Analytical device and waveform processing program for analytical device

Abstract

An analysis device according to one embodiment of the present invention comprises: a waveform deformation unit (112) that expands or compresses by a factor of N (N is a positive value other than 0 and 1) in the direction of a signal intensity axis and / or expands or compresses by a factor of M (M is a positive value other than 0 and 1 and may be the same value as N) in the direction of a prescribed parameter axis, a target waveform obtained by means of analysis and exhibiting a change in signal intensity corresponding to a change in a prescribed parameter value; a peak detection unit (113) that uses a learned model generated in advance by machine learning using a signal waveform and a starting point and an end point of a peak that is a correct answer as training data, inputs the signal waveform after deformation by the waveform deformation unit, and outputs the starting point and the end point of the peak as detection results; and a waveform inverse deformation unit (114) that compresses or expands, by a factor that is the inverse of during the deformation by the waveform deformation unit, information for the starting point and the end point of the peak, said peak being the output from the peak detection unit, and determines peak detection results for the target signal waveform.

Description

[Technical field]

[0001] The present invention relates to an analysis device and a program for processing, on a computer, a signal waveform obtained by analysis. [Background technology]

[0002] In a liquid chromatograph (LC) or gas chromatograph (GC) apparatus, a sample is introduced into a column, and various components contained in the sample are separated in the time direction while the sample passes through the column, and the components are detected by a detector provided at the outlet of the column. Peaks corresponding to various components in the sample appear in a chromatogram waveform created based on a detection signal obtained by the detector. Since the time at which a peak is observed (retention time) corresponds to the type of component, the component can be identified from the retention time of the peak, that is, a qualitative analysis can be performed. In addition, since the height and area of ​​the peak correspond to the concentration or content of the component, the concentration or content of the component can be obtained from the height and area values ​​of the peak, that is, a quantitative analysis can be performed.

[0003] In order to perform qualitative and quantitative analysis, it is necessary to accurately detect peaks on a chromatogram waveform and determine the positions of the start and end points (retention time) of the peaks. The chromatogram waveform obtained by analysis may have various noises superimposed thereon, the baseline may fluctuate, or peaks derived from multiple components may overlap. Therefore, it is not necessarily easy to accurately detect peaks from a chromatogram waveform, and various algorithms have been used in the past to detect peaks using a computer. As one of such peak detection methods, a method using AI (artificial intelligence) technology, represented by deep learning, has been put to practical use in recent years (see Non-Patent Document 1).

[0004] The peak detection method described in Non-Patent Document 1 detects peaks using a trained model generated by so-called supervised machine learning. When generating such a trained model, that is, when learning the model, a large amount of training data is usually used, which is a set of chromatogram waveforms collected by an actual analyzer and information on peaks (start and end points) detected by an expert on the waveform. One of the major advantages of such a peak detection method using AI technology is that, unlike many other peak detection methods, it does not require the operator to set cumbersome detection conditions (parameters) in advance or make decisions during the peak detection process. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] "PeakintelligenceTM Ver.2 Waveform Processing Optional Software for LabSolutions InsightTM", [Online], [Retrieved April 7, 2021], Shimadzu Corporation, Internet<URL: https: / / www.an.shimadzu.co.jp / lcms / m_package / peakintelligence.htm> Summary of the Invention [Problem to be solved by the invention]

[0006] An effective method for improving the accuracy of a trained model generated in supervised machine learning, including deep learning, is to use as much training data as possible that covers the variability (fluctuation) of the features of the detection target for training. However, collecting training data is very time-consuming and laborious. In addition, as the amount of training data increases, the processing effort required for generating a trained model also increases rapidly, and the cost required for this also increases. For this reason, the amount of training data prepared is limited, and when peak detection is performed using a trained model generated based on such training data, there are cases where a problem occurs in which a peak that should be detected and is observed in an actual chromatogram waveform is not detected and is overlooked.

[0007] The present invention has been made to solve these problems, and its purpose is to provide an analytical device and a waveform processing program for an analytical device that can effectively detect peaks that would have been overlooked in the past, while using an existing trained model. [Means for solving the problem]

[0008] In order to solve the above problems, one aspect of the present invention is an analytical device comprising: a waveform modification unit that expands or reduces a target signal waveform, which indicates a change in signal intensity corresponding to a change in a value of a predetermined parameter obtained by the analysis, by a factor of N (where N is a positive value other than 0 and 1) along a signal intensity axis and / or by a factor of M (where M is a positive value other than 0 and 1 and may be equal to N) along the predetermined parameter axis; a peak detection unit that uses a trained model that has been generated in advance by machine learning using the signal waveform and the correct peak start and end points as training data, receives the signal waveform transformed by the waveform transformation unit as an input, and outputs the peak start and end points as detection results; a waveform inverse transformation unit that reduces or expands the information on the start and end points of the peaks output from the peak detection unit at a ratio opposite to that used for transformation by the waveform transformation unit, thereby obtaining a peak detection result for the target signal waveform; It is equipped with the following.

[0009] One embodiment of the waveform processing program for an analytical device according to the present invention, which has been made to solve the above problems, is a program for processing, on a computer, a target signal waveform that indicates a change in signal intensity corresponding to a change in a value of a predetermined parameter obtained by analysis, the program comprising: a waveform modification step of enlarging or reducing the target signal waveform by N times (where N is a positive value other than 0 and 1) in the direction of a signal strength axis, and / or enlarging or reducing the target signal waveform by M times (where M is a positive value other than 0 and 1 and may be equal to N) in the direction of the predetermined parameter axis; a peak detection step of using a trained model generated in advance by machine learning using the signal waveform and the correct peak start and end points as training data, inputting the signal waveform after the waveform transformation step, and outputting the peak start and end points as detection results; a waveform inverse transformation step of reducing or expanding the information on the start and end points of the peaks output by the peak detection step at a magnification ratio opposite to that used in the waveform transformation step, thereby obtaining a peak detection result for the target signal waveform; This is what causes the following to be executed. Effect of the Invention

[0010] According to the above-described aspects of the analytical device and the waveform processing program for the analytical device of the present invention, it is possible to effectively detect peak waveforms that may have been overlooked in the past due to low peak signal intensity in a signal waveform such as a chromatogram waveform, while using an existing trained model. This makes it possible to improve the accuracy of peak detection without regenerating a trained model. [Brief description of the drawings]

[0011] [Figure 1] 1 is a schematic diagram of an LC device according to one embodiment of the present invention. [Diagram 2] FIG. 2 is a schematic diagram illustrating the configuration of a model generation device that generates a trained model used in the LC device shown in FIG. [Diagram 3] FIG. 4 is a diagram showing an example of a chromatogram waveform. [Figure 4] FIG. 4 is an explanatory diagram of the procedure of peak detection processing in the LC device of the present embodiment. [Diagram 5] FIG. 1 is a diagram illustrating a procedure for peak detection processing in a conventional LC device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] The analytical device according to the present invention may include, for example, a liquid chromatograph (LC) device including a liquid chromatograph mass spectrometer, a gas chromatograph (GC) device including a gas chromatograph mass spectrometer, a spectrometer (such as an infrared absorption spectrophotometer, a visible-ultraviolet spectrophotometer, or a fluorescence spectrophotometer), an X-ray analyzer (such as an X-ray fluorescence analyzer or an X-ray diffraction analyzer), or the like.

[0013] Furthermore, the waveform processing program for an analyzer according to the present invention can be provided to a user by storing it in a computer-readable non-transitory recording medium, such as a CD-ROM, a DVD-ROM, a memory card, or a USB memory (dongle). The program can also be provided to a user in the form of data transfer via a communication line such as the Internet. Furthermore, the program can be pre-installed in a computer that is a part of the system (strictly speaking, a storage device that is a part of the computer) when the user purchases the system.

[0014] In addition, in the present invention, the type of machine learning is not particularly important as long as it is supervised machine learning, but generally, deep learning can be used.

[0015] Hereinafter, an LC apparatus, which is one embodiment of an analytical apparatus according to the present invention, will be described with reference to the accompanying drawings.

[0016] [Configuration of the device according to this embodiment] Fig. 1 is a schematic diagram of an LC device according to the present embodiment. Fig. 2 is a schematic diagram of a trained model generation device that generates a trained model used in the LC device shown in Fig. 1.

[0017] 1, the LC device 1 of this embodiment includes an LC measurement unit 10, a data analysis unit 11, an input unit 12, and a display unit 13. Although not shown, the LC measurement unit 10 includes a mobile phase container, a liquid delivery pump, an injector, a column, a column oven, a detector, etc., and performs LC analysis on a given sample and outputs chromatogram data indicating the change over time in the signal intensity obtained by the detector. There is no particular limit to the type or type of detector, and examples that can be used include a mass spectrometer, a photodiode array (PDA) detector, and a fluorescence detector.

[0018] The data analysis unit 11 includes functional blocks such as a data collection unit 110, a waveform selection unit 111, a waveform enlargement unit 112, a peak detection unit 113, a waveform reduction unit 115, a qualitative / quantitative analysis unit 116, and a result display processing unit 117. The peak detection unit 113 includes a trained model storage unit 114 in which a trained model generated by performing learning in advance is stored. This trained model is essentially a peak discriminator that discriminates peak waveforms.

[0019] The entity of the data analysis unit 11 is a personal computer or a high-performance workstation on which a specific software is installed, or a computer system including a high-performance computer connected to such a computer via a communication line. That is, the functions of each block included in the data analysis unit 11 can be realized by processing using various data stored in a single computer or a computer system including multiple computers, which is achieved by executing software installed in the computer or computer system. Of course, processing speed can be increased by replacing part of the functions of the data analysis unit 11 with a dedicated hardware circuit such as a digital signal processor.

[0020] 2, the model generation device 2 provided separately from the LC device 1 includes, as functional blocks, a teacher data input unit 20, a learning execution unit 21, and a model construction unit 22. The trained model generated in this model generation device 2 is stored in a trained model storage unit 114 in the data analysis unit 11 of the LC device 1.

[0021] The model generating device 2 is actually a high-performance computer in which a specific software is installed. As is well known, learning a model based on a machine learning algorithm in deep learning or the like generally requires a huge amount of calculation, and therefore a high-performance computer capable of high-speed calculation is used. Therefore, in many cases, the task of learning the model is performed by the manufacturer of the LC device 1 or a company that develops software. In that case, the model generating device 2 itself is located at such a manufacturer or company. Of course, if possible, the data analysis unit 11 of the LC device 1 may be provided with a function of learning such a model.

[0022] [Steps for generating a trained model] First, the operation of the model generation device 2 when generating a trained model will be described. When generating a trained model, a large number of diverse chromatogram waveform data are prepared, and the retention times of the start and end points of one or more peaks appearing in each chromatogram waveform are accurately calculated. The diverse chromatogram waveform data referred to here refers to chromatogram waveforms that include elements such as various noise contamination, baseline fluctuation (drift), overlapping of multiple peaks, or deformation of peak shape, which may appear in a chromatogram waveform when peak detection is actually performed.

[0023] FIG. 3 is an example of a chromatogram waveform obtained by an LC device. In this example, the start and end points are (t s1 , t e1 ), (t s2 , t e2 ), (t s3 , t e3 ) are observed. The teacher data input unit 20 reads in a large amount of teacher data that is a set of such chromatogram waveform data and accurate peak information including the peak start and end points. The chromatogram waveform does not have to be the entire waveform from the start to end of the measurement as shown in Figure 3, but may be a waveform that selects the peak portion indicated by the dashed line in Figure 3.

[0024] The learning execution unit 21, for example, converts chromatogram waveform data into an image, and trains a model based on a predetermined method using the image and peak information that is the correct answer. Based on the learning result, the model construction unit 22 constructs a learning model for estimating the start and end points of peaks on the chromatogram waveform. As is well known, there are various algorithms for machine learning, and here, deep learning, which is one of general object detection algorithms in image recognition, is used, but usable machine learning methods are not limited to this.

[0025] The model construction unit 22 temporarily stores the trained model generated by learning using a large amount of teacher data. In the LC device 1, the trained model generated in the model generation device 2 as described above is transferred to and stored in the trained model storage unit 114 via, for example, a communication line or a storage medium.

[0026] [Peak detection processing for measured waveforms] Next, the operation of the peak detection process in the LC device 1 will be described with reference to FIG. The LC measurement unit 10 performs LC analysis on the target sample. In the data analysis unit 11, the data collection unit 110 acquires the chromatogram data obtained by the LC measurement unit 10 and temporarily stores it. Here, it is assumed that the chromatogram waveform shown in Fig. 4(a) was obtained by the LC analysis.

[0027] When an instruction to execute peak detection processing is given from the input unit 12, the waveform selection unit 111 receives chromatogram data to be processed from the data collection unit 110. Then, the waveform selection unit 111 determines the retention time range to be subjected to waveform expansion according to a predetermined criterion.

[0028] Specifically, for example, the difference between the minimum signal intensity and the maximum signal intensity in the chromatogram data to be processed is calculated, and if the difference is equal to or less than a predetermined threshold, the entire chromatogram (i.e., the entire retention time range) can be set as a target for waveform enlargement. Alternatively, the chromatogram can be divided into a plurality of parts along the time axis, with a position where the change in signal intensity along the time axis is small (for example, position U in FIG. 4(a)) as a boundary, and each divided waveform can be judged as to whether it is within the retention time range of the target for waveform enlargement according to the above-mentioned criteria.

[0029] The reason why the waveform selection unit 111 determines the retention time range for which the waveform is to be expanded is to expand the waveform only in the retention time range where the peak signal strength is small and there is a possibility that the waveform is overlooked, and not to expand the waveform in the retention time range where the peak signal strength is large and there is no possibility that the waveform is overlooked. As long as it is possible to determine the waveform shape corresponding to such a purpose, the method is not limited to the above. Also, if there is no practical problem even if the signal strength becomes even stronger by expanding the waveform when the peak signal strength is large, the waveform selection unit 111 may be omitted and all retention time ranges may be subject to waveform expansion.

[0030] In the example of Fig. 4(a), the entire chromatogram waveform is assumed to be the target for waveform enlargement. Next, as shown in Fig. 4(b), the waveform enlargement unit 112 performs an enlargement process (N times in the vertical axis direction) of the signal intensity of the waveform included in the retention time range determined to be the target for waveform enlargement by multiplying it by N (here, N is a value greater than 1). Two peaks P1 and P2 (however, they are not detected as peaks at this point) present on the original chromatogram waveform correspond to peaks P10 and P20, respectively, on the enlarged chromatogram waveform. The value of N may be a predetermined value, or may be appropriately selected by the user.

[0031] The peak detection unit 113 receives peak information including the start and end points of the peaks as an output by inputting the chromatogram waveform with the signal intensity expanded as described above into the trained model stored in the trained model storage unit 114. This peak information is information on the peaks automatically detected by the peak discriminator in the input chromatogram waveform. FIG. 4(c) is a diagram showing the positions of the peak start and end points automatically detected for the chromatogram waveform shown in FIG. 4(b), and a dotted line connecting the paired peak start and end points with a straight line. Here, two peaks P10 and P20 are accurately detected.

[0032] After peak detection is performed in this manner, the waveform reducing unit 115 performs the opposite process to that performed by the waveform expanding unit 112. That is, the waveform reducing unit 115 performs a reduction process of multiplying the signal intensity by 1 / N (reducing it to 1 / N in the vertical axis direction) for the chromatogram waveform included in the retention time range that is the target for waveform expansion. As a result, the signal intensity corresponding to the peak start point and end point, which are the peak detection results in the chromatogram waveform expanded as shown in Fig. 4(c), decreases together with the chromatogram waveform as shown in Fig. 4(d). As a result, a peak detection result that is substantially equivalent to that obtained when peak detection is performed on the original chromatogram waveform shown in Fig. 4(a) is obtained.

[0033] 5 is an explanatory diagram of processing when automatic peak detection is performed in peak detection unit 113 on a chromatogram waveform that is not subjected to waveform magnification. This corresponds to peak detection processing in a conventional LC device. In this case, the signal intensity of peak P2 on the chromatogram waveform is much smaller than that of peak P1. Generally, in machine learning such as deep learning, if training data that is appropriately labeled to recognize such peaks with low signal intensity as peaks is not provided during model training, the generated trained model does not have the ability to recognize peaks with low signal intensity. Therefore, as shown in Figure 5(b), peak P1 with high signal intensity is detected, but peak P2 with low signal intensity is missed.

[0034] As mentioned above, when preparing training data, possible deformations and variations of peak waveforms are assumed, but the more training data there is, the greater the cost and time required to generate a trained model, so it is often necessary to limit the amount of training data due to cost and time constraints. In addition, since the needs of users of LC devices are diverse, there may be users who want to detect small signal intensity peaks that were not anticipated when the trained model was generated. In the latter case, it is possible to address this by adding new training data and regenerating the trained model, but the amount of work required is quite large and the cost burden is high.

[0035] In contrast, the LC device 1 of this embodiment can detect peaks with low signal intensity that have been missed in the past without changing the trained model, that is, by using an existing trained model that does not have the ability to recognize peaks with low signal intensity. Adding the functions of the waveform selection unit 111, the waveform enlargement unit 112, and the waveform reduction unit 115 to the data analysis unit 11, specifically, adding a program for realizing such functions, is much easier and less costly and time-consuming than reconstructing a trained model using new teacher data. Thus, the LC device 1 of this embodiment can improve the accuracy of peak detection while saving costs and time.

[0036] In the data analysis unit 11, the qualitative / quantitative analysis unit 116 identifies the components (compounds) corresponding to each peak based on the peak information provided by the peak detection unit 113. Alternatively, the qualitative / quantitative analysis unit 116 calculates the peak height value or peak area value, and calculates the quantitative value, which is the concentration or content of each component, from the calculated values. The result display processing unit 117 receives the identification results and quantitative results of each component, and displays them in a predetermined format on the display unit 13. The result display processing unit 117 may also display a chromatogram waveform including the peak detection results as shown in FIG. 4(d) on the display unit 13.

[0037] [Variations] In the LC device 1 of the above embodiment, the waveform enlargement unit 112 and the waveform reduction unit 115 enlarge / reduce the chromatogram waveform only in the signal intensity axis direction, but the chromatogram waveform may also be enlarged / reduced in the time axis direction. If the chromatogram waveform is enlarged only in the signal intensity axis direction, the peak shape becomes vertically elongated. That is, the peak half width / peak height value becomes smaller. Depending on the method of learning using teacher data during model learning, there may be cases where a peak deformed so that the peak half width / peak height value becomes smaller cannot be properly detected. Therefore, depending on the characteristics of the trained model used, it is preferable to enlarge / reduce the chromatogram waveform in the signal intensity axis direction and also in the time axis direction.

[0038] In this case, the magnification in the signal intensity direction and the magnification in the time direction do not necessarily have to be the same. That is, before peak detection, the chromatogram waveform is magnified by N times in the signal intensity direction and by M times in the time direction (however, N=M may be a value greater than 1), and after peak detection, the chromatogram waveform is reduced by 1 / N times in the signal intensity direction and by 1 / M times in the time direction.

[0039] Furthermore, the waveform enlargement unit 112 and the waveform reduction unit 115 may enlarge or reduce the chromatogram waveform only in the time axis direction without enlarging or reducing the chromatogram waveform in the signal intensity axis direction. This is useful in a situation where the chromatogram waveform is compressed in the time axis direction because the LC separation conditions are different from the assumed conditions, for example, the speed of the mobile phase in the actual LC analysis is fast.

[0040] In the LC device 1 of the above embodiment, the chromatogram waveform is expanded before the peak is detected and reduced after the peak is detected. However, depending on the characteristics of the trained model used, the chromatogram waveform may be reduced before the peak is detected and expanded after the peak is detected to return it to its original state. This is useful in a situation where the signal intensity of the peak in the waveform to be actually processed is excessively large compared to the range of the variation of the signal intensity of the peak assumed during model training, contrary to the above description. Also, it is useful in a situation where the chromatogram waveform is extended in the time axis direction because the LC separation conditions are different from the assumed conditions, such as the speed of the mobile phase in the actual LC analysis being slow.

[0041] In addition, in the LC device 1 of the above embodiment, multiple peaks are detected by performing a single peak detection process on the chromatogram waveform, but it is also possible to detect peaks with significantly different signal intensities as thoroughly as possible by repeatedly performing the peak detection process multiple times as follows.

[0042] That is, for example, for a chromatogram waveform to be processed as shown in FIG. 4(a), the first peak detection process performs peak detection in the same manner as in the conventional method without expanding or reducing the waveform. In this case, as explained with reference to FIG. 5, peak P1 is detected, but peak P2 may be missed. Next, the peak detected in the first peak detection process is removed (for example, the start point and end point of peak P1 in FIG. 5(b) are connected with a straight line), and then peak detection with waveform expansion and reduction as described above is performed as the second peak detection process. In this case, peak P10 in FIG. 4(b) does not exist, so only the remaining peak P20 is detected. At this time, peak P20 is detected without omission because its signal intensity is large. In this way, by aggregating the peak information detected in each of the multiple peak detection processes, the final peak detection result can be created, making it possible to detect peaks with few omissions.

[0043] Naturally, by changing the zoom ratio in more stages, or by zooming in and out at different ratios in the signal intensity direction and in the time direction, it is possible to repeat the peak detection process three or more times.

[0044] Furthermore, although the above embodiment is an LC device, the present invention can be applied not only to LC devices but also to GC devices, and furthermore can be applied to various other analytical devices.

[0045] For example, it is clear that the present invention can be applied to cases where peak detection is performed on spectral waveforms such as mass spectra obtained by mass spectrometers, optical spectra obtained by various spectroscopic analyzers such as absorption spectrophotometers and fluorescence spectrophotometers, ion mobility spectra obtained by ion mobility analyzers, and X-ray spectra obtained by X-ray analyzers.

[0046] Furthermore, it is clear that the present invention is not limited to the above-described embodiment and modifications, and that any modifications, amendments, or additions made within the spirit and scope of the present invention will fall within the scope of the claims of the present application.

[0047] [Various aspects] It will be understood by those skilled in the art that the exemplary embodiments and modifications described above are examples of the following aspects.

[0048] (Item 1) One aspect of the analytical device according to the present invention is a waveform modification unit that expands or reduces a target signal waveform, which indicates a change in signal intensity corresponding to a change in a value of a predetermined parameter obtained by the analysis, by a factor of N (where N is a positive value other than 0 and 1) along a signal intensity axis and / or by a factor of M (where M is a positive value other than 0 and 1 and may be equal to N) along the predetermined parameter axis; a peak detection unit that uses a trained model that has been generated in advance by machine learning using the signal waveform and the correct peak start and end points as training data, receives the signal waveform transformed by the waveform transformation unit as an input, and outputs the peak start and end points as detection results; a waveform inverse transformation unit that reduces or expands the information on the start and end points of the peaks output from the peak detection unit at a ratio opposite to that used for transformation by the waveform transformation unit, thereby obtaining a peak detection result for the target signal waveform; It is equipped with the following.

[0049] (Item 6) One aspect of the waveform processing program for an analytical device according to the present invention is a program for processing, on a computer, a target signal waveform that indicates a change in signal intensity corresponding to a change in a value of a predetermined parameter obtained by analysis, the program comprising: a waveform modification step of enlarging or reducing the target signal waveform by N times (where N is a positive value other than 0 and 1) in the direction of a signal strength axis, and / or enlarging or reducing the target signal waveform by M times (where M is a positive value other than 0 and 1 and may be equal to N) in the direction of the predetermined parameter axis; a peak detection step of using a trained model generated in advance by machine learning using the signal waveform and the correct peak start and end points as training data, inputting the signal waveform after the waveform transformation step, and outputting the peak start and end points as detection results; a waveform inverse transformation step of reducing or expanding the information on the start and end points of the peaks output by the peak detection step at a magnification ratio opposite to that used in the waveform transformation step, thereby obtaining a peak detection result for the target signal waveform; This is what causes the following to be executed.

[0050] According to the analytical device described in paragraph 1 and the waveform processing program for an analytical device described in paragraph 6, while using an existing trained model, it is possible to satisfactorily detect peak waveforms that may have been overlooked in the past due to low peak signal intensity in a signal waveform such as a chromatogram waveform, for example. This makes it possible to improve the peak detection accuracy without regenerating a trained model.

[0051] (Item 2) In the analyzer described in item 1, the predetermined parameter may be time, and the signal waveform may be a chromatogram waveform.

[0052] In particular, chromatogram waveforms obtained by LC devices are prone to baseline drift and overlapping of multiple peaks due to incomplete separation of multiple components, making it difficult to detect all peaks using a single specific algorithm and making parameter adjustment and setting cumbersome. In contrast, the analytical device described in paragraph 2 makes it possible to detect peaks with high accuracy while simplifying the need for cumbersome parameter adjustment and setting.

[0053] (Item 3) In the analysis device described in item 1 or 2, the machine learning may be deep learning.

[0054] There are various methods for machine learning, among which deep learning in particular tends to have difficulty in obtaining sufficient performance for data with characteristics that are not included in the training data when the model is trained. In contrast, the analysis device described in paragraph 3 can detect peaks with high accuracy while using an existing trained model.

[0055] (Item 4) In the analyzer described in any one of items 1 to 3, the peak information may include a start point and an end point of a peak.

[0056] According to the analytical device described in paragraph 4, by determining the start and end points of a peak, the area value of the peak can be calculated accurately, and quantitative analysis can be performed with high accuracy.

[0057] (Item 5) The analytical device described in any one of items 1 to 4 may further include a judgment unit that judges the magnitude of fluctuation in the signal intensity level in the target signal waveform, and the waveform transformation unit may perform processing to expand or shrink the signal waveform for a range of parameter values ​​in which the judgment unit judges that the fluctuation in the signal intensity level is smaller or larger than a specified value.

[0058] In the analysis device described in paragraph 5, the enlargement or reduction process can be performed only on a partial signal waveform where the signal strength is particularly small or particularly large, rather than on the signal waveform corresponding to the entire parameter value range obtained by analysis. This makes it possible to narrow the range of peak signal strength, and to perform accurate peak detection even when using a trained model with a relatively narrow tolerance for fluctuations in signal strength. [Explanation of symbols]

[0059] 1...LC device 10...LC measurement section 11…Data Analysis Section 110…Data collection section 111...Waveform selection section 112…Waveform expansion section 113...Peak detector 114…Trained model memory unit 115...Waveform reduction section 116…Qualitative / Quantitative Analysis Department 117...Result display processing unit 12...Input section 13…Display section 2. Model generation device 20…Teacher data input section 21…Learning Execution Department 22…Model Construction Section

Claims

1. a waveform selection unit that determines a range of parameter values ​​to be expanded or contracted by a waveform modification unit (described later) in accordance with a predetermined criterion for a target signal waveform that indicates a change in signal intensity corresponding to a change in the value of a predetermined parameter obtained by the analysis; a waveform modification unit that expands or reduces the range of parameter values ​​of the target signal waveform determined by the waveform selection unit by a factor of N (where N is a positive value other than 0 and 1) along a signal strength axis and / or by a factor of M (where M is a positive value other than 0 and 1 and may be equal to N) along an axis of the predetermined parameter; a peak detection unit that uses a trained model that has been generated in advance by machine learning using the signal waveform and the correct peak start and end points as training data, receives the signal waveform transformed by the waveform transformation unit as an input, and outputs the peak start and end points as detection results; a waveform inverse transformation unit which reduces or expands the information on the start and end points of the peaks output from the peak detection unit by 1 / N times in the direction of the signal strength axis and / or reduces or expands the information by 1 / M times in the direction of the axis of the predetermined parameter in the opposite manner to the transformation performed by the waveform transformation unit, thereby obtaining a peak detection result for the target signal waveform; An analytical device comprising:

2. The analytical device according to claim 1 , wherein the predetermined parameter is time and the signal waveform is a chromatogram waveform.

3. The analysis device according to claim 1 , wherein the machine learning is deep learning.

4. 2. The analytical apparatus of claim 1, wherein the value of N and / or the value of M are predetermined or user selected values.

5. The analytical device described in claim 1, wherein the waveform selection unit includes a judgment unit that judges the magnitude of fluctuation in signal intensity level in the target signal waveform, and determines the range of parameter values ​​for which the judgment unit judges the fluctuation in signal intensity level to be smaller or larger than a specified value as the range of parameter values ​​to be expanded or contracted by the waveform transformation unit.

6. A program for processing, on a computer, a target signal waveform showing a change in signal intensity in response to a change in a value of a predetermined parameter, the program comprising: a waveform selection step of determining a range of parameter values ​​for which the target signal waveform is to be expanded or contracted by a waveform modification unit (described later) in accordance with a predetermined criterion; a waveform modification step of expanding or reducing the range of parameter values ​​of the target signal waveform determined in the waveform selection step by a factor of N (where N is a positive value other than 0 and 1) along a signal strength axis and / or by a factor of M (where M is a positive value other than 0 and 1 and may be equal to N) along an axis of the predetermined parameter; a peak detection step of using a trained model generated in advance by machine learning using the signal waveform and the correct peak start and end points as training data, inputting the signal waveform after the waveform transformation step, and outputting the peak start and end points as detection results; a waveform inverse transformation step of reducing or expanding the information on the start and end points of the peaks output by the peak detection step at a magnification ratio opposite to that used in the waveform transformation step, thereby obtaining a peak detection result for the target signal waveform; A waveform processing program for an analytical device.

Images