Waveform analysis method, waveform analysis device, and analyzation device

The waveform analysis method and device address the challenge of variable and long chromatogram waveforms by decomposing them into fixed-size segments and integrating region estimates, achieving high accuracy in peak detection across varying chromatogram lengths.

WO2024236865A1PCT designated stage expired Publication Date: 2024-11-21SHIMADZU CORP
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2024/002636
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-15
Filing Date
2024-01-29
Publication Date
2024-11-21

AI Technical Summary

Technical Problem

Existing waveform analysis methods struggle to accurately detect peaks in chromatograms with variable and long signal waveforms, such as total ion current chromatograms, due to fixed input data size limitations, which complicates calculation processing and reduces accuracy.

Method used

A waveform analysis method and device that decompose the target signal waveform into multiple processing target waveforms of predetermined data size using a sliding window method, allowing for peak detection with machine learning models trained on reference waveforms, and integrate region estimation information from overlapping windows to determine peak regions with high accuracy.

Benefits of technology

Enables accurate peak detection for chromatograms of varying lengths, improving processing efficiency and accuracy by fixing input data size and utilizing overlapping windows for enhanced region determination.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2024002636_21112024_PF_FP_ABST
    Figure JP2024002636_21112024_PF_FP_ABST
Patent Text Reader

Abstract

A waveform analysis method according to an aspect of the present invention is a method for analyzing a signal waveform such as a chromatogram. The method comprises: a step for creating a trained model for identifying a peak portion included in an input waveform by machine learning using a plurality of reference waveforms in which the positions of peak portions are known; a step (S13) for decomposing a signal waveform of interest into a plurality of waveforms to be processed each having a predetermined data size so as to cover the entirety of the waveform in the horizontal axis direction; steps (S14, S15) for dividing each of the plurality of waveforms to be processed into a plurality of partial waveforms in the horizontal axis direction, estimating, for each of the plurality of partial waveforms using the trained model, whether or not the partial waveform is classified as any of a plurality of different types of regions including a peak region and a non-peak region, and obtaining individual region estimation information reflecting the likelihood of the classification; a step (S16) for integrating the individual region estimation information obtained for each partial waveform of the plurality of waveforms to be processed and obtaining region estimation information for the signal waveform of interest as a whole; and a step (S17) for determining, for each portion of the signal waveform of interest using the region estimation information, which region of the plurality of regions the portion falls into.
Need to check novelty before this filing date? Find Prior Art

Description

Waveform analysis method, waveform analysis device, and analysis device

[0001] The present invention relates to a method and apparatus for analyzing signal waveforms acquired by an analysis device, and to an analysis device equipped with such a waveform analysis device.

[0002] Chromatograms obtained by analytical devices such as gas chromatographs (GC) and liquid chromatographs (LC) contain peaks derived from components contained in a sample. Data processing devices included in such analytical devices typically detect peaks by performing waveform processing on the chromatogram obtained by analysis, and identify peaks of target compounds by performing identification processing on the detected peaks. Furthermore, such data processing devices calculate the concentration or amount of the compound corresponding to the identified peak from the area or height of the peak.

[0003] Various methods have been put to practical use so far as methods for detecting peaks, but in recent years, a new peak detection method that uses machine learning has been proposed and put into practical use (see Patent Document 1 and Non-Patent Document 1).

[0004] In the waveform analysis method described in Patent Document 1, a reference waveform, the positions of which are known, is finely divided along the time axis to create a set of partial waveforms for each reference waveform. Machine learning is then performed using the multiple sets of partial waveforms for each reference waveform to create a trained model that identifies partial waveforms corresponding to peaks in an input waveform. The trained model is capable of semantic segmentation. Similarly to the reference waveform, the target waveform is also divided into multiple partial waveforms, and the trained model is used to determine whether each of the multiple partial waveforms is a peak. Based on the determination results, peak regions and other regions in the target waveform are determined. Furthermore, this waveform analysis method performs machine learning using partial waveforms corresponding to peak start and end points in addition to peaks, thereby enabling the partial waveforms corresponding to peak start and end points to be found among the multiple partial waveforms in the target waveform. Furthermore, by further dividing peaks into single peaks and overlapping peaks, single peak regions and overlapping peak regions in the target waveform can be determined.

[0005] International Publication No. 2021 / 064924

[0006] "Peakintelligence for GCMS LabSolutions Insight waveform processing software," [online], [searched April 18, 2023], Shimadzu Corporation, Internet <URL: https: / / www.an.shimadzu.co.jp / products / gas-chromatograph-mass-spectrometry / gc-ms-software / peakintelligence-for-gcms / index.html> "Convolutional Neural Network," [online], [searched April 18, 2023], Accelia Co., Ltd., Internet <URL: https: / / www.accelia.net / column / research / 1222 / > "Coursera Machine Learning (11): Photo Optical Character Recognition (Photo OCR)," [online], [searched April 18, 2023], Qiita Co., Ltd., Internet <URL: https: / / qiita.com / katsu1110 / items / b43a1ace7c8bbf354864 Olaf Ronneberger and 2 others, "U-Net: Convolutional Networks for Biomedical Image Segmentation," [online], [Submitted on 18 May 2015], arXiv.org, Internet <URL: https: / / arxiv.org / pdf / 1505.04597.pdf>

[0007] Generally, in algorithms widely used in deep learning, the size of input data to a classifier using a trained model (hereinafter simply referred to as a "classifier") is fixed. For example, in the waveform analysis method described in Patent Document 1, a chromatogram waveform over a 5-minute time range is divided into 1,024 partial waveform data pieces and then processed by a classifier. As shown in the example in Non-Patent Document 1, in waveform processing targeting extracted ion chromatograms (chromatograms corresponding to specific MRM transitions) obtained by multiple reaction monitoring (MRM) measurements using a gas chromatograph mass spectrometer (GC-MS), the chromatograms obtained for each target compound over a relatively narrow time range around the retention time of that compound are the target of processing. Therefore, as described above, there is no substantial problem even if the size of input data to the classifier is fixed and relatively small.

[0008] However, it is difficult to apply the waveform processing disclosed in Patent Document 1 and the like directly to, for example, the entire chromatogram (e.g., total ion current chromatogram: TIC) obtained by GC or LC or the extracted ion chromatogram obtained by selected ion monitoring (SIM) measurement using GC-MS. The main reason for this is that, whereas the size of the input data to the discriminator is fixed and is small to avoid complex calculations, chromatograms such as TIC have varying lengths along the time axis depending on the type of sample, the purpose of analysis, and the like, i.e., they are variable length, and moreover, are often quite long.

[0009] An object of the present invention is to provide a waveform analysis method and a waveform analysis device that can accurately detect peaks even for variable-length and long signal waveforms, while avoiding the complexity of calculations required to create a trained model used in a classifier and the complexity of calculation processing using the classifier.

[0010] One aspect of the waveform analysis method of the present invention is a waveform analysis method for analyzing a signal waveform that is a chromatogram or a spectrum, comprising: a model creation step of creating a trained model that identifies peak portions contained in an input waveform through machine learning using a plurality of reference waveforms whose peak positions are known; a waveform decomposition step of decomposing the target signal waveform to be analyzed into a plurality of target waveforms, each of which has a predetermined data size, so as to cover the entire waveform in the horizontal axis direction; a region estimation step of dividing each of the plurality of target waveforms into a plurality of partial waveforms in the horizontal axis direction and, for each target waveform, using the trained model to estimate whether each of the plurality of partial waveforms contained in the target waveform corresponds to one of a plurality of different types of regions, including peak regions and non-peak regions, and obtaining individual region estimation information that reflects the possibility of each being a peak region or a non-peak region; an integration step of integrating the individual region estimation information obtained for each partial waveform of the plurality of target waveforms in the region estimation step to obtain region estimation information corresponding to all or a portion of the target signal waveform; and a region determination step of using the region estimation information to determine which of the plurality of regions each portion of the target signal waveform corresponds to.

[0011] Another aspect of the waveform analysis device according to the present invention is a waveform analysis device that analyzes a signal waveform that is a chromatogram or a spectrum, and includes: a waveform decomposition unit that decomposes a target signal waveform to be analyzed into a plurality of target waveforms, each of which has a predetermined data size, so as to cover the entire waveform in the horizontal direction; a region estimation unit that divides each of the plurality of target waveforms into a plurality of partial waveforms in the horizontal direction, and, for each target waveform, estimates whether each of the partial waveforms included in the target waveform corresponds to one of a plurality of different types of regions, including peak regions and non-peak regions, using a trained model that identifies peak portions included in an input waveform and is created by machine learning using a plurality of reference waveforms whose peak positions are known, and obtains individual region estimation information that reflects the possibility of each partial waveform corresponding to the target waveform; an integration processing unit that integrates the individual region estimation information obtained for each partial waveform of the plurality of target waveforms in the region estimation unit to obtain region estimation information corresponding to all or a portion of the target signal waveform; and a region determination unit that uses the region estimation information to determine which of the plurality of regions each portion of the target signal waveform corresponds to.

[0012] Furthermore, one aspect of the analysis device according to the present invention is any one of a chromatograph device, a mass spectrometer, and an optical measurement device, which includes any aspect of the waveform analysis device according to the present invention as a data analysis section.

[0013] According to the above-described aspects of the waveform analysis method and waveform analysis device of the present invention, accurate peak detection can be performed on long signal waveforms whose waveform lengths vary along the horizontal axis, while using a trained model in which the input data size is fixed and relatively small.

[0014] 8 is a schematic configuration diagram of an embodiment of an analytical device that implements a waveform analysis method according to the present invention. FIG. 9 is a flowchart showing the procedure for a trained model creation process in the analytical device of this embodiment. FIG. 10 is a flowchart showing the procedure for a peak detection process on a chromatogram waveform to be analyzed in the analytical device of this embodiment. FIG. 11 is a waveform diagram showing an example of the relationship between a chromatogram waveform and a window set by a sliding window. FIG. 12 is a diagram showing the result of performing region estimation on a waveform to be processed cut out by window [#1] in FIG. 4. FIG. 13 is a diagram showing the result of performing region estimation on a waveform to be processed cut out by window [#2] in FIG. 4. FIG. 14 is a diagram showing the result of performing region estimation on a waveform to be processed cut out by window [#3] in FIG. 14. FIG. 15 is a diagram showing the result of integrating the region estimation results shown in FIGS. 5 to 7. FIG. 16 is a diagram showing a region determined from the integration result shown in FIG. 17.

[0015] In the waveform analysis method and waveform analysis device according to the above aspects of the present invention, the chromatogram includes a chromatogram obtained by a GC device, such as a gas chromatograph mass spectrometer (GC-MS), a chromatogram obtained by an LC device, such as a liquid chromatograph mass spectrometer (LC-MS), and an electropherogram obtained by an electrophoresis device. In a GC device or LC device in which the detector is a mass spectrometer, the chromatogram includes a total ion (total ion current) chromatogram and an extracted ion chromatogram. Furthermore, the spectrum includes a mass spectrum (a profile spectrum not subjected to centroid processing) obtained by a mass spectrometer, a time-of-flight spectrum obtained by a time-of-flight mass spectrometer before conversion to a mass spectrum, a light intensity spectrum obtained by an optical measurement device such as a spectrometer or a fluorescence measurement device, and an X-ray intensity spectrum obtained by an X-ray analysis device.

[0016]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0097]

[0098]

[0099]

[0099]

[0099]

[0099]

[0099]

[0099]

[0099]

[0099] [

[0017] The GC system A in this embodiment includes a GC measurement unit 1, a data analysis unit 2, an input unit 24, and a display unit 25. Although not shown, the GC measurement unit 1 includes an injector, a column, a column oven, a detector, etc., and performs GC analysis on a given sample to obtain chromatogram data that shows the temporal change in signal intensity obtained by the detector. There is no particular limitation on the type or type of detector, and examples that can be used include a mass spectrometer, a flame ionization detector, and a barrier discharge ionization detector.

[0018] The data analysis unit 2 includes functional blocks such as a data collection unit 20, a peak detection processing unit 21, a qualitative / quantitative analysis unit 22, and a display processing unit 23. The peak detection processing unit 21 includes functional blocks such as a resampling unit 210, a waveform division processing unit 211, a determination unit 212, a trained model storage unit 213, an overlap portion estimated value adjustment unit 214, a waveform integration processing unit 215, and a region determination unit 216.

[0019] In the data analysis unit 2, the data collection unit 20 collects and stores the chromatogram data obtained by the GC measurement unit 1. In response to instructions received from the user via the input unit 24, the peak detection processing unit 21 automatically detects peaks in a chromatogram waveform composed of the collected chromatogram data and outputs peak information including the start and end positions (retention times) of the detected peaks and the range of the peak region. Based on the peak information provided by the peak detection processing unit 21, the qualitative / quantitative analysis unit 22 identifies the components (compounds) corresponding to each peak, calculates peak heights and peak areas, and then calculates quantitative values, which are the concentrations or content amounts of each component, from these values. The display processing unit 23 displays the peak detection results and calculated values, such as quantitative values ​​calculated based on the peak detection results, on the display unit 25 in a predetermined format.

[0020] Generally, the entity of the data analysis unit 2 is a personal computer or a high-performance workstation on which predetermined software (computer programs) is installed, or a computer system including a high-performance computer connected to such a computer via a communication line. In other words, the functions of each block included in the data analysis unit 2 are realized by executing software installed on a single computer or a computer system including multiple computers. Of course, some of these functions may also be performed using hardware circuits specialized for specific calculations, such as a digital signal processor.

[0021] The computer program can be provided to the user by being stored on a computer-readable non-transitory storage medium, such as a CD-ROM, DVD-ROM, memory card, or USB memory (dongle). The program can also be provided to the user in the form of data transfer via a communication line such as the Internet. Furthermore, the program can be pre-installed on a computer that is part of the system (strictly speaking, a storage device that is part of the computer) when the user purchases the system.

[0022] 1 , system B, which is provided separately from GC system A, includes a model creation unit 3, which includes, as functional blocks, a learning data storage unit 30, a learning execution unit 31, and a model storage unit 32. The trained model created in this model creation unit 3 is stored in a trained model storage unit 213 in the data analysis unit 2 of GC system A.

[0023] As will be described in detail later, creating a trained model generally requires a huge amount of calculation. Therefore, the entity of system B is a high-performance computer, and the functions of each block are realized by executing software installed on the computer. Of course, system B may be integrated with GC system A.

[0024] Next, we will explain the peak detection process that is executed mainly by the peak detection processor 21 of the data analysis unit 2. In very general terms, the peak detection processor 21 converts a chromatogram waveform formed from chromatogram data into a one-dimensional (or two-dimensional) data array, and then detects the positions or ranges of peak regions, peak start regions, peak end regions, non-peak regions, etc. by using a semantic segmentation technique based on deep learning, which is a machine learning technique that detects the categories and positions of objects present in the image.

[0025] [Processing when creating a trained model] As is well known, machine learning methods require that a trained model be constructed in advance using a large amount of learning data (training data and validation data). As described above, the work of constructing this trained model is not performed in the data analysis unit 2 that is part of the GC system A, but is performed in the model creation unit 3 that is configured by a separate computer system, and the results are stored in the trained model storage unit 213. Figure 2 is a flowchart showing an outline of the steps of the trained model creation process performed in the model creation unit 3.

[0026] When creating a trained model, first, training data based on a reference waveform is prepared (step S1). Here, a large number of diverse chromatogram waveforms are used as reference waveforms. The diverse chromatogram waveforms referred to here preferably refer to chromatogram waveforms that include elements that may appear in chromatogram waveforms when actually performing peak detection, such as various noise contaminations, baseline fluctuations (drift), overlapping of multiple peaks, or deformation of peak shapes. However, this chromatogram waveform data does not necessarily have to be data collected by actual GC analysis, but may be data created by simulation.

[0027] For example, peak detection has been performed in advance on the chromatogram waveform, which serves as the reference waveform, by an expert, to accurately determine the start and end points of one or more peaks on the waveform. This chromatogram waveform is converted into a one-dimensional data array with the signal intensity, i.e., the vertical axis of the graph, normalized, and further divided into a predetermined number of partial waveforms along the horizontal axis, i.e., the time axis. The number of divisions is determined so that the width (length along the time axis) of each partial waveform is at least smaller than the peak width. Therefore, the number of divisions can be appropriately set depending on the minimum expected peak width.

[0028] A single chromatogram waveform (reference waveform) is composed of multiple partial waveforms. Each piece of data constituting a partial waveform is associated with characteristic information indicating which of multiple regions the partial waveform belongs to. For example, a partial waveform corresponding to a peak portion of a chromatogram waveform is associated with characteristic information indicating a peak region, a partial waveform including a peak start point is associated with characteristic information indicating a peak start region, and a partial waveform that does not include any peak portion is associated with characteristic information indicating a non-peak region (baseline region). There are at least four types of regions: peak region, peak start region, peak end region, and non-peak region. However, peak regions may be divided into single-peak regions, where only one peak exists (no overlapping peaks), and overlapping peak regions, where multiple peaks overlap, depending on the state of the peaks in the peak portion. Furthermore, overlapping peak regions can be further divided, as described below. The training data storage unit 30 stores the partial waveform data constituting each of multiple chromatogram waveforms and the associated characteristic information. The learning data may be divided into training data and verification data in advance, or may be used as either training data or verification data as appropriate when learning is performed without such division.

[0029] When an instruction to start creating a learning model is issued, the learning execution unit 31 prepares an unlearned learning model (step S2). This learning model can be any of a variety of models capable of performing semantic segmentation. While semantic segmentation is generally used to analyze images composed of pixel data distributed two-dimensionally, here, it is applied to the analysis of chromatogram waveforms composed of data arranged one-dimensionally along the time axis. Here, U-Net (see Non-Patent Document 4) is used as a learning model capable of performing semantic segmentation, but other learning models such as SeGNet and PSPNet may also be used.

[0030] Next, the learning execution unit 31 reads the learning data (partial waveform data and characteristic information) from the learning data storage unit 30 (step S3). The learning execution unit 31 performs machine learning using the read learning data and constructs a learning model for estimating to which region a given partial waveform corresponds (step S4). The learning procedure will not be described in detail here, but a trained model can be constructed according to a known procedure, for example, as described in Patent Document 1.

[0031] The model storage unit 32 stores the trained model created by machine learning using a large amount of training data (step S5). The trained model stored in the model storage unit 32 is transmitted via a communication line, for example, and stored in the trained model storage unit 213 in the GC system A.

[0032] [Peak Detection Processing for Target Signal Waveform] Next, a description will be given of the process of detecting peaks on a chromatogram waveform obtained for a target sample, which is executed by the data analysis unit 2 in the GC system A. Fig. 3 is a flowchart showing an outline of the peak detection processing performed by the peak detection processing unit 21.

[0033] First, the resampling unit 210 reads the chromatogram waveform (target signal waveform) data to be analyzed from the data collection unit 20 (step S11). For example, if the GC measurement unit 1 is a GC-MS, this chromatogram is a total ion current chromatogram or an extracted ion chromatogram obtained by SIM measurement or MRM measurement. The length of this chromatogram waveform in the time direction depends on the measurement time and measurement conditions when the data is acquired. Therefore, its length varies and can be said to be variable. The resampling unit 210 resamples the read chromatogram waveform data at a specified sampling interval (step S12).

[0034] The sampling interval during this resampling is determined by the time width of the waveform cut out by one window (described later) and the data size input to the trained model. In the example shown in FIG. 4 (described later), the chromatogram waveform is resampled to 1,024 points per 1.5 minutes, which is the input data size of the trained model using U-net. In particular, when the GC measurement unit 1 acquires chromatogram data through measurement, the sampling interval of the data is not necessarily constant, nor is the sampling interval necessarily the same for each measurement. This is because, for example, when performing SIM measurement (or MRM measurement) in a GC-MS, the sampling interval may change depending on measurement conditions such as the number of SIM measurements performed in the same time period, or in the case of scan measurement, depending on measurement conditions such as the scan m / z range and scan speed. In contrast, by performing resampling, the sampling interval can be made constant throughout the entire chromatogram, regardless of the sampling status of the original chromatogram waveform data.

[0035] Next, the waveform division processing unit 211 applies a sliding window method to the resampled chromatogram waveform data in the time axis direction to divide the data into a plurality of waveforms to be processed, each having a predetermined window width (step S13).

[0036] As described in Non-Patent Documents 2 and 3, the sliding window method is one method for decomposing (dividing) an image into a size suitable for processing when processing it with a deep learning classifier. Here, this sliding window method is applied to signal waveform processing. Specifically, by using the sliding window method, a window corresponding to each of two or more consecutive processing target waveforms in the time direction is set so that at least a portion of the processing target waveforms overlap. When decomposing an original chromatogram waveform into multiple processing target waveforms, a method can be used in which two adjacent windows in the time direction are set so that they do not overlap and there is no gap between them. However, as will be explained later, when a waveform is extracted from a continuous waveform and a region is determined for the extracted waveform using a trained model, the accuracy of the determination tends to decrease near the end of the extracted waveform, i.e., near the junction with another adjacent waveform. In contrast, by overlapping at least a portion of two or more consecutive processing target waveforms, regions such as peak regions can be determined at the overlapping portions based on the determination results for the multiple different processing target waveforms. This improves the accuracy of determination at the end of the waveform to be processed.

[0037] The width of a window and the overlap width of adjacent windows can be determined appropriately in advance depending on the expected shape of the signal waveform, measurement conditions, data processing conditions, etc. Specifically, they can be determined appropriately based on parameters such as the size of the input data of the trained model, analysis time (the longer the waveform, the greater the increase in analysis time), peak width (usually the expected maximum and minimum peak widths), and sampling rate. For example, it is preferable to set the overlap width wider than the expected maximum peak width.

[0038] Preferably, the window widths and overlap widths are determined so that three or more windows overlap near the center of the waveform in the time axis direction. Also, the windows including the ends of the chromatogram waveform may be set to extend outside the waveform, i.e., into an area where no waveform exists, so that at least two or more windows overlap at both ends of the chromatogram waveform.

[0039] FIG. 4 is a waveform diagram showing an example of the relationship between a chromatogram waveform and windows set by a sliding window. In this example, three windows, [#1], [#2], and [#3], cover the entire chromatogram waveform. The window width is W, and the overlap width is L. As described above, the leftmost window [#1], which includes the left end of the waveform, extends to the left of the waveform, and the rightmost window [#3], which includes the right end of the waveform, extends to the right of the waveform. In this example, three windows overlap near the center of the chromatogram waveform, and two windows overlap in the other areas. Note that appropriate dummy data may be inserted into parts of the windows where no waveform exists, i.e., where no actual data exists.

[0040] The waveform division processing unit 211 normalizes the signal intensity information for each of the plurality of target waveforms extracted for each window as described above, and converts the signal intensity information into a one-dimensional data array, and divides the waveform data into a predetermined number of partial waveforms along the horizontal axis (time axis) (step S14). This number of divisions may be the same as the number of divisions in the training data, but may be a different number as long as the width of the partial waveforms is smaller than the peak width.

[0041] Next, the determination unit 212 reads the trained model from the trained model storage unit 213 and sequentially inputs, for each target waveform, multiple partial waveforms constituting the target waveform into the trained model. The trained model determines whether the input partial waveform corresponds to each region, such as a peak region, a peak start region, a peak end region, or a non-peak region. Specifically, the determination unit 212 uses the trained model to calculate an estimated value for each partial waveform, indicating the likelihood that the partial waveform corresponds to each region (step S15). The larger the estimated value, the more likely the partial waveform is to be in that region. In this way, the determination unit 212 outputs all target waveforms extracted from the original chromatogram waveform, with estimated values ​​for each region, such as the peak region and peak start region, assigned to each partial waveform constituting the target waveform.

[0042] 5 to 7 each show the results of the region estimation process performed in step S15 on the target waveform extracted by windows [#1], [#2], and [#3] in FIG. 4. In this example, the peak regions are divided into single peak regions and overlapping peak regions as described above, resulting in five types of regions. Furthermore, the time range in which a peak generally exists is indicated by P in FIG. 4, and the same time range is also indicated by P in FIGS. 5 to 7. As shown in FIGS. 5 to 7, the determination results by the determination unit 212 using the trained model indicate the temporal changes in estimated values ​​for each of the five regions: single peak region, overlapping peak region, peak start region, peak end region, and non-peak region.

[0043] As described above, two or more windows overlap at any given time in the original chromatogram waveform, and the overlapping regions yield the same number of estimated values ​​as the number of overlapping windows. The estimated results for the regions in the target waveform vary depending on the waveform shape. Therefore, the estimated values ​​for the overlapping regions obtained from different windows are typically not identical. To give the simplest example, the estimates for the single-peak region are higher than the estimates for the overlapping peak region in window [#1] shown in Figure 5 from time 7.05 to 7.08. However, the estimates for the overlapping peak region are significantly higher than the estimates for the single-peak region in window [#2] shown in Figure 6 and window [#3] shown in Figure 7 from time 7.05 to 7.08. This is presumably because the second peak of the overlapping peaks is cut off (only a portion is included) in the target waveform extracted by window [#1].

[0044] The estimation results for regions near both ends of the waveform to be processed vary significantly depending on the discontinuity of the waveform (whether the entire peak is included in the waveform to be processed, or whether only a portion of the peak is included in the waveform to be processed). Therefore, in this embodiment, the system takes advantage of the fact that independent estimates can be obtained from multiple different windows in the overlapping region, performs statistical processing based on the multiple estimates for the same region, and calculates a statistical value, which is then used as the estimate for the overlapping region. That is, the overlapping region estimate adjustment unit 214 obtains estimates for each region from multiple different windows in the overlapping region, and then performs predetermined statistical processing to obtain a single statistical value from the multiple estimates for each region. The statistical processing referred to here can be very simple, and specifically, any of the following can be used as the statistical value: maximum, minimum, average, median, mode, etc. By the processing in step S15, the multiple estimates for each region are integrated into a single statistical value for each region.

[0045] The waveform integration processor 215 integrates the estimated values ​​for each region corresponding to the multiple target waveforms, and calculates information on the time change of the estimated values ​​for each region corresponding to the entire original chromatogram waveform using the above statistical values ​​for the overlapping portions (step S16). Note that the information on the time change of the estimated values ​​for each region may be calculated not for the entire chromatogram waveform, but only for a portion of the waveform desired by the user.

[0046] FIG. 8 shows the results of integrating the region estimation results shown in FIGS. 5 to 7. In this example, the minimum value of multiple estimates is used as the statistical value for non-peak regions, and the maximum value of multiple estimates is used for regions other than non-peak regions, i.e., single peak regions, overlapping peak regions, peak start regions, and peak end regions. This is done to prioritize the adoption of judgment results that indicate even the slightest possibility of a peak, with the aim of minimizing missed peaks. Because the above values ​​are used as statistical values, the estimates for the overlapping peak region are higher than the estimates for the single peak region around the aforementioned time period of 7.05 to 7.08, and the estimation results for window [#1] are not substantially reflected.

[0047] The region determining unit 216 receives the output from the waveform integration processing unit 215, and determines the region corresponding to each partial waveform by determining the region showing the highest estimated value for each partial waveform across the entire chromatogram waveform (step S17). In this way, all partial waveforms constituting the entire chromatogram waveform are classified into one of the regions.

[0048] 9 shows the integration results shown in FIG. 8 together with the regions determined by the integration. The superimposed peak region, non-peak region, peak start region, and peak end region are each indicated by a thick double-headed arrow. The peak portion indicated by P in FIG. 4 has been estimated over time in the following order: peak start region → superimposed peak region → peak end region → peak start region → superimposed peak region → peak end region, and the superimposed peak has been accurately detected.

[0049] Thereafter, the display processing unit 23 displays the peak detection results by the peak detection processing unit 21 on the screen of the display unit 25 (step S18). Furthermore, if the setting is such that qualitative analysis is automatically performed based on the peak detection results, the qualitative / quantitative analysis unit 22 calculates, for example, the retention time of the peak top of each detected peak and identifies the component corresponding to that peak based on the retention time. Furthermore, if the setting is such that quantitative analysis is automatically performed based on the peak detection results, the qualitative / quantitative analysis unit 22 calculates the peak area or height value for each detected peak and calculates the concentration (content) of the component corresponding to that peak by comparing the area or height value with a calibration curve created in advance. The display processing unit 23 displays the results of the qualitative or quantitative analysis together with the peak detection results on the screen of the display unit 25.

[0050] As described above, the GC system of this embodiment can detect peaks with high accuracy for chromatograms of various time lengths acquired by GC measurement, and can provide accurate peak information to the user.

[0051] In the above explanation, the minimum value is used as the statistical value for the non-peak region, and the maximum value is used for the region other than the non-peak region, but the statistical value can be appropriately selected from the maximum value, minimum value, average value, median value, mode value, etc. based on a plurality of estimated values. This selection may be predetermined and cannot be changed by the user, or may be made possible for the user to appropriately select from the input unit 24.

[0052] Generally, the following types of statistical values ​​can be used: As described above, if the minimum value is used as the statistical value for non-peak regions and the maximum value is used for regions other than the non-peak regions, there is a relatively high possibility that noise will be erroneously determined to be a peak, but it is possible to detect low-intensity peaks that are likely to be overlooked as peaks. Therefore, this selection is useful when you want to detect peaks as thoroughly as possible.

[0053] Furthermore, for example, if the mean, median, mode, or the like is used as a statistical value, the possibility of missing a low-intensity peak increases relatively, but the possibility of erroneously determining noise as a peak decreases. Therefore, such selection is useful when it is desired to detect only peaks that are highly likely to be peaks. However, with the median or mode, it is difficult to achieve sufficient accuracy unless the number of estimated values ​​is at least a certain level, at least three or more, and preferably five or more. Therefore, when using such statistical values, it is recommended to widen the overlap width of the sliding window (to approach the window width).

[0054] [Explanation of Various Additional Configurations] Next, several preferred additional configurations for the device of the above embodiment will be described. The configurations described below can be appropriately combined with those described in the above embodiment. Furthermore, multiple additional configurations can be combined as long as they do not perform contradictory processes.

[0055] In the above embodiment, the region where multiple peaks overlap is simply referred to as the overlapping peak region, but when determining the concentration or amount of a compound from the peak area or height, it is necessary to divide the overlapping peak into multiple peaks. Several methods for peak division are known, and the appropriate division method differs depending on the way the peaks overlap, etc.

[0056] Specifically, methods used to split overlapping peaks include tailing, which splits the peak from its start point to its end point into two peaks by superimposing another peak on top of the first peak; complete separation, which splits the peak into two peaks by connecting the start point, minimum point, and end point of the peak; and vertical splitting, which splits the peak into two peaks using a perpendicular line passing through the minimum point of the peak. Therefore, the overlapping peak region can also be set as a vertically split peak region, a completely separated peak region, and a tailing peak region. In this case, a trained model can be created by machine learning using multiple sets of partial waveforms created by dividing a reference waveform containing overlapping peaks that have been peak-separated by the tailing, complete separation, and vertical splitting methods along the time axis.

[0057] In the above embodiment, the determining unit 212 uses a trained model to estimate regions such as non-peak regions and single-peak regions for each partial waveform and obtains estimated values, but it is also possible to adjust the estimated values ​​themselves or change the rules for determining the regions in step S17 above by using, for example, numerical information reflecting the shape and characteristics of the peaks in the range corresponding to the peak regions. Examples of numerical information reflecting the shape and characteristics of the peaks that can be used here include the signal-to-noise ratio, the degree of peak separation, the symmetry coefficient, the percentage of missing peaks, the peak height, and the peak width.

[0058] For example, even if the estimated value corresponding to a peak region (single peak region or overlapping peak region) is large, if the S / N ratio calculated from the signal intensities in the peak region and non-peak region is equal to or less than a predetermined threshold (e.g., 10), there is a high possibility that noise has been recognized as a peak, and the reliability of the peak detection result itself is low. Therefore, if the S / N ratio of a partial waveform is equal to or less than a predetermined threshold, it can be considered to adjust the estimated value for the peak region corresponding to that partial waveform by a predetermined value, or to adjust the estimated value for the non-peak region by a predetermined value.

[0059] Furthermore, for example, the symmetry coefficient is an index showing the symmetry of a peak, but if the symmetry of the peak is poor, there is a possibility that a small peak is located at the base of a large peak. Therefore, when the symmetry coefficient is larger than a predetermined threshold, the estimated value for the single peak region corresponding to that peak portion may be large, or even if the estimated value for the overlapping peak region is small, the estimated value for the overlapping peak region may be adjusted to be large over the entire peak portion.

[0060] Furthermore, the Japanese Pharmacopoeia defines a resolution of 1.5 or higher as complete separation. Therefore, based on this resolution, the estimated values ​​for the complete separation peak region, tailing peak region, and vertical separation peak region can be appropriately adjusted. Furthermore, if a peak exists at the end of a chromatogram and part of the peak is missing, quantification is not a problem if the missing proportion is small, but if the missing proportion is large, it is not suitable for quantification. Therefore, if the missing proportion is large, it can be configured not to detect it as a peak (i.e., the estimated value for the non-peak region can be increased).

[0061] It is also possible to calculate the slope of the baseline based on the partial waveform determined to be a non-peak region and use this value to change the region estimation result, etc. Specifically, if the slope of the baseline changes from a positive value that is large to a small value over time, it is assumed that the baseline is convex upward. In this case, even if overlapping peaks existing on such a baseline are vertically separated peak regions, it can be estimated that the bottoms of the valleys of the overlapping peaks are in contact with the baseline, and the estimate of the completely separated peak region can be increased.

[0062] Furthermore, when making a judgment by comparing a numerical value reflecting the shape or characteristics of a peak or a numerical value reflecting the shape of a baseline with a threshold, it is preferable to set the threshold from the input unit 24. This allows the user to appropriately change the criteria for changing the region depending on the purpose of analysis, the type of sample, etc.

[0063] In the above embodiment, the statistical values ​​used in step S16 are predetermined types or types selected by the user, but it is also possible to automatically select the type of statistical value to be used depending on various factors. Furthermore, in addition to calculating statistical values ​​only from multiple estimated values ​​for each region of each partial waveform in the overlapping portion of multiple windows, it is also possible to correct or adjust statistical values ​​depending on various factors other than estimation using a trained model.

[0064] For example, compound information about the target compound in a sample, which can be known in advance, can be used to select the type of statistical value or to adjust the statistical value itself. The compound information here includes, for example, not only the concentration of the compound, but also structural information such as whether or not the compound has isomers, and information such as whether or not derivatives generated during pretreatment or the like are present.

[0065] For example, if the concentration of a compound in a sample is known to be high, the signal intensity of the peak corresponding to the compound should be high and the signal-to-noise ratio should be good. Therefore, since there is little possibility of erroneously recognizing noise as a peak, the maximum value can be used as a statistical value for the non-peak region in the overlapping region. Furthermore, if the presence of isomers in the target compound is known from the compound information, the peaks on the chromatogram often shift in time due to structural differences, even if the isomers have the same molecular weight. Therefore, for a peak estimated to correspond to the target compound, it is advisable to adjust the estimated value for the single peak region to be small, thereby making it easier to determine that the peak is an overlapping peak region.

[0066] Furthermore, mass spectrometers typically allow for the observation of multiple ions (called quantitative ions and confirmation ions) with different m / z values ​​generated from a single component. Therefore, GC-MS and LC-MS can generate an extracted ion chromatogram of a quantitative ion and one or more extracted ion chromatograms of confirmation ions for the same component. Here, the peak detection process described above is performed on at least one of these chromatograms. However, because the extracted ion chromatograms of quantitative ions and confirmation ions are chromatograms derived from the same component, their waveforms should be similar. Therefore, it is possible to compare the region estimation results in the quantitative ion chromatogram and the confirmation ion chromatogram for the same component and adjust the region estimation results based on the comparison results.

[0067] It is also possible to compare the region estimation results of signal waveforms such as chromatograms and spectra obtained for an unknown sample containing the same target component with a standard sample containing the target component at a known concentration, or to additionally use the region estimation results of signal waveforms such as chromatograms and spectra obtained for a blank sample not containing the target component.

[0068] Furthermore, the above-described embodiments and modifications are merely examples and can be modified as appropriate within the spirit of the present invention. In the above-described embodiments, the waveform analyzer for peak detection is incorporated into the same GC system A as the measurement unit, but it can also be configured as a waveform analyzer independent of the GC system A. In that case, chromatogram data previously acquired by the GC measurement unit 1 can be read and analyzed.

[0069] Furthermore, although the above embodiment has been described taking the case of processing a chromatogram waveform as an example, it is clear that the present invention can be applied to waveform analysis of signal waveforms obtained by various analytical instruments, such as electropherograms obtained by an electrophoresis apparatus, mass spectra (profile spectra) obtained by a mass spectrometer, spectroscopic spectra obtained by a spectrometer, fluorescence spectra obtained by a fluorescence measuring apparatus, and X-ray intensity spectra obtained by an X-ray analyzer, and whose signal intensity can change depending on changes in the value of a predetermined parameter.

[0070] Various Aspects It will be appreciated by those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0071] (Item 1) One aspect of the waveform analysis method of the present invention is a waveform analysis method for analyzing a signal waveform that is a chromatogram or a spectrum, comprising: a model creation step of creating a trained model that identifies peak portions contained in an input waveform through machine learning using multiple reference waveforms whose peak positions are known; a waveform decomposition step of decomposing the target signal waveform to be analyzed into multiple target waveforms, each of which has a predetermined data size, so as to cover the entire waveform in the horizontal axis direction; a region estimation step of dividing each of the multiple target waveforms into multiple partial waveforms in the horizontal axis direction and, for each target waveform, using the trained model to estimate whether each of the multiple partial waveforms contained in the target waveform corresponds to one of multiple different types of regions, including peak regions and non-peak regions, and obtaining individual region estimation information that reflects the possibility of each being a target waveform; an integration step of integrating the individual region estimation information obtained for each partial waveform of the multiple target waveforms in the region estimation step to obtain region estimation information corresponding to all or part of the target signal waveform; and a region determination step of using the region estimation information to determine which of the multiple regions each portion of the target signal waveform corresponds to.

[0072] (Item 13) One aspect of the waveform analysis device of the present invention is a waveform analysis device that analyzes a signal waveform that is a chromatogram or a spectrum, and includes: a waveform decomposition unit that decomposes the target signal waveform to be analyzed into a plurality of target waveforms, each of which has a predetermined data size, so as to cover the entire waveform in the horizontal direction; a region estimation unit that divides each of the plurality of target waveforms into a plurality of partial waveforms in the horizontal direction, and, for each target waveform, uses a trained model that identifies peak portions contained in an input waveform, created by machine learning using a plurality of reference waveforms whose peak positions are known, to estimate whether each of the plurality of partial waveforms contained in the target waveform corresponds to one of a plurality of different types of regions, including peak regions and non-peak regions, and obtains individual region estimation information that reflects the possibility of each being included; an integration processing unit that integrates the individual region estimation information obtained for each partial waveform of the plurality of target waveforms in the region estimation unit to obtain region estimation information corresponding to all or a portion of the target signal waveform; and a region determination unit that uses the region estimation information to determine which of the plurality of regions each portion of the target signal waveform corresponds to.

[0073] According to the waveform analysis method described in paragraph 1 and the waveform analysis device described in paragraph 13, accurate peak detection can be performed on long signal waveforms whose waveform lengths vary in the horizontal axis direction, while using a trained model in which the input data size is fixed and relatively small.

[0074] (Section 2) In the waveform analysis method described in Section 1, the waveform decomposition step can apply a sliding window method to the target signal waveform in the horizontal direction to obtain multiple processing target waveforms, at least some of whose ranges overlap with other processing target regions, and the integration step can obtain integrated region estimation information in the overlapping range of the multiple processing target waveforms based on individual region estimation information obtained for each of the overlapping ranges of the multiple processing target waveforms.

[0075] (Clause 14) In addition, in the waveform analysis device described in Clause 13, the waveform decomposition unit can apply a sliding window method to the target signal waveform in the horizontal direction to obtain multiple processing target waveforms, at least some of whose ranges overlap with other processing target regions, and the integration processing unit can determine integrated region estimation information in the overlapping range of the multiple processing target waveforms based on the individual region estimation information obtained for each of the overlapping ranges of the multiple processing target waveforms.

[0076] When a trained model is used to determine a region of a waveform to be processed that has been extracted from a continuous waveform using a window, the accuracy of the determination tends to decrease near the ends of the extracted waveform to be processed, i.e., near the junctions with another adjacent waveform to be processed. According to the waveform analysis method described in paragraph 2 and the waveform analysis device described in paragraph 14, since the ends of the waveform to be processed extracted using a single window are overlapping regions, region determination can be performed based on the determination results for multiple different waveforms to be processed in the overlapping regions, thereby improving the accuracy of region determination at the ends of the waveform to be processed.

[0077] (3) In the waveform analysis method described in paragraph 2, the individual region estimation information is an estimated value reflecting the possibility of falling into each region, and in the integration step, in the range where multiple waveforms to be processed overlap, statistical values ​​can be obtained by statistical processing from the estimated values ​​obtained for each corresponding partial waveform for each overlapping range of the multiple waveforms to be processed.

[0078] (4) In the waveform analysis method described in 3, the statistical value may include one of a maximum value, a minimum value, a median value, a mean value, or a mode value.

[0079] (Clause 15) In the waveform analysis device described in clause 14, the individual region estimation information is an estimated value reflecting the possibility of falling into each region, and the integration processing unit can be configured to obtain statistical values ​​by statistical processing from the estimated values ​​obtained for each corresponding partial waveform for each overlapping range of the multiple waveforms to be processed.

[0080] (Item 16) In the waveform analysis device described in item 15, the statistical value may include one of a maximum value, a minimum value, a median value, an average value, or a mode value.

[0081] According to the waveform analysis methods described in paragraphs 3 and 4 and the waveform analysis devices described in paragraphs 15 and 16, an area is estimated based on statistical values ​​calculated from multiple estimated values, making it possible to make highly reliable and reproducible estimates based on quantitative grounds.

[0082] (5) In the waveform analysis method described in 4, the integration step may use the minimum value as the statistical value for the non-peak region within the range where multiple waveforms to be processed overlap, and the maximum value as the statistical value for the region other than the non-peak region.

[0083] (Item 17) In the waveform analysis device described in Item 16, the integration processing unit may use the minimum value as the statistical value for the non-peak region within the range where multiple waveforms to be processed overlap, and the maximum value as the statistical value for the region other than the non-peak region.

[0084] According to the waveform analysis method described in paragraph 5 and the waveform analyzer described in paragraph 17, it is possible to extract, as a peak region, a partial waveform that may be a peak. Therefore, although there is a higher possibility that a region that is not actually a peak will be erroneously estimated as a peak region, it is possible to estimate, as a peak region, peaks with subtle shapes or sizes that may have been overlooked in the past, and it is possible to reduce the number of peaks that are overlooked or missed from detection.

[0085] (6) The waveform analysis method described in paragraphs 4 and 5 may further include a statistical value determination step that determines, based on a user's instruction, the type of statistical value to be used in the range where multiple waveforms to be processed overlap in the integration step.

[0086] (Clause 18) The waveform analysis device described in clauses 16 and 17 may further include a statistical value setting unit that determines, based on user instructions, the type of statistical value to be used in the range where multiple waveforms to be processed overlap in the integration processing unit.

[0087] According to the waveform analysis method described in paragraph 6 and the waveform analyzer described in paragraph 18, the user can appropriately select the type of statistical value to be used depending on the purpose of the analysis, the type of sample and compounds contained therein, the measurement environment such as the degree of disturbance and noise, or the measurement conditions, etc. This makes it possible to estimate a region with accuracy according to the conditions desired by the user.

[0088] (Clause 7) The waveform analysis method described in any one of clauses 1 to 6 may further include a resampling step in which, prior to waveform decomposition by the waveform decomposition step, data constituting the target signal waveform is resampled at a sampling interval according to the length of one waveform to be processed in the horizontal axis direction and the predetermined data size.

[0089] (Clause 19) The waveform analysis device described in any one of clauses 13 to 18 may further include a resampling unit that, prior to waveform decomposition by the waveform decomposition unit, resamples the data constituting the target signal waveform at a sampling interval corresponding to the length in the horizontal axis direction of one waveform to be processed and the specified data size.

[0090] In the waveform analysis method described in paragraph 7 and the waveform analysis device described in paragraph 19, the sampling interval of the target signal waveform is constant regardless of the sampling interval of the original signal waveform. Furthermore, the positions of the sampling points in the overlapping portions of multiple processing target waveforms are the same. This improves the accuracy of region estimation using a trained model. Furthermore, when integrating individual region estimation information in the overlapping portions, there is no deviation in the sampling points for each processing target waveform, improving the accuracy of the region estimation information after integration and also improving the accuracy of peak region estimation using this information.

[0091] (Item 8) In the waveform analysis method described in any one of Items 3 to 6, in the integration step, compound information regarding the target compound can be used when calculating the statistical value or when selecting the type of the statistical value.

[0092] The "compound information" referred to here can include information about the compound itself contained in the sample, such as concentration values, as well as information about isomers of the compound and derivatives produced by pretreatment, etc.

[0093] For example, it is expected that the S / N ratio will be low when the concentration value provided as compound information is low, and high when the concentration value is high. Therefore, when the concentration value is low, the maximum value is used as the statistical value for the non-peak region, making it less likely that noise will be mistakenly recognized as a peak. When the concentration value is high, the minimum value is used as the statistical value for the non-peak region, making it possible to estimate waveforms that may be peaks as peaks as much as possible. Furthermore, when compound information indicating the presence of isomers is provided, superimposed peaks with slightly shifted retention times are likely to appear. Therefore, using the maximum value as the statistical value for the superimposed peak region makes it easier to estimate that a peak is a superimposed peak. Thus, the waveform analysis method described in Section 8 can improve the detection rate of peaks and superimposed peaks under low-concentration conditions.

[0094] (Item 9) In the waveform analysis method described in any one of Items 1 to 8, the waveform length of the waveform to be processed can be determined based on at least one of the input size, analysis time, peak width, and sampling rate of the trained model.

[0095] According to the waveform analysis method described in Section 9, the window width, i.e., the length of the waveform to be processed, is appropriately determined, thereby improving the accuracy of area estimation using a trained model and enabling more accurate peak detection.

[0096] (Item 10) In the waveform analysis method described in any one of Items 1 to 9, the plurality of different types of regions may include, in addition to non-peak regions, single-peak regions with one peak, overlapping peak regions with multiple overlapping peaks, peak start regions including peak start positions, and peak end regions including peak end positions.

[0097] In the waveform analysis method described in paragraph 10, peaks can be detected more accurately by changing the type of statistical value used for each of the non-peak regions, namely, the peak start region, the peak end region, the single peak region, and the superimposed peak region.

[0098] (Item 11) In the waveform analysis method described in any one of Items 3 to 6, the integration step can calculate one or more of the S / N ratio, resolution, symmetry coefficient, peak missing rate, peak height, or peak width in the peak region, or the baseline slope in the non-peak region, and the calculation results can be used together when determining the individual region estimation information.

[0099] According to the waveform analysis method described in paragraph 11, the peak region can be estimated more accurately by utilizing information such as the S / N ratio in the peak region, the peak height, or the resolution that affects the peak shape, the symmetry coefficient, the percentage of missing peaks, and the peak width, etc. Furthermore, the non-peak region can be estimated more accurately by utilizing information on the baseline, such as the slope of the baseline, in the region estimation.

[0100] (12) The waveform analysis method according to any one of paragraphs 3 to 6 may further include a display step of displaying the statistical values ​​on a screen.

[0101] According to the waveform analysis method described in paragraph 12, the user can quantitatively grasp the statistical values ​​that are the basis for determining each region, which allows the user to make adjustments to improve the accuracy of peak detection, such as changing the type of statistical value used, as needed.

[0102] (Item 20) Furthermore, one aspect of the analytical device according to the present invention can be any one of a chromatograph device, a mass spectrometer, and a spectrometer, which includes a waveform analysis device according to any one of items 13 to 19 as a data analysis unit.

[0103] According to the analytical device described in paragraph 20, it is possible to achieve high qualitative and quantitative performance by utilizing highly accurate peak information.

[0104] 1...GC measurement unit 2...Data analysis unit 20...Data collection unit 21...Peak detection processing unit 210...Resampling unit 211...Waveform division processing unit 212...Determination unit 213...Trained model storage unit 214...Overlap portion estimated value adjustment unit 215...Waveform integration processing unit 216...Area determination unit 22...Qualitative / quantitative analysis unit 23...Display processing unit 24...Input unit 25...Display unit 3...Model creation unit 30...Trained data storage unit 31...Training execution unit 32...Model storage unit

Claims

1. A waveform analysis method for analyzing a signal waveform which is a chromatogram or a spectrum, comprising: a model creation step of creating a trained model which identifies peak portions contained in an input waveform by machine learning using multiple reference waveforms whose peak positions are known; a waveform decomposition step of decomposing the target signal waveform to be analyzed into multiple target waveforms, each of which has a predetermined data size so as to cover the entire waveform in the horizontal direction; a region estimation step of dividing each of the multiple target waveforms into multiple partial waveforms in the horizontal direction and estimating, for each target waveform, whether each of the multiple partial waveforms contained in the target waveform corresponds to each of multiple different types of regions including peak regions and non-peak regions using the trained model, and obtaining individual region estimation information reflecting the possibility of each corresponding to each of the multiple partial waveforms; an integration step of integrating the individual region estimation information obtained for each partial waveform of the multiple target waveforms in the region estimation step, to obtain region estimation information corresponding to the entire or part of the target signal waveform; and a region determination step of using the region estimation information to determine which of the multiple regions each portion of the entire or part of the target signal waveform corresponds to.

2. A waveform analysis method as described in claim 1, wherein in the waveform decomposition step, a sliding window method is applied to the target signal waveform in the horizontal direction to obtain multiple processing target waveforms having at least a partial range that overlaps with other processing target regions, and in the integration step, in the range where the multiple processing target waveforms overlap, integrated region estimation information is obtained based on individual region estimation information obtained for each of the overlapping ranges of the multiple processing target waveforms.

3. A waveform analysis method as described in claim 2, wherein the individual area estimation information is an estimate reflecting the possibility of falling into each area, and in the integration step, in the range where a plurality of waveforms to be processed overlap, statistical values ​​are obtained for each corresponding partial waveform by statistical processing from the estimates obtained for the respective overlapping ranges of the plurality of waveforms to be processed.

4. The waveform analysis method according to claim 3, wherein the statistical value includes one of a maximum value, a minimum value, a median value, an average value, or a mode value.

5. The waveform analysis method according to claim 4, wherein in the integration step, in the range where a plurality of waveforms to be processed overlap, a minimum value is used as a statistical value relating to non-peak regions, and a maximum value is used as a statistical value relating to regions other than the non-peak regions.

6. The waveform analysis method according to claim 4, further comprising a statistical value determination step of determining, based on a user's instruction, the type of statistical value to be used in the range where a plurality of processing target waveforms overlap in said integration step.

7. A waveform analysis method as described in claim 1, further comprising a resampling step of, prior to the waveform decomposition in said waveform decomposition step, resampling data constituting the target signal waveform at a sampling interval according to the horizontal axis length of one processing target waveform and said specified data size.

8. The waveform analysis method according to claim 3, wherein in the integration step, compound information relating to the target compound is used when calculating the statistical value or when selecting the type of the statistical value.

9. The waveform analysis method according to claim 1, wherein the waveform length of the waveform to be processed is determined based on at least one of the input size, analysis time, peak width, and sampling rate of the trained model.

10. The waveform analysis method according to claim 1, wherein the plurality of different types of regions include, in addition to non-peak regions, a single peak region with one peak, a superimposed peak region with multiple overlapping peaks, a peak start region including a peak start position, and a peak end region including a peak end position.

11. The waveform analysis method according to claim 3, wherein the integration step calculates one or more of the signal-to-noise ratio, resolution, symmetry coefficient, percentage of missing peaks, peak height, or peak width in the peak region, or baseline slope in the non-peak region, and the calculation results are used together when determining the individual region estimation information.

12. The waveform analysis method according to claim 3, further comprising a display step of displaying said statistical values ​​on a screen.

13. A waveform analysis device for analyzing a signal waveform which is a chromatogram or a spectrum, comprising: a waveform decomposition unit which decomposes a target signal waveform to be analyzed into a plurality of target waveforms each having a predetermined data size so as to cover the entire waveform in the horizontal direction; a region estimation unit which divides each of the plurality of target waveforms into a plurality of partial waveforms in the horizontal direction, and uses a trained model which identifies peak portions contained in an input waveform for each target waveform, the trained model being created by machine learning using a plurality of reference waveforms whose peak positions are known, to estimate whether each of the plurality of partial waveforms contained in the target waveform corresponds to each of a plurality of different types of regions including peak regions and non-peak regions, and obtains individual region estimation information reflecting the possibility of each corresponding portion; an integration processing unit which integrates the individual region estimation information obtained for each partial waveform of the plurality of target waveforms in the region estimation unit to obtain region estimation information corresponding to the whole or a part of the target signal waveform; and a region determination unit which uses the region estimation information to determine which of the plurality of regions each portion of the whole or a part of the target signal waveform corresponds to.

14. The waveform analysis device described in claim 13, wherein the waveform decomposition unit obtains multiple processing target waveforms having at least a partial range that overlaps with other processing target regions by applying a sliding window method in the horizontal direction to the target signal waveform, and the integration processing unit obtains integrated region estimation information in the overlapping range of the multiple processing target waveforms based on individual region estimation information obtained for each of the overlapping ranges of the multiple processing target waveforms.

15. The waveform analysis device described in claim 14, wherein the individual area estimation information is an estimated value reflecting the possibility of falling into each area, and the integration processing unit derives statistical values ​​by statistical processing from the estimated values ​​obtained for each corresponding partial waveform for each overlapping area of ​​the multiple waveforms to be processed.

16. The waveform analyzer of claim 15, wherein the statistical value includes one of a maximum value, a minimum value, a median value, a mean value, or a mode value.

17. The waveform analysis device according to claim 16, wherein the integration processing unit uses the minimum value as a statistical value relating to non-peak regions and the maximum value as a statistical value relating to regions other than the non-peak regions in the range where multiple processing target waveforms overlap.

18. The waveform analysis device according to claim 16, further comprising a statistical value setting section which determines, based on a user's instruction, the type of statistical value to be used in the range where a plurality of processing target waveforms overlap in the integration processing section.

19. The waveform analysis device of claim 13, further comprising a resampling unit that, prior to waveform decomposition by the waveform decomposition unit, resamples data constituting the target signal waveform at a sampling interval according to the horizontal axis length of one processing target waveform and the specified data size.

20. An analytical device, which is any one of a chromatograph device, a mass spectrometer device, and a spectrometer, comprising the waveform analyzer according to any one of claims 13 to 19 as a data analysis unit.

Citation Information

Patent Citations

  • Cognitive ability change prediction device, method, and program

    JP2017221416A

  • Method, apparatus, and system for periodic motion detection and monitoring

    JP2019111321A

  • Information processor and control method thereof, program, calculation device, and calculation method

    JP2020101524A

  • Waveform analysis device

    WO2019092837A1

  • Waveform analysis method and waveform analysis device

    WO2021064924A1