JPH07151677A - Densitometer - Google Patents

Abstract

PURPOSE:To realize highly accurate measurement of an abnormal value sample by providing a function for deciding the adaptability of a sample to a test set and a function for correcting the measured concentration of the sample. CONSTITUTION:The glucose concentration is measured by a definitive method for (n) blood samples thus preparing a target variable series of nX1. Another (n) specimens are measured by a spectrometer thus preparing a description variable series of nXm based on the absorbance of (m) wave number exhibiting absorption specific to glucose. Calibration and prediction of concentration are effected by a test set through aralysis using partial least square(PLS) method. Adaptability of a sample to the test set is decided by deciding the scattering of sample absorbance from the statistic value of description variables of the test set. For a sample inadaptable to the test set, error from the test set is calculated and then a correction value is calculated. The correction value is added to a predicted concentration to obtain a corrected PLS concentration which is outputted as a measured concentration.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quantitative analysis method for a plurality of components using a spectrum, and more particularly to a quantification method used in a blood biochemical test apparatus using infrared spectroscopy.

[0002]

2. Description of the Related Art Quantitative analysis of blood biochemical components using infrared absorption spectrum is described in Analytical Chemistry (1989), pages 2016 to 2023. Here, the latent variable is derived by the partial least squares method (PLS method) with the blood concentration of the known component concentration collected from the patient as the test set, the component concentration as the objective variable, and the measured spectrum as the explanatory variable. We are trying to quantify unknown concentration samples by calculating regression equations by least-squares variables and objective variables. Samples with out-of-statistical scatter values are excluded from these series of test sets by outlier detection. Therefore, samples with scattered values, or abnormal values in the clinical laboratory field,
It does not fit the PLS model, and the error is large even when measured. Further, the method of PLS and outlier detection is described in Chemometrics (1992, Tetsuro Aijima, Maruzen Co., Ltd.), pages 116 to 118, and pages 89 to 102.

[0003]

In the case of quantitative analysis using multivariate analysis in the spectrum that has been conventionally used,
As mentioned above, in order to improve the prediction accuracy, spectra with values outside the average of the test set to be used for calibration are excluded from the test set as outliers and then calibrated. There is a problem in that a value having a value cannot accurately predict the concentration. In particular, when measuring substances containing multiple components, in order to maintain quantification accuracy, the test set should be composed of mixed substances containing all of these components, and the concentration distribution should form a continuous group. As a result, it is often necessary to use ready-made substances as test sets. Therefore, it is difficult to create a test set that continuously covers expected outliers. Especially in the case of biochemical analysis of blood, it is important to detect abnormal values and measure them correctly, and therefore it is problematic to directly apply these series of quantitative methods to blood analysis.

[0004]

In order to solve the above problems, a function for determining the compatibility between the test set and the sample and a mechanism for correcting the measured concentration of the sample are provided.

[0005]

The above means operates as follows. The function of determining the compatibility between the test set and the sample is a function of checking whether or not the sample is scattered from the group formed by the test set. If no scattering of the sample is observed, this sample conforms to the test set, and the measured concentration is regarded as having no systematic error derived from the test set and does not require correction, and the measured concentration is output as it is.

When the sample is scattered from the test set, the directionality of the scattering causes a systematic error in the sample measurement value. Therefore, in order to make the measured concentration of the sample more accurate, a correction mechanism for compensating for the systematic error is provided. Samples deemed to be non-compliant with the test set are sent to a correction mechanism where the error is corrected and output as a more accurate measurement.

[0007]

EXAMPLES The present invention will be described in detail below based on examples. FIG. 1 is a device configuration of a blood glucose analyzer according to an embodiment of the present invention. This apparatus comprises a spectroscope 1 and a data processing unit 2 for numerically processing the signals measured by the spectroscope and for predicting the sample concentration from the processing result. Infrared light 4 emitted from a light source 3 in the spectroscope 1 enters a sample cell 5 using an attenuating total reflection prism. Blood 6 is introduced onto the sample cell 5, and the light 6 absorbs the light 4 only at a specific wavelength and is emitted from the cell 5. The light 4 emitted from the cell 5 causes a phase difference by the interferometer 7, is detected by the detector 8, and is converted into the digital signal 10 via the AD converter 9. The signal 10 having the phase difference generated by the interferometer 7 is sent to the data processing unit 2 and the intensity for each wavelength is calculated by Fourier transform.

FIG. 2 shows the outline of the method for quantitatively analyzing blood glucose according to the embodiment of the present invention. This analysis method includes the creation 12 of the test set 11, the calibration 13 using the test set 11, the concentration prediction 15 of the sample 14, the determination 16 of the compatibility between the test set 11 and the sample 14, and the correction when the test set and the sample are not compatible. It can be roughly divided into six stages of 17 and output of measured concentration 18. Less than,
Each stage will be described in detail.

FIG. 3 is a flow of the test set creation 12. Here, the glucose concentration of n pieces of blood 20 randomly collected from n patients 19 is measured by a reference method 21 (hereinafter, the glucose concentration measured by the reference method is referred to as a reference value for convenience), and n × The target variable sequence y22 of 1
To create. Further, n samples 20 are measured by the spectroscope 1 shown in FIG.
An n × m explanatory variable sequence X23 is created by the absorbance of m wave numbers in the wave number range of ¹ to 950 cm ⁻¹ . In the calibration 13 and the concentration prediction 15 by the test set, the analysis using the PLS method is performed. In the PLS modeling, the explanatory variable sequence X and the objective variable sequence y are represented by latent variable sequence T, P, q as shown in Formulas 1 and 2.

[0010]

## EQU1 ## X = TP '+ E (Equation 1)

[0011]

## EQU00002 ## y = Tq '+ f (Equation 2) where the sequence T is n.times.a, the sequence P is m.times.a, and the sequence q is a.times.
It has an array of 1, the sequences E and f are residuals of variables X and y, the arrays are n × m and n × 1, respectively, and a is the number of factors of the PLS model. The latent variable sequence T, P, q is calculated from the m-th order weight vector w obtained from X and y. PL of y
The S regression equation y ^ is expressed as in Equation 3 using the latent variable T. Since T is a linear combination of X, the expression 3 is expressed as a linear combination of the explanatory variables X as in the expression 4, and the coefficient b can be calculated from the a weight vectors w by the expression 5.

[0012]

## EQU3 ## y ^ = Tq + E (Equation 3)

[0013]

Y ^ = Tq + E = Xb + E (Equation 4)

[0014]

## EQU00005 ## b = W (P'W) .sup.- ¹ q (Equation 5) As described above, the PLS regression equation (Equation 4) is calculated based on the test set and used for the concentration prediction of the sample. In the concentration prediction 14, the PLS regression equation (Equation 4) obtained in the calibration 12
The concentration of the unknown concentration sample 13 is predicted using the coefficient b and the residual E. The sample 13 is measured by the spectroscope 1 and expressed as an m × 1 absorbance sequence x̂. From this absorbance sequence x ^ and equation 4, the glucose concentration in the sample y ^
(Hereinafter referred to as PLS concentration to distinguish it from the reference value)
It is represented by.

[0015]

Y ^ = x ^ b + E (Equation 6) On the other hand, in the determination 16 of the compatibility of the sample with the test set, it is determined from the statistical value of the explanatory variable of the test set whether or not the sample absorbance is scattered. When the sample is scattered from the test set, the PLS concentration y ^ of the sample calculated by the equation 6 contains a systematic error, and y ^ needs to be corrected. In judgment 16, SIMCA (soft i
A method called “independent modeling of class analogy” is used to determine whether the sample is an outlier sample.

In SIMCA, first, the explanatory variable sequence X of the test set is subjected to a principal component analysis, and is expressed by Equation 7.

[0017]

X = x ⁻ + VU + E (Equation 7) Here, the m × 1 sequence x ⁻ is the average value of the explanatory variable sequence X, and the m × d sequence V is calculated from the explanatory variable sequence X. The load amount on the principal component, the d × n sequence U is the principal component score for d principal components, and the n × m sequence E is the residual. SIMCA sets a determination range based on the variation of the residual E, and determines an abnormal value. The variation s ₀ of the residual E is expressed by
Represented by

[0018]

[Equation 8]

Next, the sample 13 is judged. Estimated value of the principal component score by the multiple regression analysis with the difference between the average value of the test set and the sample x ^ as the objective variable and the load sequence V to the principal component of the equation 7 as the explanatory variable U
^ Is calculated as in Equation 10.

[0020]

[Equation 9] ^{x ^ -x - = VU ^ +} E ... ( number 9)

[0021]

Equation 10] ^{U ^ = (V'V) -1 V} '(x ^ -x -) ... ( Equation 10) Equation 9, after calculating the error E of the sample 13 by the number 10, a dispersion s of the error E It is calculated by Equation 11.

[0022]

[Equation 11]

With the F value being the ratio of s ₀ obtained by the _equation 8 and s obtained by the _equation 11, it is determined whether the sample 13 conforms to the test set according to the F distribution.

[0024]

[Equation 12] F = s ² / s ₀ ² (Equation 12) When conforming, the PLS concentration y ^ of the sample 13 obtained in Equation 6 is output as it is without correction. If they do not match, correction 17 corrects y. FIG. 4 shows a flow of correction. The correction is made by calculating the correction formula 24 and y.
The correction 25 can be roughly divided into two stages. In the correction formula calculation 24, for t abnormal value samples 26,
Absorbance error 27 (E with the explanatory variable sequence X used in calibration
s), the concentration error 28 (Ec) between the PLS concentration y ^ and the reference value according to the equation 6 of each abnormal value sample 26 is calculated, and E
The correction equation 29 is calculated by analyzing the systematic relationship between s and Ec by multivariate analysis.

The procedure will be described in detail below. First, t outlier samples 26 are measured by the standard method and the spectroscope, and t ×
A reference value sequence Yr of 1 and an absorbance sequence Xs of t × m can be created. The absorbance sequence Xs is used as it is as t samples to predict the PLS concentration by the equation 6, and the t expected concentration sequence Y is calculated.
Create c. Further, the density error Ec of the PLS model of t × 1 is calculated by taking the difference between the two density sequences as shown in Expression 13.

[0026]

[Equation 13] Ec = Yr−Yc (Equation 13) The absorbance error Es of the t × m PLS model is calculated by using the latent variable P obtained in the calibration 12 described above and the absorbance sequence X.
The difference between s and the average value x ⁻ of the test set can be obtained by describing as in Expression 14.

[0027]

Equation 14] ^{Xs-x - = T ^ P} + Es ... ( number 14) an estimate of the latent variables T T ^ is determined by the following equation to minimize Es.

[0028]

[Number 15] ^{T ^ = (P'P) -1 P} '(Xs-x -) ... ( number 15) number 14, after obtaining the Es than the number 15, objective variable density error Ec, the absorbance error Es As an explanatory variable, the PLS regression equation Ec ^ number 16 which represents the relationship between both variables by the PLS method again
To calculate.

[0029]

[Equation 16] Ec ^ = zEs + E (Equation 16) Therefore, the correction equation 29 in FIG.

[0030]

[Mathematical formula-see original document] y * = y ^ + Es ^ (Equation 17) On the other hand, for the sample determined to be incompatible with the test set by the determination equation 16, the error Xs ^ from the test set is calculated by equation 14 in the correction 25. , The correction value Ec ^ is calculated by the equation 16. Furthermore, the concentration prediction 15 is made by the equation 17
The corrected PLS concentration y * is obtained by adding this to the y ^ obtained in step 1 and is output as the measured concentration.

5 and 6 are graphs showing the effect of the present invention. FIG. 5 shows the results of measuring the glucose concentration in blood by the blood glucose analyzer of the example without using the correction mechanism. The abscissa plots the measured value by the standard method and the ordinate plots the measured value of the device of the example, and the correlation between the two glucose concentrations was evaluated. As a result, samples with glucose concentrations outside the range of the test set used in the examples, ie, about 20
In the samples having the reference values of 0 mg / dl or more and 50 mg / dl or less, the difference between the reference value and the PLS concentration was large and the correlation was poor. FIG. 6 is a graph of the result of performing the same evaluation using the correction mechanism. According to this result, the samples scattered from the test set were corrected to have a density close to the reference value, and the correlation was good.

[0032]

According to the present invention, it is possible to measure an outlier sample with high accuracy without the need to create a test set that continuously covers outliers.

[Brief description of drawings]

FIG. 1 is a block diagram of an apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart of the quantification method used in the examples.

FIG. 3 is a flowchart of a method for creating a test set.

FIG. 4 is a flowchart showing an outline of a correction method.

FIG. 5 is a characteristic diagram showing the effect of the present invention.

FIG. 6 is a characteristic diagram showing the effect of the present invention.

[Explanation of symbols]

1 ... Spectrometer, 2 ... Data processing part, 3 ... Light source, 4 ... Infrared light, 5 ... Sample cell, 6 ... Blood, 7 ... Interferometer, 8 ... Detector, 9 ... AD converter, 10 ... Digital signal .

Claims (8)

[Claims] 1. A light source for irradiating light, a sample cell having a light receiving portion for irradiating the sample with the light of the light source, and light transmitted, reflected or scattered by the sample cell is separated into at least m wavelengths. A spectroscope which comprises a spectroscopic means, a detector for detecting light, and a data processing unit for numerically processing the signal measured by the detector, and the measurement result for one sample is output as m intensities as light intensities for m wavelengths. In total, n
Measurement is performed using a sample of known concentration as a standard substance, and n ×
Calibration is performed by regression analysis of the explanatory variable sequence consisting of m light intensities and the n objective variable sequence consisting of the concentration of the component to be measured to calculate a calibration formula, and the calibration formula and the light intensity of the unknown concentration sample are used to calculate the sample It is a quantitative method that measures the concentration of components, has a discrimination function to determine whether an unknown sample belongs to the standard substance group used for calibration, and explains the light intensity sequence and standard substance group of unknown samples that do not belong to the standard substance group. A densitometer having a mechanism for representing a deviation from a variable number sequence by a light intensity error number sequence consisting of m pieces, and correcting an unknown sample concentration by the light intensity error number sequence. 2. The correction mechanism according to claim 1, wherein the correction mechanism obtains the light intensity error of t abnormal value substances having values scattered from the n standard substance groups and the calibration formula calculated by the calibration. A densitometer that uses a regression formula obtained by regressing the error between the concentration of the t abnormal value substances and the actual concentration for correction. 3. The densitometer according to claim 1, wherein the regression analysis method uses multiple regression analysis, principal component analysis, and partial least squares method. 4. The light source according to claim 1, wherein the light source is 2.5 μm.
A densitometer that emits light with a wavelength of -25 μm. 5. The densitometer according to claim 1, wherein the spectroscopic means is a diffraction grating, a prism, or an interferometer. 6. The densitometer according to claim 1, wherein the sample cell uses an attenuated total reflection prism. 7. The densitometer according to claim 1, wherein the error sequence is calculated using principal component analysis, partial least squares method, and multiple regression analysis. 8. A blood biochemical analyzer using the concentration meter according to claim 1.