Analysis device and analysis method

Abstract

To enable computation of a certainty factor of peak picking when performing peak picking using the semantic segmentation technique.SOLUTION: An analysis device (1) provided herein is configured to: divide a target waveform into multiple partial waveforms (S12); determine a peak waveform corresponding to a peak portion from among the divided multiple partial waveforms using a trained model (S14); and compute a certainty factor of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model (S17).SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present disclosure relates to an analysis device and an analysis method for analyzing waveforms of chromatograms and spectra. [Background technology]

[0002] Chromatography has traditionally been used to identify or quantify components contained in a sample. In a chromatograph, the components in the sample are separated using a column, and the components that flow out of the column are detected in order. A chromatogram is then created, with the horizontal axis representing time and the vertical axis representing detection intensity.

[0003] To determine the height and area of ​​a peak from a chromatogram, it is necessary to identify the start and end points of the peak that rise from the baseline of the chromatogram. The process of identifying the start and end points of a peak in a chromatogram is called peak picking. Once the start and end points of the peak are identified, the height and area of ​​the peak are determined. From the peak height and area, the concentration of the compound corresponding to the peak can be calculated.

[0004] In recent years, attempts have been made to automate peak picking using deep learning. Known peak picking methods using deep learning include those that use object detection technology and those that use semantic segmentation technology.

[0005] Patent Document 1 discloses a method for displaying the confidence level of peak picking results using a Single Shot Multibox Detector (SSD) by formulating the peak picking problem as object detection in the image recognition field. The SSD outputs the peak picking result together with the confidence level for that result. In contrast, Non-Patent Document 1 discloses a method for performing peak picking using a U-Net by formulating peak picking as a semantic segmentation problem. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] International Publication No. 2020 / 225864 [Non-patent literature]

[0007] [Non-Patent Document 1] Kanazawa S, and 10 others, Fake metabolomics chromatogram generation for facilitating deep learning of peak-picking neural networks . J Biosci Bioeng. 2021 Feb;131(2):207-212. doi: 10.1016 / j.jbiosc.2020.09.013. Epub 2020 Oct 10. PMID: 33051155. Summary of the Invention [Problem to be solved by the invention]

[0008] However, there is currently no method for calculating the confidence level in peak picking using semantic segmentation technology. Therefore, in conventional peak picking methods using semantic segmentation technology, although the peak picking results are output, the confidence level of the output results is not output.

[0009] An object of the present disclosure is to make it possible to calculate the confidence of peak picking when performing peak picking using semantic segmentation technology. [Means for solving the problem]

[0010] An analysis device according to one aspect of the present disclosure is an analysis device that analyzes a target waveform that is a chromatogram or a spectrum, and includes a processor and a memory that stores a trained model created by machine learning using multiple sets of partial waveforms created by dividing a reference waveform whose peak portion positions are known. The processor divides the target waveform into multiple partial waveforms, determines the peak portions of the target waveform using the trained model, and classifies the target waveform into peak regions where peak portions are consecutive and non-peak regions other than the peak regions based on the determination results of the peak portions of the target waveform. When determining the peak portions of the target waveform using the trained model, the processor calculates the confidence level of the peak portion determination results using data output from the trained model.

[0011] An analysis method according to one aspect of the present disclosure is an analysis method for analyzing a target waveform that is a chromatogram or a spectrum, and includes the steps of: creating a trained model that identifies peak portions contained in an input waveform through machine learning using multiple sets of partial waveforms created by dividing a reference waveform whose peak portion positions are known; dividing the target waveform into multiple partial waveforms; determining the peak portions of the target waveform using the trained model; classifying the target waveform into peak regions with consecutive peak portions and non-peak regions other than the peak regions based on the determination results of the peak portions of the target waveform; and calculating the confidence of the determination results using data output from the trained model when determining the peak portions of the target waveform using the trained model. [Effects of the Invention]

[0012] According to the present disclosure, when peak picking is performed using semantic segmentation technology, it is possible to calculate the confidence level of peak picking. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 is a block diagram showing the overall configuration of an analysis device. [Figure 2]FIG. 1 is a diagram showing an example of a chromatogram. [Figure 3] FIG. 1 is a block diagram illustrating a procedure for creating a trained model. [Figure 4] 1 is a flowchart illustrating a procedure for creating a trained model. [Figure 5] 10 is a flowchart illustrating a procedure for determining chromatogram data using a trained model. [Figure 6] FIG. 10 is a diagram illustrating an example of a judgment result of a trained model. [Figure 7] FIG. 10 is a diagram showing an example of a graph on which labeling processing has been performed based on a determination result. [Figure 8] FIG. 10 is a diagram showing an example of an image displaying a determination result together with a certainty level. [Figure 9] FIG. 10 is a diagram showing an example of an image that accepts an operation to correct a determination result. [Figure 10] FIG. 10 is a diagram showing the relationship between the confidence level of a peak and the accuracy rate. [Figure 11] FIG. 10 is a diagram showing various modified examples 1 to 7 of the method for calculating the confidence factor of a peak. [Figure 12] FIG. 13 is a diagram for explaining Modification 7. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference numerals, and description thereof will not be repeated.

[0015] 1 is a block diagram showing the overall configuration of an analysis device 1. The analysis device 1 includes a processor 10 that functions as a control unit, a memory 20 that functions as a storage unit, and an input / output port 30. A mouse 40, a keyboard 50, and a display device 60 are connected to the input / output port 30. A mass spectrometer or the like may also be connected to the input / output port 30. One or more terminal devices may also be connected to the input / output port 30 via the Internet or an in-house network.

[0016] Analysis device 1 is configured, for example, based on a personal computer, or may be configured as a server that can be accessed from one or more terminal devices via a network such as the Internet.

[0017] Measurement data (chromatogram data) to be analyzed and learning data used for machine learning are input to the input / output port 30. The measurement data to be analyzed may be input via a mass spectrometer connected to the input / output port 30. A liquid chromatography mass analysis system can also be configured by the mass spectrometer, a liquid chromatograph connected to the mass spectrometer, and the analysis device 1.

[0018] The memory 20 stores at least learning data 210 input to the input / output port 30, measurement data 213 input to the input / output port 30, an estimation model 300 used for machine learning, and an analysis program 200 for executing analysis processing and machine learning processing.

[0019] The learning data 210 is classified into training data 211 and validation data 212. The training data 211 and validation data 212 are waveform data of chromatograms obtained by measuring a sample containing various components using a chromatograph mass spectrometer. The chromatogram is, for example, a total ion chromatogram that represents the change over time in the total intensity of all ions of mass-to-charge ratios detected by MS scan measurement using a mass spectrometer of components separated by liquid chromatography. The chromatogram may also be a mass chromatogram that represents the change over time in the intensity of ions of a specific mass-to-charge ratio obtained by SIM or MRM measurement.

[0020] The training data 211 and validation data 212 include data on the positions of peaks previously identified by peak picking. These waveform data are previously normalized so that the intensity values ​​fall within a predetermined range (for example, ±1.0). By normalizing multiple chromatograms with different intensity scales to a common intensity scale, the accuracy of the trained model can be improved. Here, chromatograms obtained by measuring actual samples are used as the training data 211 and validation data 212, but chromatograms created by simulation may also be used.

[0021] The waveform of a chromatogram is divided into a predetermined number of partial waveforms along the time axis. The predetermined number is, for example, 512 or 1024, and is set so that the width (length along the time axis) of each partial waveform is at least smaller than the peak width. The predetermined number is determined, for example, based on the size of the peak width and the number of data points required to form one peak.

[0022] Each waveform segment is associated with information about the characteristics of the waveform segment (characteristic information). The characteristic information associated with the waveform segment includes at least information indicating whether the waveform segment belongs to a peak region or a non-peak region.

[0023] The analysis program 200 configures a division unit 201 , a model creation unit 202 , a determination unit 203 , a calculation unit 204 , an image processing unit 205 , and an output unit 206 .

[0024] The dividing unit 201 divides the waveform of a chromatogram into a predetermined number of partial waveforms. The model creating unit 202 uses the learning data 210 to advance machine learning of the estimation model 300 and create a trained estimation model 300. The determining unit 203 uses the trained estimation model 300 to perform peak picking of the chromatogram. Hereinafter, the trained estimation model 300 may be referred to as a "trained model."

[0025] The calculation unit 204 calculates the confidence level of the determination result of the determination unit 203. The image processing unit 205 creates image data including the determination result and the confidence level. The output unit 206 outputs a display signal including the image data from the input / output port 30 to the display device 60. Note that the analysis device 1 may also be equipped with the display device 60.

[0026] FIG. 2 shows an example of a chromatogram. Here, the names of each part identified from the chromatogram will be briefly explained. A chromatogram can be classified into a baseline part and a peak region. The rising parts from the baseline are called the peak start point and the peak end point. The region between the peak start point and the peak end point is called the peak region. Within the peak region, the part with very strong detection intensity (the strongest part) is called the peak top.

[0027] The peak region includes a single peak, as exemplified in Figure 2. When an unresolved peak appears in a chromatographic waveform, the peak region includes both the single peak and the unresolved peak. For example, a region where two consecutive mountain-shaped waveforms with a peak top at the top are present and the detected intensity of the valley between the two mountain-shaped waveforms has not yet dropped to the intensity corresponding to the baseline is called an unresolved peak.

[0028] Next, a procedure for creating a trained model will be described with reference to a flowchart. FIG. 3 is a block diagram for explaining the procedure for creating a trained model. As shown in FIG. 3, the model creation unit 202 of the analysis device 1 functions as a learning device. The model creation unit 202 trains an estimation model 300 based on input learning data 210. The estimation model 300 performs deep learning by using a neural network. The estimation model 300 includes parameters such as weighting coefficients used in calculations by the neural network.

[0029] For example, a supervised learning algorithm is used to train the estimation model 300. The model creation unit 202 trains the estimation model 300 by supervised learning using the training data 210.

[0030] The semantic segmentation technique is used for training the estimation model 300. Semantic segmentation is generally used to analyze images made up of pixel data distributed two-dimensionally. In this embodiment, semantic segmentation is applied to the analysis of chromatogram waveforms made up of data arranged one-dimensionally along the time axis. As a training model capable of executing semantic segmentation, for example, U-Net, SeGNet, PSPNet, etc. can be used. In this embodiment, U-Net is used.

[0031] The model creation unit 202 receives input of partial waveforms of a chromatogram and corresponding correct data. The correct data is, for example, the result of peak picking that has already been identified. The result of peak picking may include peak tops.

[0032] The model creation unit 202 determines the result of peak picking based on the input learning data 210 and the estimation model 300, and trains the estimation model 300 based on the determination result and the correct answer data. Specifically, the model creation unit 202 trains the estimation model 300 by adjusting parameters in the estimation model 300 so that the result obtained by the estimation model 300 approaches the correct answer data.

[0033] 4 is a flowchart illustrating the procedure for creating a trained model. The processor 10 of the analysis device 1 executes a part of the analysis program 200 to realize the processing of this flowchart.

[0034] First, the processor 10 detects an operation to start learning of the estimation model 300 (step S1). For example, when a user performs an operation to start learning of the estimation model 300 using the mouse 40 and the keyboard 50, the operation is detected in step S1.

[0035] Next, the processor 10 reads out the learning data 210 (the training data 211 and the verification data 212) from the memory 20 (step S2). Next, the processor 10 inputs the training data 211 to the estimation model 300 (step S3). Next, the estimation model 300 executes a learning process by deep learning (step S4). In the U-Net used for learning the estimation model 300 in this embodiment, the weighting of the neural network is adjusted so that correct characteristic information can be obtained from the partial waveform.

[0036] More specifically, the parameters of the estimation model 300 are adjusted based on the partial waveforms of the training data 211 and the characteristic information associated with the partial waveforms. In the process of adjusting the parameters, a process of estimating a single peak, an unseparated peak, a peak start point, a peak end point, a baseline, etc., and a process of comparing the estimation results with the correct answer data are executed.

[0037] Next, the processor 10 stores the estimation model 300 created according to the result of the learning process in step S4 in the memory 20 (step S5). Next, the processor 10 checks the accuracy rate of the characteristic information assigned by the estimation model 300 by analyzing the partial waveform of the verification data 212 (step S6).

[0038] Next, the processor 10 determines whether a predetermined termination condition is met (step S7). For example, if the number of times the learning process is repeatedly performed using the training data 211 reaches a predetermined number, the processor 10 determines that the termination condition is met. If the termination condition is not met, the processor 10 repeats the processes from step S3 to step S6 until the termination condition is met.

[0039] When the termination condition is met, the processor 10 selects an appropriate estimation model 300 from among the multiple estimation models 300 stored in the memory 20, and stores the selected estimation model 300 in the memory 20 as a trained model (step S8).

[0040] This causes the processor 10 to end the series of processes shown in Fig. 4. The trained model is selected based on criteria such as the highest accuracy rate for the verification data 212 or whether overtraining has occurred. Note that here, an example has been shown in which the estimation model 300 is stored in the memory 20 for each training run. However, the same estimation model 300 may be repeatedly updated until the number of training runs reaches a predetermined number, and the estimation model 300 may be stored in the memory 20 when the number of training runs reaches the predetermined number.

[0041] Next, the procedure for analyzing the waveform of an unanalyzed chromatogram will be described with reference to a flowchart. Fig. 5 is a flowchart for explaining the procedure for determining chromatogram data using a trained model (trained estimation model 300). The processor 10 of the analysis device 1 executes a part of the analysis program 200 to realize the processing of this flowchart.

[0042] First, the processor 10 acquires chromatogram data (measurement data) (step S11). The chromatogram data is input to the analysis device 1 through a measuring instrument such as a mass spectrometer connected to the input / output port 30, or through a terminal device connected to the input / output port 30.

[0043] Next, the processor 10 divides the acquired chromatogram waveform into a predetermined number of partial waveforms (step S12). The number of divisions of the chromatogram waveform may be the same as or different from the number of the training data 211 and the validation data 212.

[0044] However, the number of divisions is determined according to the waveform length (length of the chromatographic mass analysis execution time) so that the width of each partial waveform (length along the time axis) is at least smaller than the width of the peaks expected to be included in the chromatogram. For example, the number of divisions can be set to 512 or 1024.

[0045] Next, the processor 10 inputs the partial waveform into the trained estimation model 300 (trained model) (step S13). Next, the trained model determines whether the partial waveform belongs to a peak region, and executes labeling processing (step S14). More specifically, the peak start and end points, baseline, single peak, unseparated peak, peak top, etc. are determined from the partial waveform. Furthermore, a weight for each determination result is calculated. Furthermore, in step S14, characteristic information (information on whether the partial waveform belongs to a peak region) is assigned to each partial waveform.

[0046] Next, the processor 10 calculates the confidence of the peak (step S17). The confidence of the peak is calculated by averaging the weight corresponding to the peak start point determined by the trained model and the weight corresponding to the peak end point determined by the trained model.

[0047] Next, processor 10 creates a graph showing the determination result and the confidence level (step S18). In this embodiment, multiple types of graphs are created by processor 10. Processor 10 outputs a display signal for displaying the created graph to display device 60 (step S19). As a result, the determination result and the confidence level are displayed on display device 60. For example, the screen of display device 60 displays the peak start point, peak end point, and confidence level on the waveform of the chromatogram.

[0048] Next, processor 10 determines whether or not an instruction to modify the peak start and end points has been detected (step S20). In this embodiment, the user can perform an operation to modify the peak start and end points on the screen of display device 60. If an instruction to modify is not detected, processor 10 proceeds to step S22.

[0049] When the user operates the mouse 40 and keyboard 50 to modify the peak start and end points, the processor 10 modifies the data on the screen in accordance with the modification instruction (step S21). In this way, the processor 10 accepts the user's modification instruction and modifies the peak start and end points.

[0050] After correcting the data, processor 10 determines whether an operation to confirm the data has been detected (step S22). If an operation to confirm the data has not been detected, processor 10 returns the process to step S20. If an operation to confirm the data has been detected, processor 10 stores the determination result (or the determination result after correction if the data has been corrected) in memory 20 (step S23), and ends the process based on this flowchart.

[0051] FIG. 6 shows an example of the judgment results of the trained model. The upper graph in FIG. 6 shows the waveform W0 of the input chromatogram. The lower graph in FIG. 6 shows the judgment results of the trained model for the input chromatogram. The horizontal axis (index) of both graphs corresponds to the time axis. The vertical axis of the upper graph in FIG. 6 indicates intensity. The vertical axis of the lower graph in FIG. 6 indicates the weights output by the trained model. The weights are normalized to the range of 0 to 1.

[0052] The waveforms W1 to W5 shown as the judgment results of the trained model correspond to the baseline, a single peak, an unresolved peak, a peak start point, and a peak end point, respectively. By comparing the chromatogram waveform W0 with the waveforms W1 to W5, it can be seen that, for example, in the chromatogram waveform W0, the weight corresponding to the peak start point is highest at the position of index Is. Similarly, it can be seen that in the chromatogram waveform W0, the weight corresponding to the peak end point is highest at the position of index Ie. In this case, for example, the analysis device 1 determines that, in the chromatogram waveform W0, the position of index Is is the peak start point, and the position of index Ie is the peak end point.

[0053] Here, the peak start point, peak end point, single peak, unresolved peak, and baseline are taken as examples of the determination targets, but other elements such as peak tops can also be added to the determination targets.

[0054] As shown in Figure 6, the processor 10 determines the confidence of the peak by calculating the average value of the weight Ws corresponding to the peak start point Is determined by the trained model and the weight We corresponding to the peak end point Ie determined by the trained model.

[0055] Figure 7 shows an example of a graph that has been labeled based on the determination results. The top graph in Figure 7 is the same as the bottom graph in Figure 6. The bottom graph in Figure 7 is a graph in which waveform W0 (see Figure 6) of the input chromatogram has been labeled based on waveforms W1 to W5. Labels 0 to 4 correspond to the baseline, single peak, unresolved peak, peak start point, and peak end point, respectively.

[0056] For example, the labeling process is performed in the following procedure. That is, among the waveforms W1 to W5, the waveform with the greatest weight at a certain index Ix is selected, and the value of index Ix is labeled with the selected waveform. The labeling process is completed by repeating the same process while changing x from the initial value to the final value of the index. For example, FIG. 7 shows a graph in which the section from index 0 to Is is labeled with the baseline (label = 0).

[0057] FIG. 8 is a diagram showing an example of an image 61 that displays a determination result together with a degree of certainty. Image 61 is displayed by display device 60. Image 61 displays the waveform of the chromatogram being measured as well as the peak start point Is and peak end point Ie corresponding to the determination result. Image 61 also displays the degree of certainty for the determined peak start point Is and peak end point Ie. By looking at image 61, the user can recognize the likelihood of the determination result.

[0058] In addition to image 61, processor 10 can selectively display on display device 60 an image including two graphs in the manner shown in Fig. 6, an image including two graphs in the manner shown in Fig. 7, and an image in which the three graphs included in Figs. 6 and 7 are arranged vertically. Each image also displays a confidence level in the manner shown in Fig. 8. The user can use mouse 40 and keyboard 50 to input an instruction to analysis device 1 indicating which image to display.

[0059] Fig. 9 is a diagram showing an example of an image 62 that accepts an operation to correct the determination result. The image 62 is displayed by the display device 60. In addition to the content shown in Fig. 8, the image 62 displays icons 65 and 66 for correcting the positions of the peak start point Is and the peak end point Ie.

[0060] Icon 65 corresponds to the peak start point Is. The position of the peak start point Is changes when the user operates icon 65 using the mouse 40 and keyboard 50. The position of the peak end point Ie changes when the user operates icon 66 using the mouse 40 and keyboard 50. In conjunction with the change in the positions of the peak start point Is and the peak end point Ie, the index position and confidence displayed below the graph also change.

[0061] The user corrects the positions of the peak start point Is and the peak end point Ie to appropriate positions, and then performs an operation to confirm the data. When the operation to confirm the data is detected by the processor 10, the corrected results are stored in the memory 20.

[0062] Note that, here, an example has been shown in which icons 65 and 66 are displayed based on image 61 shown in Fig. 8. However, icons 65 and 66 for correcting the determination result may also be displayed for an image including two graphs in the manner shown in Fig. 6, an image including two graphs in the manner shown in Fig. 7, and an image in which the three graphs included in Figs. 6 and 7 are arranged vertically.

[0063] As described above, in this embodiment, the judgment results and confidence levels of the trained model are displayed on the display device 60. This allows the user to visually distinguish between probable peak information and less reliable peak information. As a result, the user's visual confirmation and correction instructions are made easier, reducing the burden on the user in such work. Furthermore, when analyzing a waveform in which multiple peaks are observed, reducing the number of peaks that the user must confirm can also prevent mistakes and oversights in the confirmation work.

[0064] Next, we will explain an example of creating a trained model using actual chromatogram data and performing chromatogram waveform analysis. To create the trained model, 30 sets of chromatograms of primary metabolites were prepared. Each set contained 475 chromatograms. Manual peak picking was performed on each of the prepared chromatograms. The chromatogram waveforms were then classified into five classes: baseline, peak onset, peak end, single peak, and unresolved peak, and each was labeled. This created training data. Cross-validation evaluation was performed using the prepared training data. In the cross-validation evaluation, one of the 30 sets was used as validation data, and evaluation was performed 30 times.

[0065] The weight of the peak start point output from the trained model and the weight of the peak start point output from the trained model were added together and divided by 2 to calculate the average of the weights, which was used as the confidence of the peak.The relationship between the confidence and the accuracy rate was then verified.The verification results are shown in Figure 10.

[0066] Fig. 10 is a diagram showing the relationship between the confidence of the peak and the accuracy rate. In Fig. 10, TP indicates the number of correct answers, and FP indicates the number of incorrect answers. As shown in Fig. 10, the higher the confidence, the higher the accuracy rate. This shows that the method of calculating the confidence disclosed in this embodiment is effective.

[0067] Next, modified examples of the method for calculating the confidence factor of a peak will be described with reference to Fig. 11. Fig. 11 is a diagram showing modified examples 1 to 7 of the method for calculating the confidence factor of a peak. Note that the waveforms W1 to W5 used in the following description of the modified examples are shown in Figs. 6 and 7.

[0068] As shown in FIG. 11, the confidence of a peak can be calculated using any one of the baseline (Variant 1), single peak (Variant 2), peak start point (Variant 3), peak end point (Variant 4), and peak top (Variant 5) individually.

[0069] Variation 1 is an example of calculating the confidence of a peak using a baseline. As shown in Fig. 11, the confidence can be calculated by "1 - (average value of weights of index portions belonging to the peak region of the baseline waveform W1)". Here, the index portions belonging to the peak region refer to the range of indexes Is to Ie in Fig. 6, for example.

[0070] Variation 2 is an example of calculating the confidence level of a peak using a single peak. As shown in Fig. 11, the confidence level can be calculated by the average weight of the index portion of the waveform W2 of the single peak that belongs to the peak region.

[0071] Variation 3 is an example of calculating the confidence of a peak using the peak start point. As shown in Figure 11, the confidence can be calculated by the "average value of weights of the index portion corresponding to waveform W4 at the peak start point." For example, in Figure 6, the confidence is derived by identifying the weight corresponding to waveform W4 for each index in the range from the initial value to the terminal value of the index, and then calculating the average value of all the identified weights.

[0072] Variation 4 is an example of calculating the confidence of a peak using the peak end point. As shown in Figure 11, the confidence can be calculated by the "average value of the weights of the index portion corresponding to the waveform W5 of the peak end point."

[0073] Variation 5 is an example in which the confidence of a peak is calculated using the peak top. As shown in Fig. 11, the confidence can be calculated by the "average value of the weights of the index portion corresponding to the peak top."

[0074] Variation 6 is an example in which the confidence of a peak is calculated by combining a single peak, an unresolved peak, and a baseline. As shown in Figure 11, the confidence is calculated by "(B+C) / (A+B+C)", where A, B, and C are as follows:

[0075] A: The sum of the weights of the index parts that belong to the peak region of the baseline waveform W1 B: The sum of the weights of the index parts belonging to the peak region of the waveform W2 of the single peak C: The sum of the weights of the index parts belonging to the peak region of the waveform W3 of the unseparated peak Variation 7 is an example in which the confidence of a peak is calculated by combining the baseline, unresolved peak, peak start point, and peak end point. As shown in Figure 11, the confidence is calculated by "X / (X+Y)", where X and Y are as follows:

[0076] X: The index number corresponding to one of labels 2 to 4 in the peak area Y: Index number corresponding to label 0 in the peak region Modification 7 will be described in more detail with reference to Fig. 12. Fig. 12 is a diagram for explaining Modification 7. Fig. 12 is a diagram in which various regions Xa, Xb, and Ya are added to a graph obtained by labeling the determination result, in order to explain Modification 7. In the graph shown in Fig. 12, the baseline is included in part of the peak region. Depending on the relationship between the trained model and the measurement object, a determination result may be obtained that produces a graph such as that shown in Fig. 12.

[0077] In the formula for calculating the confidence factor according to the seventh modification, X is the number of indexes in the peak region that corresponds to any of the labels 2 to 4. This corresponds to the sum of the number of indexes in region Xa and the number of indexes in region Xb.

[0078] In the formula for calculating the confidence factor according to the seventh modification, Y is the number of indexes in the peak region that corresponds to label 0. This corresponds to the number of indexes in the region Ya.

[0079] As described above, the analysis device 1 according to this embodiment can calculate the confidence level of the determination result. In particular, the analysis device 1 according to this embodiment is characterized in that it calculates the confidence level of the determination result while performing peak picking using semantic segmentation technology.

[0080] In peak picking using deep learning, two known methods are to apply object detection technology from the image recognition field and semantic segmentation technology. Non-Patent Document 1 states that performance is improved by formulating the peak picking problem using semantic segmentation rather than object detection. However, there has not been a method to calculate confidence in peak picking using semantic segmentation technology.

[0081] The analysis device 1 according to this embodiment performs peak picking using semantic segmentation technology, calculates the confidence level of the determination result, and displays the determination result and confidence level on the display device 60. Furthermore, the analysis device 1 provides an interface that allows the user to modify the determination result. This allows the user to easily and efficiently check peak information, such as the peak start and end points detected by peak picking, and modify it as necessary. As a result, this embodiment provides an analysis device 1 that can output highly accurate peak detection results.

[0082] The present embodiment is merely an example and can be modified as appropriate within the spirit and scope of the present disclosure. Here, the processing of a chromatogram waveform obtained by chromatographic mass analysis has been described as an example. However, chromatograms obtained by a chromatograph having a detector other than a mass spectrometer (a spectrophotometer) and a gas chromatograph can also be analyzed by the analysis device 1. Furthermore, the analysis target is not limited to a chromatogram. For example, an optical spectrum (a waveform representing changes in detection intensity relative to the wavelength or wavenumber axis) obtained by measurement using a spectrophotometer may also be analyzed. Any waveform obtained by LC, GC, LC-PDA, LC / MS, GC / MS, LC / MS / MS, GC / MS / MS, LC / MS-IT-TOF, etc. may also be analyzed.

[0083] [Aspect] It will be understood by those skilled in the art that the above-described embodiment and its modifications are specific examples of the following aspects.

[0084] (Item 1) An analysis device according to one embodiment is an analysis device that analyzes a target waveform that is a chromatogram or a spectrum, and includes a processor and a memory that stores a trained model created by machine learning using multiple sets of partial waveforms created by dividing a reference waveform whose peak portion positions are known. The processor divides the target waveform into multiple partial waveforms, and uses the trained model to determine a peak waveform that is a peak portion among the multiple divided partial waveforms, and calculates the confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model.

[0085] According to the analysis device described in paragraph 1, when peak picking is performed using semantic segmentation technology, it is possible to calculate the confidence level of peak picking.

[0086] (2) In the analysis device described in paragraph 1, the processor calculates the confidence using a value identified from the data output from the trained model or data that has been subjected to a labeling process to label the data output from the trained model.

[0087] According to the analysis device described in paragraph 2, the confidence level can be appropriately calculated using values ​​identified from data output from a trained model or data that has been subjected to a labeling process to label data output from a trained model.

[0088] (Item 3) In the analysis device described in item 1, the processor labels the peak waveform and calculates the confidence factor.

[0089] According to the analysis device described in Section 3, peak waveforms are labeled and confidence levels are calculated.

[0090] (Item 4) In the analysis device according to item 2 or 3, the label includes at least one of a single peak, an unresolved peak, a peak start point, a peak end point, a peak top, and a baseline.

[0091] According to the analysis device described in item 4, at least one label from among a single peak, an unresolved peak, a peak start point, a peak end point, a peak top, and a baseline can be used.

[0092] (Item 5) In the analysis device described in item 1, the processor calculates, as the confidence level, the average value of the weight value corresponding to the peak start point of the target waveform and the weight value corresponding to the peak end point of the target waveform.

[0093] According to the analysis device described in paragraph 5, the confidence level can be calculated using a relatively simple formula by averaging the weight value corresponding to the peak start point of the target waveform and the weight value corresponding to the peak end point of the target waveform.

[0094] (Item 6) The analysis device according to any one of items 1 to 5, further comprising an output port that outputs a display signal for displaying the determination result and the certainty factor.

[0095] According to the analysis device described in paragraph 6, by inputting a display signal into the display device, the user can recognize the relationship between the determination result and the certainty factor.

[0096] (Clause 7) The analysis device described in clause 6 further includes a display device that displays the judgment result and the confidence level based on the display signal, and the processor accepts an operation to correct the judgment result when the judgment result and the confidence level are displayed on the display device.

[0097] According to the analysis device described in paragraph 7, the user can correct the determination result to a result that is considered more appropriate, taking into account the degree of certainty.

[0098] (Item 8) An analysis method according to another aspect is an analysis method for analyzing a target waveform which is a chromatogram or a spectrum, and includes the steps of: creating a trained model that identifies peak portions contained in an input waveform by machine learning using multiple sets of partial waveforms created by dividing a reference waveform whose peak portion positions are known; dividing the target waveform into multiple partial waveforms; using the trained model to determine a peak waveform that is a peak portion among the multiple divided partial waveforms; and calculating the confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model.

[0099] According to the analysis method described in Section 8, when peak picking is performed using semantic segmentation technology, it is possible to calculate the confidence level of peak picking.

[0100] In addition, when the processor defines the sum of the weights of the parts of the baseline estimation results that belong to the peak region as the first sum, the sum of the weights of the parts of the single peak estimation results that belong to the peak region as the second sum, and the sum of the weights of the parts of the unseparated peak estimation results that belong to the peak region as the third sum, the processor may calculate the confidence by calculating (second sum + third sum) / (first sum + second sum + third sum) (variation 6).

[0101] The processor may also be capable of labeling data output from the trained model, and when the total number of labels belonging to the peak region that correspond to any of the unseparated peaks, peak start points, and peak end points is defined as a first total number, and the total number of labels belonging to the peak region that correspond to the baseline is defined as a second total number, the processor may calculate the confidence by calculating (first total number) / (first total number+second total number) (variant 7).

[0102] The embodiments disclosed herein should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, not by the description of the above embodiments, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0103] 1 analysis device, 10 processor, 20 memory, 30 input / output port, 40 mouse, 50 keyboard, 60 display device, 61 image, 62 image, 65, 66 icon, 200 analysis program, 201 division unit, 202 model creation unit, 203 judgment unit, 204 calculation unit, 205 image processing unit, 206 output unit, 210 learning data, 211 training data, 212 validation data, 213 measurement data, 300 estimated model, Xa, Xb, Ya region, W0 waveform (chromatogram), W1 waveform (baseline), W2 waveform (single peak), W3 waveform (unresolved peak), W4 waveform (peak start point), W5 waveform (peak end point).

Claims

1. An analysis device for analyzing a target waveform that is a chromatogram or a spectrum, a processor; a memory in which a trained model created by machine learning using a plurality of sets of partial waveforms created by dividing a reference waveform whose peak positions are known is stored; The processor: Dividing the target waveform into a plurality of partial waveforms along a time axis; inputting the plurality of partial waveforms into the trained model; Using the trained model, a peak waveform that is a peak portion among the divided partial waveforms is determined; An analysis device that calculates the confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model.

2. An analysis device for analyzing a target waveform that is a chromatogram or a spectrum, a processor; a memory in which a trained model created by machine learning using a plurality of sets of partial waveforms created by dividing a reference waveform whose peak positions are known is stored; The processor: Dividing the target waveform into a plurality of partial waveforms; Using the trained model, a peak waveform that is a peak portion among the divided partial waveforms is determined; Calculating a confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model; The processor calculates the confidence using a value identified from the data output from the trained model or data that has been subjected to a labeling process that labels the data output from the trained model.

3. An analysis device for analyzing a target waveform that is a chromatogram or a spectrum, a processor; a memory in which a trained model created by machine learning using a plurality of sets of partial waveforms created by dividing a reference waveform whose peak positions are known is stored; The processor: Dividing the target waveform into a plurality of partial waveforms; Using the trained model, a peak waveform that is a peak portion among the divided partial waveforms is determined; Calculating a confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model; The processor labels the peak waveforms and calculates the confidence levels.

4. 4. The analysis device according to claim 2, wherein the label includes at least one of a single peak, an unresolved peak, a peak start point, a peak end point, a peak top, and a baseline.

5. An analysis device for analyzing a target waveform that is a chromatogram or a spectrum, a processor; a memory in which a trained model created by machine learning using a plurality of sets of partial waveforms created by dividing a reference waveform whose peak positions are known is stored; The processor: Dividing the target waveform into a plurality of partial waveforms; Using the trained model, a peak waveform that is a peak portion among the divided partial waveforms is determined; Calculating a confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model; The processor calculates, as the confidence level, the average value of the weight value corresponding to the peak start point of the target waveform and the weight value corresponding to the peak end point of the target waveform.

6. 6. The analysis device according to claim 1, further comprising an output port that outputs a display signal for displaying the determination result and the certainty factor.

7. a display device that displays the determination result and the certainty factor based on the display signal; The analysis device according to claim 6 , wherein the processor accepts an operation to correct the determination result while the determination result and the certainty factor are being displayed on a display device.

8. An analysis method for analyzing a target waveform that is a chromatogram or a spectrum, comprising: A step of creating a trained model that identifies peak portions included in an input waveform by machine learning using a plurality of sets of partial waveforms created by dividing a reference waveform whose peak positions are known; Dividing the target waveform into a plurality of partial waveforms in a time axis direction; inputting the plurality of partial waveforms into the trained model; determining a peak waveform that is a peak portion among the divided partial waveforms using the trained model; An analysis method comprising a step of calculating the confidence level of the peak waveform determination result using data output from the trained model when determining the peak portion of the target waveform using the trained model.

Images