Biological information acquisition device, processing device, and computer program

Abstract

To enhance interpretation properties of a result of processing biological information acquired from a subject.SOLUTION: A reception device 11 receives waveform data WD corresponding to a measurement waveform of an electrocardiogram of a subject 30 from an electrode 20, and acquires values PV of a plurality of feature parameters associated with the measurement waveform on the basis of the waveform data WD. A processing device 12 acquires an inference result RS on at least one of a plurality of classes by which the waveform data WD is classified by inputting the values PV of a plurality of feature parameters to a machine-learned model 123, and specifies a degree of importance of each of the plurality of feature parameters to the inference result RS. An output device 13 outputs an index indicating a name of at least one of the plurality of feature parameters and the degree of importance specified on at least one of the plurality of feature parameters.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a biological information acquisition device that acquires biological information of a subject. The present invention also relates to a processing device that processes biological information of a subject and a computer program that can be executed by a processing unit mounted on the processing device.

Background Art

[0002] Patent Document 1 discloses an apparatus for acquiring a pulse wave, which is an example of biological information of a subject. When it is determined that noise equal to or higher than a predetermined level is mixed in the measured waveform of the pulse wave, a notification is given to the user.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] An object of the present invention is to improve the interpretability of the processing result of biological information acquired from a subject.

Means for Solving the Problems

[0005] One aspect for achieving the above object is a biological information acquisition device that acquires biological information of a subject, a reception device that receives waveform data corresponding to a measurement waveform of the biological information from a sensor and acquires values of a plurality of characteristic parameters associated with the measurement waveform based on the waveform data; a processing device that inputs the values of the plurality of characteristic parameters into a machine-learned model to obtain an inference result for at least one of a plurality of classes into which the waveform data is classified and specifies the importance of each of the plurality of characteristic parameters with respect to the inference result; An output device that outputs an index indicating at least one name of the plurality of feature parameters and the importance specified for the at least one feature parameter is provided.

[0006] One aspect for achieving the above object is a processing device that processes biological information of a subject, an interface that receives values of a plurality of feature parameters obtained based on waveform data corresponding to a measurement waveform of the biological information and associated with the measurement waveform, a processor that inputs the values of the plurality of feature parameters into a machine-learned model to obtain an inference result for at least one of a plurality of classes into which the waveform data is classified, specifies the importance of each of the plurality of feature parameters with respect to the inference result, and outputs an index indicating at least one name of the plurality of feature parameters and the importance specified for the at least one feature parameter to an output device is provided.

[0007] One aspect for achieving the above object is a computer program executable by a processor mounted on a processing device that processes biological information of a subject, which, when executed, causes the processing device to receive values of a plurality of feature parameters obtained based on waveform data corresponding to a measurement waveform of the biological information, input the values of the plurality of feature parameters into a machine-learned model to obtain an inference result for at least one of a plurality of classes into which the waveform data is classified, specify the importance of each of the plurality of feature parameters with respect to the inference result, and output an index indicating at least one name of the plurality of feature parameters and the importance specified for the at least one feature parameter to an output device.

[0008] When unstructured data such as waveform data corresponding to a measurement waveform of biological information is the target of inference processing, it is generally difficult to visualize the basis of the inference. However, according to the configuration according to each of the above aspects, by using a visualization method for feature parameters that can be handled as structured data, it is possible to make the user recognize through an index which feature parameters in the measurement waveform of biological information contributed to the inference. Therefore, the interpretability of the processing result of biological information can be improved.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Embodiments for Carrying Out the Invention

[0010] While referring to the accompanying drawings, examples of embodiments will be described in detail below.

[0011] FIG. 1 illustrates a functional configuration of an electrocardiograph 10 according to an embodiment. The electrocardiograph 10 is a device that acquires an electrocardiogram of a subject 30 through electrodes 20. The electrocardiogram is an example of biological information. The electrocardiograph 10 is an example of a biological information acquisition device. The electrodes 20 are an example of sensors.

[0012] The electrocardiograph 10 includes a receiving device 11. The receiving device 11 is configured to receive waveform data WD corresponding to the measured waveform of the electrocardiogram from the electrode 20. The waveform data WD may be in the form of analog data or digital data. When the waveform data WD is in the form of analog data, the receiving device 11 includes an appropriate conversion circuit including an A / D converter.

[0013] As illustrated in FIG. 2, the receiving device 11 is configured to obtain values PV of a plurality of characteristic parameters associated with the measured waveform of the electrocardiogram based on the waveform data WD. The characteristic parameter is a quantity representing the shape characteristic of the measured waveform. Examples of the characteristic parameters in the measured waveform of the electrocardiogram include the polarity and amplitude of the P wave, the QRS wave width, the polarity and amplitude of the T wave, the polarity and amplitude of the U wave, the width of the P wave, the width of the Q wave, the width of the R wave, the width of the S wave, the interval between the P wave and the Q wave, the interval between the Q wave and the T wave, the interval between the P waves, the interval between the R waves, and the like.

[0014] Specifically, based on a general electrocardiogram measurement algorithm, an average waveform is calculated for each lead based on the waveform data WD obtained within a predetermined time, and the above characteristic parameter values are obtained for the average waveform.

[0015] As illustrated in FIG. 1, the electrocardiograph 10 includes a processing device 12. The processing device 12 includes an input interface 121, a processor 122, a machine learning trained model 123, and an output interface 124.

[0016] The input interface 121 is configured to receive values PV of a plurality of characteristic parameters obtained by the receiving device 11.

[0017] As illustrated in FIG. 2, the processor 122 is configured to input values PV of a plurality of characteristic parameters received by the input interface 121 into the machine learning trained model 123.

[0018] The pre-trained machine learning model 123 is an algorithm that infers, as an output, the presence or absence of the risk that the subject 30 will develop paroxysmal atrial fibrillation (pAF) based on the values PV of a plurality of input parameters. The inference results of "risk of developing pAF" and "no risk of developing pAF" are examples of a plurality of classes into which the waveform data WD is classified.

[0019] The inference result RS of the pre-trained machine learning model 123 can take various forms. Only the classification result related to a single class of "risk of developing pAF" may be output, or the classification result related to a single class may be output together with the probability of "the risk of developing pAF is 84%". Alternatively, the classification results related to a plurality of classes of "the probability of developing pAF is 84% and the probability of not developing pAF is 16%" may be output together with the probabilities.

[0020] The pre-trained machine learning model 123 is an inference algorithm generated through machine learning. The generation of the pre-trained machine learning model 123 may be performed through machine learning using a neural network such as deep learning, or may be performed through other machine learning algorithms. Examples of other machine learning algorithms include decision trees, random forests, support vector machines, and the like.

[0021] The processor 122 is configured to identify the importance of each of the values PV of a plurality of feature parameters with respect to the inference result RS. In this example, the processor 122 is configured to calculate the SHAP (SHapley Additive exPlanations) value for each of the values PV of the plurality of feature parameters. That is, the SHAP value is an example of importance.

[0022] The SHAP value is obtained by calculating how the prediction result changes due to the presence of a certain feature quantity for all combinations of feature quantities that do not include that quantity. If the prediction result changes significantly due to the presence of the value of a certain feature quantity, the SHAP value becomes large. In this case, it can be determined that the said feature quantity makes a large contribution to the prediction. That is, by calculating the SHAP value for each of a plurality of feature parameters, the degree of contribution (importance) of each feature parameter to the inference result RS can be specified. Since the method of calculating the SHAP value itself is well-known, detailed description is omitted.

[0023] The processor 122 is configured to generate an index data ID for outputting an index indicating at least one name of a plurality of feature parameters and the SHAP value specified for the at least one feature parameter, and output it from the output interface 124. The index data ID may be in the form of analog data or in the form of digital data. When the index data ID is in the form of analog data, the output interface 124 includes an appropriate conversion circuit including a D / A converter.

[0024] As illustrated in FIG. 1, the electrocardiograph 10 includes an output device 13. The output device 13 can be realized as a display that displays an index based on the index data ID output from the processing device 12. The display of the index is an example of the output of the index. The output mode of the index may be other visual presentations such as projection or printing of an image, or in addition to or instead of the visual presentation, an auditory presentation may be made.

[0025] FIG. 3 shows an example of an index output by the output device 13. In the index according to this example, the class "with risk of developing pAF" included in the inference result is shown together with its probability. In addition, in the index according to this example, the names of a plurality of feature parameters in the measured waveform of the electrocardiogram are shown together with the SHAP value specified for each feature parameter and the measured value of each feature parameter. In this example, the magnitude of the SHAP value corresponds to the length of a bar extending in the left-right direction.

[0026] In this example, it is shown that for the inference result of "the probability of developing pAF is 84%", the "negative P wave lower area in the V1 lead" in the measured electrocardiogram waveform contributed the most. On the other hand, for the inference result of "there is a possibility of not developing pAF", it is shown that the "R wave width in the V6 lead" in the measured electrocardiogram waveform contributed the most.

[0027] When unstructured data such as waveform data WD corresponding to the measured electrocardiogram waveform is the object of inference processing, it is generally difficult to visualize the basis of the inference. However, according to the above configuration, by using a visualization method for feature parameters that can be treated as structured data, it is possible to let the user recognize through an index which feature parameters in the measured electrocardiogram waveform contributed to the inference. Therefore, the interpretability of the electrocardiogram processing result can be improved.

[0028] Especially when the pre-trained machine learning model 123 is generated through machine learning using a neural network, it is fundamentally difficult to obtain a clear basis for the inference result. In the medical field, the ambiguity of the basis for judgment is generally avoided. Therefore, in such a case, the effect of enhancing the interpretability of the electrocardiogram processing result becomes more prominent.

[0029] In the index according to this example, a plurality of SHAP values specified for a plurality of feature parameters are shown in a state where the relative high and low of the importance can be compared. For example, for the inference result of "the probability of developing pAF is 84%", it is shown that the "negative P wave upper area in the V1 lead" in the measured electrocardiogram waveform has a higher importance than the "positive P wave amplitude in the V3 lead".

[0030] According to such a configuration, it is possible to let the user recognize through an index which feature parameters of the measured electrocardiogram waveform contributed more to the inference than which other feature parameters. Therefore, the interpretability of the electrocardiogram processing result can be further improved.

[0031] If it is possible to show the plurality of SHAP values specified for a plurality of feature parameters in a state where the relative levels of importance can be compared, the index output by the output device 13 can take a form as exemplified in FIG. 4. In this example, the names of the plurality of feature parameters for which importance has been specified are arranged vertically. The magnitude of the SHAP value specified for each feature parameter corresponds to the length of the bar arranged on the side of the name.

[0032] On the other hand, the index output by the output device 13 may be configured to indicate the importance of a single feature parameter. For example, the name of the feature parameter that contributed the most to the inference result may be shown. In this case, the visibility of the index can be enhanced, and information with relatively high importance can be made recognizable to the user.

[0033] For example, in the index shown in FIG. 3, the SHAP value specified for each feature parameter is shown together with the value of the feature parameter. For example, it is shown that the "positive P wave amplitude in V3 lead" is 85 mV and the "positive P wave amplitude in V1 lead" is 70 mV.

[0034] It is common for medical staff to recognize the values that each feature parameter can take. Therefore, according to the above configuration, the validity of the importance of a specific feature parameter for the inference result shown in the index can be examined while taking into account the value of the feature parameter.

[0035] However, the display of the value of at least one of the plurality of feature parameters can be appropriately omitted in view of the visibility of the index.

[0036] As exemplified in FIG. 5, the output device 13 can display an index indicating at least one name of a plurality of feature parameters and the SHAP value specified for the at least one feature parameter so as to overlap with the measured waveform WF of the electrocardiogram.

[0037] In this example, the average waveform of the electrocardiogram acquired by the reception device 11 is displayed as the measurement waveform WF. In addition, an index including a plurality of feature parameter names such as "positive P wave amplitude", "interval between P wave and Q wave", "ST elevation", and "T wave amplitude" and the importance specified for each feature parameter with respect to the estimation result of "there is a possibility of pAF occurring" is displayed so as to overlap the measurement waveform WF.

[0038] More specifically, an index indicating the importance specified for a certain feature parameter is displayed as a band overlapping the location in the measurement waveform WF where the feature parameter is relevant. The numerical value displayed so as to overlap the band represents the value of the feature parameter. For example, the index related to the "positive P wave amplitude" is displayed as a band overlapping the position of the P wave in the measurement waveform WF, together with "60 (mV)", which is the value of the "positive P wave amplitude". In this example, the magnitude of the SHAP value corresponds to the darkness of the color of the band.

[0039] According to such a configuration, it is possible to easily and intuitively grasp which feature parameters related to which part of the measurement waveform WF contributed to the inference result made for the electrocardiogram.

[0040] Note that when a plurality of feature parameters are associated with a specific location in the measurement waveform WF and a plurality of importance levels are specified for the plurality of feature parameters, the name of the feature parameter with the maximum importance and the index indicating the maximum importance level may be displayed.

[0041] For example, SHAP values are specified for each of the "positive P wave amplitude" and the "area under the positive P wave form" associated with the P wave portion of the measurement waveform WF. When the SHAP value of the "positive P wave amplitude" is higher, only the index corresponding to the SHAP value specified for the "positive P wave amplitude" is displayed so as to overlap the measurement waveform WF.

[0042] According to such a configuration, it is possible to suppress a decrease in visibility due to the display of a plurality of indices in a limited area in the measurement waveform WF.

[0043] Alternatively, as illustrated in FIG. 6, the output device 13 can display an index indicating at least one name of a plurality of feature parameters and the SHAP value specified for the at least one feature parameter so as not to overlap with the measured waveform WF of the electrocardiogram.

[0044] In this example, the measured waveform WF of the electrocardiogram corresponding to multiple heartbeats is displayed. However, similar to the example shown in FIG. 5, an average waveform may be displayed. Also, as shown in this example, an index indicating the importance specified using Grad-CAM (Gradient-weighted Class Activation Mapping) can be displayed so as to overlap with the measured waveform WF as necessary. In this example, a strip-shaped index is superimposed on the waveform portion determined to have a high importance for the inference result of "there is a possibility of pAF". The level of importance corresponds to the darkness of the color of the strip.

[0045] According to such a configuration, while referring to the measured waveform WF of the electrocardiogram, the validity of the output index can be examined. Also, other indexes indicating importance can be used in combination as necessary.

[0046] As illustrated in FIG. 7, the output device 13 can display an index in association with the atrium and the ventricle. The atrium and the ventricle are an example of a plurality of body parts of the subject.

[0047] Specifically, an index associated with the atrium is generated by adding the SHAP value specified for the feature parameter related to the P wave of the electrocardiogram. Similarly, an index associated with the ventricle is generated by adding the SHAP value specified for the feature parameter related to the QRST wave of the electrocardiogram. The addition may be performed for all feature parameters, or may be performed for a predetermined number of feature parameters in descending order of the SHAP value. Multiplication by a coefficient may be appropriately performed. In this example, the magnitude of the sum of the SHAP values corresponds to the length of the bar extending in the left-right direction.

[0048] As illustrated in FIG. 8, the output device 13 can display an index in association with each of the V1 to V6 leads. In the figure, the shape of the heart as viewed from below the human body is schematically shown, and the indices related to each of the V1 to V6 leads are arranged to correspond to the positions in the heart. That is, the positions where the V1 to V6 leads are acquired are an example of a plurality of body parts of the subject.

[0049] Specifically, by adding the SHAP value specified for the characteristic parameter related to the V1 lead, an index associated with the acquisition position of the V1 lead is generated. The same process is performed for each of the V2 to V6 leads. The addition may be performed for all the characteristic parameters, or may be performed for a predetermined number of characteristic parameters in descending order of the SHAP value. Multiplication by a coefficient may be appropriately performed. In this example, the magnitude of the sum of the SHAP values corresponds to the distance from the center of the figure simulating the heart. The greater the distance from the center, the higher the specified importance.

[0050] According to the above configuration, it is possible to easily and intuitively grasp which characteristic parameters of which part of the heart contributed to the inference result made for the electrocardiogram. In the example shown in FIG. 7, an analysis can be made that some abnormality in the atrium is suspected for the inference result of "there is a possibility of pAF occurring". In the example shown in FIG. 8, an analysis can be made that some abnormality in the side wall of the heart is suspected for the inference result of "there is a possibility of pAF occurring".

[0051] The processor 122 of the processing device 12 having the functions as described above can be realized by a general-purpose microprocessor that operates in cooperation with a general-purpose memory. Examples of the general-purpose microprocessor include a CPU, an MPU, and a GPU. Examples of the general-purpose memory include a ROM and a RAM. In this case, a computer program for executing the above-described processing can be stored in the ROM. The ROM is an example of a non-transitory computer-readable medium that stores a computer program. The general-purpose microprocessor designates at least a part of the program stored on the ROM and expands it onto the RAM, and executes the above-described processing in cooperation with the RAM. The computer program may be pre-installed in the general-purpose memory, or may be downloaded from an external server via a communication network and then installed in the general-purpose memory. In this case, the external server is an example of a non-transitory computer-readable medium that stores a computer program.

[0052] The processor 122 of the processing device 12 having the functions as described above may also be realized by an application-specific integrated circuit capable of executing the above computer program, such as a microcontroller, an ASIC, or an FPGA. In this case, the above computer program is pre-installed in a storage element included in the application-specific integrated circuit. The storage element is an example of a computer-readable medium that stores a computer program. The processor 122 of the processing device 12 having the functions as described above may also be realized by a combination of a general-purpose microprocessor and an application-specific integrated circuit.

[0053] The above embodiments are merely examples for facilitating the understanding of the present invention. The configurations according to the above embodiments can be appropriately changed and improved without departing from the gist of the present invention.

[0054] When generating the index output by the output device 13, an LIME (Local Interpretable Model-agnostic Explanation) algorithm may be used instead of the SHAP algorithm.

[0055] In the above-described embodiment, the output device 13 forms part of the electrocardiograph 10. However, if it is possible to receive the index data ID from the processing device 12 via wired communication or wireless communication, the output device 13 can be a device provided independently of the electrocardiograph 10.

[0056] In the above-described embodiment, part of the processing performed by the reception device 11 can be executed by the processor 122 of the processing device 12. For example, the process of obtaining the value PV of each characteristic parameter from the waveform data WD may be performed by the processor 122.

[0057] In the above-described embodiment, the processing device 12 infers the possibility of the subject 30 developing paroxysmal atrial fibrillation. However, in addition to or instead of paroxysmal atrial fibrillation, the possibility of developing other heart diseases may be the subject of the inference. Examples of other heart diseases include atrial premature beats, paroxysmal supraventricular tachycardia, ventricular premature beats, ventricular tachycardia, ventricular fibrillation, myocardial infarction, and the like.

[0058] The possibility of developing diseases other than heart diseases may also be the subject. An example of another disease is epileptic seizures. When the possibility of having an epileptic seizure is the subject of the inference, based on the electroencephalogram waveform obtained from the subject by an electroencephalograph, characteristic parameters related to alpha waves, beta waves, theta waves, etc. are obtained, and the importance of the characteristic parameters for the inference result is specified. That is, the electroencephalogram waveform is an example of biological information. The electroencephalograph is an example of a biological information acquisition device.

[0059] Another example of other diseases includes hypotension and hypertension. When the possibility of developing hypotension or hypertension is the subject of the inference, based on the blood pressure waveform obtained from the subject by a sphygmomanometer, characteristic parameters such as systolic peak time, diastolic time, and cardiac output are obtained, and the importance of the characteristic parameters for the inference result is specified. That is, the blood pressure waveform is an example of biological information. The sphygmomanometer is an example of a biological information acquisition device.

[0060] Another example of other diseases is apnea syndrome. When the possibility of developing apnea syndrome is the subject of inference, based on the respiratory waveform acquired from the subject by a respirometer, characteristic parameters such as exhalation time, inhalation time, and respiratory time are acquired, and the importance of the characteristic parameters for the inference result is specified. That is, the respiratory waveform is an example of biological information. The respirometer is an example of a biological information acquisition device.

[0061] As used herein, the expression "at least one of A and B" for two entities A and B means the case where only A is specified, the case where only B is specified, and the case where both A and B are specified. Each of the entities A and B may be singular or plural, unless otherwise specified.

[0062] As used herein, the expression "at least one of A, B, and C" for three entities A, B, and C means the case where only A is specified, the case where only B is specified, the case where only C is specified, the case where A and B are specified, the case where B and C are specified, the case where A and C are specified, and the case where all of A, B, and C are specified. Each of the entities A, B, and C may be singular or plural, unless otherwise specified. The same applies when there are four or more entities to be described.

Description of Reference Numerals

[0063] 10: Electrocardiograph, 11: Reception device, 12: Processing device, 121: Input interface, 122: Processor, 123: Trained machine learning model, 13: Output device, 20: Electrode, 30: Subject, ID: Index data, PV: Value of characteristic parameter, RS: Inference result, WD: Waveform data, WF: Measured waveform

Claims

1. A biological information acquisition device for acquiring biological information of a subject, comprising: a reception device that receives waveform data corresponding to a measurement waveform of the biological information from a sensor and acquires values of a plurality of characteristic parameters associated with the measurement waveform based on the waveform data; a processing device that inputs the values of the plurality of characteristic parameters into a machine-learned model to obtain an inference result for at least one of a plurality of classes into which the waveform data is classified, and specifies the importance of at least one of the plurality of characteristic parameters with respect to the inference result; an output device that displays an index indicating the importance so as to overlap the measurement waveform; and a biological information acquisition device.

2. The output device according to claim 1, wherein the output device displays at least one name of the plurality of characteristic parameters so as to overlap the measurement waveform.

3. The output device according to claim 1, wherein the output device displays the name of the characteristic parameter having the maximum importance among the plurality of characteristic parameters associated with a specific location in the measurement waveform and an index indicating the maximum importance at a position corresponding to the specific location.

4. The index according to claim 1, wherein the index includes values of the at least one characteristic parameter.

5. The machine-learned model according to claim 1, wherein the machine-learned model is generated by machine learning using a neural network.

6. A processing device for processing biological information of a subject, comprising: an interface that receives values of a plurality of characteristic parameters acquired based on waveform data corresponding to a measurement waveform of the biological information and associated with the measurement waveform; a processor that inputs the values of the plurality of characteristic parameters into a machine-learned model to obtain an inference result for at least one of a plurality of classes into which the waveform data is classified, specifies the importance of at least one of the plurality of characteristic parameters with respect to the inference result, and causes an output device to display an index indicating the importance so as to overlap the measurement waveform; and a processing device.

7. A computer program executable by a processor mounted on a processing device for processing biological information of a subject, wherein when executed, the processing device ​ ​ ​ ​ ​ ​ ​ Obtained based on waveform data corresponding to the measurement waveform of the biological information, and accepting values of a plurality of characteristic parameters associated with the measurement waveform, By inputting the values of the plurality of characteristic parameters into a machine-learned model, obtaining an inference result for at least one of a plurality of classes into which the waveform data is classified, Identifying the importance of at least one of the plurality of characteristic parameters with respect to the inference result, Causing an output device to display an index indicating the importance so as to overlap with the measurement waveform, A computer program.

Images