A method for analyzing spectral peaks using neural networks

Abstract

A method of operating a spectrometer controller is provided. The method includes obtaining an interfered peak using a detector of the spectrometer, the interfered peak being generated by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions in the interfered peak being incident on a detector at an associated detector location. For one or more of the spectral emissions of the interfered peak, an associated curve is generated using a neural network, the neural network being trained to output data indicative of a shape of the associated curve based on data representative of the associated detector location. For one or more of the spectral emissions of the interfered peak, the associated curve is output.

Description

[Technical Field]

[0001] FIELD OF THE DISCLOSURE The present disclosure relates to methods for analyzing spectral peaks, and in particular to methods for analyzing spectral peaks generated using a spectrometer. [Background technology]

[0002] Spectroscopy is an analytical technique for analyzing a sample.

[0003] Thus, a spectrometer may generate multiple spectral peaks from a sample. Part of the process of analyzing multiple spectral peaks involves identifying the spectral peaks from the measurement data. The identification process typically involves fitting a curve to the measurement data to identify the peak locations (and associated wavelengths) and peak intensities. The peak wavelengths and intensities can be used to determine the elements present in the sample and the relative amounts of each element.

[0004] The process of fitting a curve to measured data typically involves assumptions about peak shape. For example, "Quantification and deconvolution of asymmetric LC-MS peaks using the bi-Gaussian mixture model and statistical model selection," Yu T., and Peng H., BMC Bioinformatics (November 12, 2010), describes a method for quantifying asymmetric chromatographic peaks measured using a liquid chromatography-mass spectrometry system. The described method uses a bi-Gaussian peak model to fit a curve to the measured data. Summary of the Invention [Problem to be solved by the invention]

[0005] In spectroscopic measurements, the spectral peaks generated from a sample may contain two or more spectral peaks with similar wavelengths. As a result, when the spectral peaks are imaged onto a detector, the spectral peaks with similar wavelengths may overlap. When the spectral peaks overlap, interference between the overlapping spectral peaks may lead to erroneous identification.

[0006] When overlapping spectral peaks occur, a user may choose not to use the overlapping spectral peaks for further analysis. Discarding the analysis of overlapping spectral peaks increases the time it takes to analyze the sample, requires user input to review the overlapping peaks, and prevents the user from taking advantage of all of the available spectral data.

[0007] Alternatively, interelement correction algorithms can be applied to separate overlapping peaks, as described in "Interelement Corrections in Spectrochemistry," Volker, T.; Schatzlein, D.; and Mercuro, D., Spectroscopy, v. 21, n. 7, p. 32 (July 2006). Performing interelement corrections may not be feasible in some measurement situations, requires additional user effort, and does not necessarily produce accurate results.

[0008] Accordingly, the present disclosure seeks to provide a method for spectral peak analysis that addresses at least one of the problems associated with prior art methods, or at least provides a commercially useful alternative. [Means for solving the problem]

[0009] According to a first aspect, there is provided a method of operating a spectrometer controller, the method comprising: acquiring an interfered peak using a detector of the spectrometer, the interfered peak being produced by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions in the interfered peak being incident on the detector at an associated detector location; generating an associated curve for one or more of the spectral emissions of the interfered peaks using a neural network, the neural network being trained to output data indicative of the shape of the associated curve based on data representing the associated detector positions; and outputting an associated curve for one or more of the spectral emissions of the interfered peaks.

[0010] According to the method of the first aspect, an interfered peak acquired by a spectrometer controller can be processed. In particular, the method of the first aspect processes an interfered peak generated by at least two spectral emissions of different wavelengths. The method of the first aspect provides a method for generating a set of curves for the interfered peak to identify the underlying spectral emissions that form the interfered peak. Thus, the method of the first aspect allows different spectral emissions from the interfered peak to be characterized by the generated curves (whose parameters include the peak intensity and peak wavelength of each of the spectral emissions), so that information from the interfered peak can be used for further analysis. In other words, the method of the first aspect allows users to utilize a greater proportion of the sample spectrum using a peak identification process that is computationally efficient and requires less manual user intervention, thereby increasing throughput.

[0011] To process the interfered peaks, the method of the first aspect generates a set of curves for the interfered peaks. The present invention recognizes that the contribution to the overall shape of the interfered peak from each different spectral radiation depends, at least in part, on optical aberrations introduced by the spectrometer as a result of the detector and any associated optics. The degree of optical aberration depends on the detector position (detector position has an associated wavelength). Due to the position-varying nature of optical aberrations, it is difficult to accurately fit curves to the interfered peaks using any conventional technique that assumes that each spectral radiation has the same peak shape across the detector, and therefore that the peak shape is position-independent. Such any assumption leads to inaccurate analysis due to the nature of the optical aberrations introduced by the spectrometer.

[0012] By accounting for the optical aberrations of the spectrometer, the peak shape associated with each spectral radiation can be more accurately generated. Thus, the contribution of each spectral radiation to the interfered peak can be more accurately accounted for. For example, the area under the interfered peak can be more accurately attributed to one or more of the spectral radiations, thereby improving the accuracy of any subsequent analytical techniques based on the area under the associated peak.

[0013] To accurately generate a set of curves for the interfered peaks, the method of the first aspect provides a neural network technique that outputs a peak shape for each spectral emission that forms the interfered peak. In some embodiments, the set of curves associated with the interfered peaks is the set of peak shapes output by the neural network technique, while in other embodiments, the set of peak shapes output by the neural network technique can be further modified to generate a set of curves including multiple adjusted curves associated with the interfered peaks. Each curve in the set of curves associated with the interfered peaks can have an associated peak wavelength and an associated peak intensity. The curves in the set of curves can then be output by the method for further processing of the underlying spectral emissions that form the interfered peaks.

[0014] In some embodiments, the neural network technique includes deploying a trained neural network to output coded representations of peak shapes associated with the spectral emissions of the interfered peaks based on data representing the detector positions of the spectral emissions (and, in some embodiments, data representing the peak intensities of each spectral emission). In some embodiments, the neural network technique may include applying a decoder to the output of the neural network to decode the coded representations of each peak shape to generate a peak shape for each spectral emission associated with the interfered peaks.

[0015] In some embodiments, the neural network is trained to predict the peak shape of each spectral emission based on a training dataset including multiple training peaks. In some embodiments, each training peak is a single spectral emission associated with a different detector position acquired by the spectrometer. That is, the training peaks have different wavelengths to characterize the spectral peaks distributed across the detectors at multiple different detector positions. In some embodiments, at least some of the training datasets are acquired by the same spectrometer used to generate the interfered peaks. Thus, the neural network may be trained using measurement data from the same spectrometer that generates the sample peaks to be processed. Therefore, optical aberrations present in the sample spectrum to be processed will also be present in the training data used to train the neural network. Therefore, because the training dataset can reflect optical aberrations introduced by the spectrometer, the neural network can be trained to output peak shapes of the spectral emissions of the interfered peaks that more accurately characterize the interfered peaks than conventional methods.

[0016] In some embodiments, other spectrometers having the same type of detector may be used to generate one or more training peaks that form part of the training dataset. That is, the other spectrometers may have similar (preferably identical) detectors and optical arrangements. For example, spectrometers of similar models with generally similar components (detectors, optical arrangements, etc.) may have similar optical aberrations, so that training peaks generated by one spectrometer can be used as part of the training dataset for other spectrometers of similar models. Thus, the training dataset may include training peaks from multiple spectrometers.

[0017] In some embodiments, multiple training peaks in a training dataset are obtained by measuring one or more calibration samples using a spectrometer. Calibration samples with known compositions may be provided so that the resulting calibration spectrum has multiple known single-peak spectral sources (i.e., uninterfered peaks). In some embodiments, the multiple known single peaks may be distributed across the detector to allow the optical aberrations of the spectrometer to be accurately characterized by the training peaks.

[0018] In some embodiments, the calibration sample comprises a single element solution. In some embodiments, the single element solution comprises a transition metal. In particular, when the calibration sample is used to calibrate an optical emission spectrometer, the transition metal spectrum comprises multiple spectral emissions over a wide range of wavelengths. Therefore, a calibration sample comprising a single element solution of a transition metal is well suited to providing multiple training peaks for a neural network used in the method of the first aspect.

[0019] In some embodiments, the spectrometer is a topological spectrometer. That is, the spectrometer is configured to generate a signal whose feature-level representation (i.e., a function approximation) of its shape is topologically related to the detector of the topological spectrometer. For example, the topological spectrometer may be a detector, such as an array detector, configured to detect a signal that is topologically distributed across the detector. Examples of topological spectrometers include atomic emission spectrometers, optical emission spectrometers, x-ray fluorescence spectroscopy systems, and laser-induced breakdown spectroscopy systems. In some embodiments, the spectrometer comprises an echelle grating and a two-dimensional detector (i.e., a two-dimensional array detector), and the spectrometer uses the echelle grating to diffract light onto the two-dimensional detector to generate a sample spectrum. Such spectrometers distribute sample peaks across the two-dimensional detector as multiple orders. That is, the sample peaks are spatially distributed across the detector. Therefore, the effect of optical aberrations introduced by the spectrometer on the peak shapes of the spectral radiation and their associated wavelengths can be highly nonlinear. Therefore, the method of the first aspect is particularly suitable for characterizing optical aberrations introduced by echelle gratings and related optical elements.

[0020] In some embodiments, the method further comprises identifying the sample peak as an interfered peak. The step of identifying the sample peak as an interfered peak may be performed before the spectrometer controller acquires the interfered peak for purposes of performing the method of the first aspect.

[0021] In some embodiments, identifying a sample peak as an interfered peak includes calculating a first derivative of the sample peak, and the sample peak is determined to be an interfered peak when the number of zero crossings of the first derivative of the sample peak is greater than one. Thus, the method of the first aspect can use efficient analysis techniques to identify interfered peaks for further analysis. In other embodiments, a user can designate a particular peak or area of ​​the sample spectrum as an interfered peak for further processing according to the first aspect.

[0022] In some embodiments, the associated detector position of each spectral emission in the interfered peak is determined based on the zero crossing of the first derivative of the sample peak. Thus, the process of determining whether a sample peak is an interfered peak can also be used to provide a starting point for prediction of the curve forming the interfered peak. The initially determined peak wavelength and peak intensity can then be further refined by the method of the first aspect using neural network techniques.

[0023] In some embodiments, the peak intensities determined by the controller may be provided to a neural network to generate associated curves. In some embodiments, the neural network uses the peak intensities to output one or more curves associated with the spectral emissions of each of the interfered peaks. In some embodiments, the peak intensities may not be provided as input to the neural network, but instead may be applied to the output of the neural network (e.g., by scaling) to ensure that the curves in the resulting set of curves have the correct peak intensities.

[0024] In some embodiments, the detector of the spectrometer is an array detector, which should be understood to mean that the detector includes a plurality of detecting elements (e.g., pixels) arranged in an array. The array may be a one-dimensional array or a two-dimensional array.

[0025] In some embodiments, the spectrometer is an atomic emission spectrometer and the spectrometer controller is an atomic emission spectrometer controller. In particular, the spectrometer may be an optical emission spectrometer and the spectrometer controller may be an optical emission spectrometer controller. The method of the first aspect may also be applied to other types of spectrometers (and associated controllers), such as X-ray fluorescence spectroscopy systems, laser-induced breakdown spectroscopy systems, etc.

[0026] In some embodiments, a curve is output for each of the spectral emissions in the interfered peaks, and each curve output can be used to directly predict the peak shape of the associated spectral emission.

[0027] In some embodiments, the comparative curve associated with a spectral emission is obtained by subtracting the curve of another spectral emission from the interfered peak, and it will be appreciated that the comparative curve may also show the peak shape of the associated spectral emission.

[0028] Ideally, the peak shape predicted by the comparison curve for a spectral radiation should closely match the peak shape of the spectral radiation predicted directly by the neural network. This relationship between the comparison curve and the directly predicted curve can be used as a cross-check. Thus, in some embodiments, the method further includes performing a confidence analysis that includes comparing the comparison curve for the spectral radiation with a curve output by the spectrometer controller for the same spectral radiation and determining a confidence level for the curve output by the spectrometer controller based on the comparison. The confidence level may be a numeric value (e.g., mean square error) or one of several discrete flags (e.g., pass / fail, green / yellow / red) based on one or more thresholds for the difference between the two curves.

[0029] According to a second aspect of the present disclosure, there is provided a spectrometer controller for a spectrometer, the spectrometer controller comprising: acquiring an interfered peak using a detector of the spectrometer, the interfered peak being produced by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions in the interfered peak being incident on the detector at an associated detector location; generating an associated curve for one or more of the spectral emissions of the interfered peaks using a neural network, the neural network being trained to output data indicative of the shape of the associated curve based on data representing the associated detector positions; and outputting an associated curve for one or more of the spectral emissions of the interfered peaks.

[0030] The spectrometer controller of the second aspect may therefore be configured to perform the method of the first aspect, and may therefore incorporate any of the optional features and associated advantages of the first aspect.

[0031] The spectrometer controller of the second aspect may be provided using a spectrometer controller of a spectrometer measurement system. In some embodiments, the spectrometer controller may comprise a processor (e.g., a microprocessor) or the like.

[0032] According to a third aspect of the present disclosure, there is provided a spectroscopic measurement system comprising a spectrometer and a spectrometer controller. The spectrometer includes a detector. The spectrometer is configured to generate a sample spectrum from the sample using the detector. The spectrometer controller is configured to process the sample spectrum, the controller comprising: obtaining an interfered peak from the sample spectrum using a detector of the spectrometer, the interfered peak being produced by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions in the interfered peak being incident on the detector at an associated detector location; generating, for one or more of the spectral emissions of the interfered peaks, an associated curve using a neural network, the neural network being trained to output data indicative of the shape of the associated curve based on data representing the associated detector positions; and outputting an associated curve for one or more of the spectral emissions of the interfered peaks.

[0033] Thus, a spectroscopic measurement system may comprise the spectrometer controller of the second aspect. The spectroscopic measurement system may be configured to perform the method of the first aspect. It will be appreciated that the spectroscopic measurement system of the third aspect may therefore incorporate any of the optional features and associated advantages of the first or second aspects described above.

[0034] In some embodiments, the spectrometer comprises a plasma source.

[0035] According to a fourth aspect of the present disclosure, there is provided a computer program comprising instructions configured, when executed by one or more processing devices of a controller, to cause the controller of the second aspect or the spectroscopic measurement system of the third aspect to perform the steps of the first aspect.

[0036] According to a fifth aspect of the present disclosure, there is provided a computer-readable storage medium storing the computer program of the fourth aspect.

[0037] The invention may be carried out in several ways and specific embodiments will now be described, by way of example only, with reference to the accompanying drawings in which: [Brief explanation of the drawings]

[0038] [Figure 1] FIG. 1 shows a schematic diagram of a spectroscopic measurement system. [Figure 2]1 illustrates a detector imaging a spectrum produced by an echelle grating. [Figure 3] 1 shows a block diagram of a method for analyzing spectral peaks of a sample spectrum according to an embodiment of the present disclosure. [Figure 4] 1 shows an example of an interfered peak resulting from two spectral emissions of different wavelengths. [Figure 5] 1 shows an image recorded by the detector, including multiple training peaks. [Figure 6] 1 shows a schematic diagram of an autoencoder used to encode spectral peaks. [Figure 7] FIG. 1 shows a block diagram of a method for fitting multiple curves to an interfered peak. [Figure 8] 1 shows a further example of an interfered peak resulting from at least three spectral emissions of different wavelengths. [Figure 9] 9 shows a graph of three curves generated from the interfered peaks of FIG. 8. [Figure 10] 9 shows a graph of the curve generated for the central spectral emission of FIG. 8 and the associated baseline correction. [Figure 11] FIG. 1 illustrates a block diagram of a method for retraining a neural network algorithm according to the present disclosure. [Figure 12] FIG. 1 is a block diagram of a method for performing trust analysis according to the present disclosure. [Figure 13] As part of the confidence analysis, a graph of the comparison curve generated for the interfered peak in FIG. 8 is shown. [Figure 14] 10 shows a graph of a comparison curve generated for an interfered peak as part of a confidence analysis when the predicted curve has relatively low confidence. DETAILED DESCRIPTION OF THE INVENTION

[0039] According to one embodiment of the present disclosure, a spectroscopic measurement system 10 is provided. The spectroscopic measurement system 10 is configured to perform a spectroscopic measurement method on a sample to generate a sample spectrum. The spectroscopic measurement system 10 may also process sample peaks in the sample spectrum according to the methods of the present disclosure. A schematic diagram of the spectroscopic measurement system 10 is shown in FIG. 1. As shown in FIG. 1, the spectroscopic measurement system 10 includes an excitation source 11, an optical arrangement 12, a detector 13, a processor (μP) 14, a memory 15, and an input / output (I / O) unit 16. Although the spectroscopic measurement system 10 of FIG. 1 may be an optical emission spectroscopy system, the embodiments disclosed herein may be applied to any suitable spectroscopic measurement system, such as an x-ray fluorescence spectroscopy system, a laser-induced breakdown spectroscopy system, or any other suitable spectroscopic measurement system (the elements of the spectroscopic measurement system 10 may be provided by similar elements of other types of spectroscopic measurement systems, as is known in the art).

[0040] 1, excitation source 11 is a plasma source, such as an inductively coupled plasma (ICP) source. In other embodiments, excitation source 11 may be a furnace or any other high-energy electromagnetic source that generates excited species suitable for use in spectroscopy. Excitation source 11 may also be configured to receive a sample to be analyzed using spectroscopy system 10. For example, if excitation source 11 is a plasma source, the sample may be introduced into the plasma, where it interacts with the plasma. A sample in liquid form may be introduced directly into the plasma source, while a solid sample may be introduced using, for example, laser ablation or vaporization.

[0041] In the embodiment of Figure 1, optical arrangement 12 may comprise an echelle grating and a prism (and / or further gratings) to generate a two-dimensional image of the light generated by excitation source 11 (and the sample, if present). The two-dimensional image is formed on detector 13. It will be appreciated that in such an arrangement, optical arrangement 12 may be configured to direct radiation from excitation source 11 to the detector such that the radiation is suitable for detection by detector 13. The entire area of ​​detector 13 that can return radiation information to processor 14 may be referred to as a "full frame."

[0042] In the embodiment of FIG. 1, detector 13 may be a CCD (charge-coupled device) array. A typical CCD array may have at least about 1024 x 1024 pixels (i.e., 1 megapixel). In other embodiments, detector 13 may be a complementary metal-oxide semiconductor (CMOS) or charge injection device (CID) detector. The CCD array (or other detector 13) may be configured to generate a spectral intensity value for each pixel of detector 13 representing the intensity of light incident on the pixel. Detector 13 is configured to transfer the photon intensity values ​​to processor 14. As such, detector 13 may be a multi-channel detector configured to detect multiple different wavelengths. Detector 13 (as in the embodiment of FIG. 1) may be configured to detect a two-dimensional spectrum. Detector 13 is configured to output the recorded intensity for each pixel of detector 13 to processor 14 for further analysis.

[0043] The processor 14 (controller) may comprise one or more commercially available microprocessors or any other suitable processing device. The memory 15 may be a suitable semiconductor memory and may be used to store instructions that enable the processor 14 to perform method embodiments according to the present disclosure. The processor 14 and the memory 15 may be configured to control the spectroscopic measurement system 10 to perform methods according to embodiments of the present disclosure. Thus, the memory 15 may include instructions that, when executed by the processor 14, cause the spectroscopic measurement system 10 to perform methods according to embodiments of the present disclosure.

[0044] Spectroscopy system 10 can be configured to generate a sample spectrum by introducing a sample into excitation source 11. Excitation source 11 interacts with the sample, causing the sample to emit spectral radiation specific to the sample. The spectral radiation from excitation source 11 and the sample is directed to detector 13 by optical arrangement 12. An echelle diffraction grating in optical arrangement 12 diffracts different wavelengths of spectral radiation by varying amounts, causing peaks associated with the different spectral radiation to be detected at different locations on detector 13. Thus, the location of the spectral radiation on detector 13, or a pixel number (x) representing the detector location at which the spectral radiation is incident, can be converted to wavelength based on the known relationship between detector location / pixel number and wavelength in spectroscopic measurement system 10. Thus, spectroscopic measurement system 10 according to the present disclosure can interchangeably refer to the wavelength of the interfered peak as the detector location or pixel number on detector 13.

[0045] FIG. 2 shows a schematic diagram of the two-dimensional detector 13 of the embodiment of FIG. 1. The two-dimensional detector 13 of FIG. 2 is formed from an array of pixels, although each pixel is not individually represented in FIG. 2. FIG. 2 includes a schematic diagram (dashed lines) of orders of light 20 diffracted by an echelle grating and prism and imaged onto the detector 13. Each order 20 corresponds to a different wavelength range, with the wavelength varying laterally along each order. For example, in the embodiment of FIG. 2, the wavelength of light may increase along each order from left to right. The starting wavelength (on the left) of each order may also increase, from order a) to order i). FIG. 2 also shows four detailed views of an exemplary single spectral radiation imaged by groups of detector pixels at different locations on the detector 13. It will be understood that the peak shape of each of the spectral radiation will differ, at least in part, based on optical aberrations of the optical arrangement 12.

[0046] Each spectral radiation incident on the detector 13 may be detected as a peak incident across multiple pixels of the detector 13. The shape of the peak associated with the spectral radiation depends, at least in part, on the optical arrangement 12 used to diffract and focus the spectral radiation onto the detector 13. For example, if the optical arrangement 12 includes an echelle grating, the optical aberrations introduced by the echelle grating will vary depending on the position on the detector 13 at which the spectral radiation is directed. Thus, the shape of the peak measured by the spectroscopy measurement system 10 may depend on the detector position (representing the wavelength) of the peak. It will be understood that in some optical arrangements 12, the same wavelength may be diffracted to multiple positions on the detector 13. Thus, while a detector position may be associated with a wavelength, the wavelength of the peak may be associated with multiple detector positions.

[0047] When two spectral emissions have similar wavelengths, the peaks associated with each spectral emission may be directed toward similar regions of the detector. When two spectral emissions are directed toward similar regions of the detector such that at least a portion of one peak overlaps another, it may be difficult to separate the individual peaks. These peaks are known as interfered peaks. Separating the peaks may be particularly difficult due to varying optical aberrations introduced by the optical arrangement, which may cause the peak shapes of the individual spectral emissions to vary across the detector / with wavelength.

[0048] Thus, the spectroscopic measurement system 10 according to the present disclosure provides a method for analyzing the interfered peaks of a sample spectrum to separate the different spectral emissions that form the interfered peaks.

[0049] A method 100 for analyzing spectral peaks of a sample spectrum will now be described with reference to Figure 3, which shows a block diagram of method 100. Method 100 may be performed by a processor of spectroscopic measurement system 10. Alternatively, method 100 may be performed by any other processor to which the sample spectrum generated by spectroscopic measurement system 10 is provided.

[0050] In step 102 of method 100, processor 14 identifies whether a sample peak in the sample spectrum is an interfered peak. The sample spectrum may include multiple peaks generated from the spectral emissions of spectroscopic measurement system 10. An interfered peak is the result of two or more spectral emissions overlapping on the same region of the detector. That is, peaks from two or more spectral emissions may be close together on the detector (e.g., within about 20 pixels of each other in some spectroscopic measurement systems 10) such that at least a portion of the peak associated with each spectral emission overlaps with one or more other peaks of the other spectral emissions.

[0051]

[0023] Figure 4 shows an example of an interfered peak that may be processed according to an embodiment of the present disclosure. Figure 4 shows an example of an interfered peak resulting from two spectral emissions of different wavelengths. In the example of Figure 4, the peaks of the two spectral emissions are within 20 pixels of each other. Therefore, the two spectral emissions overlap.

[0052] In method 100, the interfered peak shown in FIG. 4 can be distinguished from other peaks in the sample spectrum by analyzing the first derivative of the sample spectrum. A sample peak in the sample spectrum resulting from a single spectral emission can be distinguished from an interfered peak based on, for example, the number of zero crossings of the first derivative within a specific wavelength range of the wavelength corresponding to the peak's maximum intensity (e.g., ±10 pixels from the pixel where the intensity of the sample peak is at the peak). Based on the number of zero crossings of the first derivative present, the method can determine the number of different spectral emissions that form the interfered peak. For each different spectral emission present in the interfered peak, method 100 attempts to generate a curve representing the spectral emission. Thus, if the first derivative of the sample spectrum indicates that two different spectral emissions are present in the interfered peak (as in FIG. 4), method 100 then generates a set of curves including two curves, each associated with a different one of the spectral emissions in the interfered peak.

[0053] If an interfered peak is identified in step 102, method 100 proceeds to step 104, where a set of curves associated with the interfered peak is generated. To generate the set of curves associated with the interfered peak, a neural network is used to output a peak shape for each spectral emission that forms part of the interfered peak. In the example of Figure 4, the neural network outputs two peak shapes to be fitted to the interfered peak.

[0054] It will be appreciated that the method of Figure 3 relies on neural network techniques to process interfered peaks. Before describing step 104, which uses neural network techniques to predict peak shapes in detail, a process for training the neural network used in the neural network techniques will be described.

[0055] As described above, a neural network can be trained to output a peak shape based on data representing the peak wavelength of the spectral radiation (and, in some embodiments, the peak intensity of the spectral radiation). The peak shape output by the neural network can be an encoded peak shape, which represents the peak shape using a fixed number of parameters. A particular encoded output by the neural network can be that produced by an autoencoder trained on a training data set. A diagram of an autoencoder is shown in FIG. 6. An autoencoder can include an encoder and a decoder, where the encoder can receive an uninterrupted peak as input and output a low-dimensional encoded representation of the peak's shape. Referring to FIG. 6, the number of hidden nodes in the autoencoder corresponds to the number of parameters used in the encoded representation of the input peak (e.g., three parameters in FIG. 6). The decoder can receive as input a number of parameters corresponding to the number of hidden nodes and output a peak having the same dimensionality as the input to the encoder. The training peaks in the training dataset may be normalized (e.g., scaled so that the maximum values ​​of the training peaks all equal the same constant value, such as 1) to ensure that the training of the autoencoder focuses on peak shape and to prevent the autoencoder from making inappropriate intensity-related inferences (since intensity information in the sample spectra is related to properties of the underlying samples). The autoencoder (which itself takes the form of a neural network, as known in the art) may be trained using techniques known in the art to reduce the dimensionality of the input peaks to the number of nodes in a hidden layer and then reconstruct the peaks using the reduced dimensionality information. In some embodiments, the autoencoder may be iteratively trained with successively smaller numbers of hidden nodes, the resulting encoding / decoding performance may be evaluated, and the smallest hidden layer (corresponding to the number of parameters in the encoding) that achieves a desired level of performance (e.g., a desired accuracy in the encoding / decoding) may be selected. In some embodiments, the autoencoder may be a linear autoencoder.

[0056] To train a neural network that will output peak shape information, the training peaks can be provided to a trained encoder, whose output can be a shape parameter vector (with a number of elements equal to the number of hidden nodes). Thus, each training peak can be reduced to a shape parameter vector containing a selected number of shape parameters (e.g., three shape parameters (p1, p2, p3) in the embodiment of FIG. 6). The training peak can be reconstructed by providing this shape parameter vector to a trained decoder. Of course, in other embodiments, the nature of the optical aberrations in a particular spectroscopic measurement system 10 can result in a shape parameter vector with a different number of parameters.

[0057] The neural network may then be trained on the shape parameter vectors representing the training peaks. In particular, the neural network may be trained with input-output pairs where the input is data representing the peak wavelengths of the training peaks (e.g., detector positions) and the output is the shape parameter vectors of the training peaks. When training the neural network using a large number of training peaks associated with many different peak wavelengths across the detector 13, the neural network may learn to predict the peak shape (shape parameter vector) of a peak based on the data representing the peak wavelength (e.g., input detector positions / pixel numbers).

[0058] As described above, each training peak may be generated from a single spectral radiation, with each of the multiple training peaks having a different wavelength. The multiple training peaks are taken from a range of different locations on the detector 13 (e.g., as illustrated in FIG. 2 ). In some embodiments, the multiple training peaks are generated by the spectroscopic measurement system 10 using one or more calibration samples. In some embodiments, the calibration sample may be a single-element solution (i.e., a solution containing a single element). Some such single-element solutions enable the detector 13 to generate calibration spectra containing multiple sparsely distributed individual peaks, and collecting such spectra can provide good coverage of wavelengths detectable by the detector 13. In some embodiments, one or more of the single-element solutions used for calibration contain a single element selected from transition metals. When used in spectroscopic measurements, transition metals generate multiple well-defined, well-dispersed individual peaks that are well suited to characterizing the optical aberrations of the spectroscopic measurement system 10 across the detector 13.

[0059] Thus, a neural network may be trained as described above and then deployed to generate peak shapes (e.g., in the form of a vector of shape parameters, such as the vector (p1, p2, p3) described above with reference to FIG. 6) for any detector position of interest. The neural network technique performed in step 104 of method 100 may include using the trained decoder to decode the shape parameters describing the peak shape to generate a decoded peak shape representing the spectral emission at the specified detector position. Thus, the neural network technique may be trained and then used to predict peak shapes for purposes of analyzing interfered peaks.

[0060] The method performed in step 104 of Figure 3 will now be described. As noted above, in step 104, a set of curves associated with the interfered peak is generated. To generate the set of curves, neural network techniques predict a peak shape for each spectral emission that forms part of the interfered peak based on data representing the initial wavelength (e.g., detector position) for each spectral emission and the intensity associated with each of the spectral emissions. In the example of Figure 34, the neural network generates two curves that fit the interfered peak.

[0061] FIG. 7 shows a further block diagram illustrating an exemplary set of steps performed in step 104 to generate a set of curves for the interfered peaks.

[0062] To predict the shape of the peak, the neural network is fed with the initial peak positions (x) of one or more spectral emissions incident on the detector. n ) (corresponding to wavelength as described above), and the initial peak intensity (a n ) is provided. Thus, in step 112, processor 14 processes the interfered peaks to find the initial peak position (x n ) and initial peak intensity (a n ) where n=1, 2, ... N is determined (e.g., using the first derivative technique described above, or any other suitable technique). Note that although a single variable name (x) is given for the peak location, the peak location may be specified by two-dimensional parameters (e.g., x and y pixels in a two-dimensional pixel array of the detector).

[0063] Thus, in step 112, processor 14 determines the initial parameters (x1, a1; x2, a2; ... x N ,a NIn the example of Figure 4, fit two curves (N=2) to the interfered peak with associated parameters (x1, a1; x2, a2).

[0064] In the example of Figure 4, the interfered peak data recorded by detector 13 is shown as a dashed line. In Figure 4, the data is presented using pixel numbers rather than wavelength on the horizontal axis of Figure 4. Based on this data from detector 13, processor 14 determines that there is a first curve having a first peak position (x1) equal to about 12 and a first initial peak intensity (a1) equal to about 1.0, and a second curve having a second initial peak position (x2) equal to about 21 and a second peak intensity (a2) equal to about 0.1.

[0065] Based on the initial peak locations (x) and initial peak intensities (a), in step 114, processor 14 uses a neural network to output an initial identification of the curves. As explained above, the neural network algorithm determines the shape parameters (e.g., (p) as a three-dimensional encoded shape representation) for each of the N curves to be output. n1 ,p n2 ,p n3 )) A decoder may then be used to decode the shape parameters to provide an initial identification of the curve forming the interfered peak.

[0066] In some embodiments, the peak intensity may not be provided as an input to the neural network, but instead may be applied to the output of the neural network (e.g., by scaling) to ensure that the curves in the resulting set of curves have the correct peak intensity.

[0067] For example, in Figure 4, the solid lines of Peak 1 and Peak 2 indicate two initial output curves, which are decoded from the shape parameters output by the neural network algorithm.

[0068] To further improve the fit of the fitted curves, in step 116, processor 14 may further adjust the initially output curves. For example, the operation in step 116 may include shifting the positions of peaks (e.g., their associated peak wavelengths) to attempt to minimize the root-mean-square error (RMSE) between the total curve and the measured sample spectrum. It will be understood that adjustment step 116 is optional. Thus, in some embodiments, the initially output set of curves may be suitable for use in further analysis. Thus, in some embodiments, the method may proceed directly from step 114 to step 106 of method 100.

[0069] Returning to method 100 of FIG. 3 , in step 106, the curve associated with each spectral emission in the interfered peak may be output for further analysis. Step 106 may include any of several operations. In some embodiments, the peak closest in position to its factory-calibrated location may be automatically selected, and only its position and intensity may be used to generate a single interference-free peak. In some embodiments, processor 14 may cause the curves to be displayed on a display device and prompt a user to select from among multiple curves associated with the interfered peaks. The user may, for example, select only the curve representing the analyte of interest and use its position and intensity to generate a single interference-free peak. In some embodiments, the peaks are considered to be interferences from pixel intensities of peaks associated with the analyte of interest (e.g., by identifying those furthest in position from their factory-calibrated location) to generate a single interference-free peak.

[0070] Instead of directly using a neural network to predict the peak shape of the analyte of interest, in some embodiments, processor 14 may predict the peak shape of peaks that interfere with the peak associated with the analyte of interest. The predicted peak shapes for the interfering peaks can then be subtracted from the original signal to determine the peak shape associated with the analyte of interest. That is, for example, if an interfered peak containing three spectral emissions is detected, the two predicted peak shapes (associated with the interfering peak) can be subtracted from the interfered peak, leaving only the single peak associated with the analyte of interest.

[0071] In some embodiments, the method 100 can be used to generate a spectral emission curve of the interfered peak, which can be used to improve the spectral emission background correction method, as further described below.

[0072] FIG. 8 shows a further graph of an interfered peak generated by the spectroscopic measurement system 10. FIG. 8 shows the intensity (counts per second, CPS) detected by the pixels of the array detector. In the graph of FIG. 8, the detector position of each pixel has been converted to its associated wavelength (in nm). The interfered peak in FIG. 8 includes three distinct peaks representing at least three spectral emissions close to each other on the detector. Attempts to perform baseline correction to analyze the central peak in FIG. 8 will result in inaccurate results. As shown in FIG. 8, interference from adjacent spectral emissions distorts the baseline correction line shown in FIG. 8. As a result, subsequent analysis of the central peak of the interfered peak will likely be inaccurate due to interference from the adjacent spectral emissions. In such cases, users will likely ignore the interfered peak due to low confidence in the accuracy of the baseline correction.

[0073] According to the method 100 of the present disclosure, the interfered peak may be analyzed to generate a set of curves including three curves (Curve 1, Curve 2, Curve 3). Figure 9 shows a graph of the generated set of curves superimposed on the interfered peak of Figure 8. Generating a curve for each of the three most prominent spectral emissions in the interfered peak allows each of the spectral emissions to be analyzed individually (i.e., without interference from adjacent spectral emissions in the interfered peak).

[0074] FIG. 10 shows a graph of Curve 2 superimposed on the interfered peak of FIG. 8. Using Curve 2, an improved baseline correction can be performed to analyze the central spectral emission of the interfered peak. As shown in FIG. 10, the improved baseline correction is not distorted by the other spectral emissions of the interfered peak. As a result, Curve 2 can be used to analyze the intensity of the associated spectral emission (i.e., the area under Curve 2) with improved accuracy.

[0075] In some embodiments, the neural network technique may not include an initial encoding of the peak shape, but may instead be structured to predict, using techniques known in the art (e.g., selection of an appropriate loss function), a class of possible mathematical distributions (e.g., Gaussian, Lorentzian, bi-Gaussian, Gaussian + Lorentzian, Lorentzian + Gaussian, etc.) that can describe the observed peak, with the neural network free to guess the most appropriate one during training. Additionally, in some embodiments, the neural network itself may perform the encoding of the peak shape. For example, the neural network may perform the encoding such that, through appropriate selection of an error function, the parameters that exhibit the highest rate of change with position are selected as the encoding parameters.

[0076] It will be appreciated that neural network techniques can be used to generate peak shapes for the individual spectral emissions that form part of the interfered peak. Thus, neural network-based analysis methods according to the present disclosure can be used to determine information about the individual spectral emissions that form part of the interfered peak. For example, information about the wavelengths and intensities of the different spectral emissions that form part of the interfered peak can be determined according to embodiments of the present disclosure. This information (peak wavelength, peak intensity) can then be used to assist in the identification and analysis of the sample.

[0077] In some embodiments, the neural network may be retrained after additional training data is generated and / or after corrections are received from the user during operation of the spectroscopy system 10. For example, in some embodiments, the processor 14 may cause the display to request that the user mark an area of ​​a full frame (which may include part or all of the full frame) where the user would like to retrain the peak shape (e.g., because the user is not satisfied with the current performance).

[0078] FIG. 11 shows a block diagram of a method 200 for retraining a neural network algorithm according to the present disclosure. As shown in step 202 of FIG. 11, the method includes identifying at least one detector region for retraining. If the detector is an array detector, the detector region may be, for example, one or more orders of the detector. Alternatively, the detector region may be defined as the area of ​​the detector, for example, a square or rectangular region of the detector. In some embodiments, the detector region may comprise an area extending at least 20 pixels, 30 pixels, 50 pixels, 70 pixels, or 100 pixels in at least one direction (e.g., along an order). Alternatively, a user may specify a region of a full frame of the detector 13 for retraining. The user may perform the identification step manually, or a processor may be configured to identify the region of the detector 13 for retraining.

[0079] Once one or more regions of the detector have been identified for retraining, the processor can determine the calibration samples to be used in the retraining process. For example, the processor 14 may then output to the user a recommendation of one or more calibration solutions to be prepared (based on the most likely / strongest radiation falling into the selected region). Preferably, one or more of the calibration solutions are single-element standard solutions, thereby avoiding inter-element interferences. The recommended calibration solution is selected by the processor based on the knowledge that the element in the calibration solution has an uninterfered peak within the previously identified desired detector area. The processor may determine the calibration solution by referencing a database of known spectral peaks of single-element calibration solutions.

[0080] Once the user has completed preparing these solutions, in step 206, additional training peaks may be acquired by the spectroscopic measurement system 10. For example, the processor 14 may instruct the user on how to use the spectroscopic measurement system 10 to acquire a spectrum containing the desired training peak. The processor 14 may then request the user to review the spectrum and check that the training peak is not interfered with by other peaks, or otherwise mark the training peak as "interfered" or "not interfered with."

[0081] In step 208, processor 14 may then retrain the neural network algorithm using the additional training peaks. Processor 14 may then take the peaks selected by the user as "uninterfered," preprocess them by scaling their intensities, and provide them to an encoder to generate coded representations of their shapes, as described above. The coded representation of each additional training peak, along with its peak position, is used to retrain the neural network to improve its ability to map peak positions to peak shapes. The resulting retrained model may be stored in a memory device (e.g., on the user's premises or in the cloud) and used for subsequent spectra.

[0082] As mentioned above, it will be appreciated that the peak shape (curve) associated with the spectral radiation can be obtained by directly predicting the curve from the interfered peak using a neural network. Alternatively, the peak shape associated with the spectral radiation can be obtained by predicting the curve of another spectral radiation of the interfered peak and subtracting the predicted curve from the interfered peak.

[0083] In principle, two methods for obtaining peak shapes for an analyte of interest should arrive at peak shapes with a high degree of similarity. If the two methods result in different peak shapes, such differences may indicate that further investigation is needed. Thus, comparing the curves generated by the two methods may provide an initial indication that the predicted curve is an accurate reflection of the spectral emissions forming the interfered peak (i.e., confidence that the predicted curve is accurate). Accordingly, in some embodiments, method 100 may include performing a confidence analysis 120 on the set of curves obtained in step 104. The confidence analysis may be performed on the initial prediction of the set of curves (see step 114) or on the adjusted curve output following step 116 of FIG. 7. The confidence analysis may be used to indicate whether one or more curves in the set of curves were predicted with a relatively high degree of confidence or a relatively low degree of confidence.

[0084] FIG. 12 is a block diagram of a method for performing a confidence analysis 120 according to the present disclosure. The confidence analysis 120 may be performed on a set of curves generated for an interfered peak according to the present disclosure. The confidence analysis 120 of FIG. 12 includes step 122 of selecting a first curve from the set of curves. For example, the first curve may be the curve associated with the most intense spectral radiation that may be selected. In step 124, the confidence analysis 120 may then calculate a comparison curve for the first curve by subtracting the other curves in the set of curves from the interfered peak (i.e., the original signal). In step 126, the confidence analysis may then compare the first curve with the comparison curve, and then in step 128, determine a confidence level of the first curve associated with the set of curves.

[0085] For example, in step 126, the confidence analysis 120 may compare the first curve to the comparison curve by assessing the difference between the two curves. Suitable algorithms for constructing the first curve and the comparison curve include root mean square error, mean absolute error, Frechet distance, etc. Other suitable algorithms for numerically assessing the difference between two curves may also be used.

[0086] In some embodiments, the comparison in step 126 may produce a generated numeric value (e.g., a root mean square error). In step 128, the determined confidence level may be the numeric value calculated in step 126. In some embodiments, the numeric value may be scaled to present the numeric value on a more convenient scale as a confidence value. In some embodiments, the numeric value may be compared to one or more predetermined thresholds, with different confidence levels assigned to different ranges of numeric values. For example, in one embodiment, the numeric value may be compared to a confidence threshold, and if the root mean square error (or any other suitable algorithm and associated numeric value) is at or below the confidence threshold, a first confidence value may be assigned to the set of curves, indicating that the first curve and the comparison curve are sufficiently similar. For a root mean square error that exceeds the confidence threshold, a second confidence value may be assigned to the set of curves, indicating that the first curve and the comparison curve have a relatively high degree of difference that may be further investigated.

[0087] As an example, Figure 13 shows a further graph of the interfered peaks and set of curves of Figures 8 and 9, where a confidence analysis has been performed. As shown in Figure 13, curve 2 of Figure 8 has been selected as the first curve, and a comparison curve has been generated by subtracting curves 1 and 3 from the interfered peak. It will be seen that the comparison curve and the first curve (curve 2) have a relatively high degree of similarity, indicating a high degree of confidence that the set of curves has accounted for all spectral emissions in the background and interfered peaks.

[0088] In contrast, FIG. 14 shows a graph of a set of curves generated for another interfered peak. In the example of FIG. 14, the neural network has generated two curves (Curve 1 and Curve 2) for the interfered peak. Performing a confidence analysis using Curve 1 as the first curve indicates a relatively large difference between the first curve and the comparison curve (the comparison curve substantially overlaps with the interfered peak over the wavelength range of 309.38-309.44 nm in FIG. 14). Thus, FIG. 14 is an example in which the root mean square error of the difference between the first curve and the comparison curve may exceed a first threshold, thereby indicating relatively low confidence in the set of predicted curves. Therefore, the confidence analysis may output a signal to the user that the predicted curve (Curve 1) has a relatively low confidence associated with it.

[0089] Thus, the spectroscopic measurement system 10 and method according to the present disclosure allows a user to analyze the interfered peaks generated by the spectroscopic measurement system 10. In particular, a set of curves associated with one or more of the spectral emissions forming the interfered peaks may be generated, allowing the spectral emissions to be further analyzed.

Claims

1. 1. A method of operating a spectrometer controller, comprising: acquiring an interfered peak using a detector of a spectrometer, the interfered peak being produced by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions within the interfered peak being incident on the detector at an associated detector location; generating, for the spectral emissions of one of the interfered peaks, an associated curve using a neural network, the neural network being trained to output data indicative of the shape of the associated curve based on data representing the associated detector positions; and outputting the associated curve for the one spectral emission of the interfered peak.

2. the neural network outputs a coded representation of the shape of the associated curve; The method of claim 1 , wherein generating an associated curve comprises decoding the coded representation.

3. 10. The method of claim 1, further comprising training the neural network based on a plurality of training peaks, each training peak being a single spectral emission produced by the spectrometer and associated with a different detector position.

4. The method of claim 3 , wherein training is initiated for a detector region based on a user indication of the detector region to be trained.

5. The method of claim 3 , wherein the training peaks are associated with one or more single element solutions.

6. The method of claim 5 , wherein each one of the single element solutions is a transition metal solution.

7. acquiring a further training peak generated by the spectrometer for a detector area of ​​the detector; The method of claim 3 , further comprising: repeating the training of the neural network based on the further training peaks.

8. The method of claim 7 , further comprising identifying calibration samples used to derive the further training peaks.

9. The method of claim 1 , further comprising, after causing a display device to output the associated curve, receiving a user selection of the associated curve for use in subsequent analysis.

10. 10. The method of claim 1, wherein the spectrometer comprises an echelle diffraction grating and a two-dimensional array detector, and wherein the spectrometer uses the echelle diffraction grating to diffract light onto the two-dimensional detector to produce a sample spectrum.

11. The method of claim 1 , further comprising identifying the sample peak as an interfered peak.

12. 12. The method of claim 11, wherein identifying the sample peak as an interfered peak includes calculating a first derivative of the sample peak, and determining that the sample peak is an interfered peak based on a number of zero crossings of the first derivative of the sample peak.

13. The method of claim 12 , wherein the associated detector position of each spectral emission in the interfered peak is determined based on the zero crossing of the first derivative of the sample peak.

14. The method of claim 1 , wherein the spectrometer controller causes a display device to output the associated curve.

15. The method of claim 1 , wherein the spectrometer controller calculates the concentration of an element based on the area under the associated curve.

16. The method of claim 1 , wherein the spectrometer is an optical emission spectrometer and the spectrometer controller is an optical emission spectrometer controller.

17. The method of claim 1 , wherein the detector of the spectrometer is an array detector.

18. The method of claim 1 , wherein a curve is output for each of the spectral emissions in the interfered peaks.

19. 20. The method of claim 18, wherein the comparative curve associated with the spectral radiation is obtained by subtracting curves for other spectral radiations of the interfered peak from the interfered peak.

20. comparing the comparison curve for the spectral radiation with a curve output by the spectrometer controller for the same spectral radiation; 20. The method of claim 19, further comprising determining a confidence level for the curve output by the spectrometer controller based on the comparison.

21. 1. A spectrometer controller for a spectrometer, comprising: acquiring an interfered peak using a detector of a spectrometer, the interfered peak being produced by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions within the interfered peak being incident on the detector at an associated detector location; generating, for the spectral emissions of one of the interfered peaks, an associated curve using a neural network, the neural network being trained to output data indicative of the shape of the associated curve based on data representing the associated detector positions; and outputting the associated curve for the one spectral emission of the interfered peak.

22. 1. A spectroscopic measurement system, comprising: a spectrometer comprising a detector, the spectrometer configured to generate a sample spectrum from the sample using the detector; a spectrometer controller configured to process the sample spectrum, the controller comprising: obtaining an interfered peak from the sample spectrum using the detector of the spectrometer, the interfered peak being produced by a plurality of spectral emissions of different wavelengths, each of the plurality of spectral emissions in the interfered peak being incident on the detector at an associated detector location; generating, for the spectral emissions of one of the interfered peaks, an associated curve using a neural network, the neural network being trained to output data indicative of the shape of the associated curve based on data representing the associated detector positions; and outputting the associated curve for the one spectral emission of the interfered peak.

23. 23. The spectroscopic measurement system of claim 22, wherein the spectrometer includes an excitation source, preferably a plasma source.

24. 23. A computer program comprising instructions configured to, when executed by one or more processing devices of the controller, cause the spectrometer controller of claim 21 or the spectroscopic measurement system of claim 22 to perform the steps of the method of claim 1.

25. 22. A computer-readable storage medium having the computer program of claim 21 stored thereon.

Images