JP7093093B2 - Ultrasonic urine volume measuring device, learning model generation method, learning model - Google Patents

Description

The present invention relates to an ultrasonic urine volume measuring device, a learning model generation method, and a learning model.

Bladder urine volume is measured using ultrasound for urination diary for diagnosing diseases related to excretion and for urination management at the nursing care site.
For the measurement of urine volume by ultrasonic waves, a method of analyzing the A-mode signal of the bladder obtained by transmitting and receiving ultrasonic waves is well known. For example, the bladder wall is detected from the A mode signal, and the urine volume is estimated from the calculation with various coefficients using the reflection intensity of the bladder wall and the distance between the anterior wall and the posterior wall as parameters, or from the distance between the anterior wall and the posterior wall. A method such as calculating the volume to estimate the bladder urine volume is known (Patent Document 1).

Re-table 2005-099582 JP 2016-043274

Since the conventional ultrasonic urine volume measuring device detects the bladder wall from the A mode signal and measures or estimates the urine volume using the reflection intensity of the bladder wall and the distance between the bladder walls as parameters, the bladder wall detection is performed. It was an essential technique for bladder measurement. Therefore, if an A-mode signal with signal quality capable of detecting the bladder wall cannot be obtained, accurate bladder measurement cannot be performed.

The learning model generation method according to the present invention for solving the problem comprises the ultrasonic wave reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject, and the urine volume data associated with the ultrasonic wave reception data. An acquisition process for acquiring a plurality of teacher data, and a machine learning process for generating a learning model that outputs urine volume data from the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject using the teacher data. It is characterized by having.

In order to solve the problem, the ultrasonic urine volume measuring device according to the present invention acquires ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to a subject, and generates the acquired ultrasonic reception data as the learning model. It is characterized by having a urine volume data generation means that inputs to a learning model trained by the method and outputs urine volume data from the learning model.

In order to solve the problem, the learned program according to the present invention includes an input layer for inputting ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to a subject, an output layer for outputting urine volume data, and the ultrasonic waves. It is equipped with an intermediate layer in which parameters are learned using multiple teacher data that inputs received data and outputs urine volume data, and acquires ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject. It is characterized in that the acquired ultrasonic reception data is input to the input layer, the calculation is performed in the intermediate layer in which the parameters are learned, and the computer functions so that the urine volume data is output from the output layer.

The learning model generation method according to the present invention for solving the problem is associated with the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject, the position information of the subject, and the ultrasonic reception data. The acquisition process of acquiring a plurality of teacher data consisting of the urine volume data, the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject using the teacher data, and the position information of the subject. It is characterized by having a machine learning process for generating a learning model which is input and outputs urine volume data.

In order to solve the problem, the ultrasonic urine volume measuring device according to the present invention acquires the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject and the position information of the subject, and obtains the acquisition. A urine volume data generation means for inputting the acquired ultrasonic reception data and the acquired position information of the subject into the learning model learned by the learning model generation method and outputting urine volume data from the learning model. It is characterized by having.

In order to solve the problem, the learned program according to the present invention is an input layer and urine volume in which ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to a subject and position information of the subject are input. Parameters are learned using the output layer that outputs data, the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject, the position information of the subject, and multiple teacher data that outputs urine volume data. The bladder ultrasonic reception data acquired by transmitting and receiving ultrasonic waves to and from the subject and the position information of the subject are acquired, and the acquired ultrasonic reception data and the acquired are obtained. It is characterized in that the position information of the subject is input to the input layer, the calculation is performed in the intermediate layer in which the parameters are learned, and the computer functions so that the urine volume data is output from the output layer.

According to the present invention, it is possible to generate a learning model for measuring urine volume without going through the step of detecting the bladder wall from the ultrasonic received data. Further, according to the present invention, it is possible to realize an ultrasonic urine volume measuring device having a urine volume measuring means for outputting urine volume data without going through a step of detecting a bladder wall from ultrasonic wave reception data. Further, according to the present invention, it is possible to realize a learned program that outputs urine volume data without going through the step of detecting the bladder wall from the ultrasonic reception data.

According to the present invention, it is possible to generate a learning model that outputs urine volume data by inputting not only ultrasonic reception data but also position information indicating a place where measurement is performed. Further, according to the present invention, an ultrasonic urine volume measuring device having a urine volume measuring means for outputting urine volume data by using not only ultrasonic reception data but also position information indicating a place where measurement is performed is realized. can. Further, according to the present invention, it is possible to realize a learned program that outputs urine volume data by inputting not only ultrasonic reception data but also position information indicating a place where measurement is performed.

It is a flowchart of the structure and the generation method of the 1st learning model which concerns on this invention. Explanatory drawing of ultrasonic wave reception data. Explanatory drawing of the ultrasonic urine volume measuring apparatus which concerns on one Embodiment. An example of the configuration of the first learning model according to the present invention is shown. It is a flowchart of the first learned program which concerns on this invention. It is a flowchart of the structure of the 2nd learning model which concerns on this invention, and the generation method. Explanatory drawing of the ultrasonic urine volume measuring apparatus which concerns on one Embodiment. An example of the configuration of the second learning model according to the present invention is shown. It is a flowchart of the 2nd learned program which concerns on this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. The embodiment is not limited to the configuration and means shown in the figure.

<Example 1>
FIG. 1 is a configuration (a) of a learning model generated by the first learning model generation method according to the present invention, and a flowchart (b) of the learning model generation method.
As shown in FIG. 1A, the learning model 10 has a configuration in which the ultrasonic wave reception data 11 is input and the urine volume data 12 is output.
The ultrasonic reception data 11 is ultrasonic data transmitted and received toward the bladder of the subject for measuring urine volume, and is composed of data of reception signals transmitted and received by a plurality of ultrasonic vibrators.

The ultrasonic reception data 11 is a plurality of reception data acquired by transmitting and receiving ultrasonic waves by a one-dimensional array or a two-dimensional array of oscillators, and is composed of a plurality of reception data obtained from a plurality of locations in the bladder. ..
For example, it may be two or more received data transmitted / received at two or more locations along the vertical axis of the bladder by an ultrasonic transducer arranged in one dimension, or transmitted / received at two or more locations along the horizontal axis of the bladder. It may be two or more received data. Further, it may be two or more received data at two or more places transmitted / received by the two-dimensionally arranged ultrasonic vibrator.
Further, the ultrasonic wave reception data 11 may include reception data that does not include a reception signal from the bladder.

The urine volume data 12 is data indicating the urine volume in the bladder output from the learning model 10, and is data associated with the ultrasonic reception data 11.
The urine volume data 12 is expressed in terms of volume, volume, weight, body weight ratio of subject, increase amount, decrease amount, change amount, change rate, large amount, small amount, appropriate amount, increasing, decreasing, presence / absence of change, etc. Regardless of, it is sufficient to include any one of these information, and it may be time-series data, multiple data, or a combination of these, and is data related to bladder urine volume. All you need is.

The learning model 10 is a neural network composed of an input layer, an intermediate layer, and an output layer, and may include a plurality of intermediate layers. Further, the present invention is not limited to this configuration, and AI may be an AI such as a convolutional neural network, clustering, decision tree, random forest, support vector machine, or a combination thereof. FIG. 3 is an example of the configuration of the learning model 10.

FIG. 1B is a flowchart showing the steps of the learning model generation method according to the present invention.
Step S100 is a step of acquiring a plurality of ultrasonic wave reception data and urine volume data as teacher data.
Step S200 is a machine learning step of performing machine learning by inputting ultrasonic reception data and outputting urine volume data based on the teacher data acquired in step S100. Machine learning is performed by AI technology such as deep learning, reinforcement learning, and deep reinforcement learning. However, the technique is not limited to these, and these techniques may be combined.
Machine learning is executed by an arithmetic means or an information processing device composed of a computer, a CPU, a GPU, an FPGA, or the like.

Next, the method of measuring the urine volume of the learning model according to this embodiment will be described with reference to FIG. Although the ultrasonic reception data exemplified in FIG. 2 is an RF signal waveform, it may be in any format such as full-wave rectification, logarithmic compression amplification, phase data, baseband signal, and envelope-processed signal, and may be plural. It may be composed of various types of waveforms.
2 (a1) and 2 (b1) are examples of ultrasonic wave reception data transmitted and received toward the bladder. The vertical axis of the ultrasonic wave reception data waveform represents the intensity of the received signal, and the horizontal axis represents the elapsed time based on the ultrasonic wave transmission time, and represents the distance from the body surface. 2 (a2) and 2 (b2) are diagrams schematically showing the tissues in the body of the subject, and are schematic cross-sectional views of the path through which ultrasonic waves are transmitted and received. In FIGS. 2 (a1), (b1), (a2) and (b2), the symbol S usually indicates the position of the body surface with which the ultrasonic vibrator abuts (referred to as body surface S). The symbol A represents the anterior wall of the bladder near the body surface of the approximately spherical bladder (denoted as anterior wall A). Further, the symbol P represents the posterior wall of the bladder in a portion far from the body surface of the approximately spherical bladder (denoted as posterior wall P). The symbol B indicates the back side of the subject (denoted as back B).

In FIGS. 2 (a2) and 2 (b2), the section between the body surface S and the anterior wall A represents the body tissue from the body surface to the anterior wall of the bladder. The area between the anterior wall A and the posterior wall P is inside the bladder, that is, represents urine. The area between the posterior wall P and the back B represents the body tissue from the posterior wall of the bladder to the back.

In FIGS. 2 (a1) and 2 (b1), the area between the body surface S and the anterior wall A is a received signal from the body surface to the anterior wall of the bladder. It passes through tissues with different acoustic impedances from the body surface to the skin, fat, muscles, etc., and because it is a region with little attenuation immediately after ultrasonic transmission, it is ultrasonic reception data with a large amplitude. The ultrasonic wave reception data of the portion of the anterior wall A is a large reflected wave because the difference in acoustic impedance between the bladder wall and urine is large. The signal received from the urine is between the anterior wall A and the posterior wall P, but since the reflection from the urine is small, the ultrasonic wave reception data has a small amplitude or a small change. The ultrasonic wave reception data of the portion of the posterior wall P is a large reflected wave because the difference in acoustic impedance between the urine and the bladder wall is large.

Now, let's look at the ultrasonic reception data of FIG. 2 (a1). It can be seen from this ultrasonic reception data that the front wall A and the rear wall P can be easily extracted. For example, a method such as extracting a signal having a small amplitude by threshold processing and then extracting portions A and P having a large amplitude can be considered. Further, a method of extracting the front wall A and the rear wall P by paying attention to the portion where the amount of change in the amplitude is large can be considered. Further, a method of finding the envelope of the received signal and extracting the front wall A and the rear wall P from the descending point and the ascending point of the envelope can be considered.

After extracting the anterior wall A and the posterior wall P by the above method, the distance between the anterior wall A and the posterior wall P is estimated as the approximate diameter of the bladder to calculate the volume of the bladder, or the anterior wall A- The urine volume can be obtained by an estimation formula from the distance between the posterior walls P or the respective distances from the body surface S and the signal amplitude. The conventional ultrasonic urine volume measuring device measures the urine volume by such a method.

Here, the received data of FIG. 2 (b1) is similarly viewed. In this ultrasonic reception data, the amplitudes of the front wall A and the rear wall P are relatively small, and it can be seen that it is difficult to extract the front wall A or the rear wall P by the above method. This is because the contact between the ultrasonic transducer and the subject is not sufficient, the received signal is weak depending on the constitution of the subject, and the reflected wave from urine is large, so the reflected wave on the bladder wall becomes relatively small. It can occur due to various factors such as urination. In the case of FIG. 2 (b1), the anterior wall A and the posterior wall B are not extracted, and the conventional ultrasonic urine volume measuring device cannot measure the urine volume.

The learning model according to the present invention is an AI that outputs urine volume data by inputting received data using FIGS. 2 (a1) and 2 (b1) as an example. The learning model according to the present invention is machine-learned to output urine volume data from the characteristics of the ultrasonic reception data waveform, the characteristics of the combination of ultrasonic reception data from a plurality of locations, and the like, and the calculation of the urine volume data. The process is characterized by not including the bladder wall detection step. Therefore, even the ultrasonic wave reception data as shown in FIG. 2 (b1) can output the urine volume data.

The method for measuring the urine volume of the learning model according to the present invention described above is the same for the learning model according to the other examples described below.

<Example 2>
FIG. 3 is an example of the ultrasonic urine volume measuring device according to the present invention. Examples of ultrasonic wave reception data and urine volume data in this embodiment will be described.

The ultrasonic urine volume measuring device 2 transmits ultrasonic waves to the bladder of the subject, receives ultrasonic waves reflected from the bladder or other body tissues, and measures the urine volume in the bladder from the ultrasonic reception data. It is a device for estimating and predicting.

The ultrasonic data generation means 100 includes an ultrasonic transmission means for transmitting ultrasonic waves to a subject, and an ultrasonic receiving means for receiving reflected ultrasonic waves from the subject and converting them into electrical signals for signal processing. It comprises data generation means for AD conversion and the like of the ultrasonic reception signal received by the means, and outputs the ultrasonic reception data 11. Further, it has an operation switch for controlling the ultrasonic data generation means, a control means for setting a signal from the operation switch to the transmission / reception means, and a control means for controlling the operation of the entire device.

Transmission and reception of ultrasonic waves is performed by a one-dimensional array or a two-dimensional array of ultrasonic transducers.
The ultrasonic reception data 11 is a plurality of reception data obtained by transmitting and receiving ultrasonic waves by a one-dimensional array or a two-dimensional array ultrasonic transducer, and is composed of a plurality of reception data obtained from a plurality of locations in the bladder. To.
For example, it may be two or more received data transmitted / received at two or more locations along the vertical axis of the bladder, or may be two or more received data transmitted / received at two or more locations along the horizontal axis of the bladder. However, it may be two or more received data in two or more places transmitted and received two-dimensionally. Further, the ultrasonic wave reception data may include reception data by transmission / reception to other than the bladder.

The format of the ultrasonic reception data 11 is not limited, such as A-mode data, full-wave rectified data, non-linear compressed data, filtered data, frequency spectrum data, and the like, as long as it is at least one type of data. good. Further, it may be composed of received data that has been transmitted / received to the same location a plurality of times.
Further, the ultrasonic wave reception data 11 may include array information of the ultrasonic vibrator that has performed ultrasonic wave transmission / reception, such as which position of the bladder the transmission / reception data is.

The ultrasonic wave reception data 11 described above is input to the urine volume data generation means 101. The urine volume data generation means 101 is a computer program execution means that performs a process of outputting urine volume data 12 by the learning model 10 learned by the learning model generation method described in the first embodiment.
The urine volume data generation means 101 may be executed by a calculation means such as a CPU, GPU, FPGA (not shown) incorporated in the ultrasonic urine volume measuring device. It may also be performed on a computer connected to the ultrasonic urine output measuring device or another computer connected by a network.

The urine volume data 12 includes urine volume, volume, weight, subject weight ratio, increase amount, decrease amount, change amount, change rate, large amount, small amount, appropriate amount, increasing, decreasing, no change, and unmeasurable. Regardless of the expression format, it is sufficient to include any one of these information, and it may be time-series data, multiple types of data, or a combination of these, and the amount of urine in the bladder. Any data related to

The urine volume data 12 may be displayed as characters, figures, images, sounds, or a combination thereof on a display device such as a display constituting the ultrasonic urine volume measuring device, or may be stored in a data storage means (not shown). Alternatively, the data may be transferred to a computer, server, other ultrasonic urine metering device, medical device, etc. on a network (not shown).

<Example 3>
FIG. 5 is a flowchart of the first learned program 150, which is one embodiment of the present invention. The learned program 150 may be a program executed by a computer, a CPU, or a GPU, or may be arrangement information to be executed by a hardware arithmetic element such as an FPGA. The form is not particularly limited, but any data may be used as long as it is data for operating an arithmetic unit or an information processing unit.

The trained program 150 is composed of an input layer, an intermediate layer, and an output layer. The configuration is not limited to this, but may be, for example, a neural network having a plurality of intermediate layers, or an AI such as a convolutional neural network, clustering, decision tree, random forest, or support vector machine. Further, a combination of these may be used.

Ultrasound reception data is input to the input layer, and the output layer outputs urine volume data.
The parameters of the intermediate layer are optimized by machine learning by teacher data consisting of a plurality of ultrasonic reception data and urine volume data.

Machine learning is performed by AI technology such as deep learning, reinforcement learning, and deep reinforcement learning. However, the technique is not limited to these techniques, and a plurality of techniques may be combined. Machine learning may be executed by an arithmetic means or an information processing device composed of a computer, a CPU, a GPU, an FPGA, or the like, and the means are not particularly limited.

The trained model 150 acquires ultrasonic wave reception data in the data acquisition step S400.
The acquired ultrasonic wave reception data is input to the input layer of the trained model 150 in step S410.
The input ultrasonic wave reception data is calculated in step S420, and the calculation result is output in step S430. The calculation result is output to an output means (not shown) as urine volume data in step S440.

According to the learning model generation method of Example 1, the ultrasonic urine volume measuring device of Example 2, and the trained model of Example 3 according to the first learning model described above, the present invention relates to the bladder wall. Urinary volume data can be obtained even from ultrasonic reception data that is difficult to detect.

<Example 4>
FIG. 6 is a configuration (a) of a learning model generated by the second learning model generation method according to the present invention, and a flowchart (b) of the learning model generation method.
As shown in FIG. 6A, the learning model 510 has a configuration in which ultrasonic wave reception data 511 and position information 512 are input and urine volume data 514 is output.

The ultrasonic reception data 511 is ultrasonic data transmitted and received toward the bladder of the subject for measuring urine volume, and is composed of data from a plurality of ultrasonic transducers.

The position information 512 is position information indicating the place where the urine volume measurement was performed. However, the position information may be the position information of the past subject such as the movement locus information.

The urine volume data 514 is data indicating the urine volume in the bladder output from the learning model 510, and is data associated with the ultrasonic reception data 511.
The urine volume data 514 is expressed in terms of volume, volume, weight, body weight ratio of subject, increase amount, decrease amount, change amount, change rate, large amount, small amount, appropriate amount, increasing, decreasing, presence / absence of change, etc. Regardless of, it is sufficient to include any one of these information, and it may be time-series data, multiple data, or a combination of these, and is data related to bladder urine volume. All you need is.

The learning model 510 is a neural network composed of an input layer, an intermediate layer, and an output layer, and may include a plurality of intermediate layers. Further, the present invention is not limited to this configuration, and AI may be an AI such as a convolutional neural network, clustering, decision tree, random forest, support vector machine, or a combination thereof. FIG. 8 is an example of the configuration of the learning model 510.

FIG. 6B is a flowchart showing the steps of the learning model generation method according to the present invention.
Step S610 is a step of acquiring a plurality of ultrasonic wave reception data, position information, and urine volume data as teacher data.
Step S620 is a machine learning step of performing machine learning by inputting ultrasonic reception data and position information and outputting urine volume data based on the teacher data acquired in step S610. Machine learning is performed by AI technology such as deep learning, reinforcement learning, and deep reinforcement learning. However, the technique is not limited to these, and these techniques may be combined.
Machine learning is executed by an arithmetic means or an information processing device composed of a computer, a CPU, a GPU, an FPGA, or the like.

In the second learning model as well, in the calculation process of outputting the urine volume data from the ultrasonic wave reception data as in the first learning model, the urine volume data is output by an analysis method that does not include the bladder wall detection step. It is a thing.

<Example 5>
FIG. 7 is an example of the ultrasonic urine volume measuring device according to the present invention. Examples of ultrasonic wave reception data, position information, and urine volume data in this embodiment will be described.

The ultrasonic urine volume measuring device 5 transmits ultrasonic waves to the bladder of the subject, receives ultrasonic waves reflected from the bladder or other body tissues, and measures the urine volume in the bladder from the ultrasonic reception data. It is a device for estimating and predicting.

The ultrasonic data generation means 100 includes an ultrasonic transmission means for transmitting ultrasonic waves to a subject, an ultrasonic receiving means for receiving reflected ultrasonic waves from the subject and converting them into electrical signals for signal processing, and an ultrasonic wave receiving means. It comprises a data generation means that performs AD conversion and the like of the ultrasonic wave reception signal received by the sound wave receiving means, and outputs the ultrasonic wave reception data 511. Further, it has an operation switch for controlling the ultrasonic data generation means, a control means for setting a signal from the operation switch to the transmission / reception means, and a control means for controlling the operation of the entire device.

Transmission and reception of ultrasonic waves is performed by a one-dimensional array or a two-dimensional array of ultrasonic transducers.
The received data 511 is a plurality of received data obtained by transmitting and receiving ultrasonic waves by a one-dimensional array or a two-dimensional array of ultrasonic transducers, and is composed of a plurality of received data obtained from a plurality of locations in the bladder.
For example, it may be two or more received data transmitted / received at two or more locations along the vertical axis of the bladder, or may be two or more received data transmitted / received at two or more locations along the horizontal axis of the bladder. However, it may be two or more received data transmitted and received in two or more places two-dimensionally. Further, the ultrasonic wave reception data 511 may include received data by transmission / reception to other than the bladder.

The format of the ultrasonic reception data 511 is not limited, such as A-mode data, full-wave rectified data, non-linear compressed data, filtered data, and frequency spectrum data, as long as it is at least one type of data. good. Further, it may be composed of received data transmitted / received to the same location.
Further, the ultrasonic reception data 511 may include arrangement information of the ultrasonic element that has performed ultrasonic transmission / reception, such as which position of the bladder the transmission / reception data is.

The position information 512 is position information indicating the place where the urine volume measurement was performed. For example, it is location information that specifies the location of a medical facility, an examination room in the medical facility, a hospital room, an ICU, a general outpatient department, an emergency outpatient department, or the like.
In addition, in home medical care and telemedicine, the location information may specify the location of the subject or the location of the bedroom, living room, toilet, etc. in the residence, and it may be location information not only indoors but also outdoors. May be.

The acquisition of the position information 512 includes GPS configured in the position information detecting means 513 incorporated in the ultrasonic urine volume measuring device, network connection information to which the ultrasonic urine volume measuring device is connected, and manually input information. It may be obtained from the position information of the person who measured the urine volume.
Further, when the ultrasonic urine volume measuring device is configured in combination with a portable information terminal such as a smartphone, a tablet terminal, or a personal computer, it may be acquired from the GPS included in the portable information terminal.
However, it is not limited to these, and it is acquired from the network connection information of the portable information terminal, the information manually input to the portable information terminal, the position information of the inspector who carried the portable information terminal, etc. It may be the position information specified from the movement locus of the portable information terminal.

The ultrasonic wave reception data 511 and the position information 512 described above are input to the urine volume data generation means 501. The urine volume data generation means 501 is a computer program execution means that performs a process of outputting urine volume data 514 by the learning model 510 learned by the learning model generation method described in the fourth embodiment.
The urine volume data generation means 501 may be executed by a calculation means such as a CPU, GPU, FPGA (not shown) incorporated in the ultrasonic urine volume measuring device. It may also be performed on a computer connected to the ultrasonic urine output measuring device or another computer connected by a network.

The urine volume data 514 is expressed in terms of urine volume, volume, weight, subject weight ratio, increase amount, decrease amount, change amount, change rate, large amount, small amount, appropriate amount, increasing, decreasing, no change, etc. Regardless of, it is sufficient to include any one of these information, and it may be time-series data, multiple types of data, or a combination of these, and data related to bladder urine volume. If it is good.

The urine volume data 514 may be displayed as characters, figures, images, sounds, or a combination thereof on a display device such as a display constituting the ultrasonic urine volume measuring device, or may be stored in a data storage means (not shown). It may be transferred to a computer, server, other medical device, etc. on a network (not shown).

<Example 6>
FIG. 9 is a flowchart of the second learned program 515, which is one embodiment of the present invention. The learned program 515 may be a program executed by a computer, a CPU, or a GPU, or may be arrangement information to be executed by a hardware arithmetic element such as an FPGA. The form is not particularly limited, but any data may be used as long as it is data for operating an arithmetic unit or an information processing unit.

The trained program 515 is composed of an input layer, an intermediate layer, and an output layer. The configuration is not limited to this, but may be, for example, a neural network having a plurality of intermediate layers, or an AI such as a convolutional neural network, clustering, decision tree, random forest, or support vector machine. Further, a combination of these may be used.

Ultrasound reception data and position information are input to the input layer, and the output layer outputs urine volume data.
The parameters of the intermediate layer are optimized by machine learning using a plurality of ultrasonic reception data, position information, and urine volume data as teacher data.

Machine learning is performed by AI technology such as deep learning, reinforcement learning, and deep reinforcement learning. However, the technique is not limited to these techniques, and a plurality of techniques may be combined. Machine learning may be executed by an arithmetic means or an information processing device composed of a computer, a CPU, a GPU, an FPGA, or the like, and the means are not particularly limited.

The trained model 515 acquires ultrasonic wave reception data and position information by an input means (not shown) in the data acquisition step S640.
The acquired ultrasonic wave reception data and position information are input to the input layer of the trained model 515 in step S641.
The input ultrasonic wave reception data and position information are calculated in step S642, and the calculation result is output in step S643. The calculation result is output to an output means (not shown) as urine volume data in step S644.

According to the learning model generation method of Example 4 according to the second learning model described above, the ultrasonic urine volume measuring device of Example 5, and the trained model of Example 6, the present invention provides ultrasonic reception data. More useful urine volume data can be obtained by inputting the position information obtained by measuring the urine volume as well as from.

The effect of the present invention will be described below by exemplifying the position information.

When the measurement location indicated by the position information 512 is an examination room, a general outpatient department, or the like, the subject is usually lying on the examination bed, so it is assumed that urine has accumulated in the bladder. In this case, since the ultrasonic reception data is expected to have good signal quality and the urine volume is expected to show the maximum value, the urine volume measurement adapted to this can be performed.

When the measurement location indicated by the location information 512 is a toilet, urine has accumulated in the bladder (has urinary intention) before urination, and urine has not accumulated (has no urinary intention) after urination. It is assumed that there is.
The judgment before and after urination can be determined, for example, before urination if the measurement data is immediately after moving from another room to the toilet, and after urination if the measurement data is immediately before moving from the toilet to another room.

When the measurement location indicated by the position information 512 is a hospital room, a bedroom, etc., the measurement data is expected to show an aspect in which urine gradually increases, so that the next urination time can be predicted or the urine volume can be predicted. Unintended urination can be detected by the decrease in urination, and useful information for nursing care can be provided as excretion management information.

As described above, the present invention can obtain more accurate and useful urine volume data by inputting not only ultrasonic reception data but also position information of the subject when measuring urine volume by artificial intelligence. can.

1 Subject 2 Ultrasonic urine volume measuring device 3 Bladder 5 Ultrasonic urine volume measuring device 10 First learning model 11 Ultrasonic reception data 12 Urine volume data 20 Input layer 21 Intermediate layer 22 Output layer 100 Ultrasonic data generation means 101 Urine volume data generation means 110 Transmission / reception ultrasonic 150 First learned program 501 Urine volume data generation means 510 Second learning model 511 Ultrasonic reception data 512 Position information 513 Position information detection means 514 Urine volume data 515 Second learning Completed program 520 Input layer 521 Intermediate layer 522 Output layer

Claims (3)

The position of the subject with respect to the position of the bladder ultrasonic reception data acquired by transmitting and receiving ultrasonic waves to the subject and the position information acquired at the time of acquiring the ultrasonic reception data, where the ultrasonic reception data is acquired. The information and the position information acquired before and after the acquisition of the ultrasonic reception data, including either or both of the position information of the subject that changes with respect to the position where the ultrasonic reception data acquisition is performed. An acquisition step of acquiring a plurality of teacher data consisting of the position information of the subject, the urine volume data associated with the ultrasonic reception data, and the acquisition step.
A machine learning process that uses the teacher data to generate a learning model that inputs the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject and the position information of the subject and outputs the urine volume data. When,
A learning model generation method having. The position of the subject with respect to the position where the ultrasonic reception data is acquired, which is the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject and the position information acquired at the time of acquiring the ultrasonic reception data. The information and the position information acquired before and after the acquisition of the ultrasonic reception data, which includes, or both of the position information of the subject that changes with respect to the position where the ultrasonic reception data acquisition is performed. Obtaining the position information of the subject and
The acquired ultrasonic wave reception data and the acquired position information of the subject are input to the learning model learned by the learning model generation method according to claim 1 , and urine volume data is output from the learning model.
An ultrasonic urine volume measuring device having a urine volume data generating means. The position of the subject with respect to the position where the ultrasonic reception data is acquired, which is the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject and the position information acquired at the time of acquiring the ultrasonic reception data. The information and the position information acquired before and after the acquisition of the ultrasonic reception data, which includes, or both of the position information of the subject that changes with respect to the position where the ultrasonic reception data acquisition is performed. The input layer into which the position information of the subject and the input layer are input.
Output layer that outputs urine volume data,
An intermediate layer in which parameters are learned using multiple teacher data that input the ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject and the position information of the subject and output the urine volume data.
Equipped with
The ultrasonic reception data of the bladder acquired by transmitting and receiving ultrasonic waves to the subject and the position information of the subject are acquired.
The acquired ultrasonic reception data and the acquired position information of the subject are input to the input layer.
Calculate in the intermediate layer where the parameters are learned, and output the urine volume data from the output layer.
A trained program to make your computer work.

Images