Thickness calculation method, calculation device, and program

Abstract

To provide a technique that measures the thickness of a tissue in a living body with higher accuracy using ultrasonic reception signals.SOLUTION: In a thickness calculation method for calculating the thickness of a tissue in a living body by a computer, a processor of the computer acquires reception data based on a reception signal of an ultrasonic wave for a living body of a target, classifies the living body of the target into either a first attribute or a second attribute on the basis of the acquired reception data on the target, (i) when the living body of the target is classified into the first attribute, inputs the reception data on the target into a first learning model to calculate the thickness of the tissue, and (ii) when the living body of the target is classified into the second attribute, inputs the reception data on the target into a second learning model to calculate the thickness of the tissue. The first learning model has been learned using first sample data and a thickness label for a living body sample of the first attribute as teacher data, and the second learning model has been learned using second sample data and a thickness label for a living body sample of the second attribute as teacher data.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present disclosure relates to a thickness calculation method, a calculation device, and a program. [Background technology]

[0002] Patent Document 1 discloses measuring a fat layer based on a received signal obtained by transmitting an ultrasonic wave to a living body. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 4-231945 Summary of the Invention [Problem to be solved by the invention]

[0004] A received signal obtained by transmitting an ultrasonic wave to a living body is affected not only by the thickness of the tissue to be measured but also by other physical characteristics of the living body. A technology is desired that can measure the thickness of the living body's tissue with higher accuracy using the received signal, regardless of the physical characteristics of the living body. [Means for solving the problem]

[0005] According to a first embodiment of the present disclosure, there is provided a thickness calculation method for calculating the thickness of tissue in a living body by a computer. In this calculation method, the computer has one or more processors, and the processor executes the following steps: a data acquisition step of acquiring reception data for a living body of a target based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body and received by the ultrasonic probe; a biometric attribute classification step of classifying the living body of the target into one of a first attribute related to the physique of the living body and a second attribute related to the physique based on the target reception data, which is the reception data acquired in the data acquisition step; and a thickness calculation step of (i) inputting the target reception data into a first learning model that has been trained in advance when the living body of the target is classified into the first attribute in the biometric attribute classification step, and calculating the thickness of tissue at a predetermined site of the living body of the target, and (ii) inputting the target reception data into a second learning model that has been trained in advance when the living body of the target is classified into the second attribute in the biometric attribute classification step, and calculating the thickness of the tissue. The first learning model has been trained in advance using as training data first sample data, which is the received data for a biological sample having the first attribute, and a thickness label relating to the thickness of the tissue associated with the first sample data. The second learning model has been trained in advance using as training data second sample data, which is the received data for a biological sample having the second attribute, and the thickness label associated with the second sample data.

[0006] According to a second embodiment of the present disclosure, there is provided a calculation device for calculating a thickness of tissue in a living body. The calculation device includes a storage device that stores a first learning model that has been previously trained, a second learning model that has been previously trained, and one or more processors. The processor executes a data acquisition step of acquiring reception data based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body and received by the ultrasonic probe, a biometric attribute classification step of classifying the living body of the target into one of a first attribute related to the physique of the living body and a second attribute related to the physique based on the target reception data that is the reception data acquired in the data acquisition step, and a thickness calculation step of (i) inputting the target reception data into the first learning model when the living body of the target is classified into the first attribute in the biometric attribute classification step, and calculating a thickness of tissue at a predetermined site of the living body of the target, and (ii) inputting the target reception data into the second learning model when the living body of the target is classified into the second attribute in the biometric attribute classification step, and calculating a thickness of the tissue. The first learning model has been trained in advance using as training data first sample data, which is the received data for a biological sample having the first attribute, and a label related to the tissue thickness associated with the first sample data. The second learning model has been trained in advance using as training data second sample data, which is the received data for a biological sample having the second attribute, and the label associated with the second sample data.

[0007] According to a third aspect of the present disclosure, there is provided a program for causing a computer to realize a function of calculating a tissue thickness in a living body. This program causes the computer to realize a function of executing a data acquisition step of acquiring reception data based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body and received by the ultrasonic probe, a function of executing a biometric attribute classification step of classifying the living body of the target into one of a first attribute related to the physique of the living body and a second attribute related to the physique based on the target reception data, which is the reception data acquired in the data acquisition step, and a function of executing a thickness calculation step of (i) inputting the target reception data into a first learning model that has been trained in advance when the living body of the target is classified into the first attribute in the biometric attribute classification step, and calculating a tissue thickness of a predetermined part of the living body of the target, and (ii) inputting the target reception data into a second learning model that has been trained in advance when the living body of the target is classified into the second attribute in the biometric attribute classification step, and calculating a tissue thickness. The first learning model has been trained in advance using training data including first sample data, which is the received data for a biological sample having the first attribute, and a label related to the tissue thickness associated with the first sample data, and the second learning model has been trained in advance using training data including second sample data, which is the received data for a biological sample having the second attribute, and the label associated with the second sample data. [Brief description of the drawings]

[0008] [Figure 1] FIG. 2 is a diagram illustrating a schematic configuration of a calculation device. [Diagram 2] FIG. 2 is a diagram illustrating a measuring device. [Diagram 3] 11 is a flowchart of a first learning process. [Figure 4] FIG. 2 is a diagram illustrating learning data used for learning each learning model. [Diagram 5] 13 is a flowchart of a second learning process. [Figure 6] 13 is a flowchart of a third learning process. [Figure 7] 13 is a flowchart of a measurement process. [Figure 8] FIG. 11 is a diagram illustrating an example of a segmentation result. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] A. First embodiment: FIG. 1 is a diagram showing a schematic configuration of a calculation device 20 in the first embodiment. The calculation device 20 is a device that uses ultrasound to calculate the thickness of tissue in a target living body TO. The target living body TO is, for example, a person such as a user or a test subject who uses the calculation device 20. Hereinafter, the tissue that is the target of thickness measurement is also referred to as a target tissue. The target tissue may be any tissue in any predetermined body part, for example, tissue in the abdomen or thigh. The target tissue may be any type of tissue, for example, a muscle layer or a fat layer. The target tissue in this embodiment is a layer of skeletal muscle. Specifically, the target tissue is the rectus abdominis muscle.

[0010] The calculation device 20 includes a control device 100 and a measuring device 400 having an ultrasonic probe 410. The control device 100 includes one or more processors 110, a memory device 120, an interface circuit 130, and a display unit 150 connected to the interface circuit 130. The control device 100 is, for example, a personal computer. The memory device 120 includes, for example, a ROM and a RAM. The display unit 150 displays various information. The display unit 150 is, for example, a liquid crystal monitor. The control device 100 and the measuring device 400 can communicate data with each other via wire or wirelessly.

[0011] 2 is a diagram illustrating the measuring device 400. The measuring device 400 is configured as an ultrasonic transmitting / receiving device that transmits ultrasonic waves to a living body and receives ultrasonic waves reflected by the living body. The measuring device 400 includes an ultrasonic probe 410 having ultrasonic elements 81 that transmit and receive ultrasonic waves, and a control unit 460 that controls the ultrasonic probe 410 and executes signal processing. The ultrasonic probe 410 and the control unit 460 are electrically connected by wiring 430.

[0012] The ultrasonic probe 410 includes one or more ultrasonic elements 81. The ultrasonic element 81 includes an ultrasonic transmission element 811 and an ultrasonic reception element 812. In this embodiment, one ultrasonic element 81 has a function as a transmission element 811 and a function as a reception element 812. The transmission element 811 transmits ultrasonic waves toward the target tissue. The transmitted ultrasonic waves travel through the living body along the thickness direction of the target tissue, and are reflected in the living body TO of the target, becoming reflected waves, which are received by the reception element 812. When the reception element 812 receives the reflected waves, a reception signal is output from the reception element 812. When the ultrasonic probe 410 includes a plurality of ultrasonic elements 81, the ultrasonic elements 81 are arranged at equal intervals in a row on a flat base member, for example. The ultrasonic probe 410 is used in a state where the ultrasonic elements 81 are brought close to the body surface of the living body. Specifically, the ultrasonic probe 410 is used in a state where it is pressed against or fixed to the body surface of the living body so that the ultrasonic elements 81 are brought close to the body surface of the living body. Although two ultrasonic elements 81 are shown in FIG. 2, the number of ultrasonic elements 81 may be one, or three or more.

[0013] The control unit 460 generates reception data based on the reception signals output from each ultrasonic element 81. In this embodiment, the control unit 460 generates a so-called B-mode image as the reception data. In the B-mode image, the difference in the intensity of the reflected wave is expressed as the difference in brightness in the image. In the B-mode image, at least a part including the target tissue is expressed in the B-mode image. The reception signal for obtaining such a B-mode image is obtained, for example, by using a plurality of ultrasonic elements 81, or by scanning one or a plurality of ultrasonic elements 81 along the body surface in a direction intersecting the thickness direction. Hereinafter, the part including the target tissue is also referred to as the target part. Specifically, the target part in this embodiment includes a part from the muscle layer, which is the target tissue, to the body surface located in the vicinity of the muscle layer.

[0014] The control unit 460 includes a drive pulse generating circuit 465, a transmission circuit 466, a signal processing circuit 467, a reception circuit 468, a multiplexer 469, a microcomputer 470, and a communication unit 471. The drive pulse generating circuit 465 generates a pattern of a predetermined drive frequency and wave number when transmitting ultrasonic waves. The transmission circuit 466 outputs a transmission waveform of a predetermined drive voltage according to the generated pattern. Each of the above-mentioned transmission elements 811 transmits the transmission waveform output from the transmission circuit 466 as ultrasonic waves toward the measurement target site. The reception circuit 468 amplifies the reception signals received by each of the reception elements 812. The signal processing circuit 467 generates a so-called A-mode image by performing envelope processing, LOG compression processing, etc. on the reception signals amplified by the reception circuit 468. The microcomputer 470 is configured to sequentially switch the transmission and reception operations of each ultrasonic element 81 by the multiplexer 469. That is, each ultrasonic element 81 is configured to receive each reception signal in sequence.

[0015] Ultrasound is hardly reflected by tissues such as fat layers and muscle layers, but is more likely to be reflected by fascia located at the boundary between tissues. Fascia here means a thin connective tissue that encases and protects various tissues such as muscles, bones, internal organs, nerves, and blood vessels. By utilizing this property of ultrasound, the positions of the fascia on one side of the target tissue and the fascia on the other side in the thickness direction of the target tissue can be identified based on the received data, and the positions of the first end and the second end in the thickness direction of the target tissue can be identified. The distance in the thickness direction between the positions of the first end and the second end thus identified corresponds to the thickness of the target tissue.

[0016] In addition to the fascia, ultrasonic waves are more likely to be reflected by, for example, intramuscular fat, which is fat within the muscle layer, and cellulite within the fat layer, as compared to other parts of each tissue. As a result, in addition to the feature amount related to the position of the fascia, the received data may contain the feature amount related to the position of the intramuscular fat and cellulite as noise. When the position of the fascia is specified using the received data, such noise makes it easier to erroneously specify the position of the fascia. In addition, the reflection intensity of ultrasonic waves in a living body varies depending on, for example, the state or degree of intramuscular fat and cellulite. The state or degree of intramuscular fat and cellulite in such a living body correlates with physical characteristics such as the body size of the living body. Specifically, for example, in the case of a person who is lacking in exercise and tends to be obese, the amount of intramuscular fat and cellulite tends to be greater. In addition, the distribution of intramuscular fat and cellulite differs depending on, for example, the presence or absence of exercise habits. In other words, the received data may be affected by the physical characteristics of the target living body TO.

[0017] The storage device 120 shown in FIG. 1 stores the first learning model 121, the second learning model 122, the third learning model 123, the first learning data LD1, the second learning data LD2, the third learning data LD3, and the program 129. The processor 110 executes the program 129 stored in the storage device 120 to realize various functions including the functions of the measurement data acquisition unit 111, the biometric attribute classification unit 112, the tissue thickness calculation unit 113, and the learning execution unit 119. The measurement data acquisition unit 111, the biometric attribute classification unit 112, and the tissue thickness calculation unit 113 are each a functional unit used in the measurement process described later. The learning execution unit 119 is a functional unit used in the learning process described later. In the following, when the first learning model 121, the second learning model 122, and the third learning model 123 are not distinguished from each other, they are also simply referred to as learning models. The first learning model 121 and the second learning model 122 are learning models used in a thickness calculation step described later, and are also referred to as calculation models. The third learning model 123 is a learning model used in a biological attribute classification step described later, and is also referred to as a classification model. In this embodiment, each learning model is a convolutional neural network that has been trained by supervised learning. Hereinafter, the convolutional neural network is also simply referred to as CNN. The first learning data LD1 to the third learning data LD3 will be described in detail later.

[0018] The measurement data acquisition unit 111 executes a data acquisition step. The data acquisition step is a step of acquiring reception data output from the ultrasonic probe 410 of the measurement device 400. The measurement data acquisition unit 111 stores the reception data acquired in the data acquisition step in the storage device 120. Hereinafter, the reception data acquired in the data acquisition step is also referred to as target reception data. It can also be said that the target reception data is reception data used to measure the thickness of a target tissue for a target living body TO.

[0019] The biometric attribute classification unit 112 executes a biometric attribute classification step. The biometric attribute classification step is a step of classifying the target biometric TO into either a first attribute or a second attribute based on the target reception data acquired by the measurement data acquisition unit 111. The first attribute and the second attribute are attributes related to the physique of the living body, respectively. The first attribute in this embodiment is an attribute that represents a slimmer physique compared to the second attribute. Hereinafter, classifying the target biometric TO into either the first attribute or the second attribute is also simply referred to as "classifying the target biometric TO."

[0020] In this embodiment, the first attribute and the second attribute are classified according to the magnitude of a physique parameter, which is a parameter related to the physique of a living organism. The physique parameter is preferably a parameter representing at least one of fat thickness, body fat percentage, and BMI (Body Mass Index). The physique parameter in this embodiment is a parameter representing fat thickness. Specifically, this fat thickness is the thickness of a fat layer, which is a layer of subcutaneous fat located on the epidermis side of a muscle layer, which is a target tissue. That is, the fat thickness in this embodiment is the thickness of a fat layer located on the epidermis side of the rectus abdominis muscle. The first attribute in this embodiment is an attribute that the fat thickness is equal to or less than a predetermined reference thickness. Also, the second attribute is an attribute that the fat thickness is greater than the reference thickness.

[0021] In this embodiment, in the biometric attribute classification step, the biometric attribute classification unit 112 classifies the target biometric TO using the above-mentioned third learning model 123. As described later, the third learning model 123 is a learned model different from the first learning model 121 and the second learning model 122.

[0022] The tissue thickness calculation unit 113 executes a thickness calculation step of calculating the thickness of the target tissue. In the thickness calculation step, the tissue thickness calculation unit 113 executes either a first process or a second process according to the attribute of the target living body TO classified in the biometric attribute classification step. The first process is executed when the target living body TO is classified into the first attribute in the biometric attribute classification step. The first process is a process of inputting the target reception data into the first learning model 121 and calculating the thickness of the target tissue for the target living body TO. The second process is executed when the target living body TO is classified into the second attribute in the biometric attribute classification step. The second process is a process of inputting the target reception data into the second learning model 122 and calculating the thickness of the target tissue for the target living body TO.

[0023] 3 is a flowchart of the first learning process executed in this embodiment. The first learning process is a process for learning the first learning model 121. In this embodiment, the first learning process is executed by a user performing a predetermined operation on the control device 100. The first learning process is executed prior to a measurement process, which will be described later. Note that the first learning process and the second and third learning processes, which will be described later, each represent a method of generating a learning model.

[0024] In step S110, the learning execution unit 119 acquires the first learning data LD1. The first learning data LD1 is a data set for learning the first learning model 121. The first learning data LD1 includes first sample data, which is received data on a biological sample of a first attribute, and a first thickness label associated with the first sample data. Specifically, the first learning data LD1 includes a plurality of first sample data with first thickness labels as main data. In this specification, the main data means received data that accounts for more than 50% of the total number of received data included in the learning data. The proportion of the number of main data in the learning data for learning the calculation model is preferably 70% or more, more preferably 80% or more, and even more preferably 90% or more. The first learning data LD1 in this embodiment is composed of a plurality of first sample data and a first thickness label associated with each of the first sample data. That is, the proportion of the number of first sample data in the total number of received data included in the first learning data LD1 is 100%. In step S110, the first learning data LD1 may be acquired, for example, from the storage device 120, or from an external computer or recording medium.

[0025] Specifically, the first sample data is reception data based on a reception signal derived from a biological sample of a first attribute. The biological sample is preferably the same type of biological sample as the target biological TO. The biological sample in this embodiment is a human. The first sample data is reception data for the same tissue as the target tissue. Specifically, the first sample data in this embodiment is a B-mode image showing a target site including at least the rectus abdominis muscle. The biological sample may or may not include the same person as the target biological TO. The first thickness label is a label related to the thickness of the target tissue for the first sample data to which the first thickness label is attached. Hereinafter, the label related to the thickness of the target tissue is also referred to as a thickness label.

[0026] FIG. 4 is a diagram for explaining learning data used for learning each learning model. FIG. 4 shows an example of the first sample data SD1 included in the first learning data LD1 and an example of the first thickness label LB1 associated with the first sample data SD1. FIG. 4 also shows an example of the second sample data SD2 included in the second learning data LD2 described later and an example of the second thickness label LB2 associated with the second sample data SD2. The second thickness label LB2 is a thickness label LB, as described later. In the following, the first thickness label LB1 and the second thickness label LB2 are also simply referred to as "thickness label LB". In the following, when the first sample data SD1 and the second sample data SD2 are not particularly distinguished from each other, they are also simply referred to as sample data.

[0027] In FIG. 4, the upper side of each B-mode image as received data represents the body surface side in the thickness direction d1 of the body part including the target tissue. The lower side of each B-mode image represents the back side in the thickness direction d1 of the body part including the target tissue. The back side here means the side closer to the center of the body of the living body. In each B-mode image, from the top, a fat layer FL, a muscle layer ML, and an internal organ OR are shown. In addition, at the boundary between the fat layer FL and the muscle layer ML, a first fascia FS1, which is the fascia on the epidermis side of the target tissue, appears. At the boundary between the muscle layer ML and the internal organ OR, a second fascia FS2, which is the fascia on the back side of the target tissue, appears. As described above, the position of the first fascia FS1 represents the position of the first end Eg1 in the thickness direction d1 of the target tissue. The position of the second fascia FS2 represents the position of the second end Eg2 in the thickness direction d1 of the target tissue. Hereinafter, the direction perpendicular to the thickness direction d1 is also referred to as the vertical direction d2.

[0028] Each thickness label LB in this embodiment is a label indicating whether each region included in the B-mode image of each sample data associated with the thickness label LB is a region representing the fat layer FL, the muscle layer ML, or the viscera OR. Specifically, the thickness label LB is a label indicating whether each pixel in the B-mode image represents the fat layer FL, the muscle layer ML, or the viscera OR. Therefore, the thickness label LB is a label capable of identifying the boundary between the muscle layer ML, which is the target tissue, and other tissues, that is, the first end Eg1 and the second end Eg2. Also, as described above, the thickness of the target tissue can be calculated by identifying the first end Eg1 and the second end Eg2, so the thickness label LB is a label related to the thickness of the target tissue. Note that, in order to facilitate understanding of the technology, FIG. 4 shows the thickness label LB as an image in which the muscle layer ML region is masked. Specifically, FIG. 4 shows the thickness label LB as an image in which the tissue region other than the muscle layer ML is represented in black and the muscle layer ML region is represented in white.

[0029] In step S120 of FIG. 3, the learning execution unit 119 executes learning of the machine learning model by inputting the first learning data LD1 acquired in step S110 to the machine learning model. That is, in step S120, supervised learning is executed using the first learning data LD1 as teacher data. As a result, a first learning model 121 that has been trained using the first learning data LD1 as teacher data is generated. In this embodiment, when a B-mode image is input as received data, the first learning model 121 is generated as a model that executes three-class segmentation of the B-mode image. Specifically, the first learning model 121 outputs, as a result of segmentation, information indicating whether each pixel in the input B-mode image represents a fat layer FL, a muscle layer ML, or an internal organ OR. The learning execution unit 119 stores the generated first learning model 121 in the storage device 120.

[0030] 5 is a flowchart of the second learning process executed in this embodiment. The second learning process is a process for learning the second learning model 122. The second learning process is executed prior to the measurement process, similar to the first learning process.

[0031] In step S210, the learning execution unit 119 acquires the second learning data LD2. The second learning data LD2 is a data set for learning the second learning model 122. The second learning data LD2 includes the second sample data SD2, which is received data on a biological sample having a second attribute, and the second thickness label LB2 associated with the second sample data SD2. Specifically, the second learning data LD2 includes the second sample data SD2 with the second thickness labels LB2 as main data. In this embodiment, the second learning data LD2 is composed of the second sample data SD2 and the second thickness labels LB2 associated with each of the second sample data SD2. That is, the ratio of the number of the second sample data SD2 to the total number of received data included in the second learning data LD2 is 100%. Specifically, the second sample data SD2 is received data based on a received signal derived from a biological sample having a second attribute. The second thickness label LB2 is a label related to the thickness of the target tissue for the second sample data SD2 to which the second thickness label LB2 is attached. In step S210, the second learning data LD2 may be acquired, for example, from the storage device 120, or from an external computer or recording medium.

[0032] In step S220, the learning execution unit 119 executes learning of the machine learning model by inputting the second learning data LD2 acquired in step S210 to the machine learning model. That is, in step S220, supervised learning is executed using the second learning data LD2 as teacher data. As a result, a second learning model 122 that has been trained using the second learning data LD2 as teacher data is generated. In this embodiment, the second learning model 122 is generated as a model that executes three-class segmentation when received data is input, substantially similar to the first learning model 121. The learning execution unit 119 stores the generated second learning model 122 in the storage device 120.

[0033] 6 is a flowchart of the third learning process executed in this embodiment. The third learning process is a process for learning the third learning model 123. The third learning process is executed prior to the measurement process, similar to the first learning process and the second learning process.

[0034] In step S310, the learning execution unit 119 acquires the third learning data LD3. The third learning data LD3 is a data set for learning the third learning model 123. The third learning data LD3 includes the first sample data SD1 and a physique label associated with the first sample data SD1, and the second sample data SD2 and a physique label associated with the second sample data SD2. The physique label is a label related to a physique parameter. Specifically, the physique label in this embodiment is a label indicating whether each region included in the B-mode image is a region of the fat layer FL or a region of tissue other than the fat layer FL. For example, this physique label is a label indicating whether each pixel in the B-mode image represents the fat layer FL or a tissue other than the fat layer FL. Therefore, this physique label is a label that can identify the boundary between the fat layer FL and the muscle layer ML, that is, the first end Eg1. Identifying the first end Eg1 makes it possible to calculate the thickness of the fat layer FL. That is, the physique label in this embodiment is a label related to fat thickness.

[0035] It is possible to use the same sample data as the first sample data SD1 included in the first learning data LD1 and the second sample data SD2 included in the second sample data SD2 as the first sample data SD1 included in the first learning data LD1 and the second sample data SD2 included in the second sample data SD2, as the first sample data SD1 and the second sample data SD2 included in the third learning data LD3. That is, the same sample data may be used for learning the classification model and the calculation model, and only the labels attached to the sample data may be different. In this way, the effort of preparing sample data can be reduced, and teacher data for learning the classification model and the calculation model can be efficiently prepared.

[0036] In step S320, the learning execution unit 119 executes learning of the machine learning model by inputting the third learning data LD3 acquired in step S310 to the machine learning model. That is, in step S320, supervised learning is executed using the third learning data LD3 as teacher data. As a result, a third learning model 123 that has been trained using the third learning data LD3 as teacher data is generated. In this embodiment, the third learning model 123 is generated as a model that executes two-class segmentation when received data is input. Specifically, the third learning model 123 outputs, as a result of segmentation, information indicating whether each pixel in the input B-mode image represents a fat layer FL or a pixel other than the fat layer FL. The learning execution unit 119 stores the generated third learning model 123 in the storage device 120.

[0037] 7 is a flow chart of a measurement process for implementing the thickness measurement method in this embodiment. In this embodiment, the measurement process is executed by a user performing a predetermined operation on the control device 100. In step S410, the control device 100 controls the measurement device 400 to transmit ultrasonic waves from the ultrasonic probe 410 to the target living body TO, and to cause the ultrasonic probe 410 to receive ultrasonic waves reflected in the target living body TO. As a result, a reception signal is output from the ultrasonic probe 410, and reception data based on the reception signal is generated by the control unit 460.

[0038] In step S420, the measurement data acquiring unit 111 executes a data acquiring step. Specifically, in step S420, the measurement data acquiring unit 111 acquires the reception data generated in step S410. That is, in step S420, the measurement data acquiring unit 111 acquires the target reception data.

[0039] In steps S430 and S440, the biometric attribute classifying unit 112 executes a biometric attribute classifying step. In step S430, the biometric attribute classifying unit 112 first inputs the target received data acquired in step S410 to the third learning model 123. As a result, the third learning model 123 executes segmentation using the target received data, and identifies the position of the first end Eg1 of the target tissue. Next, the biometric attribute classifying unit 112 acquires fat thickness using the identified first end Eg1. The fat thickness is calculated, for example, based on the distance between the epidermis and the first end Eg1.

[0040] In step S440, the biometric attribute classification unit 112 classifies the target biometric TO into either a first attribute or a second attribute using the fat thickness calculated in step S430. Specifically, in step S440, the biometric attribute classification unit 112 classifies the target biometric TO into the first attribute when the fat thickness calculated in step S430 is equal to or smaller than a reference thickness, and classifies the target biometric TO into the second attribute when the fat thickness calculated in step S430 is greater than the reference thickness.

[0041] In step S450, the tissue thickness calculation unit 113 determines whether the target living body TO is classified into the first attribute. If it is determined in step S450 that the target living body TO is classified into the first attribute, the tissue thickness calculation unit 113 executes a first process in step S460. If it is determined in step S450 that the target living body TO is classified into the second attribute, the tissue thickness calculation unit 113 executes a second process in step S470.

[0042] In step S460, the tissue thickness calculation unit 113 first inputs the target received data to the first learning model 121. This target received data is target received data for a living body having a first attribute. As a result, the first learning model 121 executes segmentation using the target received data, and the positions of the first end Eg1 and the second end Eg2 of the target tissue are identified. Next, the tissue thickness calculation unit 113 calculates the thickness of the muscle layer ML, which is the target tissue, using the identified first end Eg1 and second end Eg2. Specifically, the thickness of the target tissue is calculated based on the distance between the first end Eg1 and the second end Eg2.

[0043] Step S470 is similar to step S460, except that the target received data is input to the second learning model 122 instead of the first learning model 121. That is, the tissue thickness calculation unit 113 inputs the target received data about a living body having a second attribute to the second learning model 122, and calculates the thickness of the muscle layer ML, which is the target tissue.

[0044] FIG. 8 is a diagram illustrating an example of a segmentation result by the first learning model 121 and the second learning model 122. The first result R1 is a result of segmenting the received data RD1 of the living body of the first attribute by the first learning model 121. The first result R1 corresponds to the segmentation result in step S460. The second result R2 is a result of segmenting the received data RD1 by the second learning model 122. The third result R3 is a result of segmenting the received data RD1 by the fourth learning model described later. The fourth result R4 is a result of segmenting the received data RD2 of the living body of the second attribute by the first learning model 121. The fifth result R5 is a result of segmenting the received data RD2 by the first learning model 121. The fifth result R5 corresponds to the segmentation result in step S470. The sixth result R6 is a result of segmenting the received data RD2 by the fourth learning model. FIG. 8 shows a schematic image in which the muscle layer ML area is masked as in the thickness label LB in FIG. 4 as a result of each segmentation, and the Dice coefficient. The Dice coefficient represents the degree of overlap between the muscle layer ML area in the segmentation result of the received data and the muscle layer ML area in the thickness label LB associated with the received data. The correct answer ML1 in FIG. 8 also shows the area that should be identified as the muscle layer ML when segmentation is performed on the received data RD1, similarly to the thickness label LB in FIG. 4. In a similar manner, the correct answer ML2 shows the area that should be identified as the muscle layer ML when segmentation is performed on the received data RD2.

[0045] The fourth learning model is a learning model different from the first learning model 121 to the third learning model 123. Specifically, the fourth learning model is a learning model that has been trained using the fourth learning data as teacher data. The fourth learning data is a data set including a plurality of first sample data SD1 and a first thickness label LB1 associated with each of the first sample data, and a plurality of second sample data SD2 and a second thickness label LB2 associated with each of the second sample data SD2. In the fourth learning data, the proportion of the first sample data SD1 in the sample data is smaller than that in the first learning model 121. In addition, in the fourth learning model, the proportion of the second sample data SD2 in the sample data is smaller than that in the second learning model 122.

[0046] As shown in FIG. 8, among the first result R1 to the third result R3, which are segmentation results for the received data of the living body having the first attribute, the Dice coefficient of the first result R1 was the highest. This is because the main data of the first learning data LD1 used for learning the first learning model 121 is the first sample data SD1. The Dice coefficient of the third result R3 is higher than the Dice coefficient of the second result R2. This is because the second learning data LD2 used for learning the second learning model 122 does not include the first sample data SD1, whereas the fourth learning data used for learning the fourth learning model includes the first sample data SD1. However, in learning using the fourth learning data, due to the second sample data SD2 included in the fourth learning data, the feature amount of the second sample data SD2, i.e., the feature amount caused by the second attribute different from the first attribute, is more likely to be reflected in learning than in learning using the first learning data LD1. In contrast, in learning using the first learning data LD1, the feature amount caused by the first attribute is more likely to be reflected in learning. As a result, the Dice coefficient of the first result R1 is higher than the Dice coefficient of the third result R3.

[0047] Among the fourth result R4 to the sixth result R6, which are segmentation results for the received data RD2, the fifth result R5 had the highest Dice coefficient. This is because the main data of the second learning data LD2 is the second sample data SD2, similarly to the first result R1 to the third result R3. The Dice coefficient of the sixth result R6 is higher than the Dice coefficient of the fourth result R4. This is because the second learning data LD2 does not include the second sample data SD2, whereas the fourth learning data includes the second sample data SD2. However, in the learning using the fourth learning data, the feature amount caused by the first attribute, which is different from the second attribute, is more likely to be reflected in the learning compared to the learning using the second learning data LD2, due to the first sample data SD1 included in the fourth learning data. In contrast, in the learning using the second learning data LD2, the feature amount caused by the second attribute is more likely to be reflected in the learning. As a result, the Dice coefficient of the fifth result R5 is higher than the Dice coefficient of the sixth result R6. In this way, in the thickness calculation step in this embodiment, by using a calculation learning model that has been trained using teacher data corresponding to the attributes of the target living body TO, the segmentation of the target received data can be performed more appropriately. Then, the thickness of the target tissue can be calculated using the result of the appropriately performed segmentation, so that the thickness of the target tissue can be calculated with higher accuracy.

[0048] In step S480, the control device 100 controls the display unit 150 to display the calculation result of the first process in step S460 or the second process in step S470 on the display unit 150. In step S480, the display unit 150 may display, for example, only the calculated thickness, or may display the segmentation result together with the thickness.

[0049] According to the thickness calculation method in the present embodiment described above, in the biological attribute classification step, the biological TO of the target is classified into either the first attribute or the second attribute based on the target reception data acquired in the data acquisition step. Then, in the thickness calculation step, the first process is executed when the biological TO of the target is classified into the first attribute, and the second process is executed when the biological TO of the target is classified into the second attribute, thereby calculating the thickness of the target tissue. In the first process, the first learning model 121 trained by the first learning data LD1 corresponding to the first attribute is used, and in the second process, the second learning model 122 trained by the second learning data LD2 corresponding to the second attribute is used. In this way, the thickness of the target tissue is calculated using the target reception data and the learning model trained by the teacher data corresponding to the attribute of the biological TO of the target. Therefore, it is possible to increase the possibility that the thickness of the target tissue is calculated with higher accuracy.

[0050] In addition, in this embodiment, the first attribute and the second attribute are attributes classified according to the magnitude of the physique parameter. Therefore, in the biometric attribute classification step, by acquiring the physique parameter based on the target received data, the target biometric TO can be classified into either the first attribute or the second attribute. Furthermore, when preparing teacher data for training the first learning model 121 and the second learning model 122, the sample data to be used as this teacher data can be classified according to the physique parameter. Therefore, teacher data for training the first learning model 121 and the second learning model 122 can be more easily prepared.

[0051] In the present embodiment, the physique parameter is a parameter representing fat thickness. Therefore, in the biological attribute classification step, the parameter representing fat thickness is acquired based on the target received data, so that the biological TO of the target can be classified into either the first attribute or the second attribute.

[0052] In this embodiment, the thickness label LB is a label that specifies the positions of the first end Eg1 and the second end Eg2 in the thickness direction d1 of the target tissue, and the processor 110 calculates the thickness of the target tissue using the positions of the first end Eg1 and the second end Eg2 output by inputting the target reception data in the thickness calculation step. In this way, the calculation model can be generated as a learning model that specifies the positions of the first end Eg1 and the second end Eg2, and the thickness of the target tissue can be calculated using the positions of the first end Eg1 and the second end Eg2 specified by the calculation model. For example, as in this embodiment, the first learning model 121 and the second learning model 122 can be generated as learning models that perform segmentation.

[0053] In addition, in this embodiment, in the biometric attribute classification step, the target biometric TO is classified using the third learning model 123 different from the first learning model 121 and the second learning model 122. In this way, in the biometric attribute classification step, the target biometric TO can be classified using the third learning model 123 different from the calculation model. Therefore, for example, in the biometric attribute classification step, a learning model specialized for classifying the target biometric TO can be used, so that the possibility of the target biometric TO being appropriately classified in the biometric attribute classification step can be increased.

[0054] In addition, in this embodiment, the third learning model 123 has been pre-trained using the first sample data SD1 and the physique label related to the physique parameter associated with the first sample data SD1, and the second sample data SD2 and the physique label associated with the second sample data SD2 as training data. Therefore, regardless of whether the attribute of the target biological TO is the first attribute or the second attribute, the target biological TO can be appropriately classified in the biological attribute classification step. Furthermore, when preparing training data for training the third learning model 123, the sample data to be used as the training data can be classified according to the physique parameters. Therefore, training data for training the third learning model 123 can be more easily prepared.

[0055] B. Other embodiments: (B-1) In the above embodiment, the first attribute and the second attribute are attributes classified according to the magnitude of the physique parameter. In contrast, the first attribute and the second attribute do not have to be such attributes. For example, when the physique parameter is a vector quantity, the first attribute and the second attribute may be attributes classified according to the orientation of the physique parameter. In this case, the physique parameter may be a parameter represented by a multidimensional vector having various parameters such as fat thickness, body fat percentage, BMI, etc. as components. In this case, the physique label may be a label related to the parameter represented by this multidimensional vector.

[0056] (B-2) In the above embodiment, the physique parameter is a parameter representing fat thickness. In contrast, the physique parameter does not have to be a parameter representing fat thickness, and may be, for example, a parameter representing body fat percentage or BMI. Furthermore, the physique parameter representing fat thickness, body fat percentage, or BMI may be, for example, a parameter represented by a multidimensional vector having any two or all of fat thickness, body fat percentage, and BMI as components. Furthermore, the physique parameter is not limited to the above, and may be a parameter related to other physiques. In this case, the physique label may be a label related to each parameter.

[0057] (B-3) In the above embodiment, the thickness label LB is a label indicating whether each region included in the B-mode image of each sample data to which the thickness label LB is associated represents a fat layer FL, a muscle layer ML, or an internal organ OR. In contrast, the thickness label LB may be, for example, a label indicating whether each region included in the B-mode image of each sample data to which the thickness label LB is associated represents a muscle layer ML or a tissue other than the muscle layer ML. Even if the thickness label LB is such a label, the positions of the first end Eg1 and the second end Eg2 can be specified by the thickness label LB. In this case, the calculation model is generated as a learning model that performs two-class segmentation when the target received data is input. In addition, the thickness label LB does not have to be a label that specifies the positions of the first end Eg1 and the second end Eg2 as long as it is a label related to the thickness of the target tissue. In this case, for example, a value indicating the thickness of the target tissue may be used as the thickness label LB.

[0058] (B-4) In the above embodiment, the third learning model 123 is used in the biometric attribute classification step. In contrast, the third learning model 123 may not be used in the biometric attribute classification step. In this case, the target biometric TO may be classified by, for example, a rule-based model in the biometric attribute classification step.

[0059] (B-5) In the above embodiment, the third learning model 123 is a learning model that has been trained using the third learning data LD3 as teacher data. In contrast, the third learning model 123 does not have to be such a learning model, and may be, for example, a model similar to the fourth learning model described in the first embodiment. Even in this case, since the fourth learning data for training the fourth learning model includes the first sample data SD1 and the second sample data SD2, it is possible to increase the possibility that the target living body TO is appropriately classified in the living body attribute classification step.

[0060] (B-6) In the above embodiment, the calculation device 20 includes the ultrasonic probe 410, but may not include the ultrasonic probe 410. In this case, the measurement data acquisition unit 111 of the calculation device 20 may be configured to acquire target reception data transmitted from, for example, the ultrasonic probe 410, a computer, or a recording medium provided outside the calculation device 20.

[0061] (B-7) In the above embodiment, the received data is a B-mode image, but it does not have to be a B-mode image. In this case, the received data may be, for example, so-called A-mode data that represents the amplitude of the reflected wave in the received signal for each distance.

[0062] (B-8) In the above embodiment, the first learning model 121 and the second learning model 122 are each a CNN, but may be a machine learning model other than a CNN. For example, the first learning model 121 and the second learning model 122 may be a vector neural network type machine learning model having a vector neuron layer, a recurrent neural network, or a machine learning model other than a neural network. Similarly, the third learning model 123 may be a machine learning model other than a CNN.

[0063] (B-9) In the above embodiment, the calculation device 20 executes the first learning process and the second learning process, but the calculation device 20 may not execute the first learning process and the second learning process. In this case, the first learning model 121 and the second learning model 122 may be generated outside the calculation device 20. Also, in this case, the first learning data LD1 and the second learning data LD2 may not be stored in the storage device 120 of the calculation device 20. Similarly, the calculation device 20 may not execute the third learning process. In this case, the storage device 120 may not store the third learning data LD3.

[0064] (B-10) In the above embodiment, the target tissue is a muscle layer. In contrast, the target tissue may include other tissues in addition to or instead of the muscle layer. For example, the calculation device 20 may be configured to calculate the thickness of each of a muscle layer and a fat layer as the target tissue. Also, for example, the target tissue may include two or more muscle layers or two or more fat layers.

[0065] (B-11) In the above embodiment, in the biological attribute classification step, the target biological TO may be classified into three or more attributes. In this case, a learning model corresponding to each of the three or more attributes may be prepared, and in the thickness calculation step, the thickness of the target tissue may be calculated using the learning model according to the attribute of the target biological TO.

[0066] C. Other forms: The present disclosure is not limited to the above-mentioned embodiment, and can be realized in various forms without departing from the spirit of the present disclosure. For example, the present disclosure can be realized in the following forms. The technical features in the above-mentioned embodiments corresponding to the technical features in each form described below can be appropriately replaced or combined in order to solve some or all of the problems of the present disclosure, or to achieve some or all of the effects of the present disclosure. Furthermore, if the technical feature is not described as essential in this specification, it can be appropriately deleted.

[0067] (1) According to a first aspect of the present disclosure, there is provided a thickness calculation method for calculating the thickness of tissue in a living body by a computer. In this calculation method, the computer has one or more processors, and the processor executes the following steps: a data acquisition step of acquiring reception data for a living body of a target based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body and received by the ultrasonic probe; a biometric attribute classification step of classifying the living body of the target into one of a first attribute related to the physique of the living body and a second attribute related to the physique based on the target reception data, which is the reception data acquired in the data acquisition step; and a thickness calculation step of (i) inputting the target reception data into a first learning model that has been trained in advance when the living body of the target is classified into the first attribute in the biometric attribute classification step, and calculating the thickness of tissue at a predetermined site of the living body of the target, and (ii) inputting the target reception data into a second learning model that has been trained in advance when the living body of the target is classified into the second attribute in the biometric attribute classification step, and calculating the thickness of the tissue. The first learning model has been trained in advance using as training data first sample data, which is the received data for a biological sample having the first attribute, and a thickness label relating to the thickness of the tissue associated with the first sample data. The second learning model has been trained in advance using as training data second sample data, which is the received data for a biological sample having the second attribute, and the thickness label associated with the second sample data. According to this embodiment, the tissue thickness is calculated using the target reception data and a learning model that has been trained using teacher data corresponding to the attributes of the classified target living body, thereby increasing the possibility that the tissue thickness can be calculated with higher accuracy.

[0068] (2) In the above embodiment, the first attribute and the second attribute may be attributes classified according to a magnitude of a parameter related to the physique. According to this embodiment, in the biometric attribute classification step, a parameter related to the physique is acquired based on the target received data, so that the biometric information of the target can be classified into either the first attribute or the second attribute.

[0069] (3) In the above embodiment, the parameter may represent at least one of fat thickness, body fat percentage, and BMI. According to this embodiment, in the biometric attribute classification step, a parameter representing fat thickness, body fat percentage, or BMI is acquired based on the target received data, so that the target biometric entity can be classified into either a first attribute or a second attribute.

[0070] (4) In the above embodiment, the thickness label is a label that identifies the positions of a first end and a second end of the tissue in a thickness direction, and the processor may calculate the thickness of the tissue using the positions of the first end and the second end output by inputting the target received data in the thickness calculation step. According to this embodiment, the calculation model can be generated as a learning model that identifies the positions of the first end and the second end, and the thickness of the target tissue can be calculated using the positions of the first end and the second end identified by the calculation model.

[0071] (5) In the above embodiment, the biometric attribute classification step may classify the target biometric organism into either the first attribute or the second attribute using a third learning model different from the first learning model and the second learning model. According to this embodiment, the target biometric organism can be classified using a third learning model different from a calculation model in the biometric attribute classification step. Therefore, for example, a learning model specialized for classifying the target biometric organism can be used in the biometric attribute classification step, thereby increasing the possibility that the target biometric organism will be appropriately classified.

[0072] (6) In the above embodiment, the first attribute and the second attribute may be attributes classified according to the magnitude of a parameter related to the physique, and the third learning model may have been trained in advance using the first sample data and a physique label related to the parameter associated with the first sample data, and the second sample data and the physique label associated with the second sample data as teacher data. According to this embodiment, regardless of whether the attribute of the target biometric entity is the first attribute or the second attribute, the target biometric entity can be appropriately classified in the biometric attribute classification step.

[0073] (7) According to a second embodiment of the present disclosure, there is provided a calculation device for calculating a thickness of tissue in a living body. The calculation device includes a storage device that stores a first learning model that has been previously trained, and a second learning model that has been previously trained, and one or more processors. The processor executes a data acquisition step of acquiring reception data for a living body of a target based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body and received by the ultrasonic probe, a biometric attribute classification step of classifying the living body of the target into one of a first attribute related to the physique of the living body and a second attribute related to the physique based on the reception data of the target that is the reception data acquired in the data acquisition step, and a thickness calculation step of (i) inputting the target reception data into the first learning model when the living body of the target is classified into the first attribute in the biometric attribute classification step, and calculating a thickness of tissue at a predetermined site of the living body of the target, and (ii) inputting the target reception data into the second learning model when the living body of the target is classified into the second attribute in the biometric attribute classification step, and calculating a thickness of the tissue. The first learning model has been trained in advance using as training data first sample data, which is the received data for a biological sample having the first attribute, and a label related to the tissue thickness associated with the first sample data. The second learning model has been trained in advance using as training data second sample data, which is the received data for a biological sample having the second attribute, and the label associated with the second sample data.

[0074] (8) In the second embodiment, the ultrasound probe may further include an ultrasound probe that transmits ultrasound to a living body of the target and outputs the target reception data.

[0075] (9) According to a third aspect of the present disclosure, there is provided a program for causing a computer to realize a function of calculating a thickness of tissue in a living body. This program causes the computer to realize a function of executing a data acquisition step of acquiring reception data based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body and received by the ultrasonic probe, a function of executing a biometric attribute classification step of classifying the living body of the target into one of a first attribute related to the physique of the living body and a second attribute related to the physique based on the target reception data, which is the reception data acquired in the data acquisition step, and a function of executing a thickness calculation step of (i) inputting the target reception data into a first learning model that has been trained in advance when the living body of the target is classified into the first attribute in the biometric attribute classification step, and calculating a tissue thickness of a predetermined part of the living body of the target, and (ii) inputting the target reception data into a second learning model that has been trained in advance when the living body of the target is classified into the second attribute in the biometric attribute classification step, and calculating a thickness of the tissue. The first learning model has been trained in advance using training data including first sample data, which is the received data for a biological sample having the first attribute, and a label related to the tissue thickness associated with the first sample data, and the second learning model has been trained in advance using training data including second sample data, which is the received data for a biological sample having the second attribute, and the label associated with the second sample data.

[0076] In addition to the above-mentioned embodiments, the present disclosure can be realized in the form of a learning model generation method, a program, a non-transient recording medium on which a program is recorded, a program product, etc. Note that this program product may be provided as a recording medium on which a program is recorded, or may be provided as a program product that can be distributed via a network, for example. [Explanation of symbols]

[0077] 20...calculation device, 81...ultrasonic element, 100...control device, 110...processor, 111...measurement data acquisition unit, 112...biological attribute classification unit, 113...tissue thickness calculation unit, 119...learning execution unit, 120...storage device, 121...first learning model, 122...second learning model, 123...third learning model, 129...program, 130...interface circuit, 150...display unit, 400...measurement device, 410...ultrasonic probe, 430...wiring, 460...control unit, 465...drive pulse generation circuit, 466...transmission circuit, 467...signal processing circuit, 468...reception circuit, 469...multiplexer, 470...microcomputer, 471...communication unit, 811...transmission element, 812...reception element

Claims

1. A method for calculating a thickness of a tissue in a living body by a computer, comprising: the computer having one or more processors; The processor, a data acquisition step of acquiring reception data based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected within the living body of a target and received by the ultrasonic probe; a biometric attribute classification step of classifying the biometric data of the target into either a first attribute related to a physical build of the target biometric data or a second attribute related to the physical build, based on the target received data, which is the received data acquired in the data acquisition step; (i) if the target biometric data is classified into the first attribute in the biometric attribute classification step, inputting the target received data into a pre-trained first learning model to calculate a tissue thickness of a predetermined site of the target biometric data, and (ii) if the target biometric data is classified into the second attribute in the biometric attribute classification step, inputting the target received data into a pre-trained second learning model to calculate a tissue thickness. the first learning model has been trained in advance using first sample data, which is the received data for a biological sample having the first attribute, and a thickness label relating to a thickness of the tissue associated with the first sample data, as training data; A thickness calculation method in which the second learning model has been pre-trained using second sample data, which is the received data for a biological sample having the second attribute, and the thickness label associated with the second sample data as training data.

2. The thickness calculation method according to claim 1 , A thickness calculation method, wherein the first attribute and the second attribute are attributes classified according to the magnitude of a parameter related to the physique.

3. The thickness calculation method according to claim 2, The parameter represents at least one of fat thickness, body fat percentage, and BMI.

4. The thickness calculation method according to claim 1 , the thickness label is a label for identifying positions of a first end and a second end of the tissue in a thickness direction; A thickness calculation method in which, in the thickness calculation step, the processor calculates the thickness of the tissue using the positions of the first end and the second end output by inputting the target reception data.

5. The thickness calculation method according to any one of claims 1 to 4, A thickness calculation method in which, in the biometric attribute classification step, the target biometric entity is classified into either the first attribute or the second attribute using a third learning model different from the first learning model and the second learning model.

6. The thickness calculation method according to claim 5, the first attribute and the second attribute are attributes classified according to the magnitude of the physique-related parameter, A thickness calculation method in which the third learning model has been pre-trained using the first sample data and a body build label related to the parameters associated with the first sample data, and the second sample data and the body build label associated with the second sample data as training data.

7. A calculation device for calculating a thickness of tissue in a living body, comprising: A storage device that stores a first learning model that has been trained in advance and a second learning model that has been trained in advance; one or more processors; The processor, a data acquisition step of acquiring reception data based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected within the living body of a target and received by the ultrasonic probe; a biometric attribute classification step of classifying the biometric data of the target into either a first attribute related to a physical build of the target biometric data or a second attribute related to the physical build, based on the target received data, which is the received data acquired in the data acquisition step; (i) if the target biological body is classified into the first attribute in the biological attribute classification step, inputting the target received data into the first learning model to calculate a tissue thickness of a predetermined part of the target biological body, and (ii) if the target biological body is classified into the second attribute in the biological attribute classification step, inputting the target received data into the second learning model to calculate a tissue thickness, the first learning model has been trained in advance using first sample data, which is the received data for a biological sample having the first attribute, and a label related to a thickness of the tissue associated with the first sample data, as training data; A calculation device in which the second learning model has been pre-trained using second sample data, which is the received data for a biological sample having the second attribute, and the label associated with the second sample data as training data.

8. The calculation device according to claim 7, The computing device further comprises the ultrasound probe configured to transmit ultrasound waves to a living subject and output the subject received data.

9. A program for causing a computer to realize a function of calculating a thickness of tissue in a living body, A function of executing a data acquisition step of acquiring reception data based on a reception signal outputted when an ultrasonic wave transmitted from an ultrasonic probe is reflected in the living body of a target and received by the ultrasonic probe; a function of executing a biometric attribute classification step of classifying the biometric data of the subject into either a first attribute related to the physique of the subject or a second attribute related to the physique, based on the subject received data, which is the received data acquired in the data acquisition step; (i) when the biological body of the target is classified into the first attribute in the biological attribute classification step, inputting the target received data into a first learning model that has been trained in advance to calculate a tissue thickness of a predetermined part of the biological body of the target, and (ii) when the biological body of the target is classified into the second attribute in the biological attribute classification step, inputting the target received data into a second learning model that has been trained in advance to calculate a tissue thickness, the first learning model has been trained in advance using first sample data, which is the received data for a biological sample having the first attribute, and a label related to a thickness of the tissue associated with the first sample data, as training data; A program in which the second learning model has been pre-trained using teacher data including second sample data, which is the received data for a biological sample having the second attribute, and the label associated with the second sample data.

Images