Method and device for estimating flow velocity and / or flow rate of urine flow

Abstract

The present invention relates to a method for estimating the flow velocity and / or flow rate of urine flow by using a machine learning model created on the basis of training data including the following data (a) and data (b) and further including the following data (c) and / or data (d): (a) training images of liquid flow in an axis switching state; (b) the flow velocity and / or flow rate of the liquid flow; (c) lengths of major and minor axes of an abdominal cross-section during axis switching, or the aspect ratio of the major axis to the minor axis during axis switching; and (d) the angle of rotation around a central axis of the abdominal region during axis switching. The present invention also relates to a device for estimating the flow velocity and / or flow rate of urine flow by using said method.

Description

Method and apparatus for estimating the flow velocity and / or flow rate of urine.

[0001] The present invention relates to a method and apparatus for estimating the flow velocity and / or flow rate of a urine stream. Cross-reference of related applications: This application claims priority to Japanese Patent Application No. 2024-188916, filed on 28 October 2024, the entirety of which is incorporated herein by reference.

[0002] When voiding / storage disorders occur due to conditions such as benign prostatic hyperplasia, overactive bladder, and neurogenic bladder, the velocity and / or flow rate of urine may be abnormally reduced. Therefore, uroflowmetry is essential in the diagnosis, treatment, and monitoring of voiding / storage disorders.

[0003] In medical institutions, it is common practice to place a cup under a portable toilet, measure the change in urine volume using a specialized device, and record the urine volume and flow rate based on the measurement results. However, measurements are often unsuccessful because the environment is different from a toilet used in daily life, or because the timing of urination does not match. Furthermore, while routine uroflowmetry is ideal for monitoring progress, performing uroflowmetry using specialized devices like those used in medical institutions at the patient's home is not practical due to cost and hygiene concerns.

[0004] To overcome these problems, several simple methods for measuring urine flow have been proposed and developed. Such methods include a non-contact urine flow measurement method using an airborne ultrasonic Doppler system (Patent Document 1), a method for determining urine flow rate by analyzing images of overflowing water in a toilet bowl (Patent Document 2), a non-contact urine flow measurement method using a temperature sensor (Patent Document 3), a method using a waterwheel-type urine flow sensor (Patent Document 4), a method using the temperature of urination and multiple biosensors (Patent Documents 5 and 6), and a method for analyzing the volume of urination using a predictive model based on water volume information and water temperature information of the urinated water (Patent Documents 7 and 8). Although the usefulness of each method has been demonstrated, none have been completed as a system that patients can easily record on a daily basis.

[0005] One simple and routine method for recording urine flow is an application that uses a smartphone to record the sound of urination and reconstruct the urination pattern from its intensity (BE Technologies, MenHealth). This is a groundbreaking method that places almost no burden on the patient. However, because it requires a liquid level where the urine lands, its use is limited to specific locations and situations. In addition, with a weakened urine flow, the sound becomes quieter, which can significantly reduce measurement accuracy. Weakened urine flow is one of the most important abnormalities among voiding / storage disorders, and the reduced accuracy of measuring weakened urine flow is a significant problem.

[0006] The present applicants have also succeeded in developing a method and system for measuring the flow velocity and / or flow rate of urine, which can be easily measured non-contact and has excellent measurement accuracy (Patent Document 9). The measurement method in Patent Document 9 calculates the flow velocity and / or flow rate of urine based on the interval of axis switching in an image of the urine flow. This method is useful for recording urine flow on a daily basis because it can measure the flow velocity and / or flow rate of urine based on an image that has been captured by the user.

[0007] Patent Document 1: Japanese Patent Publication No. 2013-034548 Patent Document 2: Japanese Patent Publication No. 2018-109285 Patent Document 3: Japanese Patent Publication No. 2015-178764 Patent Document 4: Japanese Patent Publication No. 2020-101381 Patent Document 5: Japanese Patent Publication No. 2021-060221 Patent Document 6: Japanese Patent Publication No. 2021-018211 Patent Document 7: Japanese Patent Publication No. 2019-120553 Patent Document 8: Japanese Patent Publication No. 2018-108327 Patent Document 9: Japanese Patent Publication No. 2023-158566 The entire contents of Patent Documents 1 to 9 are incorporated herein by reference.

[0008] In recent years, machine learning with supervised data has been applied and developed in various technological fields. Machine learning with supervised data is a data analysis technique that creates a machine learning model that learns the correlation between data based on supervised data, and then has the machine learning model output predicted or estimated values ​​for unknown data for the input data.

[0009] The applicants believed that machine learning with training data would contribute to improving the efficiency of data analysis and further enhancing measurement accuracy, and attempted to apply machine learning with training data to urine flow measurement as described in Patent Document 9. Through diligent research, the applicants created a machine learning model based on training data consisting of training images of liquid flow in an axis-switching state and the flow velocity and / or flow rate of that liquid flow. They found that by inputting images of urine flow into this machine learning model, it is possible to estimate the unknown flow velocity and / or flow rate of the urine flow in the images.

[0010] However, the applicants have also found that sufficient estimation accuracy may not be obtained by simply machine learning the correlation between training images of liquid flow in an axis-switching state and the flow velocity and / or flow rate of that liquid flow. Hereinafter, an estimation model created by machine learning only the correlation between training images of liquid flow in an axis-switching state and the flow velocity and / or flow rate of that liquid flow will be referred to as the "comparative estimation model".

[0011] The object of the present invention is to provide a method and apparatus that applies machine learning with training data, which can estimate the flow velocity and / or flow rate of urine flow with higher accuracy than estimates based on a comparative estimation model.

[0012] The above problem was solved by creating a machine learning model based on training data that includes training images of the liquid flow in an axis-switching state, the flow velocity and / or flow rate of the liquid flow, and further includes specific features that affect the flow velocity and / or flow rate of the liquid flow. Specifically, the above problem was solved by the invention described in [1] below, and more preferably by the inventions described in [2] and subsequent inventions. [1] A method for estimating the flow velocity and / or flow rate of a urine flow, comprising: preparing a target image which is an image of the urine flow to be measured; and estimating the flow velocity and / or flow rate of the urine flow in the target image based on a machine learning-trained urine flow estimation model to obtain an estimated value of the flow velocity and / or flow rate: wherein the urine flow estimation model is a machine learning model created by machine learning the correlation between data based on training data which includes the following data (a) and data (b), and further includes one or both of data (c) and data (d): (a) a training image of a liquid flow in an axis-switching state; (b) the flow velocity and / or flow rate of the liquid flow; (c) the lengths of the major axis and minor axis of the abdominal cross-section in the axis-switching state or the aspect ratio of the major axis to the minor axis; and (d) the rotation angle around the central axis of the abdomen in the axis-switching state. [2] The method according to [1], wherein the training data includes both data (c) and data (d). [3] The method according to [1] or [2], wherein the estimation is performed using an estimation device that includes an arithmetic processing unit that performs the estimation calculation. [4] The method according to [3], wherein the preparation of the target image is to receive an image of the urine flow transferred from a device other than the estimation device. [5] The method according to [3], wherein the preparation of the target image is to image the urine flow using an imaging unit provided in the estimation device. [6] The method according to any one of [3] to [5], further comprising transferring the estimated value to a display device other than the estimation device. [7] The method according to any one of [3] to [5], further comprising displaying the estimated value on a display unit provided in the estimation device. [8] The method according to any one of [1] to [7], wherein the target image is a binarized image. [9] The method according to any one of [1] to [8], wherein the target image is a video or a still image.

[10] Estimation device for estimating the flow velocity and / or flow rate of a urine flow, comprising: an image preparation unit for preparing a target image which is an image of the urine flow to be measured; and a calculation processing unit for estimating the flow velocity and / or flow rate of the urine flow in the target image and obtaining an estimated value of the flow velocity and / or flow rate based on a machine learning-trained urine flow estimation model: wherein the urine flow estimation model is a machine learning model created by machine learning the correlation between data based on training data which includes the following data (a) and data (b), and further includes one or both of data (c) and data (d): (a) a training image of a liquid flow in an axis switching state; (b) the flow velocity and / or flow rate of the liquid flow; (c) the lengths of the major axis and minor axis of the abdominal cross-section in the axis switching or the aspect ratio of the major axis to the minor axis; and (d) the rotation angle around the central axis of the abdomen in the axis switching.

[11] Estimation device according to

[10] , wherein the training data includes both data (c) and data (d).

[12] The estimation device according to

[10] or

[11] , wherein the image preparation unit is a communication unit that receives images of urine flow transferred from a device other than the estimation device.

[13] The estimation device according to

[10] or

[11] , wherein the image preparation unit is an imaging unit that images urine flow.

[14] The estimation device according to any one of

[10] to

[13] , further comprising a communication unit that transfers the estimated value to a display device other than the estimation device.

[15] The estimation device according to any one of

[10] to

[13] , further comprising a display unit that displays the estimated value.

[16] The estimation device according to

[10] , wherein the image preparation unit is a communication unit that receives images of urine flow transferred from a device other than the estimation device, and the communication unit has a function of transferring the estimated value to a display device other than the estimation device.

[17] The estimation device according to

[16] , wherein the estimation device is a cloud computing type device.

[18] The estimation device according to

[10] , wherein the image preparation unit is an imaging unit for imaging urine flow, and the estimation device further includes a display unit for displaying the estimated value.

[19] The estimation device according to

[18] , wherein the estimation device is a mobile device type device.

[20] The estimation device further includes an image processing unit that binarizes the target image, and the calculation processing unit performs the estimation calculation based on the binarized target image, according to any one of

[10] to

[19] .

[0013] The urine flow estimation method and apparatus of the present invention enable the application of machine learning with training data to urine flow estimation, and allow for the estimation of urine flow velocity and / or flow rate with higher accuracy compared to estimates based on comparative estimation models.

[0014] Figure 1(a) is a simulated image of urine flow, showing the liquid flow discharged from a nozzle modeled after the human urethra. Figure 1(b) is a binarized image of the image in Figure 1(a). Figure 2(a) is a simulated image of the liquid flow obtained from a computational fluid dynamics simulation. Figure 2(b) is a cross-sectional view of the abdomen in the A-A' section shown in Figure 2(a). Figure 3 shows the line of sight direction d in the image in Figure 2(a). ｓ This shows the rotation of the liquid flow around the central axis of the abdomen a2 relative to the given point. Figure 4 is a conceptual diagram showing the dataset that constitutes the training data. Figure 5 is a block diagram showing one embodiment of the estimation device of the present invention. Figure 6 is a block diagram showing another embodiment of the estimation device of the present invention. Figure 7 shows the analysis system in computational fluid dynamics simulation. Figure 8 shows the flow rate estimation results in Example 1. Figure 9 shows the aspect ratio estimation results in Example 1. Figure 10 shows the rotation angle estimation results around the axis in Example 1. Figure 11 shows the flow rate estimation results in Example 1 and the comparative example. Figure 12 shows the flow rate estimation results in Examples 2 and 3.

[0015] [Method for estimating the velocity and / or flow rate of urine flow] The method of the present invention involves inputting an image showing the urine flow of a target object whose velocity and flow rate are unknown (target image) into a machine learning model for urine flow estimation (urine flow estimation model) created based on specific training data, and estimating the velocity and / or flow rate of the urine flow in the target image based on the urine flow estimation model. In other words, the present invention is a method for estimating the flow velocity and / or flow rate of a urine flow, comprising: preparing a target image which is an image of the urine flow to be measured; and estimating the flow velocity and / or flow rate of the urine flow in the target image based on a machine learning-trained urine flow estimation model to obtain an estimated value of the flow velocity and / or flow rate, wherein the urine flow estimation model is a machine learning model created by machine learning the correlation between data based on training data which includes the following data (a) and data (b), and further includes one or both of data (c) and data (d): (a) a training image of a liquid flow in an axis-switching state; (b) the flow velocity and / or flow rate of the liquid flow; (c) the lengths of the major axis and minor axis of the abdominal cross-section in axis switching, or the aspect ratio of the major axis to the minor axis; and (d) the rotation angle around the central axis of the abdomen in axis switching.

[0016] The flow velocity and / or flow rate of urine in the target image are estimated using an estimation device, described later, which includes a processing unit that performs the estimation calculations.

[0017] The estimation method of the present invention will now be described in detail. The estimation method of the present invention estimates the flow velocity and / or flow rate of urine based on the characteristics of axis switching of urine flow. "Axis switching" is a phenomenon in which, when liquid flows out of an elliptical or rectangular outlet (slit), the major and minor axes of the elliptical cross-section of the liquid column alternate along the flow direction, and a jet with a twisted rod shape is observed (see L. Rayleigh, "On the Capillary Phenomena of Jets", Proc. R. Soc. London, 29, 71-97, (1879)). Since the shape of the outlet at the tip of the human urethra during discharge is generally elliptical, axis switching also occurs in the urine flow during discharge. A liquid column in an axis-switched state has a belly (a flattened section in cross-section) between the outlet and the next node (twisted position), and between two adjacent nodes, with the long axes of the cross-sections of the two adjacent bellys offset from each other by 90° around the central axis. The "central axis" refers to the axis that passes through the center of the liquid flow along the flow.

[0018] The estimation method of the present invention includes preparing a target image, which is an image of the urine flow to be measured.

[0019] The target image is an image of the patient's urine flow. Figure 1(a) is a simulated image of urine flow showing the liquid flow discharged from a nozzle that mimics the human urethra. The image in Figure 1(a) reproduces the liquid flow 1 flowing out of a nozzle equipped with an elongated elliptical outlet 2a using computational fluid dynamics simulation. The simulated image in Figure 1(a) was obtained by imaging the urine flow from the side in the horizontal direction, but the imaging direction of the target image is not particularly limited. That is, the target image may be obtained by imaging from above or below in the vertical direction relative to the urine flow, or by imaging from diagonally above or diagonally below in the vertical direction relative to the urine flow. As will be described later, in the present invention, it is also possible to estimate urine flow while considering the imaging direction of the urine flow, and with such an estimation method, it is possible to estimate urine flow with high accuracy regardless of the imaging direction.

[0020] The target image preferably includes one or more abdominal sections close to the outlet, as shown in the simulated image in Figure 1(a). Since the urine flow is stronger in the abdominal section closer to the outlet, the changes in the long and short axes of the urine flow cross-section are clearer, and the shape of the urine flow tends to be more stable. Therefore, by including abdominal sections closer to the outlet in the target image, the shape characteristics of the urine flow can be grasped more accurately, and the flow velocity and / or flow rate of the urine flow can be estimated with higher accuracy. To further improve the accuracy of urine flow estimation, the target image preferably includes one or both of the first abdominal section a1 (the portion from the outlet 2a to the first section n1) and the second abdominal section a2 (the portion from the first section n1 to the second section n2). This is because the urine flow is strongest and the shape of the urine flow is most stable in the range from the first abdominal section a1 to the second abdominal section a2. However, since urine flow estimation is possible as long as any abdominal section in the axis switching of the urine flow is included in the target image, the target image does not necessarily have to include the first abdominal section a1 and the second abdominal section a2.

[0021] The target image preferably includes the tip of the urethra, particularly the outlet. The lengths of the major and minor axes of the outlet correspond to the lengths of the major and minor axes of the initial abdomen a1, respectively. Therefore, the lengths of the major and minor axes of the outlet shown in the target image can be used as the lengths of the major and minor axes of the initial abdomen a1 in the fluid flow in the target image. However, since urine flow estimation is possible if axis switching of the urine flow can be confirmed, the target image does not necessarily have to include the tip of the urethra and the outlet.

[0022] The method for preparing the target image is not particularly limited. The target image may be prepared by receiving an image transferred from another device, or by imaging the urine flow in situ before estimation.

[0023] In one aspect of the estimation method of the present invention, the preparation of the target image involves receiving an image of urine flow transmitted from a device other than the estimation device. In this aspect, the estimation device receives the image transmitted from the external device in its communication unit, sends the received image to a processing unit, where the estimation calculation is performed. The estimation device and the device that transmits the image are separate, and an image captured by another device located in a different location from the estimation device can be used as the target image. Communication between devices can be carried out via communication means such as wired LAN, wireless LAN, and Bluetooth®. Furthermore, if the distance between the source and destination is large, communication between devices can also be carried out via the Internet.

[0024] In another aspect of the estimation method of the present invention, the preparation of the target image involves imaging the urine flow using an imaging unit provided in the estimation device. Since the estimation device includes an arithmetic processing unit that performs estimation calculations using a urine flow estimation model, in this aspect, the arithmetic processing unit and the imaging unit are provided in a single device. Therefore, the target image can be prepared on-site and offline based on the captured image.

[0025] The target image can be a video or a still image, and the still image can be a single still image or two or more consecutive still images. If the target image is a single still image, one estimate can be obtained for this still image. If the target image is a series of consecutive still images, an estimate can be obtained for each still image. If the target image is a video, an estimate can be obtained for each frame that makes up the video, or for each of the frames extracted at predetermined intervals. If a video or a series of consecutive still images is used as the target image and multiple estimates are obtained, the average of these estimates can be used as the output estimate. Alternatively, if a video or a series of consecutive still images is used as the target image, the time change of urine flow velocity and / or flow rate can be determined by arranging the multiple estimates in a time series. The time change of urine flow velocity and / or flow rate can be used to diagnose voiding / storage disorders.

[0026] The target image may be a binarized image, as shown in Figure 1(b). Image binarization can be performed by known methods.

[0027] The estimation method of the present invention includes estimating the flow velocity and / or flow rate of urine in a target image based on a machine learning-prepared urine flow estimation model to obtain estimated values ​​of flow velocity and / or flow rate. The urine flow estimation model is a machine learning model created by machine learning the correlation between data, based on training data that includes the aforementioned data (a) and data (b), and further includes one or both of data (c) and data (d). The machine learning method is not particularly limited, and known methods such as deep learning, neural networks, support vector machines, and random forests can be used.

[0028] The urine flow estimation model is a machine learning model that estimates the velocity and / or flow rate of urine based on a target image showing an unknown urine flow. For each input of a target image, it outputs estimated values ​​of the urine flow velocity and / or flow rate in the target image. In addition to the estimated velocity and / or flow rate, the urine flow estimation model may also output estimated values ​​of features related to data (c) and / or (d).

[0029] The training data for performing machine learning includes data (a) and data (b) above, and also includes either or both of data (c) and data (d). That is, the training data may include any of the following datasets (i) to (iii): (i) data (a) to data (d), (ii) data (a) to data (c), and (iii) data (a), data (b), and data (d).

[0030] Data (b) through Data (d) are labeled to the training image of Data (a). To further improve the accuracy of urine flow estimation, it is preferable that the training data includes both Data (c) and Data (d), i.e., Dataset (i). Figure 4 shows an example of training data including Dataset (a) through Data (d).

[0031] The following explains data (a) through (d).

[0032] (a) The learning image data (a) of the liquid flow in the axis switching state is an image showing a liquid flow with known flow velocity and flow rate, and gives the external characteristics of the liquid flow in the axis switching state to the urine flow estimation model. The learning image is, for example, an image of a urine flow whose flow velocity and flow rate have been measured by a conventional method. Alternatively, the learning image may be an image obtained by imaging a liquid flow (water flow or pseudo-urine flow) flowing out from a nozzle having an oval or rectangular outlet at a predetermined flow velocity and flow rate, or may be a simulation image obtained by numerical fluid dynamics simulation. In the case of a simulation image, it is appropriate to select the liquid conditions in consideration of physical properties such as the viscosity of urine. The learning image may be a binary image.

[0033] The urine flow estimation model grasps the external characteristics of the liquid flow in the axis switching state based on feature amounts such as contrast in the learning image, and machine-learns the correlation between the external characteristics and the flow velocity and / or flow rate of the urine flow. In the present invention, the interval of axis switching may be used as one of the features in the internal processing of machine learning, but it is not necessary to execute an individual process for specifically calculating the interval as in Patent Document 9.

[0034] The imaging direction of the learning image is not particularly limited. That is, the learning image may be obtained by imaging from the horizontal direction with respect to the liquid flow, may be obtained by imaging from the upper side or the lower side in the vertical direction with respect to the liquid flow, or may be obtained by imaging from the upper diagonal side or the lower diagonal side in the vertical direction with respect to the liquid flow. From the viewpoint of further improving the estimation accuracy, it is preferable to prepare images taken from various directions without being biased towards images taken from a specific direction. However, when the teacher data includes the data (d) described later, learning images taken from various directions are prepared.

[0035] The training image preferably includes the liquid flow portion (urine flow portion) at a position corresponding to the position of the liquid flow portion (urine flow portion) that is expected to be included in the target image. The long axis length of the abdominal portion of the liquid flow decreases with decreasing urine flow as it is further from the outlet. By having the positions of these liquid flow portions in the flow direction correspond to each other (i.e., the distance from the outlet is approximately equal), the degree of attenuation of the long axis length can be made equal, and the flow velocity and / or flow rate of the urine flow can be estimated with higher accuracy. In the present invention, since the target image preferably includes one or more abdominal portions close to the outlet, the training image also preferably includes one or more abdominal portions close to the outlet. Furthermore, similar to the target image, the training image preferably includes one or both of the first abdominal portion a1 (the portion from the outlet 2a to the first node n1) and the second abdominal portion a2 (the portion from the first node n1 to the second node n2) (see Figure 1(a)). However, since urine flow estimation is possible as long as any of the abdominal regions in the axis switching of the liquid flow are included in the training images, the training images do not need to include the first abdominal region a1 and the second abdominal region a2.

[0036] The training image preferably includes the tip of the urethra or nozzle, particularly the outlet, and the lengths of the major and minor axes of the outlet correspond to the lengths of the major and minor axes of the initial abdomen a1, respectively. Therefore, the lengths of the major and minor axes of the outlet shown in the training image can be used as the lengths of the major and minor axes of the initial abdomen a1 in the fluid flow in the training image. However, since urine flow estimation is possible if axis switching of the fluid flow can be confirmed, the training image does not necessarily have to include the tip of the urethra or nozzle and the outlet.

[0037] (b) The flow velocity and / or flow rate data (b) of the liquid flow in the learning image is the actual flow velocity and / or flow rate of the liquid flow in the learning image. When the learning image is an image of a liquid flow (urine flow) whose flow velocity and flow rate have been measured by a conventional method, the data (b) is the measured value. When the learning image is a simulation image, the data (b) is a simulation value of the flow velocity and / or flow rate calculated by numerical calculation. The flow velocity is the flow distance of the liquid per unit time and is expressed in units such as mm / s, cm / s, etc. The flow rate is the volume of the liquid flowing per unit time and is expressed in units such as cm ３ / s, mL / s, etc. The flow velocity v (cm / s) and the flow rate Q (mL / s) have a relationship of Q = v · S, where S (cm ２ ) is the opening area of the outlet. Since the maximum flow rate of human urine flow is about 20 - 25 mL / s, the flow rate as the data (b) is preferably prepared, for example, in the range of 30 mL / s or less or 2 - 25 mL / s.

[0038] (c) The lengths of the long axis and the short axis or the aspect ratio of the long axis to the short axis of the abdominal cross-section in axis switching Fig. 2(a) is a simulation image of the liquid flow obtained by numerical fluid dynamics simulation. Fig. 2(a) shows a liquid flow 1 flowing out from a nozzle 2 having an elliptical outlet 2a. The liquid flow portion in the range from the outlet 2a to the first node n1 is the first abdomen a1, and the liquid flow portion in the range from the first node n1 to the second node n2 is the second abdomen a2. However, the long axes of two adjacent abdomens are shifted by 90° around the central axis from each other, and there are also overlapping portions of the skirts of both abdomens. The distance D from the node n1 to the node n2 is the interval of axis switching. The A - A' cross-section is a cross-section perpendicular to the central axis near the center of the abdomen a2, and within the range of a single abdomen, the long axis of the elliptical cross-section is the maximum at this position.

[0039] Fig. 2(b) is a cross-sectional view of the abdomen in the A - A' cross-section shown in Fig. 2(a). In Fig. 2(b), d ｍａｘ represents the length of the long axis of the elliptical cross-section as the data (c), and d ｍｉｎThis represents the length of the minor axis of the elongated elliptical cross-section as data (c). The aspect ratio α of the elongated elliptical cross-section as data (c) is d ｍａｘ / d ｍｉｎ It is defined as follows.

[0040] The location where the abdominal cross-section is taken is not particularly limited, as long as it is a location where the characteristics of the fluid flow in the axis-switching state are evident. To further improve the accuracy of urine flow estimation, it is preferable that the abdominal cross-section be taken at a location where the long axis is large in a range where the urine flow is relatively strong, for example, near the center of the second abdomen a2 as shown in Figure 2(a), or near the outlet of the first abdomen a1. In addition, since the lengths of the long axis and short axis of the outlet are the same as the lengths of the maximum long axis and minimum short axis of the first abdomen a1, respectively, if the lengths of the long axis and short axis of the outlet are known, these known lengths may be used as the lengths of the long axis and short axis of the first abdomen a1, and as data (c).

[0041] Data (c) is a parameter used in urine flow estimation to account for the widening of the human urethral outlet in proportion to the flow rate of urine.

[0042] Patent Document 9 calculates the velocity and / or flow rate of urine based solely on the interval of axis switching in an image of urine flow (i.e., the one-dimensional length in the flow direction). However, the elongated elliptical outlet of the human urethra (especially the width along the short axis) tends to widen depending on the flow rate of urine, and the flow rate may not be accurately estimated based solely on the interval of axis switching. For example, if the short axis length of the abdomen increases due to the widening of the outlet, the flow rate also increases, but the interval of axis switching hardly changes. This means that the flow rate is not uniquely determined in relation to the interval of axis switching.

[0043] Therefore, in one aspect of the estimation method of the present invention, the urine flow estimation model learns the correlation between data based on teacher data including the lengths of the long axis and the short axis of the abdominal cross-section in the axis switching of the liquid flow. The influence of the spread of the outlet appears in the lengths of the long axis and the short axis of the abdominal cross-section. Therefore, by including the lengths of the long axis and the short axis of the abdominal cross-section as features for machine learning, it is possible to perform more accurate urine flow estimation considering the influence of the spread of the outlet.

[0044] The length of the long axis as data (c) is preferably in the range of, for example, 5 to 15 mm or 7 to 13 mm, and the length of the short axis as data (c) is preferably in the range of, for example, 0.1 to 5 mm or 1 to 3 mm. The teacher data can include information on the lengths of the long axis and the short axis of the abdominal cross-section as data (c) in the form of an aspect ratio. The aspect ratio as data (c) is preferably in the range of, for example, 2 to 9 or 3 to 7. In order to obtain higher estimation accuracy, data (c) is preferably a set of uniform values at a predetermined interval (for example, 0.1, 0.2 or 0.5 mm) within the above range without being biased towards a specific range.

[0045] (d) Rotation angle around the central axis of the abdomen in axis switching The rotation angle around the central axis of the abdomen means the rotation angle around the central axis of the long axis of the cross-section of the abdomen. FIG. 3 shows the state of rotation of the liquid flow around the central axis of the abdominal part a2 with respect to the line-of-sight direction d in the image of FIG. 2(a). Focusing on one abdominal part a2, as shown in FIG. 3, the angle θ formed by the line-of-sight direction d ｓ and the long axis can be within the range of 0 to 180°. In FIG. 3, the line-of-sight direction d in the image ｓ is used as a reference, but the reference for the rotation angle is not limited to this. The reference for the rotation angle may be, for example, the upper side of the vertical or the lower side of the vertical. ｓ is used as a reference, but the reference for the rotation angle is not limited to this. The reference for the rotation angle may be, for example, the upper side of the vertical or the lower side of the vertical.

[0046] The long axis of the abdominal cross-section has a specific relationship of being in the same direction as the long axis of the outlet of the urethra or the nozzle or a direction rotated 90° from that direction. Therefore, when the outlet of the urethra or the nozzle is included in the learning image, the rotation angle around the central axis of the long axis of the outlet may be adopted as the rotation angle of data (d).

[0047] Data (d) is a parameter used in urine flow estimation to account for the fact that the appearance of the axis switching of the fluid flow appears different depending on the observation direction around the central axis.

[0048] The two adjacent abdominal segments are rotated 90 degrees relative to each other around the central axis, and their bases overlap near the segments during axis switching. Therefore, depending on the viewing angle of the liquid flow, the position of the segments may be difficult to determine, and it may be difficult to uniquely extract them from the image.

[0049] For example, let's compare the case where the angle between the line of sight and the long axis of the abdominal cross-section is 0° (Figures 2(a) and 3(a)) with the case where the angle is 45° (Figure 3(b)). When the angle is 0° (Figures 2(a) and 3(a)), the abdomen of interest a2 is observed from a direction parallel to the long axis, and the abdomen adjacent to abdomen a2 (e.g., abdomen a1) is observed from a direction directly to the side of the long axis. In this situation, where their bases overlap, the base of abdomen a2 is difficult to see, while the base of the adjacent abdomen is easy to see. As a result, the width of abdomen a2 may appear shorter than it actually is, and the width of the adjacent abdomen may appear longer than it actually is. On the other hand, when the angle is 45° (Figure 3(b)), both the abdomen of interest a2 and the abdomen adjacent to abdomen a2 are observed from a similar oblique direction along the long axis. In this situation, where the bases of the two flanks overlap, it is easier to confirm the relative sizes of the flank bases, thus making it easier to more accurately determine the width of the flank a2. Thus, when a single fluid flow is observed from different directions around the central axis, the morphology of axis switching may be determined to be different from one another, and different estimated values ​​may be obtained for each observation. Obtaining different estimated values ​​for a single fluid flow leads to a decrease in the accuracy of urine flow estimation.

[0050] Therefore, in one aspect of the estimation method of the present invention, the urine flow estimation model learns the correlation between data based on training data that includes the rotation angle around the central axis of the abdomen during axis switching. By including the rotation angle around the central axis of the abdomen as a feature of machine learning in this way, it is possible to perform urine flow estimation with higher accuracy that takes into account how the form of axis switching appears depending on the observation direction.

[0051] As described above, the estimation method of the present invention can perform urine flow estimation with higher accuracy compared to urine flow estimation based on comparative estimation models by using a urine flow estimation model created based on training data that includes one or both of the data (c) and (d) as further machine learning features in addition to the data (a) and (b).

[0052] The estimation method of the present invention may further include transferring the obtained estimated values ​​of flow velocity and / or flow rate to a display device other than the estimation device. In this case, the estimation method of the present invention can display the estimated values ​​on another device located in a different location from the estimation device. Communication between devices can be carried out via communication means such as wired LAN, wireless LAN, and Bluetooth®. Furthermore, if the distance between the source and destination is large, communication between devices can also be carried out via the Internet.

[0053] Alternatively, the estimation method of the present invention may further include displaying the obtained estimated values ​​of flow velocity and / or flow rate on a display unit provided in the estimation device. In this case, the estimation method of the present invention can display the estimated values ​​on the display unit provided in the estimation device on the spot and offline.

[0054] The format in which the estimated values ​​are displayed is not particularly limited; for example, the values ​​can be displayed as they are. If the target image is a video or a series of still images, the estimated values ​​can be displayed as a graph showing the time-dependent changes in flow velocity and / or flow rate.

[0055] [Device for Estimating Urine Flow Velocity and / or Flow Rate] The estimation device of the present invention is a device for estimating the flow velocity and / or flow rate of urine by performing the estimation method described above. In other words, the estimation device of the present invention is an estimation device for estimating the flow velocity and / or flow rate of a urine flow, and includes an image preparation unit that prepares a target image which is an image of the urine flow to be measured, and a calculation processing unit that estimates the flow velocity and / or flow rate of the urine flow in the target image and obtains an estimated value of the flow velocity and / or flow rate based on a machine learning-trained urine flow estimation model, wherein the urine flow estimation model is a machine learning model created by machine learning the correlation between data based on training data which includes the following data (a) and data (b), and further includes one or both of data (c) and data (d): (a) a learning image of the liquid flow in an axis switching state, (b) the flow velocity and / or flow rate of the liquid flow, (c) the lengths of the major axis and minor axis of the abdominal cross-section in the axis switching or the aspect ratio of the major axis to the minor axis, and (d) the rotation angle around the central axis of the abdomen in the axis switching.

[0056] In one embodiment of the estimation device of the present invention, the image preparation unit is a communication unit that receives images of urine flow transferred from a device other than the estimation device. That is, the estimation device includes a calculation processing unit and a communication unit. The received image may be transmitted directly to the calculation processing unit and used for estimation calculations. Alternatively, the received image may be temporarily stored in a storage unit within the estimation device and, if necessary, transmitted to the calculation processing unit and used for estimation calculations. In this embodiment, an image captured by another device located at a different location from the estimation device can be used as the target image. This communication unit may have the function of transferring estimated values ​​to a display device other than the estimation device, and may also function as a communication unit that transfers estimated values ​​as described later.

[0057] For example, the estimation device, including the processing unit and communication unit, is a cloud computer such as a cloud server provided by a medical institution, and the device that transfers images is an imaging device or cloud storage used by medical personnel. In this case, images created by medical personnel using an imaging device or images stored in cloud storage are transferred to the cloud computer, and the cloud computer performs uroflow estimation based on the received images. This method is useful from the standpoint of efficiently managing large amounts of data. Furthermore, for example, the estimation device, including the processing unit and communication unit, is the patient's personal computer (PC), and the device that transfers images is the patient's mobile device or an imaging device fixed to a wall or table. In this case, images created by the patient with a camera built into the mobile device are transferred to the patient's PC, and the PC performs uroflow estimation based on the received images. This method is useful for patients to perform uroflow estimation on a daily basis to monitor the progress of their condition and manage their physical health.

[0058] In another embodiment of the estimation device of the present invention, the image preparation unit is an imaging unit that images urine flow. That is, the estimation device includes a calculation processing unit and an imaging unit. The captured image may be transmitted directly to the calculation processing unit and used for estimation calculations. Alternatively, the captured image may be temporarily stored in a memory unit within the estimation device and, as needed, transmitted to the calculation processing unit and used for estimation calculations. In this embodiment, target images can be prepared on the spot and offline, which is useful for patients to perform urine flow estimation on a daily basis to monitor the progress of their medical condition and manage their physical health.

[0059] For example, the estimation device, which includes a processing unit and an imaging unit, is a mobile device such as the patient's smart glasses, smartphone, or smartwatch, and the imaging unit provided in the estimation device is a camera built into the mobile device. In this case, the patient can perform the imaging of the urine flow themselves using the mobile device. The obtained image is stored, for example, in the mobile device's storage unit (internal storage or external storage, etc.) or transmitted directly to the processing unit. In this embodiment of the present invention, it is possible to use a generally available mobile device, and the patient can easily generate the target image themselves.

[0060] Alternatively, for example, the estimation device including the arithmetic processing unit and the imaging unit is a medical device handled by medical professionals, and the imaging unit provided in the estimation device is an imaging device provided in the medical device. In this case, urinary flow imaging can be performed by medical professionals such as doctors, nurses, and medical technicians using the imaging device to image the patient's urinary flow. The obtained images are stored, for example, in the storage unit of the medical device (internal storage or external storage, etc.) or transmitted directly to the arithmetic processing unit. In this embodiment of the present invention, by having medical professionals perform urinary flow imaging, more accurate measurements can be made under more stringent conditions compared to when the patient themselves performs the imaging.

[0061] The arithmetic processing unit estimates the flow velocity and / or flow rate of the urine flow in the target image based on a machine learning-prepared urine flow estimation model to obtain estimated values ​​of flow velocity and / or flow rate. The urine flow estimation model may be stored in the arithmetic processing unit or in another memory unit of the estimation device. The urine flow estimation model is machine learning-prepared based on specific training data. The training data is the same as described in the estimation method of the present invention.

[0062] The estimation device of the present invention may further include a communication unit that transfers the obtained estimated values ​​of flow velocity and / or flow rate to a display device other than the estimation device. For example, the estimation device including the communication unit may be a cloud server provided by a medical institution, and the display device other than the estimation device may be a medical device with a built-in display used by medical personnel or a mobile device with a built-in display used by a patient. In this case, data processing that handles large amounts of data, such as the calculation processing for estimating flow velocity and / or flow rate, is performed on the cloud server, and only the estimated values ​​obtained as a result can be displayed on the display of the medical device or mobile device. The communication unit that transfers the estimated values ​​and the aforementioned communication unit that receives images as an image preparation unit may be a bidirectional communication unit having a transmission and reception function configured as an integrated unit.

[0063] Alternatively, the estimation device of the present invention may further include a display unit that displays the obtained estimated values ​​of flow velocity and / or flow rate, in addition to a calculation processing unit that performs calculations for estimating urine flow. For example, an estimation device including a calculation processing unit and a display unit may be a medical device equipped with a calculation processing unit and a display, or a mobile device equipped with a calculation processing unit and a display. In this case, the estimated values ​​can be displayed on-site and offline.

[0064] The estimation device of the present invention may further include an image processing unit that binarizes a target image. In this case, the calculation processing unit can perform estimation calculations based on the binarized target image.

[0065] The estimation device of the present invention may further include a model creation unit that creates a urine flow estimation model based on training data. The training data can be updated as needed, and the model creation unit can update the urine flow estimation model by performing machine learning based on the updated training data.

[0066] The estimation apparatus of the present invention will be described below with reference to specific embodiments. However, the embodiments described below are only a part of the present invention, and the estimation apparatus of the present invention is not limited to the embodiments described below.

[0067] <First Embodiment> The estimation device of the first embodiment is an estimation device in which the image preparation unit is a communication unit that receives images of urine flow transferred from a device other than the estimation device, and the communication unit has the function of transferring estimated values ​​to a display device other than the estimation device. Such a device is configured, for example, as a cloud computing type device.

[0068] Figure 5 is a block diagram of a cloud computing-type estimation device 10 and related devices. Reference numeral 20 denotes a patient's mobile device such as a smartphone, and reference numeral 30 denotes a medical device used by medical personnel. The estimation device 10, the mobile device 20, and the medical device 30 are each equipped with control devices such as a CPU and GPU (not shown), and the operation of each device is controlled by its respective control device.

[0069] The estimation device 10 includes, for example, a calculation processing unit 11, a communication unit 12, a model creation unit 13, an image processing unit 14, and a storage unit 15. The calculation processing unit 11 can perform calculations to estimate the flow velocity and / or flow rate of urine in a target image based on a urine flow estimation model. The communication unit 12 can exchange electrical signals with other devices via a telecommunication line (bidirectional). The model creation unit 13 can create a urine flow estimation model based on training data. The image processing unit 14 can perform image processing to binarize the target image. The storage unit 15 can store data such as training data, the urine flow estimation model, and the target image.

[0070] The mobile device 20 includes, for example, a communication unit 21, an imaging unit 22, and a display unit 23. The communication unit 21 can exchange electrical signals with other devices via a telecommunication line (bidirectional). The imaging unit 22 is, for example, a built-in camera that can capture images of urine flow and generate images. The display unit 23 is, for example, a built-in display that can display information such as estimated values ​​of urine flow velocity and / or flow rate.

[0071] The medical device 30 includes, for example, a communication unit 31 and a display unit 32. The communication unit 31 can exchange electrical signals with other devices via a telecommunication line (bidirectional). The display unit 32 is, for example, a desktop display and can display information such as estimated values ​​of urine flow velocity and / or flow rate.

[0072] This explains how these devices work.

[0073] The model creation unit 13 retrieves the training data stored in the memory unit 15, performs machine learning based on the training data, and creates or updates a urine flow estimation model. The urine flow estimation model is stored in the memory unit 15.

[0074] The imaging unit 22 captures the urine flow through the patient's actions, generating an image of the urine flow. The urine flow image is stored in the storage unit 15 via the communication units 21 and 12. The image processing unit 14 retrieves the urine flow image stored in the storage unit 15, binarizes the image, and generates a binarized image of the urine flow. The binarized image of the urine flow is stored in the storage unit 15. Image binarization is optional.

[0075] The arithmetic processing unit 11 retrieves the urine flow estimation model and the binarized image stored in the storage unit 15, inputs the binarized image into the urine flow estimation model, and performs calculations to estimate the flow velocity and / or flow rate of the urine flow shown in the binarized image. The obtained estimated values ​​of the urine flow velocity and / or flow rate are stored in the storage unit 15.

[0076] Upon request from the medical device 30, the estimation device 10 transfers the estimated values ​​stored in the storage unit 15 to the medical device 30 via the communication units 12 and 31, and the medical device 30 receives them. The medical device 30 displays the received estimated values ​​on the display unit 32.

[0077] Furthermore, the estimation device 10 can also transfer the estimated values ​​stored in the storage unit 15 to the mobile device 20 via the communication units 12 and 21 upon request from the mobile device 20. The mobile device 20 can receive these values ​​and display the received estimated values ​​on the display unit 23.

[0078] <Second Embodiment> The estimation device of the second embodiment is an estimation device in which the image preparation unit is an imaging unit that images urine flow, and the estimation device further includes a display unit that displays the estimated value. Such a standalone device can perform image preparation, urine flow estimation, and display of the estimated value in a single device. The estimation device of the second embodiment may be a medical device used by medical professionals, or it may be a mobile device used by patients.

[0079] Figure 6 is a block diagram of a mobile device-type estimation device 40 used by the patient. The estimation device 40 is equipped with control devices such as a CPU and GPU (not shown), and the operation of the device is controlled by the control devices.

[0080] The estimation device 40 includes, for example, a calculation processing unit 41, a model creation unit 43, an image processing unit 44, a storage unit 45, an imaging unit 46, and a display unit 47. The calculation processing unit 41, the model creation unit 43, the image processing unit 44, and the storage unit 45 are the same as those in the estimation device 10 in the first embodiment. The imaging unit 46 is, for example, a built-in camera that can capture images of urine flow and generate images. The display unit 47 is, for example, a built-in display that can display information such as estimated values ​​of urine flow velocity and / or flow rate.

[0081] This explains how this device works.

[0082] The model creation unit 43 creates or updates a urine flow estimation model, similar to the first embodiment, and the urine flow estimation model is stored in the storage unit 45.

[0083] The imaging unit 46 captures the urine flow through the patient's actions, generating an image of the urine flow. The image of the urine flow is stored in the storage unit 45. The image processing unit 44 retrieves the image of the urine flow stored in the storage unit 45, binarizes the image, and generates a binarized image of the urine flow. The binarized image of the urine flow is stored in the storage unit 45. Image binarization is optional.

[0084] The arithmetic processing unit 41 retrieves the urine flow estimation model and the binarized image stored in the storage unit 45, inputs the binarized image into the urine flow estimation model, and performs calculations to estimate the flow velocity and / or flow rate of the urine flow shown in the binarized image. The obtained estimated values ​​of the urine flow velocity and / or flow rate are stored in the storage unit 45.

[0085] The display unit 47 displays the estimated value stored in the storage unit 45.

[0086] <Other Embodiments> In the first embodiment, the imaging device was a mobile device, but the imaging device may also be a medical imaging device.

[0087] In the first embodiment, the imaging device and the display device were separate, but a single device may be configured to perform both imaging of the urine flow and display of the estimated value. For example, the medical device 30 may be equipped with an imaging unit and configured to transfer the image generated by the medical device 30 to the estimation device 10.

[0088] In the second embodiment, a mobile device was given as an example, but urine flow estimation can also be performed using a medical device in the same way.

[0089] The present invention will be described in more detail below with reference to examples. However, the scope of the present invention is not limited to the specific examples shown below.

[0090] <Example 1> 1. Creation of training images 1295 images were created using computational fluid dynamics simulation. Among the input conditions, the flow rate, the aspect ratio of the major and minor axes of the nozzle outlet, and the angle of the major axis of the nozzle outlet were changed within the ranges of 2.4 to 22.0 mL / s, 3 to 7 (1 interval), and 0 to 180° (5° interval), respectively, to set various conditions. As described above, in this example, the aspect ratio of the major and minor axes of the nozzle outlet was used as the aspect ratio of the major and minor axes of the first abdomen, and the rotation angle around the central axis of the major axis of the nozzle outlet was used as the rotation angle around the central axis of the first abdomen. The detailed conditions and execution method of the simulation are as follows.

[0091] 1-1. Diagram 7 of the analysis system shows a schematic diagram of the analysis system. The coordinate system is the Cartesian coordinate system, with the base being the zx plane (depth direction is the x-axis) and the height direction being the y-axis. A rectangular domain space with a width of 15 cm, a depth of 1.6 cm, and a height of 2 cm was set as the calculation domain. The domain space is air (density ρ ｃ and viscosity μ ｃ It was assumed that the domain space was filled with ρ. A cylinder 10 mm long and 10 mm in diameter, simulating a nozzle, was placed on the left side of the domain space such that the center of the cylinder was 1.4 cm from the bottom of the domain space and 0.8 cm from the front. An elongated elliptical slit was provided on the right side of the cylinder as a liquid outlet, and a flow path of the same shape as the slit was provided inside the cylinder. To simulate the widening of the outlet of the human urethra, the major and minor axes of the slit were adjusted as appropriate within a range where the aspect ratio was 3 to 7. In addition, to simulate the rotation around the axis of the urethral outlet, the angle of the major axis of the slit with respect to the x axis was changed within a range of 0 to 180°. At time 0, the flow path inside the cylinder had a density ρ ｄ and viscosity μ ｄ The device is filled with a liquid simulating urine, and a velocity v is applied to the inlet of the channel (the surface z=0). ｄ A uniform flow was given. Gravity was set to zero.

[0092] Air and liquids were assumed to be incompressible Newtonian fluids. By considering these fluids as a single fluid with different physical properties (single-fluid approximation) and describing the governing equations, the Navier-Stokes equations corresponding to the fluid's equations of motion and the continuity equation corresponding to the law of conservation of mass are given by equations 1 and 2, respectively.

[0093]

[0094] Here, t is time, ρ is density, u is velocity vector, P is pressure, g is gravitational acceleration vector, μ is viscosity, σ is surface tension coefficient, κ is curvature, and F is the VOF function (volume fraction of the liquid phase).

[0095] The boundary conditions were set as follows: Excluding the area where the nozzle is located, the boundaries of the left (z=0 cm), right (z=15 cm), top (y=2 cm), bottom (y=0 cm), front (x=1.6 cm), and back (x=0 cm) surfaces of the domain space were given an open condition to allow the outflow of air or liquid equal to the volume of liquid introduced. The outer surface of the cylinder and the inner surface of the internal flow path were given a no-slip condition. At the inlet of the flow path, the flow velocity v was set to have a uniform velocity distribution. ｄ He gave it.

[0096] These governing equations were approximated by finite volume methods using difference approximations and solved using OpenFOAM (Open Source Field Operation and Manipulation). OpenFOAM is a C++ toolbox for numerical analysis of continuous fluid dynamics, including computational fluid dynamics, and for pre- and post-processing (see OpenFOAM Foundation, https: / / openfoam.org / , Open CAE Society, Numerical Analysis of Heat Transfer and Flow with OpenFOAM, Morikita Publishing (2016)). The VOF method used in this example also utilizes an algorithm already implemented in OpenFOAM. The computational domain was uniformly divided into a structured grid (rectangular mesh), with a mesh count of 384,000 (32 × 40 × 300).

[0097] 1-2. Simulation Execution Table 1 shows the physical properties of the fluids. The density and viscosity of air were set to values ​​at room temperature. Typically, the specific gravity and specific viscosity of urine relative to water are reported to be 1.003–1.030 and 1.037–1.142, respectively. Although the density of urine is several percent higher and the viscosity is about 10% higher than that of water, it was assumed that there was almost no difference in fluid dynamics, and therefore the density and viscosity of urine were set to the physical properties of water at room temperature. For the surface tension of urine, considering that urine is a mixture, a value of about 70% of the surface tension of pure water (72 mN / m) (50 mN / m) was applied. This is roughly in line with the reported surface tension values ​​of urine, which are 48.2–65.1 mN / m (see Yosuke Ohara, "Study on Urine Surface Tension and Surfactants in Urine," Journal of the Japan Medical University Medical Society, 38(5), 216–224 (1971)).

[0098]

[0099] The flow rate Q of a liquid is equal to the flow velocity v. ｄ Using the area S of the slit as the outlet, Q = v ｄ It can be calculated using S. For example, the flow velocity v ｄ = 4 m / s (400 cm / s), slit area S = 5 mm ２ (0.05 cm) ２ When ) is applied, the flow rate Q is 20 mL / s.

[0100] 2. Creation of training data Based on the training images obtained from the simulation described above, a total of 1295 datasets were prepared, and training data was created by labeling the images with features from each dataset using a conventional method. The dataset for Example 1 is dataset (i). That is, one dataset includes (a) a training image and, as features of that image, (b) the liquid flow rate at the time of image generation, (c) the aspect ratio of the outlet (i.e., the aspect ratio of the major axis and minor axis in the first abdomen), and (d) the rotation angle around the central axis of the outlet (i.e., the first abdomen).

[0101] 3. Creation of a Urine Flow Estimation Model Based on the training data obtained above, a urine flow estimation model was constructed using the following procedure. First, the image data was resized and normalized. OpenCV was used for this image processing. TensorFlow and Keras were used to construct the machine learning model. The machine learning model used a convolutional neural network (CNN) consisting of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. Normalized image data was loaded into the input layer, and the liquid flow rate, the aspect ratio of the outlet, and the rotation angle around the outlet axis were obtained from the output layer. The value of the loss function was calculated from the mean squared error, and training was performed. These processes were executed using the programming language Python, and the numpy and pandas libraries were used for numerical calculations.

[0102] 4. Flow rate estimation using a urine flow estimation model The target images were input into the urine flow estimation model obtained above, and the model was made to perform calculations to estimate the features incorporated into the dataset, and these estimated values ​​were output. The target images were simulation images of known flow rates created by numerical simulation, similar to the training images.

[0103] <Example 2> Dataset (ii) was used as the dataset for creating the training data. That is, one dataset includes (a) training images, (b) the liquid flow rate at the time of image generation, and (c) the aspect ratio of the outlet, but does not include (d) the rotation angle around the central axis of the outlet. Except for the training data, it is the same as in Example 1.

[0104] <Example 3> Dataset (iii) was used as the dataset for creating the training data. That is, one dataset includes (a) training images, (b) the liquid flow rate at the time of image generation, and (d) the rotation angle around the central axis of the outlet, but does not include (c) the aspect ratio of the outlet. Except for the training data, it is the same as in Example 1.

[0105] <Comparative Example> A comparative dataset was used as the dataset for creating the training data. Specifically, one dataset includes (a) training images and (b) the liquid flow rate during image generation, but does not include (c) the aspect ratio of the outlet and (d) the rotation angle around the central axis of the outlet. Except for the training data, it is the same as in Example 1. The estimation model created here is a comparative estimation model.

[0106] <Explanation of Results> Table 2 below shows some of the estimation results in Example 1. Figure 8 is a graph showing the relationship between the actual flow rate (mL / s) and the estimated value in Example 1, Figure 9 is a graph showing the relationship between the actual aspect ratio and the estimated value in Example 1, and Figure 10 is a graph showing the relationship between the actual angle (deg.) and the estimated value in Example 1. The results of Example 1 show that when a urine flow estimation model is created using the training data of dataset (i), the liquid flow features can be estimated within an error range of ±10% (the range enclosed by the dashed line), and even within an error range of ± a few percent.

[0107]

[0108] Figure 11 shows the flow rate estimation results for Example 1 and the Comparative Example. The estimated values ​​in Example 1 were within a ±10% error range, while the estimated values ​​in the Comparative Example sometimes fell significantly outside this range. From the comparison of these results, it can be seen that including not only the flow rate but also the aspect ratio of the outlet and the rotation angle around the central axis of the outlet as features in the training data improves the accuracy of estimating the flow velocity and / or flow rate of urine flow.

[0109] Figure 12 shows the flow rate estimation results in Examples 2 and 3. These results demonstrate that even when the training data includes either the aspect ratio of the outlet or the rotation angle around the central axis of the outlet as features in addition to the flow rate, the urine flow velocity and / or flow rate can be estimated within an error range of approximately ±10%. Furthermore, since the error range of the estimated value in Example 2 is narrower than that of Example 3 in the low-volume range (approximately 10 mL / s or less) where the liquid flow tends to be unstable, it is recognized that the aspect ratio of the outlet contributes more to improving estimation accuracy than the rotation angle around the central axis of the outlet.

[0110] The urine flow estimation method and apparatus of the present invention can be used in the field of diagnosis, treatment, and monitoring of voiding / storage disorders. In particular, when using the urine flow estimation method and apparatus of the present invention, it is possible to measure urine flow based on images of urine flow taken by the patient in their daily life, for example, thus enabling daily monitoring of the patient's condition.

[0111] In addition, the urine flow estimation method and apparatus of the present invention can be easily used not only in the medical field, such as the diagnosis of voiding / storage disorders, but also in healthcare, nursing care, and rehabilitation fields, such as appropriate daily voiding management for the elderly.

[0112] 10 Estimation device (cloud computing type) 11 Calculation processing unit 12 Communication unit 13 Model creation unit 14 Image processing unit 15 Storage unit 20 Mobile device 21 Communication unit 22 Imaging unit 23 Display unit 30 Medical device 31 Communication unit 32 Display unit 40 Estimation device (standalone type) 41 Calculation processing unit 43 Model creation unit 44 Image processing unit 45 Storage unit 46 Imaging unit 47 Display unit

Claims

1. A method for estimating the flow velocity and / or flow rate of a urine flow, comprising: preparing a target image which is an image of the urine flow to be measured; and estimating the flow velocity and / or flow rate of the urine flow in the target image based on a machine learning-trained urine flow estimation model to obtain an estimated value of the flow velocity and / or flow rate: wherein the urine flow estimation model is a machine learning model created by machine learning the correlation between data based on training data which includes the following data (a) and data (b), and further includes one or both of data (c) and data (d): (a) a training image of a liquid flow in an axis-switching state; (b) the flow velocity and / or flow rate of the liquid flow; (c) the lengths of the major axis and minor axis of the abdominal cross-section in the axis-switching state, or the aspect ratio of the major axis to the minor axis; and (d) the rotation angle around the central axis of the abdomen in the axis-switching state.

2. The method according to claim 1, wherein the training data includes both data (c) and data (d).

3. The method according to claim 1, wherein the estimation is performed using an estimation device that includes an arithmetic processing unit that performs estimation calculations.

4. The method according to claim 3, wherein the preparation of the target image is to receive an image of the urine flow transferred from a device other than the estimation device.

5. The method according to claim 3, wherein the preparation of the target image is performed by imaging the urine flow using the imaging unit provided in the estimation device.

6. The method according to any one of claims 3 to 5, further comprising transferring the estimated value to a display device other than the estimation device.

7. The method according to any one of claims 3 to 5, further comprising displaying the estimated value on a display unit provided in the estimation device.

8. The method according to any one of claims 1 to 5, wherein the target image is a binarized image.

9. The method according to any one of claims 1 to 5, wherein the target image is a video or a still image.

10. Estimation device for estimating the flow velocity and / or flow rate of urine flow, comprising: an image preparation unit for preparing a target image which is an image of the urine flow to be measured; and a calculation processing unit that estimates the flow velocity and / or flow rate of the urine flow in the target image and obtains an estimated value of the flow velocity and / or flow rate based on a machine learning-trained urine flow estimation model: wherein the urine flow estimation model is a machine learning model created by machine learning the correlation between data based on training data which includes the following data (a) and data (b), and further includes one or both of data (c) and data (d): (a) a learning image of the liquid flow in an axis switching state; (b) the flow velocity and / or flow rate of the liquid flow; (c) the lengths of the major axis and minor axis of the abdominal cross-section in the axis switching or the aspect ratio of the major axis to the minor axis; and (d) the rotation angle around the central axis of the abdomen in the axis switching.

11. The estimation apparatus according to claim 10, wherein the training data includes both data (c) and data (d).

12. The estimation device according to claim 10 or 11, wherein the image preparation unit is a communication unit that receives an image of urine flow transferred from a device other than the estimation device.

13. The estimation device according to claim 10 or 11, wherein the image preparation unit is an imaging unit for imaging urine flow.

14. The estimation device according to claim 10 or 11, further comprising a communication unit for transferring the estimated value to a display device other than the estimation device.

15. The estimation device according to claim 10 or 11, further comprising a display unit for displaying the estimated value.

16. The estimation device according to claim 10, wherein the image preparation unit is a communication unit that receives an image of urine flow transferred from a device other than the estimation device, and the communication unit has a function of transferring the estimated value to a display device other than the estimation device.

17. The estimation device according to claim 16, wherein the estimation device is a cloud computing type device.

18. The estimation device according to claim 10, wherein the image preparation unit is an imaging unit for imaging urine flow, and the estimation device further includes a display unit for displaying the estimated value.

19. The estimation device according to claim 18, wherein the estimation device is a mobile device type device.

20. The estimation device according to claim 10 or 11, further comprising an image processing unit that binarizes the target image, and the calculation processing unit that performs the estimation calculation based on the binarized target image.

Images