removing abnormal data from the spectrum data using anomaly detection technology.

[0019] At this time, the target corresponds to a human or animal in which the concentration value of the blood compound is predicted. Effects of the invention

[0020] It is possible to accurately predict the concentration of a target's blood compound using a specific independent prediction model based on the target. Brief explanation of the drawing

[0021] FIG. 1 is an exemplary environment including a device for predicting the blood compound concentration of a target according to one embodiment. FIG. 2a is a block diagram of a blood compound monitoring device according to one embodiment. FIGS. 2b to 2d are graphs of dimensionally reduced data according to one embodiment. FIGS. 2e and FIGS. 2f are graphs of prediction results according to one embodiment. FIG. 3 is a flowchart illustrating a method for predicting the blood compound concentration of a target according to one embodiment. FIG. 4 is a block diagram of a computing device for implementing embodiments. Specific details for implementing the invention

[0022] Specific details of other embodiments are included in the detailed description and drawings. The advantages and features of the described technology and the methods for achieving them will become clear by referring to the embodiments described in detail below together with the drawings. Throughout the specification, the same reference numerals refer to the same components.

[0023] Terms such as "first," "second," etc., may be used to describe various components, but the components should not be limited by these terms. Terms are used solely for the purpose of distinguishing one component from another. A singular expression includes a plural expression unless the context clearly indicates otherwise. Furthermore, when a part is described as "comprising" a component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Additionally, terms such as "part," "module," etc., as used in the specification refer to a unit that performs at least one function or operation, which may be implemented in hardware or software, or as a combination of hardware and software.

[0024] In addition, as is customary in the art, functions may be performed by blocks in the embodiments, and the blocks referred to as "parts" or "modules" are physically implemented by analog or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, and circuits embedded in hardware, and may optionally be driven by firmware and software.

[0025] Additionally, the circuit may be implemented on one or more semiconductor chips or on a substrate support such as a printed circuit board. The circuit constituting the block may be implemented by dedicated hardware or a processor, or by a combination of dedicated hardware to perform some functions of the block and a processor to perform other functions of the block.

[0026] The present invention relates to a method for predicting the concentration of a target's blood compound and a blood compound monitoring device.

[0027] The target subjects correspond to humans or animals whose blood compound concentration values ​​are predicted by a blood compound monitoring device.

[0028] Spectral data related to the target body region of the learning subject can be measured by a spectrometer using near-infrared spectroscopy.

[0029] The learning subject may be a human or animal that accommodates various observations to monitor and collect historical spectrum data related to the learning subject.

[0030] Spectral data is measured or collected at multiple time intervals as multiple data instances, and each data instance of spectral data is classified into one of multiple labeled classes.

[0031] The labeled class is determined in advance using historical spectrum data related to the learning target and historical blood compound concentration values ​​related to the historical spectrum data.

[0032] Based on classification, a predefined number of optimal models are determined from multiple prediction models corresponding to the learning target.

[0033] In addition, blood compound concentration values ​​are determined using each best-fit model, and the blood compound concentration of the target can be predicted by calculating the weighted average of the blood compound concentration values.

[0034] Hereinafter, a method and apparatus for predicting the blood compound concentration of a target according to one embodiment will be described in detail with reference to the attached drawings.

[0035] FIG. 1 illustrates an exemplary environment (100) including a blood compound monitoring device (101) for predicting the blood compound concentration of a target according to one embodiment.

[0036] The exemplary environment (100) illustrated in FIG. 1 may include a blood compound monitoring device (101), a spectrum data storage (105), a blood compound value storage (115), and a data storage (111), each of which is connected to a communication network (103). The communication network (103) may include, but is not limited to, a direct interconnection, a local area network (LAN), a wide area network (WAN), a wireless network (e.g., using a wireless application protocol), the internet, etc.

[0037] The spectrum data storage (105) can store historical spectrum data (107) related to multiple learning targets. The historical spectrum data (107) includes spectrum data measured and collected for various learning targets using near-infrared spectroscopy.

[0038] Historical spectrum data (107) can be measured by a spectrometer (109). The spectrometer (109) may be any measuring device capable of collecting spectrum data (107) from a target body region of a learning subject in a specific wavelength range.

[0039] The spectrum data storage (105) can be integrated with the blood compound monitoring device (101) and can also be configured as a standalone device independent of the blood compound monitoring device (101).

[0040] The blood compound value storage (115) can store multiple blood compound concentration values ​​(117) associated with multiple learning targets. The blood compound concentration values ​​(117) can be measured by a blood compound measuring device (119). The blood compound concentration values ​​(117) can be measured for multiple learning targets and stored in the blood compound value storage (115).

[0041] The blood compound value storage (115) can be integrated with the blood compound monitoring device (101) and can also be configured as a standalone device independent of the blood compound monitoring device (101).

[0042] The spectrum data storage (105) and the blood compound value storage (115) may be configured as a single storage that stores all historical data related to the learning target and may be combined with the blood compound monitoring device (101).

[0043] The data storage (111) can store a prediction model (113) related to multiple learning targets. The prediction model (113) is a model that establishes a relationship or association between historical spectrum data (107) measured by a spectrometer (109) and blood compound concentration values ​​(117) measured by a blood compound measuring device (119).

[0044] The data storage (111) can be integrated with the blood compound monitoring device (101) and can also be configured as a standalone device independent of the blood compound monitoring device (101).

[0045] The spectrum data storage (105), blood compound value storage (115), and data storage (111) may be composed of a single storage combined with the blood compound monitoring device (101).

[0046] FIG. 2a is a detailed block diagram of a blood compound monitoring device according to one embodiment.

[0047] A blood compound monitoring device (101) may include a processor (201), an input / output (I / O) interface (203), and a memory (205). The input / output interface (203) may receive historical spectrum data (107) from a spectrum data storage (105) associated with the blood compound monitoring device (101). Additionally, the input / output interface (203) may receive a prediction model (113) and a blood compound concentration value (117), respectively, from a data storage (111) and a blood compound value storage (115) associated with the blood compound monitoring device (101). The memory (205) may be communicably connected to the processor (201) and may store data (207) and one or more modules (209). Additionally, the processor (201) may perform various functions of the blood compound monitoring device (101) using the data (207) and one or more modules (209).

[0048] Data (207) may include spectrum data (107), labeled classes (211), optimal models (213), blood compound concentration values ​​(117), and other data (115). Data (207) may be stored in memory (205) in the form of various data structures and may be configured using data models such as relational or hierarchical data models. Other data (215) may store various temporary data and files generated by one or more modules (209) while performing various functions of the blood compound monitoring device (101). Additionally, other data (215) may include multiple prediction models (113) and current and historical spectrum data (107) of the target.

[0049] Data stored in memory (205) can be processed by a module (209) of the blood compound monitoring device (101). The module (209) may be stored within the memory (205), may exist externally, or may be connected communicably to the processor (201) as an independent hardware configuration. In this specification, a module may refer to an Application Specific Integrated Circuit (ASIC), electronic circuit, or processor (shared, dedicated, or group) executing one or more software or firmware programs, combinational logic circuits, and / or other suitable components. Additionally, the module (209) may include, for example, a classification module (217), a best model determination module (219), a concentration prediction module (221), and other modules (223). Other modules (223) may be used to perform various other functions of the blood compound monitoring device (101). Additionally, the aforementioned modules may be a single module or a combination of different modules. One or more modules may be stored in memory (205) and may be new hardware.

[0050] The classification module (217) can classify each of a plurality of data instances of spectrum data of a target into one of a plurality of labeled classes (211). The plurality of data instances may correspond to spectrum data values ​​collected at a plurality of time intervals, and each of the plurality of labeled classes (211) may correspond to and represent one of a plurality of learning targets.

[0051] The optimal model determination module (219) can determine one or more optimal models from a plurality of prediction models (113) based on classification. For example, the optimal model may be a prediction model that includes the labeled class having the largest number of data instances.

[0052] The concentration prediction module (221) can calculate a blood compound concentration value (117) for each of a plurality of optimal models, and can determine the blood compound concentration value of a target using the calculated blood compound concentration value (117).

[0053] A spectrometer (109) associated with a blood compound monitoring device (101) can capture historical spectral data (107) associated with a target body region of a learning subject using near-infrared spectroscopy. The spectral data (107) may include multiple data instances collected at multiple time intervals, and when the spectral data (107) is collected, it can be transmitted to the blood compound monitoring device (101). The blood compound monitoring device (101) can classify the spectral data (107) into one of a plurality of labeled classes (211) using a classification module (217). The plurality of labeled classes (211) may correspond to a plurality of learning subjects and may be the output of a multi-class classifier. The plurality of labeled classes (211) may be predetermined using the historical spectral data (107) associated with a plurality of learning subjects. Historical spectrum data (107) for multiple targets and historical blood compound values ​​can be obtained from a spectrum data repository (105) and a blood compound value repository (115) associated with a blood compound monitoring device (101).

[0054] When the classification module (217) receives multiple data instances of spectrum data (107), it can classify each data instance into one of the labeled classes (211) using multi-class classification. Initially, the classification module (217) can determine relationship data between multiple historical spectrum data (107) and multiple historical blood compound values. Additionally, the classification module (217) can label each relationship data into a labeled class, and each of the labeled classes (211) corresponds to a single learning target. One or more anomalies of the multiple historical spectrum data (107) can be identified using a Gaussian distribution. The data store (111) can be updated by removing the historical spectrum data (107) associated with one or more anomalies and the labeled class associated with the historical spectrum data (107).

[0055] When classifying each data instance of spectrum data (107) into one of the labeled classes (211), the optimal model detection module can determine multiple optimal models from multiple prediction models (113). The prediction models (113) can be determined in advance for each learning target and can be stored in a data repository (111). Additionally, each prediction model (113) can be obtained as a machine learning regression model that is learned using a reference blood compound concentration value (117) for the historical spectrum data (107) related to the learning target. The optimal model determination module (219) can determine the number of data instances of spectrum data (107) to which the corresponding labeled class is assigned for each of the multiple labeled classes (211). Additionally, the optimal model determination module (219) can identify one or more labeled classes (211) that have the largest number of data instances. Consequently, the optimal model determination module (219) can obtain one or more prediction models (113) corresponding to one or more identified labeled classes (211) as one or more optimal models.

[0056] The concentration prediction module (221) can calculate the blood compound concentration value (117) using each determined optimal model. Additionally, the concentration prediction module (221) can determine the final blood compound concentration of the target by calculating the weighted average of the blood compound concentration values ​​(117).

[0057] In one embodiment, the step of predicting the blood compound concentration of the target is as follows.

[0058] 1. Acquire Prediction Model

[0059] Assuming there are 'N' targets in the training set, multiple prediction models from the data store (111) This can be searched. 'k' can be a value between 1 and N. As an example, the prediction model used may be a Partial Least Squares (PLS) regression model. Additionally, if a new training target exists, multiple prediction models based on multi-class classification One of them can be assigned to a new learning target.

[0060] 2. Obtaining an optimal model for each of multiple targets

[0061] For the first time, a multi-class classifier including class 'N' with label 'k', , is k in the training set th The spectrum data (107) of the target can be trained. Next, multi-class classification of the training target data can be performed to select the optimal prediction model from the multi-class classifier. Finally, the outputs of each optimal prediction model can be combined to determine the optimal model for a new training target.

[0062] In one embodiment, the N-class classifier can correspond to the 'N' targets of the training set. Additionally, labeled data for training the multi-class classifier can be obtained as follows.

[0063] [Mathematical Formula 1]

[0064] Spectral data matrix of target 'k' =

[0065] =

[0066] The features matrix is ​​the spectrum data matrix It can be obtained as a transpose of.

[0067] [Mathematical Formula 2]

[0068]

[0069] And then, the labeled dataset target Is It can be obtained, and the N-class classifier can be trained using the labeled data.

[0070] In one embodiment, unlabeled data of a new training target consisting of instances of 129 unlabeled data points can be input as a multi-class classifier. Here It corresponds to the following.

[0071] [Mathematical Formula 3]

[0072]

[0073] 129 instances were predicted by the multi-class classifier Prediction vector with . ...can be obtained. In one embodiment, the class with the maximum number of predictions may be selected as the first optimal class for the learning target. Additionally, the next class with the maximum predictions may be selected to obtain the second optimal class. This process may continue until the required number of optimal classes is obtained.

[0074] In one embodiment, the learning target is the standard target M prediction model corresponding to This can be assigned.

[0075] [Mathematical Formula 4]

[0076]

[0077] Here, can be a prediction vector with the b1 value removed. Is It can be a prediction vector with values ​​removed. In the current settings, the number of selected optimal models is 2, i.e., M=2.

[0078] 3. Prediction of blood compound concentration

[0079] In one embodiment, any anomalies in the training data can be detected and removed. Additionally, regression analysis of the training data is performed using ensemble learning and combining predictions from the acquired optimal model. Finally, outliers in the training data are corrected for accurate prediction of blood compound concentrations.

[0080] In one embodiment, the original labeled data Assuming that it is given as and can be viewed as a horizontal connection of the glucose array g and the data matrix x as given below,

[0081] [Mathematical Formula 5]

[0082]

[0083] The data matrix x can be used to fit a multivariate Gaussian distribution of the following form.

[0084] [Mathematical Formula 6]

[0085]

[0086] The probability for all vectors is a vector The probability p(k) associated with can be calculated using the above distribution. The vector corresponding to the lowest probability and the corresponding glucose label can be declared as anomalies and removed. For example, the scatter plot of dimensionality-reduced data points from the random subjects in FIGS. 2b to 2d is indicated by 'white circles' in the figure. Dimensionality reduction can be performed by Principal Component Analysis (PCA).

[0087] In one embodiment, regression using ensemble learning can be performed by the following steps. First, the learning results can be ensembled with better Standard Error of Prediction (SEP) and Mean Absolute Relative Difference (MARD) values. Then, ensemble learning is performed using a MATLAB function in the machine learning toolbox fitrensemble () This can be performed by. Then, the function can return a model containing 100 regression tree boosting results using Least Squares (LS) boosting.

[0088] In one embodiment, combining the outputs from a plurality of optimal models can be performed by the following steps. First, two optimal models can be obtained in the current setup. Then, the optimal models can produce two sets of predictions for the unlabeled data to be trained. For example, This optimal model It is a prediction using, The second optimal model In the case of a prediction using [the method], the following final prediction can be reached by using the weighted average.

[0089] [Mathematical Formula 7]

[0090]

[0091] For example, in the current settings is set to 0.6.

[0092] In one embodiment, outlier correction can help improve the accuracy of the prediction by removing unlikely values ​​from the predicted glucose values. For example, outliers may be detected in the predicted glucose vector at exponent 'k' in the following cases.

[0093] [Mathematical Formula 8]

[0094]

[0095] Here, 'y' is the glucose label of the training set, and ' ' is the predicted value and Is It is the first difference of.

[0096] If an outlier is detected in the predicted glucose vector, it can be updated with the average of the previous and subsequent values ​​of the predicted glucose vector. Additionally, outliers detected at the exponent 'k' in the predicted glucose vector can be updated as follows.

[0097] [Mathematical Formula 9]

[0098]

[0099] For example, the prediction results obtained based on ensemble learning may be as shown in Table 1 below.

[0100] [Ensemble Learning] Learning Target R SEP MARD Optimal model Anomaly (No. of anomalies) Model 1 Model 2 Target 2 0.79 318 70 14 6 12 Target 5 0.74 205 211 6 10 12 Target 6 0.69 280 62 7 14 12 Target 7 0.59 108 75 11 13 12 Target 10 0.89 90 53 7 11 12 Target 11 0.94 115 109 14 6 12 Target 12 0.82 149 126 2 13 12 Target 13 0.86 75 54 11 10 12 Target 14 0.92 81 66 10 2 12

[0101] Similarly, Table 2 below shows the results obtained from PLS regression.

[0102] [PLS Regression] Learning Target R SEP MARD Optimal model Anomaly (No. of anomalies) Model 1 Model 2 Target 2 0.78 283 68 14 6 12 Target 5 0.78 302 304 6 10 12 Target 6 0.73 321 88 7 14 12 Target 7 0.85 88 75 11 13 12 Target 10 0.84 122 79 7 11 12 Target 11 0.93 163 198 14 6 12 Target 12 0.77 318 282 2 13 12 Target 13 0.89 89 81 11 10 12 Target 14 0.70 366 223 10 2 12

[0103] At this time, as observed in Tables 1 and 2 above, ensemble learning can show better performance in terms of SEP and MARD.

[0104] Figures 2e and 2f show exemplary prediction results obtained for random subjects (subject 5 and subject 6) using the ensemble results of Table 1.

[0105] Figure 3 shows a flowchart illustrating a method for predicting the concentration of a target compound in the blood.

[0106] As illustrated in FIG. 3, a method for predicting a blood compound concentration includes one or more steps implemented by a processor (201) to predict a blood compound concentration using a blood compound monitoring device (101).

[0107] The above method may be described in relation to computer-executable instructions. Generally, computer-executable instructions may include routines, programs, objects, components, data structures, procedures, modules, and functions that perform specific functions or implement specific abstract data types.

[0108] The above method may be performed in the presented order, in a different order, or simultaneously. Additionally, the above order may be omitted, added, modified, etc. Furthermore, the above method may be implemented with any suitable hardware, software, firmware, or a combination thereof.

[0109] First, the method includes the step (301) of receiving spectral data (107) associated with a target region by a blood compound monitoring device (101). The spectral data (107) may be collected using a near-infrared (NIR) spectrometer associated with the blood compound monitoring device (101). For example, the spectral data (107) may include, without limitation, multiple data instances collected at multiple time intervals.

[0110] Next, the method includes the step (303) of classifying each of a plurality of data instances of spectrum data (107) into one of a plurality of labeled classes (211) by a blood compound monitoring device (101). Each of the plurality of labeled classes (211) may be predetermined using historical spectrum data associated with each of the plurality of learning targets. In one embodiment, each of the plurality of labeled classes (211) may represent one of the plurality of learning targets.

[0111] Next, the method includes the step (305) of obtaining one or more optimal models from a plurality of prediction models (113) based on classification by a blood compound monitoring device (101).

[0112] In one embodiment, a plurality of prediction models (113) may be models corresponding to a plurality of learning targets. Additionally, a plurality of prediction models (113) may be obtained as machine learning regression models learned using reference blood compound concentration values ​​(117) for historical spectrum data (107) related to the learning targets.

[0113] Next, the method includes the step (307) of determining a blood compound concentration value (117) corresponding to each of one or more optimal models by a blood compound monitoring device (101).

[0114] In one embodiment, one or more optimal models can be obtained by the following.

[0115] First, for each labeled class (211), the number of data instances of the spectrum data (107) to which the corresponding labeled class is assigned is determined, the labeled class (211) having the largest number of data instances is identified, and finally, the prediction model (113) corresponding to the labeled class is obtained as one or more optimal models.

[0116] Next, the method includes a step (309) of predicting the blood compound concentration of a target using the weighted average of the blood compound concentration values ​​(117) by the blood compound monitoring device (101).

[0117] FIG. 4 is a block diagram of a computing device for implementing embodiments.

[0118] The computing device (400) may be a blood compound monitoring device (101) used to predict the blood compound concentration shown in FIG. 1. The computing device (400) may include a central processing unit ('CPU' or 'processor'). The processor (402) may include at least one data processor for executing components of a program for executing a business process.

[0119] The user may include any device / sub-device operating in parallel with a person, an object, a caregiver of the object, or a computing device (400).

[0120] The processor (402) may include special processing units such as an integrated device (bus) controller, a memory management control unit, a floating point unit, a graphics processing unit, or a digital signal processing unit.

[0121] The processor (402) may be configured to communicate with one or more input / output (I / O) devices (411 and 412) through an input / output (I / O) interface (401).

[0122] The input / output interface (401) is, for example, audio, analog, digital, stereo, IEEE ® -1394, serial bus, USB (Universal Serial Bus), infrared, PS / 2, BNC, coaxial, component, composite, DVI (Digital Visual Interface), HDMI (high-definition multimedia interface), RF (Radio Frequency) antenna, S-Video, VGA (video Graphic Array), IEEE ®Communication protocols / methods such as 802.n / b / g / n / x, Bluetooth, and cellular (e.g., CDMA (Code-Division Multiple Access), HSPA+ (High-Speed ​​Packet Access), GSM (Global System for Mobile Communications), LET (Long-Term Evolution)) may be used. Using the input / output interface (401), the computing device (400) may communicate with one or more input / output (I / O) devices (411 and 412).

[0123] In one embodiment, the processor (402) may be configured to communicate with a communication network (409) through a network interface (403). The network interface (403) may communicate with the communication network (409).

[0124] The network interface (403) is a direct connection, Ethernet (e.g., twisted pair 10 / 100 / 1000 Base T), TCP / IP (Transmission Control Protocol / Internet Protocol), Token Ring, IEEE ® Connection protocols including but not limited to 802.11a / b / g / n / x may be used.

[0125] By using the network interface (403) and the communication network (409), the computing device (400) can receive historical spectrum data (107) from the spectrum data storage (105). Additionally, the network interface (403) may be used to establish communication with the blood compound value storage (115) to receive blood compound concentration values ​​(117).

[0126] In one embodiment, the communication network (409) may be implemented as one of several types of networks within an organization, such as an intranet or a Local Area Network (LAN). The communication network (409) may be a private network or a shared network, which represents an association of several types of networks using various protocols for communicating with each other, such as HTTP (Hypertext Transfer Protocol), TCP / IP (Transmission Control Protocol / Internet Protocol), and WAP (Wireless Application Protocol). Additionally, the communication network (409) may include various network devices, such as routers, bridges, servers, computing devices, storage devices, etc.

[0127] In one embodiment, the processor (402) may be configured to communicate with memory (405) (e.g., RAM (413), ROM (414), etc., as illustrated in FIG. 4) through a storage interface (404).

[0128] The storage interface (404) can be connected to memory (405), including memory drives and removable drives, using connection protocols such as SATA (Serial Advanced Technology Attachment), IDE (Integrated Drive Electronics), IEEE-1394, USB (Universal Serial Bus), Fiber Channel, SCSI (Small Computer Systems Interface), etc. The memory drive may further include drums, magnetic disk drives, optical magnetic drives, optical drives, RAID (Redundant Array of Independent Discs), solid-state memory devices, solid-state drives, etc.

[0129] Memory (405) includes, but is not limited to, user applications (406), operating systems (407), web browsers (408), etc., and can store collected programs or database components. Additionally, the computing device (400) can store user application (406) data such as data, variables, records, etc. Additionally, such a database is Oracle ® or Sybase ® It can be implemented as a fault-tolerant, relational, scalable, and secure database.

[0130] The operating system (407) can facilitate resource management and operation of the computing device (400). An example of an operating system is APPLE ® MACINTOSH ® OS X ® , UNIX ® , UNIX-like system distributions (EG, BERKELEY SOFTWARE DISTRIBUTION ® (BSD), FREEBSD ® , NETBSD ® , OPENBSD, etc.), LINUX ® DISTRIBUTIONS (EG, RED HAT ® , UBUNTU ® , KUBUNTU ® , etc.), IBM ® OS / 2 ® , MICROSOFT ® WINDOWS ® (XP ® , VISTA ® / 7 / 8, 10 etc.), APPLE ® iOS ® , GOOGLE TM ANDROID TM , BLACKBERRY ® There are OS, etc.

[0131] A user application (406) can facilitate the display, execution, interaction, manipulation, or operation of program components through text or graphic functions. For example, the user application (406) can provide computer interaction interface elements, such as a cursor, icon, checkbox, menu, scroller, window, widget, etc., on a display operably connected to a computing device (400). Also, APPLE ® MACINTOSH ® operating systems' Aqua ® , IBM ® OS / 2 ® , MICROSOFT ® WINDOWS ® (eg, Aero, Metro, etc.), web interface libraries (eg, ActiveX ® , JAVA ® , JAVASCRIPT ® , AJAX, HTML, ADOBE ® FLASH ® GUIs (graphical User Interfaces) such as , etc.) can be used.

[0132] The web browser (408) may be a hypertext viewing application. Secure web browsing may be provided using HTTPS (Secure Hypertext Transport Protocol), SSL (Secure Sockets Layer), TLS (Transport Layer Security), etc. Additionally, the web browser (408) may use AJAX, DHTML, ADOBE ® FLASH ® , JAVASCRIPT ® , JAVA ® Features such as Application Programming Interfaces (APIs) can be used.

[0133] The computing device (400) can implement a mail server in which program components are stored. The mail server is ASP, ACTIVEX ® , ANSI ® C++ / C#, MICROSOFT ® , .NET, CGI SCRIPTS, JAVA ® , JAVASCRIPT ® , PERL ® , PHP, PYTHON ® , WEBOBJECTS ® You can utilize functions such as these. In addition, mail servers support the Internet Message Access Protocol (IMAP), Messaging Application Programming Interface (MAPI), and Microsoft ® Communication protocols such as Exchange, Post Office Protocol (POP), and Simple Mail Transfer Protocol (SMTP) can be used.

[0134] In one embodiment, the computing device (400) may implement a mail client in which program components are stored. The mail client is APPLE ® MAIL, MICROSOFT ® ENTOURAGE ® , MICROSOFT ® OUTLOOK ® , MOZILLA ® THUNDERBIRD ® It could be a mail viewing application like this.

[0135] One or more computer-readable storage media may be used to implement embodiments according to the present invention. A computer-readable storage medium refers to any type of physical memory in which information or data readable by a processor may be stored. Accordingly, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions that cause a processor to perform a step or step according to an embodiment disclosed herein.

[0136] The term 'computer-readable medium' is understood to include items of the type and exclude carriers, transient signals, i.e., non-transient signals. Examples include Random Access Memory (RAM), Read-Only Memory (ROM), volatile memory, nonvolatile memory, hard drives, Compact Disc (CD) ROMs, Digital Video Disc (DVDs), flash drives, disks, and other known physical storage media.

[0137] Those skilled in the art to which this disclosure pertains will understand that the disclosed technical concept or essential features may be implemented in other specific forms without altering them. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. Explanation of the symbols

[0138] 100: Environment 101: Blood Compound Monitoring Device 103: Communication Network 105: Spectrum Data Repository 107: Historical spectrum data 109: Spectrometer 111: Data Storage 113: Prediction Model 115: Blood Compound Value Repository 117: Blood compound concentration value 119: Blood compound measuring device 201: Processor 203: Input / Output Interface 205: Memory 207: Data 209: Module 211: Labeled class 213: Optimal Model 215: Other data 217: Classification Module 219: Optimal Model Determination Module 221: Concentration Prediction Module 223: Other modules 400: Computing device 401: Input / Output Interface 402: Processor 403: Network Interface 404: Save Interface 405: Memory 406: User Application 407: Operating System 408: Web browser 409: Communication Network 411: Input device 412: Output device 413: RAM 414: ROM

Claims

Claim 1 A method for a blood compound monitoring device to predict blood compound concentration comprises: receiving spectral data including data instances collected at multiple time intervals related to a target region using near-infrared spectroscopy; classifying each data instance of the spectral data into one of a plurality of predetermined labeled classes using historical spectral data related to a plurality of training subjects; obtaining one or more optimal models from a plurality of prediction models corresponding to the plurality of training subjects based on the classification; and determining a blood compound concentration value corresponding to each of the plurality of optimal models, wherein the step of obtaining one or more optimal models includes, for each labeled class, determining the number of data instances of the spectral data to which the corresponding labeled class is assigned; identifying the labeled class having the largest number of data instances; and obtaining a prediction model corresponding to the labeled class having the largest number of data instances as the optimal model. Claim 2 A method for predicting a blood compound concentration, further comprising the step of predicting the blood compound concentration of the target using the weighted average of the blood compound concentration values ​​in claim 1. Claim 3 A method for predicting blood compound concentration according to claim 1, wherein the plurality of prediction models are obtained as machine learning regression models learned using reference blood compound concentration values ​​for the historical spectrum data related to the learning target. Claim 4 A method for predicting blood compound concentration according to claim 1, wherein each of the plurality of labeled classes represents one of the plurality of learning targets. Claim 5 delete Claim 6 A method for predicting blood compound concentration according to claim 1, wherein the prediction model corresponds to a model that establishes a relationship or association between the historical spectrum data and the blood compound concentration value. Claim 7 A method for predicting blood compound concentration according to claim 1, wherein the spectrum data is obtained by near-infrared spectroscopy by removing abnormal data from the spectrum data using an anomaly detection technique. Claim 8 A method for predicting a blood compound concentration, wherein the target corresponds to a human or animal in which the blood compound concentration value is predicted. Claim 9 A blood compound monitoring device for predicting blood compound concentration, comprising: a processor; and a memory connected to communicate with the processor and storing processor-executable instructions, wherein the processor receives spectrum data including data instances collected at multiple time intervals related to a target region using near-infrared spectroscopy, classifies each data instance of the spectrum data into one of a plurality of predetermined labeled classes using historical spectrum data related to a plurality of learning targets, obtains one or more optimal models from a plurality of prediction models corresponding to the plurality of learning targets based on the classification, and determines a blood compound concentration value corresponding to each of the plurality of optimal models, wherein for each labeled class, the processor determines the number of data instances of the spectrum data to which the corresponding labeled class is assigned, identifies the labeled class having the largest number of data instances, and obtains one or more optimal models by obtaining a prediction model corresponding to the labeled class having the largest number of data instances. Claim 10 In claim 9, the blood compound monitoring device for predicting blood compound concentration, wherein the processor predicts the blood compound concentration of the target using the weighted average of the blood compound concentration values. Claim 11 A blood compound monitoring device for predicting blood compound concentration, wherein the processor obtains the plurality of prediction models as machine learning regression models learned using reference blood compound concentration values ​​for the historical spectrum data related to the learning target. Claim 12 A blood compound monitoring device for predicting blood compound concentration, wherein each of the plurality of labeled classes represents one of the plurality of learning targets. Claim 13 delete Claim 14 A blood compound monitoring device for predicting blood compound concentration, wherein the prediction model corresponds to a model that establishes a relationship or association between the historical spectrum data and the blood compound concentration value. Claim 15 A blood compound monitoring device for predicting blood compound concentration, wherein the spectrum data is obtained by near-infrared spectroscopy by removing abnormal data from the spectrum data using an abnormality detection technique. Claim 16 In claim 9, the target is a blood compound monitoring device for predicting blood compound concentration, corresponding to a human or animal in which the concentration value of the blood compound is predicted.

Images