WO2000052433A1 - Hybrid least squares multivariate spectral analysis methods - Google Patents
Hybrid least squares multivariate spectral analysis methods Download PDFInfo
- Publication number
- WO2000052433A1 WO2000052433A1 PCT/US2000/005750 US0005750W WO0052433A1 WO 2000052433 A1 WO2000052433 A1 WO 2000052433A1 US 0005750 W US0005750 W US 0005750W WO 0052433 A1 WO0052433 A1 WO 0052433A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- sample
- spectral
- calibration
- analysis
- squares
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 156
- 238000010183 spectrum analysis Methods 0.000 title claims description 16
- 230000003595 spectral effect Effects 0.000 claims abstract description 191
- 239000000470 constituent Substances 0.000 claims abstract description 32
- 238000012628 principal component regression Methods 0.000 claims abstract description 14
- 238000000556 factor analysis Methods 0.000 claims abstract description 11
- 230000036961 partial effect Effects 0.000 claims abstract description 11
- 238000004458 analytical method Methods 0.000 claims description 88
- 230000000694 effects Effects 0.000 claims description 23
- 230000004044 response Effects 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000002790 cross-validation Methods 0.000 claims description 10
- 238000012417 linear regression Methods 0.000 claims description 4
- 238000010238 partial least squares regression Methods 0.000 abstract description 8
- 238000001228 spectrum Methods 0.000 description 142
- 239000000523 sample Substances 0.000 description 87
- 239000012491 analyte Substances 0.000 description 35
- XSQUKJJJFZCRTK-UHFFFAOYSA-N Urea Chemical compound NC(N)=O XSQUKJJJFZCRTK-UHFFFAOYSA-N 0.000 description 29
- 239000004202 carbamide Substances 0.000 description 29
- KFZMGEQAYNKOFK-UHFFFAOYSA-N Isopropanol Chemical compound CC(C)O KFZMGEQAYNKOFK-UHFFFAOYSA-N 0.000 description 23
- 230000009021 linear effect Effects 0.000 description 15
- 239000000243 solution Substances 0.000 description 15
- 239000013598 vector Substances 0.000 description 13
- 230000006872 improvement Effects 0.000 description 12
- 238000004611 spectroscopical analysis Methods 0.000 description 12
- 230000008859 change Effects 0.000 description 11
- 238000002474 experimental method Methods 0.000 description 11
- 238000011068 loading method Methods 0.000 description 10
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 10
- 239000007864 aqueous solution Substances 0.000 description 9
- 239000000203 mixture Substances 0.000 description 9
- CSCPPACGZOOCGX-UHFFFAOYSA-N Acetone Chemical compound CC(C)=O CSCPPACGZOOCGX-UHFFFAOYSA-N 0.000 description 8
- 238000000701 chemical imaging Methods 0.000 description 8
- DDRJAANPRJIHGJ-UHFFFAOYSA-N creatinine Chemical compound CN1CC(=O)NC1=N DDRJAANPRJIHGJ-UHFFFAOYSA-N 0.000 description 8
- 230000001627 detrimental effect Effects 0.000 description 8
- 101000596046 Homo sapiens Plastin-2 Proteins 0.000 description 7
- 102100035182 Plastin-2 Human genes 0.000 description 7
- 230000004075 alteration Effects 0.000 description 7
- 238000013459 approach Methods 0.000 description 7
- 239000000126 substance Substances 0.000 description 7
- 238000000862 absorption spectrum Methods 0.000 description 6
- 238000002329 infrared spectrum Methods 0.000 description 6
- 238000003780 insertion Methods 0.000 description 6
- 230000037431 insertion Effects 0.000 description 6
- 238000010926 purge Methods 0.000 description 6
- 230000003190 augmentative effect Effects 0.000 description 5
- 238000012937 correction Methods 0.000 description 5
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000002493 microarray Methods 0.000 description 5
- 239000007858 starting material Substances 0.000 description 5
- 238000005033 Fourier transform infrared spectroscopy Methods 0.000 description 4
- 238000005102 attenuated total reflection Methods 0.000 description 4
- 235000013405 beer Nutrition 0.000 description 4
- 230000000875 corresponding effect Effects 0.000 description 4
- 229940109239 creatinine Drugs 0.000 description 4
- 238000006073 displacement reaction Methods 0.000 description 4
- 230000002452 interceptive effect Effects 0.000 description 4
- 238000002360 preparation method Methods 0.000 description 4
- 239000002904 solvent Substances 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000013401 experimental design Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000013450 outlier detection Methods 0.000 description 3
- 241000894007 species Species 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 241000428199 Mustelinae Species 0.000 description 2
- 238000002835 absorbance Methods 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 238000010348 incorporation Methods 0.000 description 2
- 238000003760 magnetic stirring Methods 0.000 description 2
- 238000004949 mass spectrometry Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009022 nonlinear effect Effects 0.000 description 2
- 230000001151 other effect Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 108090000623 proteins and genes Proteins 0.000 description 2
- 238000004445 quantitative analysis Methods 0.000 description 2
- 238000000985 reflectance spectrum Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000004566 IR spectroscopy Methods 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000004497 NIR spectroscopy Methods 0.000 description 1
- 238000010222 PCR analysis Methods 0.000 description 1
- 238000001069 Raman spectroscopy Methods 0.000 description 1
- 239000004809 Teflon Substances 0.000 description 1
- 229920006362 Teflon® Polymers 0.000 description 1
- 238000004833 X-ray photoelectron spectroscopy Methods 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 235000013339 cereals Nutrition 0.000 description 1
- 239000013626 chemical specie Substances 0.000 description 1
- 238000011208 chromatographic data Methods 0.000 description 1
- 239000000356 contaminant Substances 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002149 energy-dispersive X-ray emission spectroscopy Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000000763 evoking effect Effects 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 238000009615 fourier-transform spectroscopy Methods 0.000 description 1
- 238000003052 fractional factorial design Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000005251 gamma ray Effects 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 229910052736 halogen Inorganic materials 0.000 description 1
- 230000002209 hydrophobic effect Effects 0.000 description 1
- WPYVAWXEWQSOGY-UHFFFAOYSA-N indium antimonide Chemical compound [Sb]#[In] WPYVAWXEWQSOGY-UHFFFAOYSA-N 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 229930014626 natural product Natural products 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 239000010453 quartz Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000012488 sample solution Substances 0.000 description 1
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N silicon dioxide Inorganic materials O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 1
- 238000012421 spiking Methods 0.000 description 1
- 238000003756 stirring Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000002411 thermogravimetry Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000004876 x-ray fluorescence Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
- G01J3/00—Spectrometry; Spectrophotometry; Monochromators; Measuring colours
- G01J3/28—Investigating the spectrum
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
- G01J3/00—Spectrometry; Spectrophotometry; Monochromators; Measuring colours
- G01J3/28—Investigating the spectrum
- G01J2003/2866—Markers; Calibrating of scan
Definitions
- This invention relates to methods for performing multivariate spectral analyses. More particularly, this invention relates to such methods which combine various features of both classical least squares analysis and the more modern inverse analysis methods such as partial least squares and principle components regression.
- the method may further include improved prediction ability enabled by adding spectral shapes (or selected spectral intensity data) for certain chemicals or other factors that affect the spectral response that may not have been present when spectral calibration data were obtained.
- CLS Classical least squares
- CLS is more restrictive than inverse methods such as PLS and PCR, since CLS methods require that information be known about all spectral sources of variation in the samples (i.e., component concentrations and/or spectral shapes must be known, estimated, or derived.
- Inverse multivariate methods such as PLS and PCR can empirically model interferences and can approximate nonlinear behavior though their inverse soft- modeling approach.
- a method for estimating the quantity of at least one known constituent or property in a sample comprising first forming a classical least squares calibration model to estimate the responses of individual pure components of at least one of the constituents or parameters affecting the optical response of the sample and employing a cross validation of the samples in the calibration data set, then measuring the response of the mixture to the stimulus at a plurality of wavelengths to form a prediction data set, then estimating the quantity of one of the known constituents or parameters affecting the calibration data set by a classical least squares analysis of the prediction data set wherein such analysis produces residual errors, and then passing the residual errors to a partial least squares, principal components regression, or other inverse algorithm to provide an improved estimate of the quantity of the one known constituent or parameter affecting the sample.
- the estimation can be repeated for more of the known constituents in the calibration data set by repeating the last two steps for the other constituents or parameters. Overfitting of the prediction data set by the factor analysis algorithm can be minimized by using only factors derived from each step of the cross validation that are most effective in identifying the constituent or parameter. Also, the accuracy and precision of the classical least square estimation or prediction ability can be improved by adding spectral shapes to either or both of the calibration step or the prediction step that describe the effects on the sample response from constituents that are present in the sample or parameters that affect the optical response of the sample but whose concentrations or values are not in the calibration data base. BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING
- Figure 1 is a graph of cross-validated PLS predictions for urea based upon the 31 dilute aqueous calibration samples collected at a constant temperature of 23 ° C wherein the line of identity is also presented in this and all subsequent prediction plots.
- Figure 2 is a graph of urea PLS predictions obtained when applying the constant-temperature PLS model to variable-temperature solution spectra.
- Figure 3 is a graph of urea predictions of variable-temperature solution spectra obtained by applying the constant-temperature PACLS/PLS model with the addition of the temperature pure-component spectral shape of water with only urea reference concentration information added during calibration.
- Figure 4 is a graph of urea predictions of variable-temperature solution spectra obtained by applying the constant-temperature PACLS/PLS model with the addition of the temperature pure-component spectral shape of water wherein all analyte reference concentration information plus run order were added during calibration.
- Figure 5 is a graph of urea predictions of variable-temperature solution spectra obtained when applying the constant-temperature PACLS/PLS model with the addition of the 5 non-urea CLS-estimated pure-component spectral shapes obtained from the variable-temperature subset samples.
- Figure 6 is a graph of linear-baseline-corrected spectra of the repeat sample spectra (upper plot) and the calibration sample spectra (lower plot).
- Figure 7 is a graph of cross-validated PLS predictions of isopropanol using all the calibration sample spectra collected in the morning.
- Figure 8 is a graph of isopropanol prediction of the afternoon calibration sample spectra based upon a PLS model generated on the morning calibration and morning repeat sample spectra wherein the line above the line of identity is the linear-least squares fit of the predictions.
- Figure 9 is a graph of isopropanol prediction of the afternoon calibration sample spectra based upon a PLS model generated from the morning calibration sample spectra and morning and afternoon repeat sample spectra wherein the line above the line of identity on the left is the linear-least squares fit of the predictions.
- Figure 10 is a graph of isopropanol prediction of the afternoon calibration sample spectra based upon a PACLS/PLS model generated on the morning calibration sample spectra and the first 8 eigenvectors obtained from the morning and afternoon repeat sample spectra wherein the line above the line of identity on the left is the linear-least squares fit of the predictions.
- Figure 11 is a graph of urea predictions for sample spectra obtained on a secondary spectrometer based upon a urea PLS calibration model using sample spectra obtained on the primary spectrometer wherein the line below the line of identity is the linear-least squares fit of the predictions.
- Figure 12 is a graph of urea predictions for sample spectra obtained on a secondary spectrometer based upon a PACLS/PLS calibration model using sample spectra obtained on the primary spectrometer plus spectral difference shapes from 5 subset samples run on both primary and secondary spectrometers wherein the line of identity and the linear-least squares fit of the predictions are indistinguishable on this plot.
- Figure 13 outlines the various steps of the hybrid calibration method using the example of urea analysis in constant-temperature aqueous solutions where the known spectral shape of a temperature change is added during the hybrid calibration.
- Figure 14 outlines the various steps of the hybrid prediction method using the example of predicting urea concentrations in variable temperature aqueous solutions where the spectral shapes added during prediction represent spectral shapes of spectrometer drift obtained from a subset of samples measured during both calibration and prediction experiments.
- the new hybrid method can accommodate any appropriate inverse factor analysis method.
- the invention also includes any multivariate calibration method that uses or contains an inverse method and allows spectral shapes (continuous or discontinuous) to be added directly to the method during calibration or prediction or both, such addition performed without solely weighting by a covariance matrix.
- Another approach to implementing the hybrid algorithm involves modification of the first two CLS-like steps in the PLS calibration to include all the concentration information available about the calibration samples. Therefore, the better qualitative information of CLS is achieved in the first PLS weight-loading vector for the analyte of interest (i.e., this first weight-loading vector is the CLS-estimated pure-component spectrum for the analyte with all available concentration information included).
- This procedure rotates the first PLS weight-loading vector of the analyte in the direction of the pure-component spectrum of the analyte to insure that the first PLS weight- loading vector will have a higher correlation with the analyte concentrations of the calibration samples.
- the hybrid algorithm should yield a more parsimonious model with potentially fewer factors since more concentration information is placed in the early loading vectors.
- the hybrid algorithm can be implemented in the two standard implementations of the PLS algorithm where either the scores or the loading vectors are constructed to be orthogonal.
- This form of the hybrid CLS/PLS algorithm has the advantage of retaining the qualitative interpretation of CLS while improving the parsimony of the PLS algorithm.
- this implementation does not allow the new PACLS procedure to be fully incorporated into the analysis.
- hybrid CLS/PLS algorithm incorporates an initial set of full CLS analysis steps that then passes the CLS concentration residuals and CLS spectral residuals to the PLS algorithm.
- the CLS and PLS portions of the hybrid method are combined in each step of the cross-validation procedure.
- the overfitting present in the first suggested hybrid method is alleviated, and the new PACLS features of CLS can be fully incorporated into both the cross-validated calibration and true prediction portions of the hybrid analysis. Therefore, known spectral shapes present in either the calibration or unknown samples that represent spectral components of unknown concentration in the calibration can be included in the hybrid CLS/PLS analysis.
- a further variation of this third implementation of the hybrid algorithm includes a regression step after the CLS steps that relates the CLS-estimated concentrations with the reference concentrations in this initial pass of the CLS algorithm.
- the additional regression step allows for a slope adjustment to the CLS- estimated concentrations to achieve a better fit to the reference concentrations of the calibration samples.
- the latter capability can serve to adjust the hybrid model for unmodeled components, drifting spectrometers, or changes in spectrometers between calibration and prediction samples.
- These extra spectral shapes can be added during calibration, during prediction, or separate spectral shapes can be added during both calibration and prediction.
- all the features of the PACLS method described in the companion patent application on the PACLS method can be achieved with the CLS/PLS hybrid method.
- Center A and C (preferred but optional). However, if all spectrally active chemical components add to 1 , then to avoid singularity in step 2, either leave out one component in C, do not mean center the data, or vary pathlengths.
- C is derived from C by removing the columns of the C matrix corresponding to the added spectral shapes.
- Step 5 Strip all but single analyte from E c to form concentration residual e c for that analyte.
- the following PLS-like steps are to be performed one selected analyte at a time.
- Step 6 Estimate weight-loading vector ( ⁇ ) for selected analyte.
- E A e c ⁇ +E' A
- E' A is used to differentiate the spectral residuals of the PLS model from the CLS model residuals E A .
- Step 7 Form score vector (t h ) for selected analyte.
- E A t*wj + E' A
- Step 8 Relate score vector ( i h ) to the concentration residual.
- Model: e c v h X h + e' c where e' c represents the new concentration residual for the PLS portion of the hybrid algorithm.
- Step 9 Formation of b h , the PLS loading vector for the CLS spectral residuals, E A .
- E A tX + E' A
- Step 11 Increment h, substitute e' c for e c and E' A for E A and repeat Steps 6 through 10 for the desired number of loading vectors.
- Step 12 Repeat from Steps 5 to 11 for other analytes in the calibration samples.
- Step 1 Center the unknown sample spectrum, a, if A was centered in the calibration. Also scale a if A was scaled.
- Step 2 Augment the rows of K further if more spectral shapes are to be added during prediction (e.g., to accommodate system drift with repeat sample spectral differences). Further augment the unknown sample concentration elements of c that correspond to the spectral shapes added to K .
- Step 3. Estimate the initial CLS-estimated concentrations for the analyte in the unknown sample.
- Step 4 Calculate CLS spectral residuals using model in Step 3.
- e a a-K T c
- Step 6 Calculate the PLS score from CLS spectral residual
- Step 8. e ⁇ e ⁇ _, -b A t ⁇
- the hybrid methods disclosed here can be used for any quantitative spectral analysis independent of the type of spectroscopy used. These methods can be used in any spectroscopy where the spectral data are related to the parameters of interest. In addition to quantitation, qualitative and classification analyses can be performed with the hybrid algorithms.
- Types of spectroscopy where these methods might be used include, but are not restricted to, 1) infrared, 2) near-infrared, 3) visible, 4) ultraviolet, 5) X-ray, 6) gamma-ray, 7) Raman, 8) mass spectroscopy, 9) ion-mobility mass spectroscopy, 10) Auger, 11) fluorescence, 12) phosphorescence, 13) ESCA, 14) far-infrared, 15) microwave, 16) x-ray fluorescence, 17) NMR, 18) energy loss spectroscopy, 19) EDAX, 20) ESR, and 21) multi- and hyper-spectral imaging. These spectroscopic methods can be used in absorption, transmission, reflection, or emission modes.
- the hybrid methods can be applied to other forms of multivariate data such as seismic data, chromatographic data, thermal gravimetric analysis, and image data.
- the hybrid algorithm can also be used for outlier or anomaly detection by examining various outlier metrics such as Mahalanobis distance, spectral residuals, spectral F ratios, or other statistical indications of spectra that lie outside the hybrid model. This additional capability is ideally suited for multi- and hyperspectral image analysis. With the outlier metrics obtained from hybrid algorithm, the hybrid method can also be used for classification purposes.
- spectral components can include either chemical components (molecular or elemental) or other sources of spectral change that include but are not limited to spectrometer/system drift, chromatic aberrations, diffraction effects, temperature changes, sample insertion and alignment effects, purge changes, alignment changes, detector changes and nonlinearities, source changes, and changes in spectrometer components to include even substitution of whole spectrometers.
- this newly disclosed method allows for known spectral shapes to be substituted for chemical composition information during the CLS analysis.
- the spectral components not known during calibration can still be included in the CLS prediction model if their shapes are known. Therefore, either the concentrations or the spectral shapes (derived from theory or from empirical measurements) of spectral components can be used in the CLS model.
- concentrations or the spectral shapes derived from theory or from empirical measurements
- CLS would fail to yield accurate results.
- the CLS analysis result will, in fact, be correct for the components whose concentrations are known during the calibration if all the shapes of components whose concentrations are not known in the calibration samples are added during CLS prediction.
- the NAS is that portion of the analyte signal that is orthogonal to all other sources of spectral variation and is the only portion of the analyte signal that is available for analysis.
- spectral shapes can include the cases where the spectral intensities are intermittent or even scrambled in frequency or wavelength space. The only requirement is that all calibration spectra, prediction spectra, and added spectral shapes have the same ordering and correspondence of frequencies or wavelengths.
- the spectral shapes to be added during cross-validated CLS calibration or during CLS prediction can be determined by a variety of methods. Here, it is assumed that the spectral shapes are a result of the presence of chemical components in the calibration or prediction samples.
- spectral shapes are the result of factors other than molecular or elemental species. If the component shape is unaffected by matrix effects, the spectrum of the pure component can be obtained by simply measuring the spectrum of the pure component. However, quite often, there are matrix effects or interactions between components. In this latter case, the spectral shapes can be obtained by spiking a representative calibration sample with a known (or even unknown) amount of the component. Then spectral differences will yield the spectral shape of the component that also includes displacement of the sample by the component. Another method to determine the spectral shape that is generally preferred is to estimate the component by CLS calibration methods by adding known amounts of the analyte to one or more calibration or known samples.
- the CLS calibration of this generally small set of samples will yield the desired CLS estimated pure-component spectrum to be included in the PACLS algorithm. If the data are not mean centered, the CLS calibration will yield the pure-component spectrum directly. If the data are mean centered or if an intercept is fit during the CLS calibration, then the pure-component spectrum will represent the pure component with displacement of the sample. Either • pure-component spectrum can be used in the analysis since the net-analyte signal for multi-component systems is not affected by the use of the either pure-component spectra or pure-component spectra with displacement.
- spectral shapes are due to spectrometer drift, temperature changes, purge gas changes, sample insertion effects, diffraction effects, or other sources of spectral change that are not due to the chemical components in the system, then the required spectral shapes can be determined through the use of repeat samples.
- the best single repeat sample is generally the sample representing the center of the calibration space.
- the sample spectrum of the repeat sample can be obtained during the period of the calibration. This repeat sample can then represent all the environmental changes occurring during the calibration as reflected by the mid-level sample. It is known that the drift of the spectrometer looks different on different samples. Therefore, a sample that represents the calibration data is the preferred sample to use.
- any change in the sample spectrum will represent spectral shapes that generally have not been explicitly included in the CLS calibration.
- the addition of these spectral changes to the CLS calibration model will compensate for their detrimental influence on predictions.
- it is best to perform an eigenvector analysis see C. L. Lawson and R. J. Hanson, "Solving Least Squares Problems," Prentice-Hall, Englewood Cliffs, NJ (1974)) on the repeat sample spectra and to add only those eigenvector shapes that are detrimental to the CLS calibration. In this manner, the detrimental effects of noise on the analysis can be minimized.
- spectrometer/system changes can be systematically included in the PACLS analysis. It is preferred that these perturbations be performed in a statistically designed manner such as a factorial or fractional factorial design. Examples of changes for infrared and near infrared spectrometers that can be included in the perturbation are system temperature, rate of temperature change, sample temperature, sample insertion effects, source changes (intensity, lifetime, bulb changes, etc.), purge gas composition, alignment changes, etc.
- spectral changes between calibration sample spectra are large; it may be desirable to also take repeat spectra of other samples in the calibration, e.g., samples at the extreme levels of the calibration design. These repeat sample spectra should be mean centered by sample.
- the spectra can be combined, and if desired, an eigenvector analysis can be performed on all mean-centered spectra in order to select only those eigenvectors that are important for reducing errors in the CLS analysis. This procedure minimizes the effects of spectral noise from the added spectral shapes.
- a repeat sample representing the absorbing solvent e.g., water
- the correction of the model for spectrometer/system drift can be obtained by collecting the repeat sample spectrum during true CLS prediction of unknown samples.
- the spectral shape of the difference of the repeat sample spectrum obtained during CLS calibration and prediction can be generated from the spectral difference of these repeat sample spectra. Again, if multiple repeat spectra are obtained or if multiple repeat samples are used for monitoring spectral drift of the system, then mean-centered differences and eigenvector analysis can be employed to generate the shapes added during CLS predictions. Repeat spectra taken as close as possible in time to the unknown sample spectrum should provide the best correction for drift of the system.
- the spectral shapes to add to the CLS analysis can be obtained from the spectra of one or more samples obtained on both spectrometers.
- the mean difference between the sample spectra obtained on the two spectrometers will represent the CLS bias concentration error on the secondary spectrometer.
- the mean-centered spectral differences will model other spectrometer effects that decrease the precision of the analysis on the secondary spectrometer when using a CLS calibration obtained on the primary spectrometer. These spectral differences can also be used to represent and to help model spectrometer drift on the two spectrometers.
- an eigenvector analysis with selection of eigenvectors can improve the quality of the prediction on the secondary instrument by minimizing spectral noise and other spectral variations that may not be important for CLS predictions.
- An additional method to obtain shapes for inclusion in the new PACLS method is to collect spectra from a variety of samples that vary all spectral components not analyzed while all components to be analyzed with CLS are kept invariant at either zero or constant levels. Of course the best approach is to use experimental designs in the generation of these samples or to select the sample spectra based upon spectral extremes and obtaining a good spectral distribution. Again, eigenvector analysis and selection of eigenvectors will improve the final CLS analysis by selecting only important shapes for reducing prediction errors.
- an alternative method to generate the appropriate spectral shapes is to first obtain the spectral shapes of each analyte to be determined in the CLS analysis (using methods as described above) or by any other independent means including measuring the spectrum of the pure analyte directly or by using a library spectrum of the analyte. Then if the analyte spectral shapes are subtracted from the calibration spectra in their known reference concentrations, the resulting spectra represent the best linear estimate of the sample spectra at zero analyte concentration.
- the resulting difference spectra will represent linear combinations of the interfering spectral species and other effects such as nonlinearities, matrix effects, and mispecified pure-component spectrum of the analyte. Adding these resulting difference spectral shapes directly or adding selected eigenvectors from the eigenvector decomposition of these difference spectra to the hybrid calibration or prediction will result in improved predictions.
- the CLS calibration can be adapted to the changes in the starting materials by simply obtaining the starting material spectra for the calibration materials and the comparable materials used in the later CLS prediction analysis whenever the starting materials have changed.
- the difference spectra again represent those spectral differences that were not included in the original calibration.
- the inclusion of these shapes in the prediction phase of the CLS analysis should compensate for these unmodeled spectral differences if the analyte concentrations do not change. If the analyte concentrations change, then these analyte concentration changes must be determined and included in the analysis.
- the spectral shapes to add to the PACLS algorithm can be generated artificially.
- spectral shapes due to the nonlinear effects of stray light in absorbance spectra or specular reflection in reflection-absorbance spectra can be generated and added to the PACLS analysis.
- various portions of the single-beam spectrum of the background can be added to the single-beam sample spectrum.
- These single-beam- augmented spectra are then converted to absorbance (or -log(R) for reflectance spectra).
- the original absorbance or reflectance spectrum is then subtracted from the series of augmented spectra.
- These difference spectra represent the spectral shapes of the nonlinear effects of stray light or specular reflectance. Again eigenvector analysis of these shapes should improve the signal-to-noise of the added shapes.
- the linear component of the shape of the analyte of interest not be present in the spectral shapes that are added in the CLS analysis. If the spectral shapes of the analyte are present in the spectral shapes added during CLS prediction, then the resulting CLS analysis can be significantly degraded due to the loss of net-analyte signal.
- Shapes of nonlinear portions of the component spectra can be determined by assuming nonlinear functional forms for the concentration matrix during calibration. Any explicit functional form of the concentrations may be used. As an example, squared component concentrations can be added to the concentration matrix for all those components thought to experience nonlinear behavior. Concentration cross products can be added for interacting components. The CLS- estimated pure-component spectra for the squared and/or cross-product components can be included during prediction to minimize the influence of nonlinearities in the analysis.
- This method of determining and adding spectral shapes has the advantage that additional experiments do not have to be performed to obtain the spectral shapes for augmentation.
- the hybrid method is also ideally suited for general quantitative multi- and hyperspectral image analysis. It is desired that the pure spectra be generated from the spectral analysis of the spectral images using any one of a variety of methods of determining pure-component spectra from the spectral images (e.g., pixel purity index, see J. W. Boardman, F. A. Kruse, and R. O. Green, "Mapping target signatures via partial unmixing of AVIRIS data: in Summaries," Fifth JPL Airborne Earth Science Workshop, JPL Publication 95-1 , v. 1 , p.
- SIMPLISMA see W. Windig, "Spectral data files for self-mideling curve resolution with examples using the SIMPLISMA approach,” Chemom. and Intell. Lab. Syst. 36, 3-16, 1997, or various methods of multivariate curve resolution (MCR), see R. Tauler, A. Smilde, and B. Kowalski, "Selectivity, Local Rank, Three-Way Data Analysis and Ambiguity in Multivariate Curve Resolution,” J. Chemom. 9, 31-58, 1995).
- MCR multivariate curve resolution
- Independent measures of the potential interfering spectra can be added in a hybrid analysis of the spectral image data.
- Sources of spectral variation that may not be adequately included in the pure-component spectral estimates include but are not limited to, purge or atmospheric variations, spectrometer drift, view angle changes, chromatic aberration, or any spectral component that is not in the calibration image(s) but may be present in the prediction image.
- any multi- and hyperspectral imaging is an important potential beneficiary of the hybrid algorithm.
- the use of multi- and hyperspectral imaging can have a very beneficial impact on the scanning of microarrays for studying the expression of genes.
- univariate measurements of fluorescence are used in a scanning mode to identify and quantify the fluorescence tags of hybridized gene fragments on the microarray slides.
- Performing multi- or hyperspectral imaging of the fluorescence signals on the microarray can increase sensitivity, speed, accuracy, reliability, and the numbers of fluorescence tags that can be simultaneously monitored on the microarray slide.
- the hybrid algorithm is ideally suited to perform this analysis and can be used to minimize or even eliminate the detrimental influences of stray light and contaminant and substrate fluorescence on the quantitative analyses.
- linear estimates of the pure- component spectra of the aberrations can be determined from the difference between the spectra from a homogeneous region and the spectra at the edges of the homogeneous region. If the functional form of the aberration is known, then these functions can be used to obtain pure-component spectra of the aberrations that are better than the linear approximations described above.
- the PACLS- estimated component concentration of the added shape should be a good estimate of the added component. If this spectral shape is present and variable in the calibration spectra, then its PACLS estimated component concentration might not be reliable. All methods previously developed for CLS should also work for PACLS.
- Examples of added features for CLS are weighted least squares, generalized least- squares, base-line fitting, pathlength correction, multi-band analysis (i.e., spectra are separated into multiple spectral regions with CLS analysis applied separately to each spectral region and the results pooled into a final result, see D. M. Haaland and R. G. Easteriing, "Application of New Least Squares Methods for the Quantitative Infrared Analysis of Multicomponent Samples," Applied Spectroscopy 36, 665-673 (1982)). It should be noted that any spectral preprocessing should be applied to the original calibration spectra, the prediction spectra, and the spectral shapes added in the PACLS method.
- Preprocessing steps include but are not limited to baseline correction, pathlength scaling, smoothing, derivatives, normalization, variance or auto scaling, and Fourier transforms.
- pathlength scaling pathlength adjustments are normally applied by dividing the spectra by the pathlength to compensate for the intensity variations expected for pathlength changes in the individual samples. However, this procedure scales all components of the spectra equally. Not all sources of spectral variation should have this pathlength correction applied to them. Examples of spectral variation than should not be scaled include, but are not limited to, drift of the spectrometer, spectrally active purge gas components, and spectral changes external to the sample.
- the effect of pathlength scaling can be removed from the appropriate components by scaling the non-pathlength component concentrations in the CLS calibration by the inverse of the pathlength. This procedure effectively removes the effect of inappropriate scaling of the non-pathlength spectral components.
- a method to select the eigenvectors to include in the PACLS prediction from any of the analyses discussed above is based upon the use of the CLS calibration model to perform a CLS prediction on the eigenvector spectral shapes. Since the eigenvectors are orthogonal basis vectors, the errors from the eigenvectors are independent. Thus, the total error variance of all the spectral shapes included in the eigenvector decomposition is the sum of the error variances of the individual errors from each eigenvector. Therefore, the eigenvectors can be selected from those that represent the largest error sources in the standard CLS calibration (or PACLS calibration if spectral shapes are included during the building of the calibration model and additional shapes are to be included during true prediction or validation).
- the influence of this decrease in NAS can also be incorporated into the selection of eigenvector shapes to include in the PACLS analysis.
- the eigenvectors to include in the PACLS analysis can be selected based upon the inverse of the ratio of the squared change in net-analyte signal (NAS) and squared prediction error.
- CLS and PACLS can be modified to follow any functional form for relating spectra to parameters of interest. Therefore, series of spectra that represent kinetic runs where the kinetics equations are known can use CLS and PACLS methods to extract kinetic rate constants as well as component concentration information. Other effects such as the influence of pH or scattering on the data can be included in the CLS hard modeling if functional forms of the relationship between the spectra and the varying parameter are known.
- Outlier detection has always been an important feature of multivariate spectral analysis methods, and the hybrid algorithm shares this feature.
- outlier metrics such as spectral F ratio, concentration F ratio, Mahalanobis distances, and examination of concentration and spectral residuals are all possible outlier detection methods using the various new forms of the hybrid algorithm.
- Outlier detection significantly improves the reliability of the analysis with the hybrid algorithm.
- Wentzell and coworkers have developed an alternative inverse least squares method where spectral shapes representing estimates of the covariance matrix are incorporated into a PCR analysis through the use of a weighting by the estimated covariance matrix (see P. D. Wentzell, D. T. Andrews, and B. R. Kowalski, Anal.
- PACLS/PLS can also be used to compensate for unmodeled spectral components that are present in the prediction samples but were not present in the calibration samples.
- An example of PACLS/PLS accommodating unmodeled components can be found in spectral data collected from a series of calibration samples consisting of multicomponent mixtures of dilute aqueous samples. The experimental details for the system can be found in D. M. Haaland and H. D. T. Jones, "Multivariate Calibration Applied to Near-Infrared Spectroscopy for the Quantitative Analysis of Dilute Aqueous Solutions," 9th International Conference on Fourier Transform Spectroscopy, J. E. Bertie and H. Wisser, Editors, Proc. SPIE Vol. 2089, pp. 448-449 (1993).
- the 31 calibration samples were prepared in a Latin-Hypercube experimental design with urea, creatinine, and NaCI varying in the range from 0 to ⁇ 3000 mg/dL in water. The concentrations were corrected for displacement of the water solvent.
- the near infrared spectra (7500 - 11000 cm "1 ) of the samples in 10-mm pathlength cuvettes were obtained with a Nicolet 800 Fourier transform infrared (FT-IR) spectrometer.
- FT-IR Fourier transform infrared
- the samples were temperature equilibrated with stirring to 23 ° C with the use of a Hewlett-Packard Peltier temperature controller that includes the capability of using magnetic stirring.
- sample solutions were sealed along with a small Teflon-coated magnetic stirring bar in the cuvettes using a glass cover slip and optical cement.
- the sample spectra were obtained in random run order operating the spectrometer at a nominal 16 cm '1 resolution and signal averaging 256 interferograms.
- the spectrometer source was a 75 W tungsten-halogen lamp, the beamsplitter was made of quartz, and the detector was a liquid-nitrogen-cooled InSb detector.
- the near-infrared spectra of the same set of samples were obtained with each sample temperature controlled to randomly selected 1 ° C intervals from 20 to 25 ° C.
- This constant-temperature PLS model was then used to obtain predictions for the urea component in the variable temperature spectra collected several days after the calibration data.
- the results of these PLS predictions are presented in Fig. 2 where SEP is the standard error of prediction, BCSEP is the bias-corrected SEP, and R 2 is the squared correlation coefficient for the prediction of urea.
- the urea concentration predictions are degraded both by the presence of the unmodeled temperature component and the effects of unmodeled spectrometer drift that occurred during the collection of the variable temperature samples and between each set of experiments. If the shape of the pure-component of temperature obtained during CLS calibration of the 11 variable temperature water spectra is added to the PACLS/PLS calibration, the prediction of urea in the variable temperature data improves considerably. As demonstrated in Fig. 3, these results are even better than was achieved with the PACLS algorithm applied to the same data (see SD-6227). In addition, only urea concentrations and the temperature pure-component spectral shape of water were given to the hybrid algorithm. Therefore, there is a clear relaxation in the information that the hybrid algorithm requires for good prediction results. However, the detrimental effects of instrument drift are still present in the predictions. The instrument drift causes a bias error as well as a slight decrease in precision of the prediction results.
- the hybrid algorithm is given all the concentration information of all the components plus the temperature pure-component spectral shape, a slight further improvement is achieved (see Fig. 4).
- the desired feature of the hybrid model to improve predictions when additional concentration information is present has been realized and the results are an improvement over the PLS2 method previously used to incorporate the additional available concentration information into the calibration.
- a further improvement in analysis precision can be realized if the detrimental effects of spectrometer drift during the collection of the unknown samples can be included in the PACLS/PLS analysis.
- An improvement in prediction ability can be realized by the use of a subset of sample spectra obtained on the spectrometer during both the constant and variable temperature experiments.
- a subset of 5 samples representing the center-point and extreme samples in the concentration calibration space were removed from the variable temperature data set along with repeat measurement spectra of some of these same samples.
- the spectra were then used in a CLS calibration to estimate the 6 pure-component spectra of the analytes (i.e., urea, creatinine, and NaCI), plus the solvent, temperature, and linear spectrometer drift.
- run order was included in the CLS calibration to approximate the linear effects of instrument drift. Since the spectra were obtained at approximately constant time intervals, run order represents time of sample collection. Also, the drift of the system was relatively monotonic with time, so this procedure compensates for some of the detrimental effects of system drift on the CLS calibration.
- All these pure-component spectra include at least a portion of the effects of spectrometer drift between the constant and variable temperature experiments.
- the 5 CLS-estimated pure-component spectra (excluding the urea pure- component spectrum) obtained from the subset of variable temperature samples were added to the PACLS/PLS calibration of the constant temperature data. In this case, urea concentrations from the calibration samples represented the only concentration information presented to the model.
- PACLS/PLS prediction results for urea improve dramatically as shown in Fig. 5.
- the prediction on the subset of samples used for the CLS estimated pure- component spectra from the variable temperature experiment were excluded from the prediction results to eliminate the potential for overfitting of the prediction data.
- Similar improvements in prediction ability were achieved for all the molecular species in the samples using similar PACLS/PLS prediction procedures.
- the PACLS/PLS prediction of temperature for the temperature varying samples can also be achieved.
- the temperature and other pure-component spectral shapes could be added only in the prediction phase of the PACLS/PLS analysis rather than during the building of the original hybrid calibration model. No recalibration of the original constant temperature data is required, and the analysis can be performed simply and rapidly in the prediction phase of the analysis.
- spectrometer/sensor drift can be represented with the use of the center-point repeat sample spectra collected throughout this study. If the repeat sample is truly invariant over time, then the variations in the spectra must represent the spectral variations of spectrometer/sensor drift or spectral effects of insertion variation of the sensor into the spectrometer and the reintroduction of the sample onto the sensor.
- All the repeat spectra are mean-centered (i.e., the average of the repeat spectra was subtracted from each repeat sample spectrum) to generate the spectral shapes representing spectrometer/sensor drift and sample insertion variation.
- the data set was separated into morning and afternoon sets of sample spectra. These two data sets can then be used to test the abilities of PACLS/PLS to maintain a calibration in the presence of a strongly drifting spectrometer/sensor system. Because system drift is the dominant source of spectral variance, this data set represents a challenging test for maintaining a calibration with the PACLS/PLS method.
- a PLS model for isopropanol was built based upon all the morning calibration and morning repeat data.
- the cross- validated PLS calibration results for isopropanol are shown in Fig. 7.
- the PLS model was then used to predict all the non-repeat sample spectra from the afternoon.
- the resulting PLS predictions are shown in Fig. 8.
- the PLS prediction exhibits both a bias and some loss of precision relative to the PLS calibration on the cross-validated PLS analysis of the morning data.
- the expected line of identity and the linear-least- squares fit of the prediction data are included in Fig. 8.
- the first 8 eigenvectors were selected for inclusion in the PACLS/PLS analysis based upon a visual examination of the eigenvectors. Only the first 8 eigenvectors appeared to represent information where the spectral information exceeded the noise. Selection of eigenvectors can be made more systematic by using the CLS/PLS model (without repeat sample spectral shapes added) applied to the eigenvector spectra obtained from the repeat samples. Only those eigenvectors that contribute significantly to the prediction error variance can be selected and added to the PACLS/PLS analysis.
- the eigenvectors to include in the PACLS/PLS analysis can be selected based upon the inverse of the ratio of the squared change in net-analyte signal (NAS) and squared prediction error. Since PACLS was given spectral information from the afternoon repeat spectra, this same information was given to the PLS model to achieve a fair comparison of the two methods. Therefore, PLS and PACLS/PLS calibrations were built using all the morning calibration samples and the repeat spectral information from both the morning and the afternoon parts of the experiment. The PLS calibration included all 43 repeat sample spectra while the PACLS/PLS calibration model used only the first 8 eigenvectors developed from the 43 morning and afternoon repeat spectra.
- NAS net-analyte signal
- the hybrid method can be performed in a different manner than presented above.
- the spectral shapes added in the hybrid analysis were the first 8 eigenvectors from the morning repeats and the difference between the average morning repeat spectrum and the selected afternoon repeat spectrum or spectra.
- the SEP of the afternoon spectra was found to improve relative to performing PLS predictions that were given the same spectral information in the form of added repeat spectra in a recalibrated PLS calibration.
- the final demonstration of the PACLS/PLS method is its use for calibration transfer between two spectrometers.
- the near infrared spectra of the 31 samples of dilute aqueous solutions of urea, creatinine, and NaCi described in the first example in this discussion were obtained on both primary and secondary spectrometers.
- the standard PLS calibration model was built using the spectra obtained on the primary spectrometer at a constant temperature of 23 ° C.
- the cross- validated PLS predictions are shown in Fig. 1 for these calibration data.
- the standard PLS model was then applied directly to spectra obtained on a second Nicolet 800 spectrometer using the same samples again equilibrated to 23 ° C.
- Figure 13 outlines the various steps of the hybrid calibration method using the example of urea analysis in constant-temperature aqueous solutions where the known spectral shape of a temperature change is added during the hybrid calibration.
- Figure 14 outlines the various steps of the hybrid prediction method using the example of predicting urea concentrations in variable temperature aqueous solutions where the spectral shapes added during prediction represent spectral shapes of spectrometer drift obtained from a subset of samples measured during both calibration and prediction experiments.
- spectral shapes is meant to include not only continuous shapes but can also include discontinuous sections or even discrete points within its definition.
- Set can include many elements or a few as a single element.
- Sources of spectral variation will normally mean chemical constituents but can also mean other sources of spectral variation such as temperature/pH shifts, diffraction effects, spectrometer or system drift, sample insertion effects, chromatic aberrations, changes in spectrometers or spectrometer components, purge gas variations, changes in multi- and hyperspectral images, changes in spectral returns from a satellite image evoked by a change in the scanned real estate as the satellite passes over the scene, etc.
- sources of spectral variation such as temperature/pH shifts, diffraction effects, spectrometer or system drift, sample insertion effects, chromatic aberrations, changes in spectrometers or spectrometer components, purge gas variations, changes in multi- and hyperspectral images, changes in spectral returns from a satellite image evoked by a change in the scanned real estate as the satellite passes over the scene, etc.
Landscapes
- Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Physics & Mathematics (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
Description
Claims
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP00913744A EP1166060A1 (en) | 1999-03-05 | 2000-03-03 | Hybrid least squares multivariate spectral analysis methods |
CA002363417A CA2363417A1 (en) | 1999-03-05 | 2000-03-03 | Hybrid least squares multivariate spectral analysis methods |
AU35132/00A AU3513200A (en) | 1999-03-05 | 2000-03-03 | Hybrid least squares multivariate spectral analysis methods |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US12311499P | 1999-03-05 | 1999-03-05 | |
US60/123,114 | 1999-03-05 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2000052433A1 true WO2000052433A1 (en) | 2000-09-08 |
Family
ID=22406791
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2000/005750 WO2000052433A1 (en) | 1999-03-05 | 2000-03-03 | Hybrid least squares multivariate spectral analysis methods |
Country Status (4)
Country | Link |
---|---|
EP (1) | EP1166060A1 (en) |
AU (1) | AU3513200A (en) |
CA (1) | CA2363417A1 (en) |
WO (1) | WO2000052433A1 (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1525534B1 (en) * | 2001-07-23 | 2016-03-30 | Eurofins QTA Inc. | Method of predicting a value of a property of interest using NIR spectroscopy |
CN111195131A (en) * | 2018-11-20 | 2020-05-26 | 三星电子株式会社 | Apparatus for measuring spectrum and apparatus and method for estimating analyte concentration |
CN112149858A (en) * | 2019-06-27 | 2020-12-29 | 中国石油化工股份有限公司 | Method for predicting residual oil fraction SARA composition by using crude oil spectral information |
CN113109318A (en) * | 2021-03-26 | 2021-07-13 | 中国科学院西安光学精密机械研究所 | Raman spectrum quantitative analysis method and system based on spectral peak height direct extraction |
CN113340874A (en) * | 2020-03-02 | 2021-09-03 | 中国科学院沈阳自动化研究所 | Quantitative analysis method based on combined ridge regression and recursive feature elimination |
CN112907033B (en) * | 2021-01-22 | 2024-01-23 | 国能锅炉压力容器检验有限公司 | Heat-resistant steel degradation state evaluation method based on partial least square method |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20210022319A (en) | 2019-08-20 | 2021-03-03 | 삼성전자주식회사 | Apparatus and method for estimating bio-information |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5435309A (en) * | 1993-08-10 | 1995-07-25 | Thomas; Edward V. | Systematic wavelength selection for improved multivariate spectral analysis |
US5596992A (en) * | 1993-06-30 | 1997-01-28 | Sandia Corporation | Multivariate classification of infrared spectra of cell and tissue samples |
-
2000
- 2000-03-03 EP EP00913744A patent/EP1166060A1/en not_active Withdrawn
- 2000-03-03 CA CA002363417A patent/CA2363417A1/en not_active Abandoned
- 2000-03-03 AU AU35132/00A patent/AU3513200A/en not_active Abandoned
- 2000-03-03 WO PCT/US2000/005750 patent/WO2000052433A1/en not_active Application Discontinuation
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5596992A (en) * | 1993-06-30 | 1997-01-28 | Sandia Corporation | Multivariate classification of infrared spectra of cell and tissue samples |
US5435309A (en) * | 1993-08-10 | 1995-07-25 | Thomas; Edward V. | Systematic wavelength selection for improved multivariate spectral analysis |
Non-Patent Citations (1)
Title |
---|
HAALAND ET. AL.: "Partial Least-Squares Methods for Spectra Analyses 1: Relation to Other Multivatiate Calibration Methods and the Extraction of Qualitative Information", ANALYTICAL CHEMISTRY, vol. 60, no. 11, 1 June 1988 (1988-06-01), pages 1193 - 1202, XP002929493 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1525534B1 (en) * | 2001-07-23 | 2016-03-30 | Eurofins QTA Inc. | Method of predicting a value of a property of interest using NIR spectroscopy |
CN111195131A (en) * | 2018-11-20 | 2020-05-26 | 三星电子株式会社 | Apparatus for measuring spectrum and apparatus and method for estimating analyte concentration |
CN112149858A (en) * | 2019-06-27 | 2020-12-29 | 中国石油化工股份有限公司 | Method for predicting residual oil fraction SARA composition by using crude oil spectral information |
CN113340874A (en) * | 2020-03-02 | 2021-09-03 | 中国科学院沈阳自动化研究所 | Quantitative analysis method based on combined ridge regression and recursive feature elimination |
CN113340874B (en) * | 2020-03-02 | 2023-07-18 | 中国科学院沈阳自动化研究所 | Quantitative analysis method based on combination ridge regression and recursive feature elimination |
CN112907033B (en) * | 2021-01-22 | 2024-01-23 | 国能锅炉压力容器检验有限公司 | Heat-resistant steel degradation state evaluation method based on partial least square method |
CN113109318A (en) * | 2021-03-26 | 2021-07-13 | 中国科学院西安光学精密机械研究所 | Raman spectrum quantitative analysis method and system based on spectral peak height direct extraction |
Also Published As
Publication number | Publication date |
---|---|
EP1166060A1 (en) | 2002-01-02 |
CA2363417A1 (en) | 2000-09-08 |
AU3513200A (en) | 2000-09-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US6341257B1 (en) | Hybrid least squares multivariate spectral analysis methods | |
US6415233B1 (en) | Classical least squares multivariate spectral analysis | |
Haaland et al. | New prediction-augmented classical least-squares (PACLS) methods: application to unmodeled interferents | |
Roger et al. | EPO–PLS external parameter orthogonalisation of PLS application to temperature-independent measurement of sugar content of intact fruits | |
US6842702B2 (en) | Augmented classical least squares multivariate spectral analysis | |
Rossel et al. | Visible, near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties | |
US7620674B2 (en) | Method and apparatus for enhanced estimation of an analyte property through multiple region transformation | |
Chen et al. | The combined optimization of Savitzky-Golay smoothing and multiplicative scatter correction for FT-NIR PLS models | |
US5568400A (en) | Multiplicative signal correction method and apparatus | |
Kriesten et al. | Identification of unknown pure component spectra by indirect hard modeling | |
US9207182B2 (en) | Analysis method of vibrational spectra | |
Workman Jr | NIR spectroscopy calibration basics | |
Gallagher et al. | Application of extended inverse scatter correction to mid‐infrared reflectance spectra of soil | |
Nørgaard | A multivariate chemometric approach to fluorescence spectroscopy | |
Divya et al. | Combining synchronous fluorescence spectroscopy with multivariate methods for the analysis of petrol–kerosene mixtures | |
Boysworth et al. | Aspects of multivariate calibration applied to near-infrared spectroscopy | |
Kompany-Zareh et al. | Spectrophotometric resolution of ternary mixtures of pseudoephedrine hydrochloride, dextromethorphan hydrobromide, and sodium benzoate in syrups using wavelength selection by net analyte signals calculated with hybrid linear analysis | |
DeBraekeleer et al. | Influence and correction of temperature perturbations on NIR spectra during the monitoring of a polymorph conversion process prior to self-modelling mixture analysis | |
WO2000052433A1 (en) | Hybrid least squares multivariate spectral analysis methods | |
Nørgaard | Direct standardisation in multi wavelength fluorescence spectroscopy | |
Skibsted et al. | New indicator for optimal preprocessing and wavelength selection of near-infrared spectra | |
Wehlburg et al. | New hybrid algorithm for maintaining multivariate quantitative calibrations of a near-infrared spectrometer | |
Boysworth et al. | Aspects of multivariate calibration applied to near-infrared spectroscopy | |
Workman Jr et al. | Calibration transfer chemometrics, part II: a review of the subject | |
Verdú-Andrés et al. | Correction of non-linearities in spectroscopic multivariate calibration by using transformed original variables and PLS regression |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AL AM AT AU AZ BA BB BG BR BY CA CH CN CR CU CZ DE DK DM EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT TZ UA UG UZ VN YU ZA ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): GH GM KE LS MW SD SL SZ TZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
DFPE | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101) | ||
WWE | Wipo information: entry into national phase |
Ref document number: 2000913744 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2363417 Country of ref document: CA Ref country code: CA Ref document number: 2363417 Kind code of ref document: A Format of ref document f/p: F |
|
WWP | Wipo information: published in national office |
Ref document number: 2000913744 Country of ref document: EP |
|
REG | Reference to national code |
Ref country code: DE Ref legal event code: 8642 |
|
WWW | Wipo information: withdrawn in national office |
Ref document number: 2000913744 Country of ref document: EP |