Specimen component content prediction device, specimen component content prediction method, and program

Abstract

To provide a specimen component content prediction device capable of highly accurately estimating the concentration of a specific component in a sample to be measured from spectrum data of the sample, a specimen component content prediction method, and a program.SOLUTION: A specimen component content prediction device comprises: an acquisition unit that acquires spectral data acquired by spectroscopic measurement by electromagnetic waves with respect to a sample to be measured; and a concentration estimation unit that estimates the concentration of a specific component in the sample to be measured by inputting a feature amount of spectral data of the sample to be measured acquired by the acquisition unit to a machine learning model obtained by learning, as teacher data, the concentration of the specific component in a sample and a feature amount of spectral data acquired by spectroscopic measurement with respect to the sample.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a specimen component content prediction device, a specimen component content prediction method, and a program. [Background technology]

[0002] Patent Document 1 describes a technology for determining the presence of a disease by having a neural network learn using data on the expression levels of miRNA (micro RiboNucleic Acid), and for using the neural network to extract miRNA that is a characteristic biomarker for the disease. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2018 / 079840 Summary of the Invention [Problem to be solved by the invention]

[0004] An object of the present disclosure is to provide a specimen component content prediction device, a specimen component content prediction method, and a program capable of estimating the concentration of a specific component in a sample with high accuracy from spectral data of the sample to be measured. [Means for solving the problem]

[0005] The specimen component content prediction device disclosed herein includes an acquisition unit that acquires spectral data acquired by spectroscopic measurement of a sample to be measured using electromagnetic waves, and a concentration estimation unit that estimates the concentration of a specific component in the sample to be measured by inputting feature amounts of the spectral data of the sample to be measured acquired by the acquisition unit into a machine learning model trained using the concentration of the specific component in the sample and feature amounts of the spectral data acquired by spectroscopic measurement of the sample as teacher data.

[0006] The specimen component content prediction method disclosed herein comprises the steps of acquiring spectral data obtained by spectroscopic measurement of a sample to be measured using electromagnetic waves, and estimating the concentration of the specific component in the sample to be measured by inputting the feature values ​​of the acquired spectral data of the sample to be measured into a machine learning model trained using the concentration of the specific component in the sample and the feature values ​​of the spectral data acquired by spectroscopic measurement of the sample as training data.

[0007] The program disclosed herein causes a computer to execute the steps of acquiring spectral data obtained by spectroscopic measurement of a sample to be measured using electromagnetic waves, and estimating the concentration of a specific component in the sample to be measured by inputting the features of the acquired spectral data of the sample to a machine learning model trained using the concentration of the specific component in the sample and the features of the spectral data acquired by spectroscopic measurement of the sample as training data. Effect of the Invention

[0008] According to the specimen component content prediction apparatus, specimen component content prediction method, and program of the present disclosure, the concentration of a specific component in a sample to be measured can be estimated with high accuracy from the spectral data of the sample. [Brief description of the drawings]

[0009] [Figure 1] 1 is a schematic configuration diagram of a spectroscopic measurement system according to an embodiment. [Diagram 2] FIG. 1 is a diagram for explaining an AI model. [Diagram 3] 1 is a flowchart for explaining the processing flow in an AI model. [Figure 4] 1 is a graph showing an example of spectrum data. [Diagram 5] 11 is a graph for explaining a sampling process of spectrum data. [Figure 6] 11 is a table showing an example of teacher data. [Figure 7] 1 is a graph showing the prediction accuracy of an AI model. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present disclosure will now be described with reference to the drawings. Fig. 1 is a schematic diagram of a spectroscopic measurement system 1 according to the present embodiment.

[0011] As shown in FIG. 1, the spectroscopic measurement system 1 includes a computer 10, a light emitting unit 20, an optical system 21, a prism 22, an optical system 23, a light receiving unit 24, and a detection unit 25.

[0012] The computer 10 functions as a control unit that controls the light emitter 20, the light receiver 24, and the detector 25 in the spectroscopic measurement system 1. The computer 10 also functions as a specimen component content prediction device in the technology of the present disclosure. The computer 10 will be described in detail later.

[0013] The light emitting unit 20 is composed of, for example, a laser diode, and emits an electromagnetic wave for measurement.

[0014] The optical system 21 is an optical system for making the measurement electromagnetic wave emitted from the light emitting unit 20 enter the prism 22, and is composed of one or more lenses.

[0015] The prism 22 is made of, for example, silicon that transmits the electromagnetic waves to be measured, and holds a sample 30 to be measured on its upper surface.

[0016] The optical system 23 is an optical system for making the measurement electromagnetic wave emitted from the prism 22 incident on the light receiving unit 24, and is composed of one or more lenses.

[0017] The light receiving unit 24 is composed of, for example, a photodetector, and receives the measurement electromagnetic wave incident thereon by the optical system 23 .

[0018] The detection unit 25 is composed of, for example, a lock-in amplifier, and detects the amount of light of the measurement electromagnetic wave received by the light receiving unit 24 .

[0019] When the electromagnetic wave for measurement is irradiated onto the interface between the prism 22 and the sample 30, attenuation of the light energy (absorption spectrum) occurs according to the characteristics of the sample 30.

[0020] In spectroscopic measurement system 1, a measurement electromagnetic wave emitted from light emitting unit 20 is incident on prism 22, and the measurement electromagnetic wave totally reflected at the interface between the upper surface of prism 22 and sample 30 is received by light receiving unit 24, and the amount of received light is detected by detection unit 25. The detection results of detection unit 25 acquired by such spectroscopic measurement are input to computer 10, and the concentration of a specific component in sample 30 is measured based on the spectrum data of the measurement electromagnetic wave received by computer 10. The operation of spectroscopic measurement system 1 is controlled by computer 10.

[0021] Computer 10, which functions as a control unit of spectroscopic measurement system 1 and a specimen component content prediction device, includes control unit 11, storage unit 12, display unit 13, operation unit 14, communication unit 15, and connection unit 16. Control unit 11, storage unit 12, display unit 13, operation unit 14, communication unit 15, and connection unit 16 are interconnected via bus 17.

[0022] The control unit 11 includes a CPU (Central Processing Unit) 11A, a ROM (Read Only Memory) 11B, and a RAM (Random Access Memory) 11C. The control unit 11 has a function of controlling each part of the computer 10 and a function of controlling an external device connected to a connection unit 16 of the computer 10.

[0023] For example, a hard disk drive (HDD), a solid state drive (SSD), a flash memory, etc. are used as the storage unit 12. A program 12A for performing a sample component content prediction process, etc., which will be described later, is stored in the storage unit 12. This program 12A may be stored in the ROM 12. The storage unit 12 may have an external memory or a memory that can be added later.

[0024] The program 12A may be pre-installed in the computer 10, for example. The program 12A may be realized by storing it in a non-volatile non-transient storage medium or distributing it via a network line and installing or upgrading it appropriately in the computer 10. Note that examples of the non-volatile non-transient storage medium include a CD-ROM (Compact Disc Read Only Memory), a magneto-optical disk, a HDD, a DVD-ROM (Digital Versatile Disc Read Only Memory), a flash memory, a memory card, and the like.

[0025] For example, a liquid crystal display (LCD) or an organic EL (Electro Luminescence) display is used as the display unit 13. The display unit 13 may be integrated with a touch panel.

[0026] The operation unit 14 is provided with devices for inputting operations, such as a keyboard and a mouse. A user communicates user instructions to the computer 10 through operations on the operation unit 14. The display unit 13 displays various information such as results of processing executed in response to instructions received from the user and notifications regarding the processing.

[0027] The communication unit 15 is connected to a network line such as the Internet, a local area network (LAN) and / or a wide area network (WAN), and is capable of communicating with external devices via the network line.

[0028] The connection unit 16 is an input / output port to which external devices such as the light-emitting unit 20, the light-receiving unit 24, and the detection unit 25 of the spectroscopic measurement system 1 are connected. The light-emitting unit 20, the light-receiving unit 24, and the detection unit 25 are controlled by the control unit 11 described above.

[0029] The computer 10 of this embodiment functions as a specimen component content prediction device using AI (Artificial Intelligence). The control unit 11 of this computer 10 has a function as an acquisition unit that acquires spectral data acquired by spectroscopic measurement of a sample to be measured using electromagnetic waves for measurement, and a function as a concentration estimation unit that estimates the concentration of a specific component in a sample to be measured by inputting the feature amount of the spectral data of the sample to be measured acquired by the acquisition unit into a machine learning model that is trained using the concentration of the specific component in the sample and the feature amount of the spectral data acquired by spectroscopic measurement of the sample as teacher data.

[0030] In this embodiment, terahertz waves are used as an example of the electromagnetic waves to be measured. Terahertz waves are electromagnetic waves with a frequency of about 1 THz (wavelength 300 μm), and are electromagnetic waves with a frequency of 100 GHz (wavelength 3 mm) or more and 30 THz (wavelength 10 μm) or less. In this embodiment, the electromagnetic waves to be measured have a spectral width of 0.5 THz to 7 THz. Here, the spectral width means the full width at half maximum in the spectrum of the electromagnetic waves to be measured.

[0031] The frequency of the electromagnetic wave for measurement also depends on the principle of generation of the terahertz wave. For example, when the terahertz wave is generated by wavelength conversion of the electromagnetic wave generated from the excitation light source, the frequency characteristic of the terahertz wave is determined by the wavelength component or pulse width of the electromagnetic wave generated from the excitation light source. In this embodiment, as an example, a laser light source having a pulse wave of 50 femtoseconds to 100 femtoseconds is used as the excitation light source, so that the terahertz wave has a spectral width of 0.5 THz to 7 THz that can be generated by this excitation light source, but is not limited thereto.

[0032] It should be noted that the "spectral width from 0.5 THz to 7 THz" refers to, for example, the spectral width of terahertz waves when a laser light source having a pulse wave of 50 femtoseconds to 100 femtoseconds is used as the excitation light source, and is not limited to the spectral width strictly from 0.5 THz to 7 THz, but includes at least 90% or more of the range from 0.5 THz to 7 THz, taking into account error, and also includes a total width of 7.15 THz or less, which is 110% of 6.5 THz.

[0033] In this embodiment, the concentration of glucose in the culture serum is estimated as an example of the concentration of a specific component in a sample.

[0034] Here, the AI ​​model of the computer 10 of this embodiment will be described with reference to FIG.

[0035] As shown in FIG. 2, in the learning part of the AI ​​model of computer 10, first, in the input layer, the glucose concentration in a sample and feature quantities of spectral data acquired by spectroscopic measurement of the sample are input, and in the learning layer, a machine learning model is generated that is trained using the data input in the input layer as training data.

[0036] Next, in the prediction part, the AI ​​model of computer 10 inputs features of the spectral data of the sample to be measured in the input layer, predicts the glucose content using the learning model generated in the learning part in the calculation layer, and outputs the glucose concentration in the sample to be measured as the predicted result of the glucose content in the output layer.

[0037] Next, a specific process flow in the AI ​​model of the computer 10 of this embodiment will be described with reference to the flowchart shown in FIG.

[0038] First, the learning part of the AI ​​model will be described. In step S11, the control unit 11 of the computer 10 performs spectroscopic measurement using terahertz waves on a sample having a known glucose concentration, and acquires spectral data.

[0039] Next, in step S12, the control unit 11 performs pre-processing on the spectrum data acquired in step S11.

[0040] Figure 4 is a graph showing an example of spectrum data, with the horizontal axis showing frequency (THz) and the vertical axis showing signal strength (unitless). The gray solid line shows the raw spectrum data, and the gray circles on the gray solid line are the raw spectrum data sampled every 0.01 THz. The black circles show the median values ​​of multiple signal strengths obtained in each interval every 0.1 THz, and the black solid line is a line connecting the medians every 0.1 THz.

[0041] When generating a machine learning model trained using training data, the more the training data is subdivided, the higher the prediction accuracy can be. On the other hand, the more the training data is subdivided, the more the processing load increases, and the longer it takes to generate a training model and make predictions using the training model.

[0042] In this embodiment, terahertz waves having a spectral width of 0.5 THz to 7 THz are used as the electromagnetic waves for measurement, and in this case, in order to improve the prediction accuracy and reduce the processing load at the same time, the signal strength is obtained every 0.1 THz from the spectrum data. However, since the acquisition width of this signal strength is greatly influenced by the performance of the computer, it is desirable to obtain it every frequency as low as possible at equally spaced frequencies of 0.1 THz or less, if the performance of the computer allows.

[0043] Furthermore, since the raw spectral data contains noise, in order to suppress the effects of noise, the control unit 11 samples the raw spectral data every 0.01 THz as preprocessing of the spectral data, and further samples the median as a representative value of the multiple signal intensities acquired in each interval every 0.1 THz, as shown in an enlarged view in FIG. 5 .

[0044] Since the raw spectral data contains noise, some data will have values ​​higher than the true measured values. Therefore, if the raw spectral data is sampled every 0.01 THz, and then the "average" is sampled as a representative value of the multiple signal intensities obtained in each interval every 0.1 THz, the contents of all the data will be reflected, which may result in values ​​higher than the true measured values.

[0045] In contrast, in the present embodiment, when the "median" is sampled as a representative value of multiple signal intensities acquired in each interval every 0.1 THz, the contents of the data that is highly noisy and superimposed are not reflected, and the median of multiple signal intensities acquired in each interval every 0.1 THz is acquired, so that a value close to the true measurement value can be obtained.

[0046] Next, in step S13, the control unit 11 extracts, as a feature amount, data on the signal intensity for every 0.1 THz from the spectrum data preprocessed in step S12.

[0047] Next, in step S14, the control unit 11 inputs the known glucose concentration in the sample and the feature quantities (signal intensity for each frequency) of the spectral data acquired by spectroscopic measurement of the sample as teacher data to the learning layer, as shown as an example in FIG. 6, to create a learning model.

[0048] Furthermore, the control unit 11 repeats the processes from step S11 to step S14 for a plurality of samples in which the known glucose concentrations are changed, and creates a learning model using a plurality of teacher data.

[0049] Next, the process of the prediction part in the AI ​​model will be described. In step S15, the control unit 11 of the computer 10 inputs the feature amount of the spectral data of the sample to be measured to the input layer, thereby estimating (predicting) the glucose concentration in the sample to be measured.

[0050] Here, the sample (sample with known glucose concentration) used to obtain the training data in the learning part is subjected to spectroscopic measurement in the prediction part, and the estimation result of the glucose concentration is verified when the feature values ​​of the spectral data of the measured sample are input into the input layer.

[0051] Figure 7 is a graph showing the most accurate results among the analysis results obtained using the AI ​​model described above, ranked by mean square deviation. The horizontal axis shows the predicted value (unitless), and the vertical axis shows the actual measured value (unitless).

[0052] In the results of this embodiment, the slope was 1.0217 and the coefficient of determination (square value) was 0.8521, indicating that the glucose concentration could be estimated with extremely high accuracy.

[0053] [Variations]

[0054] The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit and scope of the present invention.

[0055] For example, the measurement electromagnetic wave used in spectroscopic measurement is not limited to a terahertz wave, and other electromagnetic waves may be used.

[0056] Furthermore, the concentration of a specific component to be estimated in a sample is not limited to the glucose concentration in the culture serum, but may be in any form, such as the concentration of each component in breast milk, or the fructose concentration (sugar content) in fruit juice.

[0057] [Note] Preferred embodiments of the present disclosure will be described below.

[0058] [Appendix 1] an acquisition unit for acquiring spectral data acquired by spectroscopic measurement of a sample to be measured using electromagnetic waves; a concentration estimation unit that estimates a concentration of a specific component in a sample to be measured by inputting feature values ​​of the spectral data of the sample to be measured acquired by the acquisition unit into a machine learning model that has been trained using the concentration of the specific component in the sample and feature values ​​of the spectral data acquired by spectroscopic measurement of the sample as training data; A specimen component content prediction device comprising:

[0059] [Appendix 2] The electromagnetic wave is a terahertz wave. 2. A sample component content prediction device as described in appendix 1.

[0060] [Appendix 3] When the terahertz wave is generated by wavelength conversion of an electromagnetic wave generated from an excitation light source, The terahertz wave has a spectral width that can be generated by the wavelength components or pulse width of the electromagnetic wave generated from the excitation light source. 3. A sample component content prediction device as described in appendix 2.

[0061] [Appendix 4] The terahertz wave has a spectral width of 0.5 THz to 7 THz. 4. A sample component content prediction device according to claim 2 or 3.

[0062] [Appendix 5] The feature of the spectrum data is the signal strength at each equally spaced frequency below 0.1 THz. 5. A sample component content prediction device according to any one of claims 1 to 4.

[0063] [Appendix 6]

[0064] The feature amount of the spectrum data is data of the median value of a plurality of signal intensities acquired in each interval for each equally spaced frequency of 0.1 THz or less. 6. A sample component content prediction device as described in appendix 5.

[0065] [Appendix 7] The specific component is glucose. 7. A sample component content prediction device according to any one of claims 1 to 6.

[0066] [Appendix 8] acquiring spectral data obtained by spectroscopic measurement of a sample to be measured using electromagnetic waves; a step of estimating a concentration of a specific component in a sample to be measured by inputting the feature amount of the spectral data of the sample to be measured acquired by the acquisition unit into a machine learning model trained using the concentration of the specific component in the sample and the feature amount of the spectral data acquired by spectroscopic measurement of the sample as training data; A method for predicting the content of a sample component comprising:

[0067] [Appendix 9] acquiring spectral data obtained by spectroscopic measurement of a sample to be measured using electromagnetic waves; a step of estimating a concentration of a specific component in a sample to be measured by inputting the feature amount of the spectral data of the sample to be measured acquired by the acquisition unit into a machine learning model trained using the concentration of the specific component in the sample and the feature amount of the spectral data acquired by spectroscopic measurement of the sample as training data; A program for causing a computer to execute the following. [Explanation of symbols]

[0068] 1. Spectroscopic measurement system 10. Computers 11 Control section 11A CPU 11B ROM 11C RAM 12 Storage section 12A Program 13 Display section 14 Control section 15 Communications Department 16 Connection 17 Bus 20 Light emitting part 21 Optical system 22 Prism 23 Optical system 24 Light receiving part 25 Detection unit 30 Samples

Claims

1. an acquisition unit that acquires spectral data acquired by spectroscopic measurement of a sample to be measured using electromagnetic waves; a concentration estimation unit that estimates a concentration of a specific component in a sample to be measured by inputting feature values ​​of the spectral data of the sample to be measured acquired by the acquisition unit into a machine learning model that has been trained using the concentration of the specific component in the sample and feature values ​​of the spectral data acquired by spectroscopic measurement of the sample as training data; A specimen component content prediction device comprising:

2. The electromagnetic wave is a terahertz wave. The apparatus for predicting a content of a component in a specimen according to claim 1 .

3. When the terahertz wave is generated by wavelength conversion of an electromagnetic wave generated from an excitation light source, The terahertz wave has a spectral width that can be generated by the wavelength components or pulse width of the electromagnetic wave generated from the excitation light source. The apparatus for predicting a content of a component in a specimen according to claim 2 .

4. The terahertz wave has a spectral width of 0.5 THz to 7 THz. The apparatus for predicting a content of a component in a specimen according to claim 2 .

5. The characteristic quantities of the spectrum data are the signal strengths at equal intervals of frequencies below 0.1 THz. The apparatus for predicting a content of a component in a specimen according to claim 1 .

6. The feature amount of the spectrum data is data of the median value of a plurality of signal intensities acquired in each section for each equally spaced frequency of 0.1 THz or less. The apparatus for predicting a content of a component in a specimen according to claim 5 .

7. The specific component is glucose. The apparatus for predicting a content of a component in a specimen according to claim 1 .

8. acquiring spectral data obtained by spectroscopic measurement of a sample to be measured using electromagnetic waves; a step of estimating a concentration of a specific component in a sample to be measured by inputting the acquired feature values ​​of the spectral data of the sample to a machine learning model trained using the concentration of the specific component in the sample and the feature values ​​of the spectral data acquired by spectroscopic measurement of the sample as training data; A method for predicting the content of a sample component comprising:

9. acquiring spectral data obtained by spectroscopic measurement of a sample to be measured using electromagnetic waves; a step of estimating a concentration of a specific component in a sample to be measured by inputting the acquired feature values ​​of the spectral data of the sample to a machine learning model trained using the concentration of the specific component in the sample and the feature values ​​of the spectral data acquired by spectroscopic measurement of the sample as training data; A program for causing a computer to execute the following.

Images