Data processing device, program and characteristic evaluation method

Abstract

To provide a new method for analyzing the characteristics or the like of a sample from a spectrogram generated by NMR measurement.SOLUTION: A multivariate analysis unit 18 performs multivariate analysis on spectrograms of multiple samples to generate principal components for separating the multiple samples for each attribute of the samples. A distribution generation unit 20 generates a distribution of loading values of specific principal components with respect to indices determined by time information and frequency information contained in the multiple spectrograms on the basis of the results of the multivariate analysis. A plot generation unit 22 generates a loading plot that represents the loading values of the specific principal components on a graph defined by time and frequency from the distribution of the loading values.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a technique for processing data generated by measurements using an NMR (nuclear magnetic resonance) device. [Background technology]

[0002] There are known methods for evaluating the properties of mixtures, such as biological samples such as serum, and mixed solution samples containing polymer compounds.

[0003] Patent Documents 1 and 2 describe methods for evaluating the characteristics of biological samples. Specifically, the methods include the steps of acquiring an FID (Free Induction Decay) signal derived from the biological sample using an NMR (Nuclear Magnetic Resonance) device and calculating a spectrogram by repeating short-time frequency analysis over the entire FID signal. Furthermore, the methods describe a method of generating a score plot (i.e., a principal component plot) by performing multivariate analysis on the spectrogram and identifying the attributes of the biological sample based on the score plot.

[0004] Patent Documents 3 and 4 describe devices that perform bucket integration on target spectrum data using a bucket set, thereby reducing target spectrum data into histogram data. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2015-114157 [Patent Document 2] Japanese Patent Application Publication No. 2019-158868 [Patent Document 3] Patent No. 5415476 [Patent Document 4] Patent No. 5020491 Summary of the Invention [Problem to be solved by the invention]

[0006] As an example of time-frequency analysis, the short-time Fourier transform (STFT) is sometimes used. One of the parameters that must be set in the short-time Fourier transform is the frame length. In the short-time Fourier transform, there is a trade-off between the time resolution and frequency resolution of the analysis, and the parameter is controlled by setting the frame length.

[0007] The techniques described in Patent Documents 1 and 2 can identify sample attributes from score plots, but cannot determine the characteristics of the sample and feed the results of that determination back to the parameters of time-frequency analysis. Furthermore, the techniques described in Patent Documents 1 and 2 cannot consider factors that cause differences in sample attributes (e.g., components, structures, etc. that affect differences in attributes). The same can be said for the techniques described in Patent Documents 3 and 4.

[0008] An object of the present disclosure is to provide a new method for examining the properties of a sample from a spectrogram generated by NMR measurement. [Means for solving the problem]

[0009] One aspect of the present disclosure is a data processing device including: an acquisition means for acquiring a plurality of spectrograms on a first coordinate system defined by a time axis and a frequency axis, which are generated by performing NMR measurements on each of a plurality of samples; a multivariate analysis means for generating principal components for separating the plurality of samples by sample attributes by performing multivariate analysis on a dataset consisting of the plurality of spectrograms; a distribution generation means for generating a distribution of loading values ​​corresponding to the principal components as a result of the multivariate analysis by the multivariate analysis means; and a plot generation means for generating a loading plot representing each loading value on a second coordinate system defined by a time axis and a frequency axis from the distribution of loading values, wherein the second coordinate system is the same as the first coordinate system. One aspect of the present disclosure is a data processing device comprising: an acquisition means for acquiring a plurality of spectrograms defined by time and frequency, generated by performing NMR measurements on each of a plurality of samples; a multivariate analysis means for performing multivariate analysis on the plurality of spectrograms to generate principal components for separating the plurality of samples by their attributes; a distribution generation means for generating a distribution of loading values ​​of a specific principal component for an index defined by time information and frequency information included in the plurality of spectrograms based on the results of the multivariate analysis by the multivariate analysis means; and a plot generation means for generating, from the distribution of loading values, a loading plot representing the loading values ​​of the specific principal component on a graph defined by time and frequency.

[0010] The data processing device may further include a display control means for displaying the spectrogram of the specified sample and the loading plot side by side on the display.

[0011] The data processing device may further include an identification means for identifying an analysis region on the loading plot using a time range and a frequency range, and the multivariate analysis means, the distribution generation means, and the plot generation means may each perform processing on the analysis region.

[0012] When the time range and the frequency range are specified by a user on the loading plot, the identification means may identify an area defined by the time range and the frequency range specified by the user as the analysis area.

[0013] The specifying means may specify, as the analysis region, a region defined by a time range and a frequency range, in which a loading value falls within a specific range.

[0014] The plot generating means may generate the loading plot by performing coordinate transformation on the distribution of the loading values.

[0015] One aspect of the present disclosure is a program that causes a computer to function as: an acquisition means that acquires multiple spectrograms on a first coordinate system defined by a time axis and a frequency axis, generated by performing NMR measurements on each of multiple samples; a multivariate analysis means that generates principal components for separating the multiple samples by sample attribute by performing multivariate analysis on a dataset consisting of the multiple spectrograms; a distribution generation means that generates a distribution of loading values ​​corresponding to the principal components as a result of the multivariate analysis by the multivariate analysis means; and a plot generation means that generates a loading plot representing each loading value on a second coordinate system defined by a time axis and a frequency axis from the distribution of loading values, wherein the second coordinate system is the same as the first coordinate system. One aspect of the present disclosure is a program that causes a computer to function as: an acquisition means that acquires multiple spectrograms defined by time and frequency, generated by performing NMR measurements on each of multiple samples; a multivariate analysis means that generates principal components for separating the multiple samples by sample attributes by performing multivariate analysis on the multiple spectrograms; a distribution generation means that generates a distribution of loading values ​​of a specific principal component for an index defined by time information and frequency information included in the multiple spectrograms based on the results of the multivariate analysis by the multivariate analysis means; and a plot generation means that generates, from the distribution of loading values, a loading plot that represents the loading values ​​of the specific principal component on a graph defined by time and frequency.

[0016] One aspect of the present disclosure is a characteristic evaluation method including: a first step of acquiring a plurality of spectrograms on a first coordinate system defined by a time axis and a frequency axis, the spectrograms being generated by performing NMR measurements on each of a plurality of samples; a second step of generating principal components for separating the plurality of samples by sample attribute by performing multivariate analysis on a dataset consisting of the plurality of spectrograms; a third step of generating a distribution of loading values ​​of the principal components as a result of the multivariate analysis performed in the second step; and a fourth step of generating a loading plot representing each loading value on a second coordinate system defined by a time axis and a frequency axis from the distribution of loading values, wherein the second coordinate system is the same as the first coordinate system. One aspect of the present disclosure is a characteristic evaluation method including: a first step of acquiring a plurality of spectrograms defined by time and frequency, each of which is generated by performing NMR measurements on a plurality of samples; a second step of generating principal components for separating the plurality of samples by sample attributes by performing multivariate analysis on the plurality of spectrograms; a third step of generating a distribution of loading values ​​of specific principal components for indices defined by time information and frequency information included in the plurality of spectrograms based on the results of the multivariate analysis in the second step; and a fourth step of generating a loading plot representing the loading values ​​of the specific principal components on a graph defined by time and frequency from the distribution of loading values. [Effects of the Invention]

[0017] According to the present disclosure, a new method can be provided for examining the characteristics of a sample from a spectrogram generated by NMR measurement. [Brief explanation of the drawings]

[0018] [Figure 1] 1 is a block diagram showing a configuration of a data processing system according to an embodiment. [Figure 2] 1 is a block diagram showing a hardware configuration of a data processing device according to an embodiment; [Figure 3] FIG. 10 is a diagram showing an FID signal. [Figure 4] FIG. 1 shows a spectrogram of HSA. [Figure 5] FIG. 1 shows a spectrogram of HDL. [Figure 6] FIG. 1 shows a spectrogram of LDL. [Figure 7] FIG. 1 shows a score plot. [Figure 8] FIG. 10 is a diagram showing the distribution of loading values. [Figure 9] FIG. 1 shows a loading plot of the first principal component (PC-1). [Figure 10] FIG. 2 is a diagram for explaining coordinate conversion between two-dimensional data and one-dimensional data. [Figure 11] FIG. 1 shows a spectrogram and loading plot of HSA. [Figure 12] FIG. 1 shows a loading plot. [Figure 13] FIG. 1 shows a score plot. [Figure 14] FIG. 1 shows a loading plot. [Figure 15] FIG. 1 shows a spectrogram of BKS18. [Figure 16] FIG. 1 shows a spectrogram of Jcl. [Figure 17] FIG. 1 shows a score plot. [Figure 18] FIG. 1 shows a loading plot. [Figure 19] FIG. 1 shows a score plot. [Figure 20] FIG. 1 shows a loading plot. [Figure 21] FIG. 1 shows a spectrogram of PD. [Figure 22] FIG. 10 is a diagram showing a spectrogram of non-PD. [Figure 23] FIG. 1 shows a score plot. [Figure 24] FIG. 1 shows a loading plot. [Figure 25] FIG. 1 shows a loading plot. DETAILED DESCRIPTION OF THE INVENTION

[0019] A data processing system according to an embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the data processing system according to the embodiment. The data processing system according to the embodiment includes an NMR apparatus 10 and a data processing device 12.

[0020] The NMR instrument 10 irradiates a sample placed in a static magnetic field with a radio frequency signal and detects the radio frequency signal emitted from the sample. In this embodiment, the NMR instrument 10 performs NMR measurements on each of a plurality of different samples, thereby detecting FID signals for each of the samples. The FID signals represent the time change in amplitude of the observed signal.

[0021] The sample is a mixture, for example, a biological sample such as serum, or a mixed solution sample containing a polymer compound, etc. Of course, mixtures other than these may also be used as the sample.

[0022] The data processing device 12 receives the FID signals detected by the NMR device 10 and processes the FID signals. The data processing device 12 processes the FID signals of each of the multiple samples. The data processing device 12 may be included in the NMR device 10, may be a device physically separate from the NMR device 10, or may be composed of multiple physically separated devices.

[0023] For example, the data processing device 12 includes a reception unit 14, a frequency analysis unit 16, a multivariate analysis unit 18, a distribution generation unit 20, a plot generation unit 22, a memory unit 24, a display control unit 26, a display unit 28, an operation unit 30, and an identification unit 31.

[0024] The receiving unit 14 receives the FID signal of each sample detected by the NMR instrument 10. For example, the receiving unit 14 may receive the FID signal of each sample from the NMR instrument 10 by using wired or wireless communication. As another example, if the FID signal of each sample is stored in a storage device such as a memory or a hard disk and input to the data processing device 12 via the storage device, the receiving unit 14 may receive the input FID signal.

[0025] The frequency analysis unit 16 generates spectrograms for each of the multiple samples by repeating time-frequency analysis on the FID signals of each of the multiple samples. For example, a short-time Fourier transform (STFT) is used as the time-frequency analysis. A spectrogram is expressed by time, frequency, and signal intensity. The frequency analysis unit 16 generates a spectrogram by passing a composite signal such as a time-frequency signal through a window function to calculate a frequency spectrum. For example, a spectrogram is defined by a time axis and a frequency axis. As the first coordinate system A spectrogram is an image that represents the strength (amplitude) of a signal component on a two-dimensional graph using color or brightness. A spectrogram is a two-dimensional data format defined by time and frequency.

[0026] The spectrogram of each sample may be generated by the NMR device 10. In this case, the receiving unit 14 receives the spectrogram of each sample.

[0027] The multivariate analysis unit 18 analyzes a plurality of samples. Multiple spectrogram (Dataset) By performing multivariate analysis on the sample, a score that reflects the characteristics of the sample's attributes is calculated, and the composite variable ( MultipleA multivariate analysis is performed to generate a set of variables for separating the spectrogram into individual sample attributes. For example, the multivariate analysis may be principal component analysis (PCA), part-lead Least Squares (PLS-DA), or soft independent modeling of class analogy (SIMCA: subspace method). For example, two composite variables are generated. Of course, the number of composite variables generated is not particularly limited.

[0028] For example, the multivariate analysis unit 18 executes principal component analysis, which is an example of multivariate analysis. , many In a multidimensional space consisting of coordinate axes of a number of variables, a small number of new coordinate axes are defined on which the differences (variance) between samples are most evident, and the coordinate values ​​of each sample on the new coordinate axes are calculated. The new coordinate axes are called "principal component axes," the variables along the principal component axes are called "principal components," and the coordinate values ​​on the principal component axes are called "scores." Each principal component is defined by a linear equation of a number of variables, and the coefficients of each variable term in the linear equation are called "loadings." or "loading value" The loading of each variable on a principal component represents the contribution (i.e., weight) of each variable to that principal component. The multivariate analysis unit 18 generates a score plot by plotting the scores of each of the multiple samples on the principal component space.

[0029] The distribution generation unit 20 generates a distribution as a result of the multivariate analysis by the multivariate analysis unit 18. as , corresponding to the index column, Generate a distribution of loadings for a particular principal component. each The index is determined by the time and frequency information contained in the spectrogram.

[0030] The distribution of loading values ​​is explained in detail. Each spectrogram is a two-dimensional matrix (e.g., the horizontal axis is time m indicates , the vertical axis is frequency n Show The number of samples to be analyzed is expressed as a matrix. S If so, the multivariate analysis unit 18 S spectrograms Each component The data (i.e., signal strength) is stored in vector format (s , m , n) and the data is expressed in vector format. In a set Here, as an example, the multivariate analysis unit 18 performs a principal component analysis. , for each sample Each principal component individual The loading value is number( m×n) is identified by, i.e., one-dimensional For example, the distribution generating unit 20 is expressed in the form of a vector. ,index (m×n) column Against Response do 、 Generate a distribution of loadings for the first principal component (PC-1). The distribution of loadings is specified by the exponent (m × n). Consists of loading values It is a one-dimensional data.

[0031] The plot generator 22 ,finger number column Against Response do 、 The distribution of loading values ​​of a specific principal component (one-dimensional data) is used to create a two-dimensional graph (defined by the time axis and frequency axis). Second coordinate system ) on the basis of color or brightness of a specific main component each A loading plot representing the loading values ​​is generated. The plot generation unit 22 generates the loading plot by plotting each loading value on a two-dimensional graph defined by a time axis and a frequency axis. For example, the plot generation unit 22 ,finger number column Against Response A loading plot is generated by coordinate transformation of the distribution of loading values ​​of a specific principal component. The loading plot is a two-dimensional data defined by time m and frequency n.

[0032] The storage unit 24 is realized by a storage device. For example, the storage unit 24 stores FID signals, spectrogram data, results of multivariate analysis (e.g., scores, loading values, etc.), data on the distribution of loading values, loading plots, etc.

[0033] The display control unit 26 controls the display of each piece of information. For example, the display control unit 26 causes the display unit 28 to display the FID signal, spectrogram, results of multivariate analysis, distribution of loading values, loading plots, etc.

[0034] In this embodiment, the display control unit 26 causes the spectrogram and the loading plot to be displayed side by side on the display unit 28. For example, when a spectrogram to be displayed is designated by the user, the display control unit 26 causes the designated spectrogram and the loading plot to be displayed side by side on the display unit 28.

[0035] The display unit 28 is a display such as a liquid crystal display, an EL display, etc. The operation unit 30 is an input device such as a keyboard, a mouse, input keys, and an operation panel.

[0036] The identification unit 31 identifies a region to be analyzed (hereinafter referred to as an "analysis region"). For example, the identification unit 31 identifies the analysis region by a time range and a frequency range on a loading plot. For example, when a user specifies a time range and a frequency range on a loading plot, the identification unit 31 identifies the region defined by the time range and the frequency range specified by the user as the analysis region. As another example, the identification unit 31 may identify a region defined by the time range and the frequency range, in which the loading value falls within a specific range, as the analysis region. The specific range is, for example, a range in which the loading value is equal to or greater than a threshold, or a range in which the loading value is equal to or greater than a lower limit and less than an upper limit. The specific range may be specified by the user or may be predetermined. For example, the identification unit 31 identifies a region on the loading plot in which the loading value is equal to or greater than a threshold as the analysis region. Note that the identification unit 31 does not have to be included in the data processing device 12.

[0037] The hardware configuration of the data processing device 12 will be described below with reference to Fig. 2. Fig. 2 is a block diagram showing the hardware configuration of the data processing device 12.

[0038] For example, the data processing device 12 includes a communication device 32, a UI (user interface) 34, a storage device 36, and a processor 38.

[0039] The communication device 32 includes one or more communication interfaces having a communication chip, a communication circuit, etc., and has a function of transmitting information to other devices and a function of receiving information from other devices. The communication device 32 may have a wireless communication function or a wired communication function.

[0040] The UI 34 is a user interface and includes a display and an input device. The display is a liquid crystal display, an EL display, or the like. The input device is a keyboard, a mouse, input keys, an operation panel, or the like. The display unit 28 and the operation unit 30 are realized by the UI 34. The UI 34 may be a UI such as a touch panel that combines a display and an input device.

[0041] The storage device 36 is a device that configures one or more storage areas for storing data. The storage device 36 is, for example, a hard disk drive (HDD), a solid state drive (SSD), various types of memory (e.g., RAM, DRAM, NVRAM, ROM, etc.), other storage devices (e.g., optical disks, etc.), or a combination thereof. The storage unit 24 is realized by the storage device 36.

[0042] The processor 38 controls the operation of each unit of the data processing device 12. The frequency analysis unit 16, the multivariate analysis unit 18, the distribution generation unit 20, the plot generation unit 22, the display control unit 26, and the identification unit 31 are realized by the processor 38. A storage device 36 may be used for realizing these units.

[0043] For example, the processor 38 may be configured by a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), or other programmable logic device, or an electronic circuit.

[0044] Each function of the data processing device 12 may be realized by cooperation between hardware resources and software resources. For example, each function is realized by a CPU constituting the processor 38 reading and executing a program stored in the storage device 36. The program is stored in the storage device 36 via a recording medium such as a CD or a DVD, or via a communication path such as a network. As another example, each function of the data processing device 12 may be realized by hardware resources such as electronic circuits.

[0045] The processing by the data processing device 12 will be described in detail below.

[0046] 3 shows an example of an FID signal. The FID signal 40 is an FID signal of a certain sample. The FID signal 40 is a signal detected by the NMR apparatus 10, and is a signal that represents a change in the amplitude of the observed signal over time. When the receiving unit 14 receives the FID signal 40, the frequency analysis unit 16 performs a short-time Fourier transform on the FID signal 40 to generate a spectrogram.

[0047] In this example, aqueous solutions of serum standards are used as samples. Specifically, three types of samples are used: HSA (human serum albumin solution), HDL (high density human plasma lipoproteins), and LDL (low density human plasma lipoproteins).

[0048] A spectrogram 42 of HSA is shown in Figure 4. A spectrogram 44 of HDL is shown in Figure 5. A spectrogram 46 of LDL is shown in Figure 6.

[0049] The HSA, HDL, and LDL are each measured by the NMR device 10, and the FID signals of the HSA, HDL, and LDL are detected. The receiving unit 14 receives the FID signals of the HSA, HDL, and LDL. The frequency analysis unit 16 performs a short-time Fourier transform on the FID signals of the HSA, HDL, and LDL, thereby generating spectrograms of the HSA, HDL, and LDL.

[0050] In the spectrograms 42, 44, and 46, the horizontal axis represents time and the vertical axis represents frequency. The signal strength is expressed by color or brightness. The spectrogram can represent the characteristics of the sample.

[0051] The multivariate analysis unit 18 generates a plurality of composite variables by performing multivariate analysis on the spectrograms 42, 44, and 46. Here, as an example, the multivariate analysis unit 18 performs principal component analysis to generate a first principal component (PC-1) and a second principal component (PC-2), and calculates the loading values ​​of each principal component. Distribution of and the scores of each sample for the two principal components are calculated.

[0052] Specifically, the number of samples S If so, the multivariate analysis unit 18 S of spectrograms each The data is stored in vector format (s , m , n) and the data is expressed in vector format. In a set As mentioned above, time m corresponds to the horizontal axis of the spectrogram, and frequency n corresponds to the vertical axis of the spectrogram. individual The loading value is exponential (m×n) Identified by It is expressed in vector format.

[0053] The multivariate analysis unit 18 generates a score plot by plotting the scores of each sample for the two principal components on a two-dimensional graph defined by the first principal component (PC-1) axis and the second principal component (PC-2) axis. An example of the score plot is shown in Figure 7.

[0054] In the score plot 48 shown in FIG. 7, the horizontal axis indicates the first principal component (PC-1), and the vertical axis indicates the second principal component (PC-2). As an example, the contribution rate of the first principal component (PC-1) is 21%, and the contribution rate of the second principal component (PC-2) is 11%. The contribution rate is an index that indicates the proportion of the principal component that accounts for the entire data. By referring to the contribution rate, the user can intuitively recognize the importance of the principal component. The black circle mark indicates the HSA score. The black triangle mark indicates the HDL score. The white triangle mark indicates the LDL score.

[0055] By referring to the score plot 48, the user can recognize the degree of separation of each sample. The shorter the distance between multiple plots on the score plot 48, the more similar the samples can be evaluated to be in characteristics. On the other hand, the longer the distance between multiple plots, the more different the samples can be evaluated to be in characteristics. In the example shown in Figure 7, the marks representing HSA are distributed on the score plot 48 separately from the marks representing HDL and LDL. Therefore, the score plot 48 can be evaluated as separating HSA from HDL and LDL. Furthermore, the larger the variance value of the scores in a principal component, the higher the degree to which that principal component contributes to the classification of the samples.

[0056] The distribution generation unit 20 generates the result of the principal component analysis by the multivariate analysis unit 18. As a finger number column Against Response vinegar Each owner A distribution of loading values ​​of the components is generated. Here, as an example, the distribution generation unit 20 generates a distribution of loading values ​​of the components with an index (m×n) column Against Response A distribution of loading values ​​of the first principal component (PC-1) is generated. Figure 8 shows this distribution 50. The horizontal axis represents the index (m × n), and the vertical axis represents the loading values ​​of the first principal component (PC-1).

[0057] The plot generating unit 22 plots the loading values ​​of the first principal component (PC-1) by color or brightness on a two-dimensional graph defined by the time axis and the frequency axis from the distribution 50 of the loading values. Distribution of The plot generator 22 generates a loading plot that expresses the exponent (m×n). The loading plot 52 is shown in FIG. 9. column Against Response A loading plot 52 is generated by converting the distribution 50 of the loading values ​​of the first principal component (PC-1) into a distribution on a two-dimensional graph.

[0058] In the loading plot 52, the horizontal axis represents time and the vertical axis represents frequency. The magnitude of the loading value of the first principal component (PC-1) is represented by color or brightness. The time information in the loading plot 52 is the time information in the spectrogram, and the frequency information in the loading plot 52 is the frequency information in the spectrogram. That is, the second coordinate system of the loading plot is the same as the first coordinate system of the spectrogram.

[0059] Coordinate conversion between two-dimensional data and one-dimensional data will be described with reference to FIG. 10. As described above, the spectrogram is two-dimensional data, while the data input to and output from the multivariate analysis unit 18 are one-dimensional data. Therefore, before performing multivariate analysis on the spectrogram, the two-dimensional spectrogram must be converted into one-dimensional data. Furthermore, to generate a two-dimensional loading plot, the distribution of loading values, which is one-dimensional data, must be converted into two-dimensional data.

[0060] Specific examples of these coordinate transformations are shown in Figure 10. Reference numeral 54 indicates two-dimensional data, and reference numeral 56 indicates one-dimensional data. As reference numeral 54 indicates, the vertical index of the two-dimensional data is defined as m, and the horizontal index is defined as n. As reference numeral 56 indicates, the index of the one-dimensional data is defined as k.

[0061] In two-dimensional data, (n, m) indicates a position in two-dimensional space or a value at that position. For example, (1, 1) indicates a position in two-dimensional space or a value at that position (1, 1). In the example shown in Figure 10, the total number of m is 3 and the total number of n is 2.

[0062] In one-dimensional data, k indicates a position in one-dimensional space or a value at that position. For example, (1) indicates a position in one-dimensional space or a value at that position (1).

[0063] The relationship between the indexes m and n and the index k is expressed by the following equation (1). k = m + (n - 1) × M (1) M is the total number of points in the vertical direction of the two-dimensional data. In the example shown in Fig. 10, M = 3. For example, when m = 1 and n = 1, k = 1.

[0064] When converting two-dimensional data to one-dimensional data, one-dimensional data is generated by referencing the position and value indicated by index k in one-dimensional space.When converting one-dimensional data to two-dimensional data, two-dimensional data is generated by referencing the position and value indicated by indexes m and n in two-dimensional space.

[0065] For example, when performing principal component analysis on a two-dimensional spectrogram, the multivariate analysis unit 18 performs the following calculation in accordance with the above transformation: each The spectrogram is converted into one-dimensional data (e.g., the vector form mentioned above). In the formula data to be represented), Number of samples The distribution generation unit 20 performs principal component analysis on the one-dimensional data. For each sample,A distribution of loading values ​​is generated based on the results of the principal component analysis (one-dimensional data). When generating a loading plot from the distribution of loading values, the plot generation unit 22 converts the one-dimensional data into a two-dimensional loading plot according to the above conversion.

[0066] The display control unit 26 displays the spectrogram and the loading plot side by side on the display unit 28. For example, when the user uses the operation unit 30 to specify the HSA spectrogram 42 as a comparison target and gives an instruction to display the comparison, the display control unit 26 displays the spectrogram 42 and the loading plot 52 side by side on the display unit 28, as shown in FIG.

[0067] As described above, the loading distribution 50 is converted into a loading plot 52. The loading plot 52 is a two-dimensional graph with the horizontal axis representing the time axis and the vertical axis representing the frequency axis, and represents the loading values, which are feature quantities, using color or brightness. That is, the horizontal axis of the loading plot 52 is the same time axis as the horizontal axis of the spectrogram 42, and the vertical axis of the loading plot 52 is the same frequency axis as the vertical axis of the spectrogram 42. Furthermore, on the spectrogram 42, the signal strength, which is a feature quantity, is represented using color or brightness, and on the loading plot 52, the loading values, which are feature quantities, are represented using color or brightness. In this way, the loading plot 52 is represented using the same dimensions and exponents as the spectrogram 42. This makes it easy for the user to compare the spectrogram 42 and the loading plot 52.

[0068] 11, the spectrogram 42 and the loading plot 52 are displayed side by side, but this display example is merely an example. The display control unit 26 may cause the display unit 28 to display the spectrogram 44 or the spectrogram 46 and the loading plot 52 side by side, instead of the spectrogram 42. For example, when the user specifies the spectrogram 44 or the spectrogram 46 as the comparison target, the display control unit 26 causes the display unit 28 to display the specified spectrogram and the loading plot 52 side by side.

[0069] The display control unit 26 may cause the display unit 28 to display a plurality of spectrograms and the loading plot 52 side by side. For example, when the spectrograms 44 and 46 are specified by the user, the display control unit 26 causes the display unit 28 to display the spectrograms 44 and 46 and the loading plot 52 side by side. This makes it easy for the user to compare a plurality of spectrograms and the loading plots.

[0070] The user can make various considerations based on the distribution of loading values ​​on the loading plot 52. For example, the user can make various considerations by referring to changes in loading values ​​with respect to the frequency axis or the time axis (for example, changes in color or brightness), differences in positions on the frequency axis or the time axis, etc.

[0071] Furthermore, a user can make various observations by comparing a spectrogram and a loading plot. For example, a user can identify a position on the frequency axis where the signal strength is high on the spectrogram and observe the type of distribution formed at that specified position on the frequency axis on the loading plot. Conversely, a user can identify a position (e.g., a position on the frequency axis) where the loading value is high or low on the loading plot and observe the type of distribution formed at that specified position (e.g., a position on the frequency axis) on the spectrogram. The same can be said for positions on the time axis. Additionally, a user can observe the type of distribution formed by the spectrogram and the loading plot in the same frequency range, or the type of distribution formed by the same time range. Spectrograms and loading plots are easily compared because they are expressed using the same dimensions, i.e., the frequency axis and the time axis, and the same exponents.

[0072] In a spectrogram, the position on the frequency axis where the signal intensity is high varies depending on the components and structure of the sample. For a given sample, the frequency range in which a distribution appears is known, and by referencing this distribution, the components contained in the sample can be identified. Furthermore, the decay rate (time change) of the NMR signal differs depending on the properties of the components, so the position of the distribution on the time axis also differs depending on the properties of the components contained in the sample.

[0073] By looking at the distribution of loading values ​​on the loading plot (e.g., changes in loading values ​​relative to the frequency axis, differences in position on the frequency axis, etc.) and comparing that distribution with the distribution on the spectrogram (e.g., changes in signal intensity relative to the frequency axis, differences in position on the frequency axis, etc.), users can consider differences in components and structures that affect differences in sample attributes.

[0074] For example, by comparing the attenuation characteristics of the NMR signals of each component shown in the spectrogram with the distribution of loading values ​​shown in the loading plot, users can evaluate the characteristics of each component of the sample, and analyze and consider factors (e.g., sample components and structure) that cause changes or differences in the position on the frequency axis on the loading plot.

[0075] For example, the position on the frequency axis where the signal intensity is high in the spectrogram is identified. Since it is known what kind of signal intensity distribution appears in which frequency region, the components contained in the sample can be identified by identifying the frequency position where the signal intensity is high. If the loading value at that position on the frequency axis in the loading plot is large or small, the component corresponding to that position on the frequency axis (i.e., the component identified in the spectrogram) is presumed to make a large contribution to separating each sample. In other words, the component is presumed to have a high contribution rate to the separation of each sample. By comparing the spectrogram and the loading plot in this way, the components that may contribute to the separation of each sample can be presumed.

[0076] The analysis results of the loading plot may be fed back to the time-frequency analysis by the frequency analysis unit 16. For example, if the distribution of loading values ​​with respect to positions on the frequency axis on the loading plot is distinctive (e.g., if the loading values ​​are large or small), it may be desirable to increase the frequency resolution and examine the distribution at that position in more detail. In this case, the frame length of the short-time Fourier transform is set to a value that increases the frequency resolution, and the short-time Fourier transform is performed on the FID signal. This generates a spectrogram with improved frequency resolution, and a score plot and a loading plot are generated based on that spectrogram. In this way, by examining the loading plot, it is possible to determine whether emphasis should be placed on frequency or time in the short-time Fourier transform.

[0077] The processing by the specification unit 31 will be described below with reference to Fig. 12. In Fig. 12, a loading plot 52 is shown.

[0078] For example, a loading plot 52 is displayed on the display unit 28. The user operates the operation unit 30 to specify a time range and a frequency range on the loading plot 52. The identification unit 31 identifies, as the analysis region, a region 58 defined by the time range and frequency range specified by the user.

[0079] Once the analysis region is identified, the frequency analysis unit 16 performs a short-time Fourier transform on the analysis region of the FID signal to generate a spectrogram of the analysis region. The multivariate analysis unit 18 performs multivariate analysis on the spectrogram of the analysis region. The distribution generation unit 20 generates a distribution of loading values ​​based on the results of the multivariate analysis, and the plot generation unit 22 generates a loading plot from the distribution. This generates a loading plot for the analysis region.

[0080] For example, the user refers to the loading plot 52 and specifies as the analysis region a region defined by a frequency range and a time range corresponding to components that are presumed to contribute to the separation of each sample. In other words, the user specifies the analysis region excluding a frequency range and a time range corresponding to components that are presumed not to contribute to the classification of each sample. In this way, the analysis can be performed excluding data that is not necessary for the analysis.

[0081] 12, the analysis region is specified by a rectangular region 58, but the shape of the analysis region may be specified by a region of a shape other than a rectangle (for example, a circle, an ellipse, or a region of any shape). For example, the user may specify the analysis region along the distribution of loading values.

[0082] The identification unit 31 may identify an area where the loading value falls within a specific range (for example, a range where the loading value is equal to or greater than a threshold, a range where the loading value is less than a threshold, or a range where the loading value is equal to or greater than a lower limit and less than an upper limit) as the analysis area. For example, in the example shown in Fig. 12, the identification unit 31 may identify the analysis area along the distribution of loading values ​​equal to or greater than a threshold.

[0083] Each example will be described below.

[0084] Example 1 The samples in Example 1 are HSA, HDL, and LDL. The purpose of Example 1 is to perform NMR measurements of albumin and lipoprotein in human serum and visualize the differences in the time-frequency characteristics of the NMR signals of these samples.

[0085] Details of HSA, HDL and LDL are shown below. HSA (human serum albumin solution) Manufacturer: NMIJ Lot:144 Material number:NMIJ CRM 6202-a ·HDL(Lipoproteins High Density, Human Plasma) Manufacturer: Calbiochem Batch number:3816722 Material number: 437641-10MG ·LDL(Lipoproteins Low Density, Human Plasma) Manufacturer: Calbiochem Batch number:3883470&3912892 Material number: 437644-10MG

[0086] (1) Equipment configuration A JNM-ECZ400R manufactured by JEOL was used as the NMR instrument 10. DELTA software (version 5.3.2 (JEOL Ltd.)) was used as the software for data processing and control of the NMR instrument 10. An application program (hereinafter referred to as "STFT tool") developed for MATLAB (registered trademark) (The MathWorks, Inc.) and Unscrambler X version 11 (manufactured by Camo Software) were installed in the data processing device 12.

[0087] (2) Sample preparation and measurement The three serum standard samples (100 μL) described above were mixed with 500 μL of deuterated water (heavy water, Sigma-Aldrich) and injected into a 5 mm outer diameter NMR sample tube. The probe temperature of the NMR instrument 10 was set to 30°C. 1H-NMR measurements were performed on these three samples to detect FID signals. The resonance frequency of the signal derived from the light water present in the sample was set as the center frequency of observation. The signal derived from the light water was suppressed by using the DANTE (Delays Alternating with Nutation for Tailored Excitation) pulse method, an example of a pulse program. This facilitated subsequent analysis and resulted in high-quality results. Depending on the purpose, multiple tubes of the same sample or multiple tubes of samples with different dilution concentrations may be prepared.

[0088] (3) Analysis (3-1) Time-frequency analysis The STFT tool was launched, and (2) the FID signals of all samples obtained in the measurement were loaded. The sampling frequency and the number of data points were entered, and time-frequency analysis was performed. To shorten the analysis time, the frequency range and time range to be analyzed may be narrowed to the frequency range and time range in which the signal was detected. Spectrogram 42 shown in Figure 4 is an example of a spectrogram of HSA. Spectrogram 44 shown in Figure 5 is an example of a spectrogram of HDL. Spectrogram 46 shown in Figure 6 is an example of a spectrogram of LDL.

[0089] (3-2) Multivariate analysis 1 Unscrambler X was launched and all data obtained in (3-1) time-frequency analysis was loaded. The data file name, labels for later plot display, and labels for grouping processing were entered, and then principal component analysis (PCA) was performed. The generated score plots were used to confirm whether each sample was well separated. First, a score plot defined by the axis with the largest variance (first principal component axis = PC-1) and the axis with the second largest variance (second principal component axis = PC-2) was confirmed. If sufficient separation was not confirmed in the score plot, other plots such as PC-3 and PC-4 were confirmed. The loading values ​​of the obtained plots were confirmed and saved. The score plot 48 shown in Figure 7 is an example of a score plot related to Example 1.

[0090] (3-3) Loading plot 1 The STFT tool was launched, one data set subjected to time-frequency analysis was opened, and the loading values ​​(PC-1 loading values, PC-2 loading values) saved in (3-2) Multivariate Analysis 1 were loaded. The loading values ​​of the principal components from PC-3 onward were loaded and confirmed while checking the quality of separation and contribution rates. A loading plot was generated and displayed using the loaded loading values. The frequency information and time information in the loading plot were obtained from the spectrogram. The regions in the loading plot where features appeared were confirmed. The loading plot 52 shown in Figure 9 is an example of a loading plot related to Example 1. The frequency information of the portion where the feature appears can be used to determine which component in the sample is related to the feature, and the time information can be used to determine the nature of the feature.

[0091] (3-4) Multivariate analysis 2 In multivariate analysis 2, partial least squares discriminant analysis (PLS-DA) was performed instead of principal component analysis (PCA). Unscrambler X was launched, and partial least squares discriminant analysis (PLS-DA) was performed using the same procedure as in (3-2) multivariate analysis 1. Figure 13 shows a score plot 60 generated by partial least squares discriminant analysis (PLS-DA).

[0092] (3-5) Loading plot 2 The STFT tool was launched, and one of the data sets subjected to time-frequency analysis was opened using the same procedure as in (3-3) Loading Plot 1. The loading values ​​(PC-1 loading values, PC-2 loading values) saved in (3-4) Multivariate Analysis 2 were then loaded. A loading plot was generated and displayed using the loaded loading values. The regions in the loading plot where features appeared were confirmed. Figure 14 shows an example of such a loading plot, loading plot 62. From the frequency information of the portion where the feature appears, it is possible to determine which component in the sample is related to the feature, and from the time information, it is possible to determine the nature of the feature.

[0093] In Example 1, each sample could be clearly separated on the score plot. Furthermore, differences could be detected in the frequency positions attributable to each component of HSA, HDL, and LDL. In other words, it was demonstrated that the loading plot is effective as information for analyzing which component in the sample is responsible for the separation on the score plot.

[0094] Example 2 The purpose of Example 2 is to perform serum mode analysis on diabetic model mice (BKS. Cg db / db) and healthy mice (Jcl:ICR) and to determine the onset of arteriosclerosis.

[0095] Background and Objectives Diabetes mellitus causes atherosclerosis due to persistent hyperglycemia and subsequently leads to various cardiovascular events. Clinically, detailed examination of atherosclerotic lesions requires additional testing in addition to blood tests. It would be clinically useful if atherosclerotic lesions could be assessed using blood samples. However, diabetes and the subsequent atherosclerotic lesions involve complex molecular biological processes, and it is not easy to detect all of these substances. Currently, no testing method has been established that can detect and assess atherosclerotic lesions from blood samples of diabetic patients. NMR analysis is an analytical method that can evaluate the physical and chemical properties of serum. The inventors of this application hypothesized that this analytical method could be used to distinguish blood conditions associated with the progression of atherosclerotic lesions in diabetes from healthy conditions. Based on this hypothesis, they analyzed the differences in serum of diabetic model mice (BKS.Cg db / db) and healthy mice (Jcl:ICR). The diabetic model mice (BKS.Cg db / db) are a strain of mice that exhibit hyperglycemic pathology early on. Serum and carotid arteries were collected from diabetic model mice (BKS. Cg db / db) and healthy mice (Jcl:ICR) at 18 weeks of age. The serum samples were analyzed using NMR, and carotid artery samples were subjected to pathological examination. By comparing the results of both, we investigated whether NMR analysis of serum samples could be used for early detection and progression assessment of pathologies related to atherosclerotic lesions.

[0096] · Laboratory animal care Diabetic model mice (BKS. Cg db / db) and healthy mice (Jcl:ICR) (both purchased from CLEA Japan) were housed in a clean room in the experimental animal care room of Nippon Medical School. They were fed a standard solid diet (MF, manufactured by Oriental Yeast Co., Ltd.) and allowed free access to food and water.

[0097] -Sample collection and pathological examination methods Mice were anesthetized with isoflurane to achieve sufficient analgesia, and then the chest was opened, cardiac blood (approximately 1 mL) was collected, and the mice were then euthanized. After cardiac blood collection, the carotid artery was harvested for histological examination using HE staining. Immediately after collection, the carotid artery was infiltrated and fixed in formalin solution, and then embedded and subjected to HE staining before microscopic observation.

[0098] The blood was centrifuged to separate the serum, and the samples for NMR measurement were stored at −80°C until NMR measurement.

[0099] Using serum samples with arteriosclerosis symptoms, the same operations and processes as in Example 1 were carried out to generate spectrograms, score plots, and loading plots for each sample.

[0100] Figure 15 shows a spectrogram 64 of BKS. Figure 16 shows a spectrogram 66 of Jcl. Figure 17 shows a score plot 68. The score plot 68 is a score plot generated by performing principal component analysis (PCA). Figure 18 shows a loading plot 70. The loading plot 70 is a loading plot generated based on the results of principal component analysis (PCA).

[0101] Another score plot 72 is shown in Figure 19. The score plot 72 is a score plot generated by partial least squares discriminant analysis (PLS-DA). A loading plot 74 is shown in Figure 20. The loading plot 74 is a loading plot generated based on the results of partial least squares discriminant analysis (PLS-DA).

[0102] Significant features (separation) were confirmed in the score plot of diseased samples. When factors that significantly influence the differences in features were confirmed using spectrograms, significant features were confirmed in the signals derived from HDL, LDL, glucose, etc. These results are consistent with the medical viewpoint of the causes of arteriosclerosis.

[0103] Pathology test results Pathological images of the carotid arteries were observed. In healthy mice (Jcl:ICR), mild diffuse intimal thickening was observed in all individuals, but the presence of atherosclerotic plaques was not confirmed. In diabetic model mice (BKS.Cg db / db), the degree of intimal thickening was more severe than in 18-week-old healthy mice (Jcl:ICR), and the presence of atherosclerotic plaques was evident in more than half of the individuals, demonstrating that arteriosclerosis progresses rapidly at this age.

[0104] In Example 2, each sample could be clearly separated on the score plot. Furthermore, the differences in frequency positions resulting from each component, such as glucose, could be read from the loading plot. This is consistent with the medical viewpoint of the causes of arteriosclerosis. In other words, Example 2 also demonstrated that the loading plot is effective as information for analyzing which component in the sample is responsible for the separation on the score plot.

[0105] Example 3 The purpose of Example 3 is to identify Parkinson's disease (PD) patients using serum samples.

[0106] ·the purpose Differentiation of Parkinson's disease (PD) from various diseases presenting with parkinsonism, such as multiple system atrophy and progressive supranuclear palsy, is not necessarily easy in the early stages of onset, and no disease-specific blood biomarkers for PD have yet been reported. DAT-SPECT and MIBG myocardial scintigraphy are examples of markers currently capable of distinguishing PD from non-PD patients with parkinsonism (non-PD) with the highest accuracy. We performed DAT-SPECT and MIBG myocardial scintigraphy on PD and non-PD patients, and investigated whether PD and non-PD could be distinguished using serum samples by using the NMR analysis method developed by the inventors.

[0107] ·subject With approval from the ethics committee, serum samples were collected from 10 patients who visited Nippon Medical School Chiba Hokuso Hospital between October 2020 and February 2022 and presented with symptoms of tremor, bradykinesia, muscle rigidity, or impaired postural reflexes. A database was created from blood biochemistry data, head MRI, MIGB myocardial scintigraphy, and DAT-SPECT. NMR analysis was performed on the serum samples from the 10 patients who successfully completed diagnosis and testing, using the same procedure as in Example 1. Spectrograms, score plots, and loading plots were generated for each sample.

[0108] FIG. 21 shows a spectrogram 76 of PD. FIG. 22 shows a spectrogram 78 of non-PD. FIG. 23 shows a score plot 80. The score plot 80 is a score plot generated by partial least squares discriminant analysis (PLS-DA). FIG. 24 shows a loading plot 82. FIG. 25 shows a loading plot 84. The loading plots 82 and 84 are loading plots generated based on the results of partial least squares discriminant analysis (PLS-DA). The loading plot 84 displays contour lines for the loading values.

[0109] ·diagnosis A diagnosis of PD was made if the patient met the International Parkinson and Movement Disorder Society (MDS) diagnostic criteria (2015) for "clinically definite PD" or "clinically probable PD." For MIBG myocardial scintigraphy, a cardiac longitudinal ratio of 2.2 or less in either the early or late phase was considered abnormal. For DAT-SPECT, visual assessment and quantitative assessment using SBR were performed to confirm the presence or absence of abnormalities.

[0110] In Example 3, PD patients (abnormal MIBG and abnormal DAT-SPECT) and non-PD patients (normal MIBG and normal DAT-SPECT) were analyzed using the PLS-DA method. As a result, each group formed a cluster on the score plot, and the two groups were distributed in clearly different regions. NMR analysis demonstrated the possibility of distinguishing PD patients from non-PD patients based on their serum.

[0111] As described above, according to this embodiment, it is possible to compare a spectrogram with a loading plot and correlate and consider the distribution on the spectrogram with the distribution on the loading plot. This allows for analysis of factors that cause differences in sample characteristics and attributes. As a result, for example, in the medical field, it is expected that preemptive medicine (i.e., medicine that predicts disease before a case appears and performs therapeutic intervention to prevent or delay the onset of the disease) will be realized. For example, it is expected that ultra-early diagnosis, determination of treatment guidelines, assessment of treatment effectiveness, prognosis prediction, etc. will be realized. Furthermore, according to this embodiment, it may become possible to search for cases without using known attribute identification images. [Explanation of symbols]

[0112] 10 NMR device, 12 data processing device, 14 reception unit, 16 frequency analysis unit, 18 multivariate analysis unit, 20 distribution generation unit, 22 plot generation unit, 26 display control unit.

Claims

1. an acquisition means for acquiring a plurality of spectrograms on a first coordinate system defined by a time axis and a frequency axis, the spectrograms being generated by performing NMR measurements on each of a plurality of samples; a multivariate analysis means for performing multivariate analysis on a data set consisting of the plurality of spectrograms to generate principal components for separating the plurality of samples according to sample attributes; a distribution generating means for generating a distribution of loading values ​​corresponding to the principal components as a result of the multivariate analysis by the multivariate analysis means; a plot generating means for generating a loading plot representing each loading value on a second coordinate system defined by a time axis and a frequency axis from the distribution of the loading values; Including, the second coordinate system is the same as the first coordinate system; A data processing device characterized by:

2. 2. The data processing device according to claim 1, and a display control means for displaying the spectrogram of the designated sample and the loading plot side by side on a display. A data processing device characterized by:

3. 2. The data processing device according to claim 1, Further comprising a specifying means for specifying an analysis region on the loading plot by a time range and a frequency range, the multivariate analysis means, the distribution generation means, and the plot generation means each perform a process on the analysis region; A data processing device characterized by:

4. 4. The data processing device according to claim 3, When the time range and the frequency range are specified by a user on the loading plot, the specifying means specifies a region defined by the time range and the frequency range specified by the user as the analysis region. A data processing device characterized by:

5. 4. The data processing device according to claim 3, the specifying means specifies, as the analysis region, a region defined by a time range and a frequency range, in which a loading value falls within a specific range; A data processing device characterized by:

6. 2. The data processing device according to claim 1, the plot generation means generates the loading plot by performing coordinate transformation on the distribution of the loading values. A data processing device characterized by:

7. Computer, an acquisition means for acquiring a plurality of spectrograms on a first coordinate system defined by a time axis and a frequency axis, the spectrograms being generated by performing NMR measurements on each of a plurality of samples; a multivariate analysis means for performing multivariate analysis on a data set consisting of the plurality of spectrograms to generate principal components for separating the plurality of samples according to sample attributes; a distribution generating means for generating a distribution of loading values ​​corresponding to the principal components as a result of the multivariate analysis by the multivariate analysis means; a plot generating means for generating a loading plot representing each loading value on a second coordinate system defined by a time axis and a frequency axis from the distribution of the loading values; A program that functions as the second coordinate system is the same as the first coordinate system; A program characterized by:

8. A first step executed by a computer includes acquiring a plurality of spectrograms on a first coordinate system defined by a time axis and a frequency axis, the spectrograms being generated by performing NMR measurements on each of a plurality of samples; a second step executed by the computer, performing multivariate analysis on the data set consisting of the plurality of spectrograms to generate principal components for separating the plurality of samples by sample attributes; a third step executed by the computer, which generates a distribution of loading values ​​of the principal components as a result of the multivariate analysis performed in the second step; a fourth step executed by the computer, generating a loading plot from the distribution of the loading values, the loading plot representing each loading value on a second coordinate system defined by a time axis and a frequency axis; Including, the second coordinate system is the same as the first coordinate system; A characteristic evaluation method characterized by:

Images