KR20190101867A - Apparatus and method for estimation concentration of blood compound - Google Patents

Abstract

A method for predicting concentrations of blood components comprises the following steps: removing drifts from NIR spectrum data and obtaining drift-free spectrum properties; obtaining a global property set based on the obtained drift-free spectrum properties; and predicting concentrations of blood components through regression analysis. According to the present invention, drifts are removed from near infrared spectrum data and the global property set is obtained to be used for predicting the concentrations of blood components so accuracy for predicting the concentrations of blood components can be improved.

Description

Apparatus and method for estimation concentration of blood compound

A device and method for non-invasive estimation of concentrations of blood components using near infrared spectral data.

Monitoring the concentration of constituents in the blood is always a subject of great interest. Monitoring the concentration of blood components is typically performed by an invasive method of obtaining blood samples from the skin of a subject (human or animal). However, non-invasive methods do not require the collection of blood samples. Thus, non-invasive methods allow monitoring of certain components in the blood without pain for those who need to check the concentration of certain components in the blood several times a day. Common methods used to monitor the concentration of blood components non-invasively are mid-infrared, near-infrared and Raman spectroscopy.

Near-infrared spectroscopy among the above-mentioned methods is widely used to monitor the concentration of blood components. However, the prediction of concentrations of blood components based on near infrared spectral data is very difficult when the concentration of one component is calculated in the presence of other components of interest. For example, when monitoring blood glucose levels, other components in the blood, such as water, collagen, keratin, cholesterol, and the like, act as interference components. Another major challenge is to remove drift components from near-infrared spectral data that negatively affect the features used for prediction. This affects the prediction accuracy of blood components based on near infrared spectral data.

One of the prior art uses an optimal filter to eliminate drift. However, this method requires an error covariance matrix that cannot be calculated accurately. Another prior art uses a baseline scatter removal algorithm that can calculate drifts associated with multiple spectra simultaneously. However, this method requires continuous measurement at the same concentration of components in the mold. This is not possible because the concentration of components in the blood usually changes over time.

Thus, there is a need for a method of removing drift from near infrared spectral data to monitor blood component concentrations. In addition, there is a need to obtain a global feature set that can be used to predict concentrations of blood components using regression analysis.

An object of the present invention is to provide an apparatus and method for predicting the concentration of blood components.

According to an aspect, a method of predicting concentration of blood component may include removing drift from NIR spectral data to obtain drift-free spectral features, obtaining a global feature set based on the obtained drift-free spectral features; And predicting a concentration of blood components by regression analysis using the obtained global feature set.

Acquiring the drift-free spectral features may remove the drift using principal component analysis.

Acquiring the drift-free spectral features may include calculating a plurality of principal components of the NIR spectral data, obtaining a drift approximation from the plurality of principal components, and determining each spectral feature according to the magnitude of the spectral feature. Obtaining spectral drift approximation from the drift approximation, and obtaining drift-free spectral features by removing each spectral drift approximation from each spectral feature.

Acquiring a drift approximation from the plurality of principal components, selecting a principal component including a drift from the plurality of principal components, and obtaining a polynomial approximation of a predetermined order of the selected principal component as the drift approximation It may include.

Selecting a main component including drift from the plurality of main components may select a first main component of the plurality of main components as a main component including drift.

Obtaining a predefined polynomial approximation of the selected principal component as a drift approximation may obtain a polynomial approximation that minimizes least square error as a drift approximation.

Obtaining a spectral drift approximation from the drift approximation comprises: normalizing the drift approximation and scaling the normalized drift approximation to an amplitude-span of the spectral feature to obtain the spectral drift approximation. It may include a step.

Normalizing the drift approximation may normalize the drift approximation by dividing the drift approximation by the amplitude-span of the drift approximation.

Acquiring the global feature set includes obtaining similarity values of each drift-free spectral feature for a compound vector, and using the obtained similarity values, a similarity metric for each drift-free spectral feature. ), Ranking the drift-free spectral features according to the similarity metric, and selecting a predefined number of optimal drift-free spectral features as a global feature set. have.

According to another aspect, an apparatus for predicting concentration of blood component may include: a drift removing unit configured to remove drift from NIR spectroscopic data to obtain drift-free spectral features, and to obtain a global feature set based on the obtained drift-free spectral features. A global feature extractor may include a predictor configured to predict concentrations of blood components by regression analysis using the obtained global feature set.

The drift removing unit may remove the drift using principal component analysis.

The drift removing unit calculates a plurality of principal components of the NIR spectral data, obtains a drift approximation from the plurality of principal components, and obtains a spectral drift approximation from the drift approximation for each spectral feature according to the magnitude of the spectral feature. In addition, each spectral drift approximation may be removed from each spectral feature to obtain drift-free spectral features.

The drift removing unit may select a main component including a drift from the plurality of main components, and obtain a polynomial approximation of a predetermined order of the selected main component as the drift approximation.

The drift removing unit may select a first main component of the plurality of main components as a main component including drift.

The drift removing unit may obtain a polynomial approximation that minimizes least square error as a drift approximation.

The drift eliminator may normalize the drift approximation and obtain the spectral drift approximation by scaling the normalized drift approximation to an amplitude-span of the spectral feature.

The drift removing unit may normalize the drift approximation by dividing the drift approximation by the amplitude-span of the drift approximation.

The global feature extracting unit obtains similarity values of each drift-free spectral feature with respect to a compound vector, obtains a similarity metric for each drift-free spectral feature using the obtained similarity values, and The drift-free spectral features may be ranked according to a similarity metric, and a predefined number of optimal drift-free spectral features may be selected as a global feature set.

By removing drift from the near-infrared spectroscopic data and obtaining a global feature set and using it to predict the concentration of blood components, the accuracy of the concentration prediction of blood components can be improved.

1A is a flow diagram illustrating a method of non-invasively predicting the concentration of a blood component of interest using near infrared spectral data, in accordance with an embodiment.
1B is a block diagram illustrating an apparatus for predicting blood component concentration according to an embodiment.
FIG. 2A is an exemplary diagram showing a graphical representation of a first principal component and corresponding linear approximation of NIR spectral data of a subject S11, according to one embodiment. FIG.
FIG. 2B is an exemplary diagram showing a graphical representation of a first principal component and corresponding linear approximation of NIR spectral data of a subject S61, according to an embodiment.
FIG. 2C is an exemplary diagram showing a graphical representation of a first principal component and corresponding linear approximation of NIR spectral data of a subject S71, according to one embodiment. FIG.
3 is an exemplary diagram illustrating a first principal component of 4-decimation of NIR spectral data corresponding to a subject S11 according to an embodiment.
4 is a flowchart illustrating a method of eliminating drift using principal component analysis, according to an embodiment.
5 is a flowchart illustrating a method of extracting global features for predicting the concentration of blood components according to an embodiment.
6 is a block diagram illustrating an apparatus for predicting blood component concentration according to another embodiment.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even though they are shown in different drawings. In addition, when it is determined that description of related known functions or configurations may unnecessarily obscure the understanding of the embodiments, detailed description thereof will be omitted.

On the other hand, in each step, each step may occur differently from the stated order unless the context clearly indicates a specific order. That is, each step may be performed in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

Terms such as first and second may be used to describe various components, but the components should not be limited by the terms. The terms are only used to distinguish one component from another. Singular expressions include plural expressions unless the context clearly indicates otherwise, and the terms 'comprise' or 'have', etc., refer to features, numbers, steps, operations, components, parts, or combinations thereof described in the specification. It is to be understood that the present invention is intended to indicate that there is, and does not preclude the existence or addition of one or more other features or numbers, steps, operations, components, parts or combinations thereof in advance.

In addition, the division of the components in the present specification is only divided by the main function of each component. That is, two or more components may be combined into one component, or one component may be divided into two or more functions for each subdivided function. In addition, each component may perform additionally some or all of the functions of other components in addition to its main functions, and some of the main functions of each component are dedicated by other components. It may also be performed. Each component may be implemented in hardware or software, or a combination of hardware and software.

Hereinafter, a method of non-invasively predicting the concentration of blood components using NIR spectroscopy will be described. In addition, the drift cancellation algorithm for the drift removal process is described using information from the principal components of the NIR spectral data. Regression analysis is also used to describe the extraction of a global feature set for concentration prediction of blood mixtures.

FIG. 1A is a flowchart illustrating a method of non-invasively predicting a concentration of blood components using near infrared spectral data according to an embodiment.

Referring to FIG. 1A, in step 102, a plurality of principal components are calculated from a near infrared spectral data set. In step 104, a drift approximation is obtained on the plurality of principal components. Further, in step 106, a spectral drift approximation is obtained from the drift approximation for each spectral feature according to the magnitude of the spectral feature. Then, in step 108, drift is removed from the spectral features by subtracting each spectral drift approximation from each spectral feature. In step 110, once the drift is removed from the spectral features, a global feature set for a plurality of subjects is obtained. Finally, in step 112, the concentration of blood components is predicted through regression analysis using the obtained global feature set.

1B is a block diagram illustrating an apparatus for predicting blood component concentration according to an embodiment.

Referring to FIG. 1B, the blood component concentration prediction apparatus 150 may include a drift remover 152, a global feature extractor 154, and a predictor 156. The drift remover 152, the global feature extractor 154, and the predictor 156 may be implemented by one or more processors. One or more functions of each module are described in detail below.

The drift remover 152 may calculate and remove the drift from the NIR spectral data. NIR spectral data can be obtained as follows.

First, blood component values are obtained using a standard invasive procedure. Next, a non-invasive spectral scan is performed on the subject using a near infrared spectrometer to obtain a raw NIR spectrum. Raw NIR spectra are labeled with values of blood components obtained by invasive procedures. The raw NIR spectrum obtained is preprocessed to obtain a compound spectrum. Compound spectra and related component values can be arranged in the form of a matrix X using data obtained from successive measurements, hereinafter referred to as data matrix.

는 콤파운드 벡터(compound vector)이다. 행렬

는 NIR 분광 데이터이다. NIR 분광 데이터는 관심 성분의 예측 정확도에 영향을 미치는 드리프트에 의해 영향을 받을 수 있다. 행렬

의 각 열(column)은 파장

와 관련된 흡수 스펙트럼이며 벡터

로 나타낼 수 있다. 일부 실시예에서 흡수 스펙트럼

은 "스펙트럼 특징" 또는 "특징"으로 지칭될 수 있다.here,

Is a compound vector. procession

Is NIR spectral data. NIR spectral data can be affected by drift that affects the prediction accuracy of the component of interest. procession

Each column of is a wavelength

The absorption spectrum associated with the vector

It can be represented as. In some embodiments an absorption spectrum

May be referred to as a "spectral feature" or a "feature".

은 다음과 같이 표현될 수 있다.Absorption spectrum

Can be expressed as

는 실제 흡수 스펙트럼이고,

는 실제 흡수 스펙트럼에 영향을 주는 드리프트일 수 있다.here,

Is the actual absorption spectrum,

May be a drift that affects the actual absorption spectrum.

를 획득하고, 홉수 스펙트럼

에서 드리프트 성분의 추정치

를 감산하여 다음과 같이 드리프트-프리 스펙트럼

을 획득할 수 있다.The drift removing unit 152 estimates the drift component

To obtain the hop number spectrum

Estimate of the drift component in

Subtract the drift-free spectrum as

Can be obtained.

을 획득하기 위해 PCA 연산을 수행할 수 있다.According to an embodiment, the drift removing unit 152 may remove the drift using principal component analysis (PCA). The drift removing unit 152 is the i-th main component as follows

PCA operation may be performed to obtain.

로 표기하기로 한다.Here, the variables follow standard notation. If the drift component of a data set is significant enough to affect prediction based on the data set, it may appear in the first principal component of the data set. Otherwise, drift may appear in the i th principal component. In addition, since all the principal components are not correlated, it can be reasonably assumed that if the drift component is trapped in the i-th principal component, the drift component is unlikely to be prominent in other principal components. The i th principal component represented by the drift

It is written as.

2A-2C are graphical representations showing the first principal component and corresponding linear approximation of NIR spectral data of different subjects labeled S11, S61 and S71, according to one embodiment. As shown in FIGS. 2A-2C, the drift component predominates, and the slope of the linear approximation can provide a rate of drift change over time.

3 illustrates the first principal component of 4-decimation of NIR spectral data of Subject S11, according to one embodiment. In this case, the following NIR spectral data can be considered.

의 d-데시메이션은 다음과 같이 정의될 수 있다.Spectral data

The d-decimation of may be defined as follows.

는 행렬

의 모든 d 번째 행(row)을 포함함으로써 획득될 수 있다. 도 3에 도시된 바와 같이,

의 첫 번째 주성분은 기울기가 도 2a의 원본 행렬

의 기울기의 약 4배인 선형 근사에 의해 특징지어질 수 있다. 이것은 드리프트 성분이 첫 번째 주성분에 본질적으로 포착됨을 나타낼 수 있다.set

Is a matrix

It can be obtained by including every d-th row of. As shown in FIG. 3,

The first principal component of is the slope of the original matrix of Figure 2a.

It can be characterized by a linear approximation of about four times the slope of. This may indicate that the drift component is essentially trapped in the first principal component.

를 복수의 주성분들에서 선택한다. 단계 404에서, 선택된 주성분

의 다항식 근사

를 드리프트 근사로 획득한다. 일 실시예에 따르면, 드리프트 근사

는 최소 제곱 에러

를 최소화하는

로 획득될 수 있다. 또한, 단계 406에서, 각 홉수 스펙트럼

에 대하여, 드리프트 근사를

의 크기에 따라 스케일링하여 스펙트럼 드리프트 근사

를 획득할 수 있다. 스펙트럼 드리프트 근사

는 다음과 같이 획득될 수 있다.4 is a flowchart illustrating a method of eliminating drift using principal component analysis according to an embodiment. According to an embodiment, the drift removing unit 152 may remove the drift using principal component analysis. In step 402, the principal component characterized by the drift

Is selected from a plurality of principal components. In step 404, the selected principal component

Polynomial approximation

Is obtained by drift approximation. According to one embodiment, a drift approximation

Is the least squared error

To minimize

It can be obtained as. Further, in step 406, each hop number spectrum

Against, drift approximation

Approximation of spectral drift by scaling according to the size of

Can be obtained. Spectral Drift Approximation

Can be obtained as follows.

는

에 의해 주어지는

의 진폭 스팬(amplitude span)이고,

는

에 의해 주어지는

의 진폭 스팬일 수 있다.here,

Is

Given by

Is the amplitude span of,

Is

Given by

It can be the amplitude span of.

에 대해 각각의

에서 스펙트럼 드리프트 근사

를 감산하여 드리프트 제거가 수행된다. 이것은 다음과 같이 표현될 수 있다.Finally at step 408, all wavelengths

For each

Spectral Drift Approximation

Subtraction is performed to remove drift. This can be expressed as

는 드리프트-프리 스펙트럼 특징 또는 트리프트-프리 특징으로 호칭될 수 있다.In some embodiments,

May be referred to as a drift-free spectral feature or a trift-free feature.

로 표시되는 P 개의 피검체가 존재하며, 피검체

에 대한 대응 드리프트-프리 스펙트럼 특징들은

로 표시하고, 혼합물의 농도는

로 표시하면, 일 실시예에 따르면 유사도 값은 드리프트-프리 스펙트럼 특징

과 콤파운드 벡터

의 상관관계로 획득될 수 있다. 이를 수학식으로 표현하면 다음과 같다.FIG. 5 is a flow diagram illustrating one or more steps associated with extracting global features for predicting concentration of a blood component, according to one embodiment. In this embodiment, one or more steps performed by the global feature extractor 154 to extract global features are described below. In step 502, a similarity value of each drift-free spectral feature for each compound vector is obtained.

There are P subjects represented by

The corresponding drift-free spectral features for

And the concentration of the mixture

In one embodiment, the similarity value is a drift-free spectral feature.

And compound vector

It can be obtained by the correlation of. This is expressed as an equation.

In step 504, a similarity metric for each drift-free spectral feature may be obtained using the similarity value obtained across all the subjects. According to one embodiment, the similarity metric may be calculated as follows.

In step 506, the drift-free spectral features are ranked according to the similarity metric. In step 508, regression analysis is used to select K optimal drift-free spectral features for prediction of blood component concentrations. The value of K can be determined by the performance of the particular regression method used for the prediction. These K optimal drift-free spectral features may be referred to as “global features”.

Based on the obtained global features, the prediction unit 156 may estimate the concentration of the blood mixture using regression analysis from the drift-free spectral data.

6 is a block diagram illustrating an apparatus for predicting blood component concentration according to another embodiment.

Referring to FIG. 6, the processor 610 may include an input unit 620, a storage unit 630, a communication unit 640, and an output unit 650. Since the processor 610 may perform the functions of the drift removing unit 152, the global feature extracting unit 154, and the predicting unit 156 of FIG. 2, a detailed description thereof will be omitted.

The input unit 620 may receive NIR spectral data. In addition, the input unit 620 may receive various operation signals from the user. According to an embodiment, the input unit 620 may include a key pad, a dome switch, a touch pad (static pressure / capacitance), a jog wheel, a jog switch. , H / W button and the like. In particular, when the touch pad forms a mutual layer structure with the display, this may be referred to as a touch screen.

The storage unit 630 may store a program or commands for the operation of the blood component concentration predicting apparatus 600, and the data input to the blood component concentration predicting apparatus 600 may be processed by the blood component concentration predicting apparatus 600. Data and data output from the blood component concentration prediction apparatus 600 may be stored.

The storage unit 630 may include a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (eg, SD or XD memory), RAM Random Access Memory (RAM), Static Random Access Memory (SRAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Programmable Read Only Memory (PROM), Magnetic Memory, Magnetic Disk, Optical Disk And at least one type of storage medium. In addition, the blood component concentration prediction apparatus 600 may operate an external storage medium such as a web storage that performs a storage function of the storage unit 630 on the Internet.

The communication unit 640 may communicate with an external device. For example, the communication unit 640 may transmit data inputted to the blood component concentration predicting apparatus 600, stored data, processed data, etc. to an external device, or may receive various data to help estimate blood component concentration from the external device. Can be.

In this case, the external device may be a medical device using data inputted to the blood component concentration predicting device 600, stored data, processed data, or the like, or a print or display device for outputting a result. In addition, the external device may be a digital TV, a desktop computer, a mobile phone, a smart phone, a tablet, a notebook, a personal digital assistant (PDA), a portable multimedia player (PMP), a navigation device, an MP3 player, a digital camera, a wearable device, and the like. It doesn't work.

The communication unit 640 may include Bluetooth communication, BLE (Bluetooth Low Energy) communication, Near Field Communication (Near Field Communication), WLAN communication, Zigbee Communication, Infrared Data Association (IrDA) communication, WFD (Wi-Fi Direct) communication, ultra-wideband (UWB) communication, Ant + communication, WIFI communication, Radio Frequency Identification (RFID) communication, 3G communication, 4G communication and 5G communication can be used to communicate with external devices. However, this is only an example and is not limited thereto.

The output unit 650 may output data input to the blood component concentration prediction apparatus 600, stored data, processed data, and the like. According to one embodiment, the output unit 650 outputs the data, stored data, processed data, etc. input to the blood component concentration prediction apparatus 600 by at least one method of an auditory method, a visual method and a tactile method. can do. To this end, the output unit 650 may include a display, a speaker, a vibrator, and the like.

The above-described embodiments can be embodied as computer readable codes on a computer readable recording medium. The computer-readable recording medium may include all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media may include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, and the like. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is created and executed in a distributed fashion.

So far I looked at the preferred embodiments. It will be understood by those skilled in the art that the present invention may be implemented in a modified form without departing from the essential characteristics of the invention. Therefore, the scope of the present invention should be construed to include various embodiments that are not limited to the above-described embodiments and fall within the scope equivalent to the contents described in the claims.

150: blood component concentration prediction device
152: drift removing unit
154: global feature extraction unit
156: prediction unit

Claims (18)

Removing drift from the NIR spectral data to obtain drift-free spectral features;
Obtaining a global feature set based on the obtained drift-free spectral features; And
Predicting concentrations of blood components by regression analysis using the obtained global feature set; Including,
Method for predicting the concentration of blood components. The method of claim 1,
Obtaining the drift-free spectral features,
To remove the drift using principal component analysis,
Method for predicting the concentration of blood components. The method of claim 1,
Obtaining the drift-free spectral features,
Calculating a plurality of principal components of the NIR spectral data;
Obtaining a drift approximation from the plurality of principal components;
Obtaining a spectral drift approximation from the drift approximation for each spectral feature according to the magnitude of the spectral feature; And
Removing each spectral drift approximation from each spectral feature to obtain drift-free spectral features; Including,
Method for predicting the concentration of blood components. The method of claim 3,
Obtaining a drift approximation from the plurality of principal components,
Selecting a principal component including drift from the plurality of principal components; And
Obtaining a polynomial approximation of a predefined order of the selected principal component as the drift approximation; Including,
Method for predicting the concentration of blood components. The method of claim 4, wherein
Selecting a main component including drift from the plurality of main components is to select the first main component of the plurality of main components as a main component including drift,
Method for predicting the concentration of blood components. The method of claim 4, wherein
Acquiring a predefined polynomial approximation of the selected principal component as a drift approximation, obtaining a polynomial approximation that minimizes least square error as a drift approximation,
Method for predicting the concentration of blood components. The method of claim 3,
Obtaining a spectral drift approximation from the drift approximation,
Normalizing the drift approximation; And
Scaling the normalized drift approximation to an amplitude-span of the spectral feature to obtain the spectral drift approximation; Including,
Method for predicting the concentration of blood components. The method of claim 7, wherein
Normalizing the drift approximation comprises: normalizing the drift approximation by dividing the drift approximation by the amplitude-span of the drift approximation,
Method for predicting the concentration of blood components. The method of claim 1,
Acquiring the global feature set,
Obtaining similarity values of each drift-free spectral feature for the compound vector;
Obtaining a similarity metric for each drift-free spectral feature using the obtained similarity values;
Ranking the drift-free spectral features according to the similarity metric; And
Selecting a predefined number of optimal trift-free spectral features as a global feature set; Including,
Method for predicting the concentration of blood components. A drift eliminator for removing drift from the NIR spectral data to obtain drift-free spectral features;
A global feature extracting unit obtaining a global feature set based on the obtained drift-free spectrum features; And
A prediction unit for predicting the concentration of blood components by regression analysis using the obtained global feature set; Including,
Device for predicting the concentration of blood components. The method of claim 10,
The drift removing unit removes the drift using principal component analysis,
Device for predicting the concentration of blood components. The method of claim 10,
The drift removing unit,
Calculating a plurality of principal components of the NIR spectral data,
Obtain a drift approximation from the plurality of principal components,
Obtain a spectral drift approximation from the drift approximation for each spectral feature according to the magnitude of the spectral feature,
Removing each spectral drift approximation from each spectral feature to obtain drift-free spectral features,
Device for predicting the concentration of blood components. The method of claim 12,
The drift removing unit,
Selecting a principal component including a drift from the plurality of principal components, and obtaining a polynomial approximation of a predetermined order of the selected principal component as the drift approximation,
Device for predicting the concentration of blood components. The method of claim 13,
The drift removing unit,
Selecting a first principal component from the plurality of principal components as a principal component including drift,
Device for predicting the concentration of blood components. The method of claim 13,
The drift removing unit,
Obtain a polynomial approximation with drift approximation that minimizes least squares error,
Device for predicting the concentration of blood components. The method of claim 12,
The drift removing unit,
Normalizing the drift approximation and scaling the normalized drift approximation to an amplitude-span of the spectral feature to obtain the spectral drift approximation,
Device for predicting the concentration of blood components. The method of claim 16,
The drift removing unit,
Dividing the drift approximation by the amplitude-span of the drift approximation to normalize the drift approximation,
Device for predicting the concentration of blood components. The method of claim 10,
The global feature extraction unit,
Obtain similarity values of each drift-free spectral feature for the compound vector,
Using the obtained similarity values, obtain a similarity metric for each drift-free spectral feature,
Rank the drift-free spectral features according to the similarity metric,
Selecting a predefined number of optimal trit-free spectral features as a global feature set,
Device for predicting the concentration of blood components.

Images