Estimation device, estimation system, and computer-readable non-transitory recording medium recording estimation program

Abstract

The estimation device includes an input unit and an approximator. The input unit inputs input information including an image of a bone. The approximator can estimate an estimation result associated with the bone density of the bone based on the input information input to the input unit. The approximator has a learned parameter for obtaining an estimation result associated with the bone density of the bone based on the input information.

Description

[0001] Cross-references of related applications

[0002] This application claims the priority of Japanese Patent Application No. 2018-168502 (filed on September 10, 2018) and Japanese Patent Application No. 2018-220401 (filed on November 26, 2018), the disclosures of which are incorporated herein in their entirety for purposes of reference. Technical Field

[0003] The present disclosure relates to the estimation of bone density. Background Art

[0004] Patent Document 1 describes a technique for determining osteoporosis, and Patent Document 2 describes a technique for estimating bone strength.

[0005] Prior Art Literature

[0006] Patent Literature

[0007] Patent Document 1: Japanese Patent Application Publication No. 2008-36068

[0008] Patent Document 2: Japanese Patent Publication No. 2002-523204 Summary of the invention

[0009] An estimation device, an estimation system, and an estimation program are disclosed. In one embodiment, the estimation device comprises: an input unit that inputs input information having an image of a bone; an approximator that can estimate an estimation result associated with the bone density of the bone based on the input information input to the input unit; and an output unit that outputs the estimation result estimated by the approximator, the approximator having a learned parameter for obtaining an estimation result associated with the bone density of the bone based on the input information.

[0010] In addition, in one embodiment, the estimation system comprises: an input unit, which inputs input information having an image captured of a bone; and an approximator, which has learned parameters for obtaining an estimation result associated with the bone density of the bone based on the input information, and can estimate the estimation result associated with the bone density of the bone based on the input information input to the input unit, and when the input information is input to the input unit, the approximator performs calculation processing on the input information.

[0011] In addition, in one embodiment, the estimation program is an estimation program for causing the device to function as a neural network, which performs operations based on learned parameters and outputs an estimated value of the bone density of the bone captured in the image, and the learned parameters are used to obtain an estimation result associated with the bone density of the bone based on input information of an image having the bone captured. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a diagram showing an example of the configuration of a computer device (estimation device).

[0013] Figure 2 It is a diagram for explaining the operation of the estimation device.

[0014] Figure 3 is a diagram showing an example of the structure of a neural network.

[0015] Figure 4 This is a diagram showing an example of how learning image data and reference bone density are mutually associated with each other.

[0016] Figure 5 It is a diagram for explaining the operation of the estimation device.

[0017] Figure 6 It is a diagram showing an example of an estimation system.

[0018] Figure 7 It is a diagram showing an example of the configuration of an estimation device.

[0019] Figure 8 It is a diagram showing an example of the configuration of an estimation device.

[0020] Fig. 9 It is a diagram showing an example of the configuration of an estimation device.

[0021] Fig.10 It is a diagram showing an example of the configuration of an estimation device.

[0022] Fig.11 It is a diagram showing an example of the configuration of an estimation device.

[0023] Fig.12 It is a diagram showing an example of the configuration of an estimation device.

[0024] Fig.13 It is a conceptual diagram schematically showing the structure of the estimation system.

[0025] Fig.14 This is a conceptual diagram schematically showing the structure of a part of the estimation system.

[0026] Fig.15 This is a conceptual diagram schematically showing the structure of a part of the estimation system.

[0027] Fig.16 This is a conceptual diagram schematically showing the structure of a part of the estimation system.

[0028] Fig.17This is a conceptual diagram schematically showing the structure of a part of the estimation system.

[0029] Fig.18 This is a conceptual diagram schematically showing a part of the structure of another embodiment of the estimation system. DETAILED DESCRIPTION

[0030] Implementation method 1.

[0031] Figure 1 This is a block diagram showing an example of the configuration of the computer device 1 according to Embodiment 1. The computer device 1 functions as an estimation device for estimating bone density. Hereinafter, the computer device 1 may be referred to as an “estimation device 1”.

[0032] like Figure 1 As shown, the estimation device 1 includes, for example, a control unit 10, a storage unit 20, a communication unit 30, a display unit 40, and an input unit 50. The control unit 10, the storage unit 20, the communication unit 30, the display unit 40, and the input unit 50 are electrically connected to each other via a bus 60, for example.

[0033] The control unit 10 can centrally manage the operation of the estimation device 1 by controlling other components of the estimation device 1. The control unit 10 is also referred to as a control device or a control circuit. As described in further detail below, the control unit 10 includes at least one processor to provide control and processing capabilities for performing various functions.

[0034] According to various embodiments, at least one processor may also be implemented as a single integrated circuit (IC) or as multiple integrated circuits (ICs) and / or discrete circuits that are communicatively connected. At least one processor may be implemented according to various known techniques.

[0035] In one embodiment, the processor includes, for example, one or more circuits or units configured to execute one or more data computing programs or processes by executing instructions stored in an associated memory. In another embodiment, the processor may also be firmware (e.g., discrete logic elements) configured to execute one or more data computing programs or processes.

[0036] According to various embodiments, the processor may also include one or more processors, controllers, microprocessors, microcontrollers, application-specific integrated circuits (ASICs), digital signal processing devices, programmable logic devices, field programmable logic devices, or any combination of these devices or structures, or other known devices and combinations of structures, and perform the functions described below. In this example, the control unit 10, for example, has a CPU (Central Processing Unit).

[0037] The storage unit 20 includes a non-temporary storage medium such as a ROM (Read Only Memory) and a RAM (Random Access Memory) that can be read by the CPU of the control unit 10. The storage unit 20 stores a control program 100 for controlling the estimation device 1. Various functions of the control unit 10 are realized by the CPU of the control unit 10 executing the control program 100 in the storage unit 20. It can also be said that the control program 100 is a bone density estimation program for making the computer device 1 function as an estimation device. In this example, the control unit 10 executes the control program 100 in the storage unit 20, so that Figure 2 As shown, an approximator 280 capable of outputting an estimated value 300 of bone density is formed in the control unit 10. The approximator 280 includes, for example, a neural network 200. The control program 100 can also be said to be a program for causing the computer device 1 to function as the neural network 200. Hereinafter, the estimated value of bone density may be referred to as a "bone density estimated value". In addition, a configuration example of the neural network 200 will be described in detail later.

[0038] In addition to the control program 100, the storage unit 20 also stores learned parameters 110, estimation data 120 (hereinafter also referred to as "input information"), learning data 130, and training data 140 related to the neural network 200. The learning data 130 and the training data 140 are data used when learning the neural network 200. The learned parameters 110 and the estimation data 120 are data used by the learned neural network 200 when estimating bone density.

[0039] The learning data 130 is data input to the input layer 210 of the neural network 200 when the neural network 200 is being learned. The learning data 130 is also referred to as learning data. The training data 140 is data indicating the correct value of the bone density. The training data 140 is compared with the output data output from the output layer 230 of the neural network 200 when the neural network 200 is being learned. The learning data 130 and the training data 140 are sometimes collectively referred to as learning data with training.

[0040] The estimation data 120 is data input to the input layer 210 of the learned neural network 200 when estimating bone density. The learned parameters 110 are learned parameters in the neural network 200. The learned parameters 110 may also be referred to as parameters adjusted by learning the neural network 200. The learned parameters 110 include weighting coefficients representing weights of connections between artificial neurons. Figure 2As shown, calculations based on the learned parameters 110 are performed on the estimation data 120 input to the input layer 210 , and the bone density estimation value 300 is output from the output layer 230 .

[0041] The data input to the input layer 210 may be input to the input layer 210 via the input unit 50 or directly input to the input layer 210. When directly input to the input layer 210, the input layer 210 may be a part or the whole of the input unit 50. Hereinafter, the estimated bone density value 300 may be referred to as an estimated result 300.

[0042] The communication unit 30 is connected to a communication network including the Internet, etc., by wire or wireless. The communication unit 30 can communicate with other devices such as cloud servers and network servers through the communication network. The communication unit 30 can input information received from the communication network to the control unit 10. In addition, the communication unit 30 can output information received from the control unit 10 to the communication network.

[0043] The display unit 40 is, for example, a liquid crystal display or an organic EL display, and can display various information such as characters, symbols, and graphics by being controlled by the control unit 10 .

[0044] The input unit 50 can receive input from the user to the estimation device 1. The input unit 50 includes, for example, a keyboard and a mouse. The input unit 50 may include a touch screen that can detect an operation performed by the user on the display surface of the display unit 40.

[0045] In addition, the structure of the estimation device 1 is not limited to the above-mentioned example. For example, the control unit 10 may also include a plurality of CPUs. In addition, the control unit 10 may also include at least one DSP. In addition, all functions of the control unit 10 or a part of the functions of the control unit 10 may also be implemented by a hardware circuit that does not require software in the implementation of its functions. In addition, the storage unit 20 may also include a non-temporary storage medium that can be read by a computer other than ROM and RAM. The storage unit 20 may also include, for example, a small hard disk drive and an SSD (Solid State Drive). In addition, the storage unit 20 may also include a memory such as a USB (Universal Serial Bus) memory that is pluggable and unpluggable relative to the estimation device 1. Hereinafter, a memory that is pluggable and unpluggable relative to the estimation device 1 is sometimes referred to as a "pluggable memory".

[0046] <Structure example of neural network>

[0047] Figure 32 is a diagram showing an example of the structure of the neural network 200. In this example, the neural network 200 is, for example, a convolutional neural network (CNN). Figure 3 As shown, the neural network 200 includes, for example, an input layer 210, a hidden layer 220, and an output layer 230. The hidden layer 220 is also called an intermediate layer. The hidden layer 220 includes, for example, a plurality of convolutional layers 240, a plurality of pooling layers 250, and a fully connected layer 260. In the neural network 200, the fully connected layer 260 exists at the front stage of the output layer 230. Moreover, in the neural network 200, the convolutional layers 240 and the pooling layers 250 are alternately arranged between the input layer 210 and the fully connected layer 260.

[0048] In addition, the structure of the neural network 200 is not limited to Figure 3 For example, the neural network 200 may include one convolutional layer 240 and one pooling layer 250 between the input layer 210 and the fully connected layer 260. In addition, the neural network 200 may be a neural network other than a convolutional neural network.

[0049] <An example of estimation data, learning data, and training data>

[0050] The estimation data 120 includes image data of a simple X-ray image of a bone of an estimation object of bone density. The estimation object of bone density is, for example, a human. Therefore, it can be said that the estimation data 120 includes image data of a simple X-ray image of a human bone. The learning data 130 includes image data of a plurality of simple X-ray images of a human bone. A simple X-ray image is also referred to as a two-dimensional image, a general X-ray image, or a Roentgen image. In addition, the estimation object of bone density may be other than a human. For example, the estimation object of bone density may also be an animal such as a dog, a cat, or a horse. In addition, the bones set as the object are mainly cortical bones and spongy bones from a living body, but the bones set as the object may also include artificial bones with calcium phosphate as the main component or regenerated bones artificially manufactured by regenerative medicine, etc.

[0051] Hereinafter, the image data included in the estimation data 120 is sometimes referred to as "estimation image data". In addition, the simple X-ray image represented by the image data included in the estimation data 120 is sometimes referred to as "estimation simple X-ray image". In addition, the image data included in the learning data 130 is sometimes referred to as "learning image data". In addition, the simple X-ray image represented by the image data included in the learning data 130 is sometimes referred to as "learning simple X-ray image". The learning data 130 includes a plurality of learning X-ray image data each representing a plurality of learning simple X-ray images.

[0052] As the imaging part of the simple X-ray image for estimation, for example, the head, neck, chest, waist, hip joint, knee joint, ankle joint, foot, toe, shoulder joint, elbow joint, wrist joint, hand, finger or jaw joint can be used. In other words, as the estimation data 120, the following may be used: image data of a simple X-ray image obtained by irradiating the head with X-rays, image data of a simple X-ray image obtained by irradiating the neck with X-rays, image data of a simple X-ray image obtained by irradiating the chest with X-rays, image data of a simple X-ray image obtained by irradiating the body with X-rays, image data of a simple X-ray image obtained by irradiating the hip joint with X-rays, image data of a simple X-ray image obtained by irradiating the knee joint with X-rays, image data of a simple X-ray image obtained by irradiating the ankle joint with X-rays, image data of a simple X-ray image obtained by irradiating the foot with X-rays, image data of a simple X-ray image obtained by irradiating the toe with X-rays, image data of a simple X-ray image obtained by irradiating the shoulder joint with X-rays, image data of a simple X-ray image obtained by irradiating the elbow joint with X-rays, image data of a simple X-ray image obtained by irradiating the hand joint with X-rays, image data of a simple X-ray image obtained by irradiating the hand with X-rays, image data of a simple X-ray image obtained by irradiating the finger with X-rays, or image data of a simple X-ray image obtained by irradiating the jaw joint with X-rays. The simple X-ray image obtained by irradiating the chest with X-rays includes a simple X-ray image of the lungs and a simple X-ray image of the thoracic spine. In addition, the type of the imaging part of the simple X-ray image for estimation is not limited to this. In addition, the simple X-ray image for estimation can be a frontal image of the target part from the front or a side image of the target part from the side.

[0053] The shooting parts of the multiple simple X-ray images for learning, respectively represented by the multiple learning image data included in the learning data 130, include, for example, at least one of the head, neck, chest, waist, hip joint, knee joint, ankle joint, foot, toe, shoulder joint, elbow joint, hand joint, hand, finger and jaw joint. In other words, the learning data 130 includes at least one of 15 types of image data, including image data of a simple X-ray image obtained by irradiating the head with X-rays, image data of a simple X-ray image obtained by irradiating the neck with X-rays, image data of a simple X-ray image obtained by irradiating the chest with X-rays, image data of a simple X-ray image obtained by irradiating the waist with X-rays, image data of a simple X-ray image obtained by irradiating the hip joint with X-rays, image data of a simple X-ray image obtained by irradiating the knee joint with X-rays, image data of a simple X-ray image obtained by irradiating the ankle joint with X-rays, image data of a simple X-ray image obtained by irradiating the foot with X-rays, image data of a simple X-ray image obtained by irradiating the toe with X-rays, image data of a simple X-ray image obtained by irradiating the shoulder joint with X-rays, image data of a simple X-ray image obtained by irradiating the elbow joint with X-rays, image data of a simple X-ray image obtained by irradiating the hand joint with X-rays, image data of a simple X-ray image obtained by irradiating the hand with X-rays, image data of a simple X-ray image obtained by irradiating the finger with X-rays, and image data of a simple X-ray image obtained by irradiating the jaw joint with X-rays. The learning data 130 may include image data of some of the 15 types of image data or all types of image data. In addition, the types of the shooting parts of the simple X-ray images for learning are not limited thereto. In addition, the multiple simple X-ray images for learning may include both frontal images and side images. In addition, the multiple simple X-ray images for learning may include both frontal images and side images of the same shooting part.

[0054] In the training data 140, for each of the plurality of learning image data included in the learning data 130, a measured value of the bone density of a person having a bone photographed in the simple learning X-ray image represented by the learning image data is included. Among the plurality of bone density measured values ​​included in the training data 140, for example, at least one of the bone density measured by irradiating X-rays with respect to the lumbar spine, the bone density measured by irradiating X-rays with respect to the proximal part of the femur, the bone density measured by irradiating X-rays with respect to the radius, the bone density measured by irradiating X-rays with respect to the metacarpal bone, the bone density measured by applying ultrasound with respect to the wrist, and the bone density measured by applying ultrasound with respect to the heel is included. Hereinafter, the bone density measured values ​​included in the training data 140 are sometimes referred to as "reference bone density".

[0055] Here, as a method for measuring bone density, the DEXA (dual-energy X-ray absorptiometry) method is known. In a DEXA device that uses the DEXA method to measure bone density, when measuring the bone density of the lumbar spine, X-rays (specifically, two types of X-rays) are irradiated to the lumbar spine from the front. In addition, in the DEXA device, when measuring the bone density of the proximal part of the femur, X-rays are irradiated to the proximal part of the femur from the front.

[0056] The training data 140 may include the bone density of the lumbar spine measured by the DEXA device, and the bone density of the proximal femur measured by the DEXA device. In addition, the training data 140 may also include the bone density measured by irradiating the target part with X-rays from the side thereof. For example, the training data 140 may also include the bone density measured by irradiating the lumbar spine with X-rays from the side thereof.

[0057] In addition, as another method for measuring bone density, an ultrasonic method is known. In a device that uses the ultrasonic method to measure bone density, for example, ultrasonic waves are applied to the wrist to measure the bone density of the wrist, and ultrasonic waves are applied to the heel to measure the bone density of the heel. The training data 140 may also include bone density measured by the ultrasonic method.

[0058] In the plurality of simple X-ray images for learning represented by the plurality of learning image data included in the learning data 130, bones of different persons are captured. Figure 4As shown, in the storage unit 20, for each of the plurality of learning image data included in the learning data 130, a correspondence is established with a reference bone density of a person having a bone photographed in the simple X-ray image for learning represented by the learning image data. It can also be said that for each of the plurality of simple X-ray images for learning used in the learning of the neural network 200, a correspondence is established with a reference bone density of a person having a bone photographed in the simple X-ray image for learning. The reference bone density associated with the learning image data is a bone density measured for the same person as the person having a bone photographed in the simple X-ray image for learning, at a period that is almost the same as the period when the simple X-ray image for learning represented by the learning image data was photographed.

[0059] The parts captured in the simple X-ray image for learning represented by the learning image data (in other words, the shooting parts of the simple X-ray image for learning) may include or exclude the parts (i.e., bones) at which the reference bone density corresponding to the learning image data has been measured. In other words, the parts captured in the simple X-ray image for learning may include or exclude the parts at which the reference bone density corresponding to the simple X-ray image for learning has been measured. As an example of the former, consider the case where the learning image data representing the simple X-ray image for learning captured of the waist is corresponding to the reference bone density of the lumbar vertebrae. As another example, consider the case where the learning image data capturing the hip joint is corresponding to the reference bone density of the proximal part of the femur. On the other hand, as an example of the latter, consider the case where the learning image data capturing the chest is corresponding to the reference bone density of the lumbar vertebrae. As another example, consider the case where the learning image data capturing the knee joint is corresponding to the reference bone density of the heel.

[0060] In addition, the orientation of the part captured in the simple X-ray image represented by the learning image data and the orientation of the X-ray irradiation relative to the target part in the determination of the reference bone density corresponding to the learning image data may be the same or different. In other words, the orientation of the part captured in the learning simple X-ray image and the orientation of the X-ray irradiation relative to the target part in the determination of the reference bone density corresponding to the learning simple X-ray image may be the same or different. As an example of the former, consider the case where a correspondence is established between the learning image data representing the simple X-ray image captured from the front to the chest (hereinafter, sometimes referred to as the "chest front simple X-ray image") and the reference bone density measured by irradiating the lumbar spine from the front. As another example, consider the case where a correspondence is established between the learning image data representing the simple X-ray image captured from the front to the waist (hereinafter, sometimes referred to as the "lumbar front simple X-ray image") and the reference bone density measured by irradiating the proximal part of the femur from the front. On the other hand, as an example of the latter, consider the case where a correspondence is established between learning image data representing a simple X-ray image of the waist taken from the side (hereinafter, sometimes referred to as a "waist side simple X-ray image") and a reference bone density measured by irradiating X-rays from the front with respect to the lumbar spine. As another example, consider the case where a correspondence is established between learning image data representing a simple X-ray image of the knee joint taken from the side (hereinafter, sometimes referred to as a "knee side simple X-ray image") and a reference bone density measured by irradiating X-rays from the front with respect to the proximal part of the femur.

[0061] In addition, among the multiple learning simple X-ray images respectively represented by the multiple learning image data included in the learning data 130, both simple X-ray images of the same type of parts as the estimated simple X-ray image and simple X-ray images of different types of parts from the estimated simple X-ray image may be included. As an example of the former, consider the case where the estimated simple X-ray image is a simple X-ray image of the front of the chest, and the case where the chest front simple X-ray image is included in the multiple learning simple X-ray images. As another example, consider the case where the estimated simple X-ray image is a simple X-ray image of the knee joint taken from the front (hereinafter sometimes referred to as a "knee front simple X-ray image"), and the case where the knee side simple X-ray image is included in the multiple learning simple X-ray images. On the other hand, as an example of the latter, consider the case where the estimated simple X-ray image is a simple X-ray image of the waist from the front, and the case where the chest front simple X-ray image is included in the multiple learning simple X-ray images. As another example, consider a case where the estimation simple X-ray image is a lateral simple X-ray image of the waist, and a plurality of learning simple X-ray images include a frontal simple X-ray image of the knee.

[0062] In addition, the multiple simple X-ray images for learning may include simple X-ray images of the parts that are photographed in the same direction as the simple X-ray images for estimation, and may also include simple X-ray images of the parts that are photographed in a different direction from the simple X-ray images for estimation. As an example of the former, consider the case where the simple X-ray image for estimation is a simple X-ray image of the front of the lumbar spine, and the case where the simple X-ray image of the front of the waist is included in the multiple simple X-ray images for learning. As another example, consider the case where the simple X-ray image for estimation is a simple X-ray image of the front of the knee, and the case where the simple X-ray image of the front of the chest is included in the multiple simple X-ray images for learning. On the other hand, as an example of the latter, consider the case where the simple X-ray image for estimation is a simple X-ray image of the side of the knee, and the case where the simple X-ray image of the front of the knee is included in the multiple simple X-ray images for learning. As another example, consider the case where the simple X-ray image for estimation is a simple X-ray image of the side of the waist, and the case where the simple X-ray image of the front of the chest is included in the multiple simple X-ray images for learning.

[0063] In addition, the training data 140 may also include a reference bone density measured from a part (bone) included in the part captured in the simple X-ray image for estimation, and may also include a reference bone density measured from a part (bone) not included in the part captured in the simple X-ray image for estimation. As an example of the former, consider a case where the simple X-ray image for estimation is a simple X-ray image of the front of the waist, and the reference bone density of the lumbar spine is included in the training data 140. On the other hand, as an example of the latter, consider a case where the simple X-ray image for estimation is a simple X-ray image of the front of the chest, and the reference bone density of the metacarpal head is included in the training data 140.

[0064] In addition, the training data 140 may include a baseline bone density measured by irradiating the target part with X-rays from the same direction as the direction of the part photographed in the simple X-ray image for estimation, or may include a baseline bone density measured by irradiating the target part with X-rays from a direction different from the direction of the part photographed in the simple X-ray image for estimation. As an example of the former, consider a case where the simple X-ray image for estimation is a simple X-ray image of the front of the waist, and the training data 140 includes a baseline bone density measured by irradiating the lumbar spine from the front. On the other hand, as an example of the latter, consider a case where the simple X-ray image for estimation is a simple X-ray image of the side of the waist, and the training data 140 includes a baseline bone density measured by irradiating the proximal part of the femur from the front.

[0065] In this example, image data representing the grayscale of a simple X-ray image obtained by a simple X-ray imaging device (in other words, a general X-ray imaging device or a x-ray imaging device) is compressed, and the image data obtained by reducing the number of grayscales is used as learning image data and estimation image data. For example, when the number of multiple pixel data constituting the image data obtained by the simple X-ray imaging device is greater than (1024×640) and the number of bits of its pixel data is 16 bits. In this case, the number of multiple pixel data constituting the image data obtained by the simple X-ray imaging device is compressed to, for example, (256×256), (1024×512) or (1024×640), and the number of bits of its pixel data is reduced to 8 bits, so as to be used as learning image data and estimation image data. In this case, each of the learning simple X-ray image and the estimation simple X-ray image includes (256×256), (1024×512), or (1024×640) pixels, and the value of the pixel is expressed by 8 bits.

[0066] The image data for learning and the image data for estimation may be generated by the control unit 10 of the estimation device 1 based on the image data obtained by the simple X-ray imaging device, or may be generated by a device other than the estimation device 1 based on the image data obtained by the simple X-ray imaging device. In the former case, the image data obtained by the simple X-ray imaging device may be received by the communication unit 30 through the communication network, or may be stored in the plug-in memory included in the storage unit 20. In the latter case, the communication unit 30 may receive the image data for learning and the image data for estimation from other devices through the communication network, and the control unit 10 may store the image data for learning and the image data for estimation received by the communication unit 30 in the storage unit 20. Alternatively, the image data for learning and the image data for estimation generated by other devices may be stored in the plug-in memory included in the storage unit 20. In addition, the training data 140 may be received by the communication unit 30 through the communication network, and the control unit 10 may store the training data 140 received by the communication unit 30 in the storage unit 20. Alternatively, the training data 140 may be stored in a plug-in memory included in the storage unit 20. The number of pixel data and the number of bits of the learning image data and the estimation image data are not limited to those described above.

[0067] <Neural network learning example>

[0068] Figure 5 2 is a diagram for explaining an example of learning of the neural network 200. When the control unit 10 performs learning of the neural network 200, as shown in FIG. Figure 5As shown, the learning data 130 is input to the input layer 210 of the neural network 200. Then, the control unit 10 adjusts the variable parameter 110a in the neural network 200 so that the error of the output data 400 output from the output layer 230 of the neural network 200 relative to the training data 140 is small. More specifically, the control unit 10 inputs each learning image data in the storage unit 20 to the input layer 210. When the learning image data is input to the input layer 210, the control unit 10 inputs a plurality of pixel data constituting the learning image data to a plurality of artificial neurons constituting the input layer 210. Then, when the learning image data is input to the input layer 210, the control unit 10 adjusts the parameter 110a so that the error of the output data 400 output from the output layer 230 relative to the reference bone density corresponding to the learning image data is small. As a method for adjusting the parameter 110a, for example, an error back propagation method can be used. The adjusted parameter 110a becomes the learned parameter 110, which is stored in the storage unit 20. The parameter 110a includes, for example, parameters used in the hidden layer 220. Specifically, the parameter 110a includes filter coefficients used in the convolution layer 240 and weight coefficients used in the fully connected layer 260. In addition, the adjustment method of the parameter 110a, in other words, the learning method of the parameter 110a is not limited to this.

[0069] As described above, the storage unit 20 stores the learned parameters 110 obtained by learning the relationship between the learning data 130 including image data of a plurality of simple X-ray images for learning and the measured values ​​of the bone density as the training data 140 using the neural network 200 .

[0070] In the above example, the estimation device 1 performs learning of the neural network 200, but other devices may also perform learning of the neural network 200. In this case, the learned parameters 110 generated by other devices are stored in the storage unit 20 of the estimation device 1. In addition, it is no longer necessary to store the learning data 130 and the training data 140 in the storage unit 20. The learned parameters 110 generated by other devices may be received by the communication unit 30 through the communication network, and the control unit 10 may store the learned parameters 110 received by the communication unit 30 in the storage unit 20. Alternatively, the learned parameters 110 generated by other devices may be stored in a plug-in memory included in the storage unit 20.

[0071] In the neural network 200 learned as described above, image data of a plurality of simple X-ray images for learning are input to the input layer 210 as the learning data 130, and the learned parameters 110a learned using the reference bone density as the training data 140 are included. Figure 2As shown, the neural network 200 performs calculations based on the learned parameters 110a for the estimation data 120 input to the input layer 210, and outputs the bone density estimation value 300 from the output layer 230. When the estimation image data as the estimation data 120 is input to the input layer 210, a plurality of pixel data constituting the estimation image data are respectively input to a plurality of artificial neurons constituting the input layer 210. Then, the convolution layer 240 performs calculations using the filter coefficients included in the learned parameters 110a, and the fully connected layer 260 performs calculations using the weighting coefficients included in the learned parameters 110a.

[0072] For example, when estimation image data representing a chest front simple X-ray image is input to the input layer 210, an estimated value 300 of the bone density of a person having the chest bones captured in the chest front simple X-ray image represented by the estimation image data is output from the output layer 230. Furthermore, when estimation image data representing a waist front simple X-ray image is input to the input layer 210, an estimated value 300 of the bone density of a person having the lumbar vertebrae included in the waist captured in the waist front simple X-ray image represented by the estimation image data is output from the output layer 230. Furthermore, when estimation image data representing a waist lateral simple X-ray image is input to the input layer 210, an estimated value 300 of the bone density of a person having the lumbar vertebrae included in the waist captured in the waist lateral simple X-ray image represented by the estimation image data is output from the output layer 230. Furthermore, when estimation image data representing a simple X-ray image of the front side of the knee is input to the input layer 210, an estimated value 300 of the bone density of a person having a bone of the knee joint captured in the simple X-ray image of the front side of the knee represented by the estimation image data is output from the output layer 230. Furthermore, when estimation image data representing a simple X-ray image of the lateral side of the knee is input to the input layer 210, an estimated value 300 of the bone density of a person having a bone of the knee joint captured in the simple X-ray image of the lateral side of the knee represented by the estimation image data is output from the output layer 230.

[0073] The estimated value 300 output from the output layer 230 can be calculated using the bone mineral density per unit area (g / cm 2 ), bone mineral density per unit volume (g / cm 3 ), YAM, T score, and Z score. YAM is the abbreviation of "Young Adult Mean", sometimes referred to as the average percentage of young adults. For example, from the output layer 230, the bone mineral density expressed as per unit area (g / cm 2) and the estimated value 300 expressed as YAM, and the estimated value 300 expressed as YAM, the estimated value 300 expressed as a T score, and the estimated value 300 expressed as a Z score can also be output.

[0074] In addition, the storage unit 20 may store a plurality of estimation data 120. In this case, the plurality of estimation simple X-ray images respectively represented by the plurality of estimation data 120 in the storage unit 20 may include a plurality of simple X-ray images capturing the same type of site or a plurality of simple X-ray images capturing sites of different types. In addition, the plurality of estimation simple X-ray images may include a plurality of simple X-ray images capturing the site from the same direction or a plurality of simple X-ray images capturing the site from different directions. In other words, the plurality of estimation simple X-ray images may include a plurality of simple X-ray images capturing the site in the same direction or a plurality of simple X-ray images capturing the site in different directions. The control unit 10 inputs each of the plurality of estimation data 120 in the storage unit 20 into the input layer 210 of the neural network 200, and outputs the bone density estimation value 300 corresponding to each estimation data 120 from the output layer 230 of the neural network 200.

[0075] As described above, in this example, the neural network 200 is trained and the bone density in the neural network 200 is estimated using the image data of simple X-ray images. The image data of simple X-ray images, in other words, the image data of Roentgen images, are easily available because they are used in various medical examinations in many hospitals. Therefore, the bone density can be easily estimated without using an expensive device such as a DEXA device.

[0076] Furthermore, by using image data of a simple X-ray image captured for a medical examination or the like as estimation image data, bone density can be easily estimated by taking advantage of the medical examination or the like. Thus, by using the estimation device 1, services for hospital users can be improved.

[0077] In addition, in a frontal simple X-ray image of the chest, etc., it may be difficult to capture bones due to the influence of internal organs. On the other hand, in many hospitals, the possibility of capturing a frontal simple X-ray image is high. In this example, even when a frontal simple X-ray image in which bones may be difficult to capture is used as an estimation simple X-ray image or a learning simple X-ray image, bone density can be estimated. Therefore, bone density can be simply estimated using image data of an easily available frontal simple X-ray image. In addition, a chest frontal simple X-ray image is often captured in health diagnosis, etc., and can be said to be a simple X-ray image that is particularly easy to obtain. By using a chest frontal simple X-ray image as an estimation simple X-ray image or a learning simple X-ray image, bone density can be further simply estimated.

[0078] Furthermore, in this example, even when the plurality of learning simple X-ray images include a simple X-ray image that captures a different type of part from the estimation simple X-ray image, the bone density can be estimated based on the image data of the estimation simple X-ray image. This can improve the convenience of the estimation device 1 (in other words, the computer device 1).

[0079] Furthermore, in this example, even when the plurality of learning simple X-ray images include a simple X-ray image that captures a part in a different orientation from that of the estimation simple X-ray image, the bone density can be estimated based on the image data of the estimation simple X-ray image. This can improve the convenience of the estimation device 1.

[0080] Furthermore, in this example, even when the parts captured in the simple X-ray image for learning do not include the parts (bone) for which the reference bone density corresponding to the simple X-ray image for learning has been measured, the neural network 200 can estimate the bone density based on the learned parameters 110. Thus, the convenience of the estimation device 1 can be improved.

[0081] Furthermore, in this example, even when the orientation of the part captured in the simple X-ray image for learning and the orientation of the X-ray irradiation of the target part in the measurement of the reference bone density corresponding to the simple X-ray image for learning are different from each other, the neural network 200 can estimate the bone density based on the learned parameter 110. Thus, the convenience of the estimation device 1 can be improved.

[0082] Furthermore, in this example, even when the training data 140 includes a reference bone density measured from a site not included in the site captured by the estimation simple X-ray image, the bone density can be estimated based on the image data of the estimation simple X-ray image. This can improve the convenience of the estimation device 1.

[0083] In addition, the estimated bone density value 300 obtained by the estimation device 1 may be displayed on the display unit 40. In addition, the estimated bone density value 300 obtained by the estimation device 1 may be used by other devices.

[0084] Figure 6 1 is a diagram showing an example of a bone density estimation system 600 including an estimation device 1 and a processing device 500 that performs a process using the bone density estimated value 300 obtained by the estimation device 1. Figure 6 In the example of , the estimation device 1 and the processing device 500 can communicate with each other through the communication network 700. The communication network 700 includes, for example, at least one of a wireless network and a wired network. The communication network 700 includes, for example, a wireless LAN (Local Area Network) and the Internet.

[0085] In the estimation device 1, the communication network 700 is connected to the communication unit 30. The control unit 10 causes the communication unit 30 to send the bone density estimated value 300 to the processing device 500. The processing device 500 processes the bone density estimated value 300 received from the estimation device 1 via the communication network 700. For example, the processing device 500 is a display device such as a liquid crystal display device, which displays the bone density estimated value 300. At this time, the processing device 500 may also display the bone density estimated value 300 in a table or in a graph. In addition, when a plurality of estimation devices 1 are connected to the communication network 700, the processing device 500 may also display the bone density estimated values ​​300 obtained by the plurality of estimation devices 1. The structure of the processing device 500 may be the same as that of the processing device 500. Figure 1 The structure of the estimation device 1 shown is the same as that of the estimation device 1 , but may be different from that of the estimation device 1 .

[0086] The processing using the estimated bone density value 300 executed by the processing device 500 is not limited to the above-described example. In addition, the processing device 500 may communicate directly with the estimation device 1 wirelessly or by wire without passing through the communication network 700 .

[0087] <Other Examples of Estimation Data and Learning Data>

[0088] <Other Example 1>

[0089] In this example, in the learning data 130, for each learning image data, information related to the health status of a person having bones captured in the simple learning X-ray image represented by the learning image data is included. In other words, in the learning data 130, for each learning image data, information related to the health status of the subject (subject) of the simple learning X-ray image represented by the learning image data is included. Hereinafter, information related to the health status of the subject of the simple learning X-ray image is sometimes referred to as "learning health-related information." In addition, information related to the health status of the subject of the simple learning X-ray image represented by the learning image data is sometimes referred to as learning health-related information corresponding to the learning image data.

[0090] The health-related information for learning includes, for example, at least one of age information, gender information, height information, weight information, drinking habit information, smoking habit information, and information on the presence or absence of fracture history. The health-related information for learning is databased for each person and generated as a CSV (Comma-Separated Value) format file or a text format file. Each of the age information, height information, and weight information is represented as multiple-digit numerical data, for example. In addition, with respect to gender information, for example, "male" or "female" is represented as 1-digit data, and with respect to drinking habit information, "drinking habit" or "no drinking habit" is represented as 1-digit data. In addition, with respect to smoking habit information, 1-digit data is represented as "smoking habit" or "no smoking habit", and with respect to information on the presence or absence of fracture history, "fracture" or "no fracture" is represented as 1-digit data. In addition, the health-related information for learning may also include the body fat percentage or subcutaneous fat percentage of the subject.

[0091] When the learning data 130 includes the learning image data and the corresponding learning health-related information, the reference bone density corresponding to the learning image data (see Figure 4) is corresponded to the learning health-related information corresponding to the learning image data. That is, the measured value of the bone density of a certain person (baseline bone density) is corresponded to the learning image data representing a simple X-ray image for learning in which a certain person's bones are photographed, and the information related to the health status of the certain person (learning health-related information). Then, in the learning of the neural network 200, the learning image data and the learning health-related information corresponding thereto are simultaneously input to the input layer 210. Specifically, the learning image data is input to a part of the multiple artificial neurons constituting the input layer 210, and the learning health-related information is input to the other parts of the multiple artificial neurons. Then, when the learning image data and the learning health-related information corresponding thereto are input to the input layer 210, the output data 400 output from the output layer 230 and the baseline bone density corresponding to the learning image data and the learning health-related information are compared.

[0092] Furthermore, in this example, the estimation data 120 includes estimation image data and information related to the health status of a person having bones captured in the estimation simple X-ray image represented by the estimation image data. In other words, the estimation data 120 includes estimation image data and information related to the health status of the subject of the estimation simple X-ray image represented by the estimation image data. Hereinafter, the information related to the health status of the subject of the estimation simple X-ray image is sometimes referred to as "estimation health-related information (hereinafter, in another embodiment, sometimes also referred to as "individual data")". Furthermore, the information related to the health status of the subject of the estimation simple X-ray image represented by the estimation image data is sometimes referred to as estimation health-related information corresponding to the estimation image data.

[0093] The estimated health-related information is similar to the learning health-related information, and includes, for example, at least one of age information, gender information, height information, weight information, drinking habit information, smoking habit information, and information on the presence or absence of a fracture history. The estimated health-related information includes the same types of information as the learning health-related information. In addition, the estimated health-related information may also include the subject's body fat percentage or subcutaneous fat percentage, similar to the learning health-related information.

[0094] In this example, when estimating bone density, image data for estimation and health-related information for estimation corresponding thereto are simultaneously input to the input layer 210. Specifically, image data for estimation is input to a part of the plurality of artificial neurons constituting the input layer 210, and health-related information for estimation is input to the other parts of the plurality of artificial neurons. If image data for estimation and health-related information for estimation for a certain person are input to the input layer 210, an estimated value of the bone density of the certain person is output from the output layer 230.

[0095] In this way, by using not only the image data of a simple X-ray image but also the information on the health status of the subject of the simple X-ray image, it is possible to improve the estimation accuracy of the bone density.

[0096] <Other 2nd example>

[0097] In this example, the learning data 130 includes image data of N (N ≥ 2) simple X-ray images for learning, which capture parts of the same person and in which the directions of the captured parts are different from each other. Hereinafter, the N simple X-ray images for learning are sometimes collectively referred to as a "simple X-ray image group for learning".

[0098] The learning simple X-ray image group includes, for example, a front image and a side image of the same person. The learning simple X-ray image group includes, for example, a front simple X-ray image of the chest and a side simple X-ray image of the waist of a certain person. The image sizes of the front image and the side image included in the learning simple X-ray image group may also be different from each other. For example, the lateral width of the image size of the side image may also be smaller than the lateral width of the image size of the front image. Hereinafter, the image data of each learning simple X-ray image in the learning simple X-ray image group is sometimes collectively referred to as a "learning image data group".

[0099] The learning data 130 includes learning image data sets for different people. Thus, the learning data 130 includes a plurality of learning image data sets. Moreover, a correspondence is established with respect to one learning image data set and one reference bone density. That is, a correspondence is established with respect to a learning image data set for a certain person and a measured value of the bone density of the certain person (reference bone density).

[0100] In the learning of the neural network 200 of this example, each learning image data group is input to the input layer 210. When one learning image data group is input to the input layer 210, N learning image data constituting the one learning image data group are simultaneously input to the input layer 210. For example, it is assumed that the learning image data group includes the first learning image data and the second learning image data. In this case, the first learning image data (for example, image data of a simple X-ray image of the front of the chest) is input to a part of the multiple artificial neurons constituting the input layer 210, and the second learning image data (for example, image data of a simple X-ray image of the side of the waist) is input to another part of the multiple artificial neurons. Then, when the learning image data group is input to the input layer 210, the output data 400 output from the output layer 230 is compared with the reference bone density corresponding to the learning image data group.

[0101] In this example, the estimation data 120 includes image data of N estimation simple X-ray images that capture parts of the same person and in which the directions of the captured parts are different from each other. Hereinafter, the N estimation simple X-ray images may be collectively referred to as a "estimation simple X-ray image group".

[0102] The simple X-ray image group for estimation includes, for example, a front image and a side image of the same person. The simple X-ray image group for estimation includes, for example, a front simple X-ray image of a waist and a side simple X-ray image of a knee of a certain person. The image sizes of the front image and the side image included in the simple X-ray image group for estimation may also be different from each other. For example, the lateral width of the image size of the side image may also be smaller than the lateral width of the image size of the front image. Hereinafter, the image data of each simple X-ray image for estimation in the simple X-ray image group for estimation is sometimes collectively referred to as an "image data group for estimation".

[0103] In this example, when the estimation data 120 is used to estimate the bone density, N estimation image data constituting the estimation image data group are simultaneously input to the input layer 210. For example, it is assumed that the estimation image data group includes the first estimation image data and the second estimation image data. In this case, the first estimation image data is input to a part of the plurality of artificial neurons constituting the input layer 210, and the second estimation image data is input to another part of the plurality of artificial neurons. If the estimation image data group for a certain person is input to the input layer 210, the estimated value of the bone density of the certain person is estimated from the output layer 230.

[0104] In this way, by using image data of a plurality of simple X-ray images in which a part of the same subject is imaged and the orientations of the imaged parts are different from each other, it is possible to improve the estimation accuracy of the bone density.

[0105] In addition, the learning data 130 may also include a learning image data set and learning health-related information. In this case, in the learning of the neural network 200, the learning image data set and the learning health-related information for the same person are simultaneously input to the input layer 210. Similarly, the estimation data 120 may also include an estimation image data set and estimation health-related information. In this case, the estimation image data set and the estimation health-related information are simultaneously input to the input layer 210.

[0106] In each of the above examples, the same learned parameter 110 is used regardless of the type of bone captured in the X-ray image represented by the estimation image data, but the learned parameter 110 corresponding to the type of bone captured in the X-ray image represented by the estimation image data may be used. In this case, the neural network 200 has a plurality of learned parameters 110 corresponding to the types of bones captured in the X-ray image represented by the input estimation image data. The neural network 200 estimates the bone density using the learned parameter 110 corresponding to the type of bone captured in the X-ray image represented by the input estimation image data. For example, when the lumbar vertebra is captured in the X-ray image represented by the input estimation image data, the neural network 200 estimates the bone density using the learned parameter 110 for estimating the bone density of the lumbar vertebra. In addition, when the proximal part of the femur is captured in the X-ray image represented by the input estimation image data, the neural network 200 estimates the bone density using the learned parameter 110 for estimating the bone density of the proximal part of the femur. The neural network 200 uses, for example, a learned parameter 110 instructed by the user through the input unit 50, among the plurality of learned parameters 110. In this case, the user instructs the learned parameter 110 to be used by the neural network 200 based on the type of bone captured in the X-ray image represented by the estimation image data input to the neural network 200.

[0107] In the learning of the neural network 200, a plurality of learning image data each representing a plurality of X-ray images of the same type of bone is used to generate learned parameters 110 corresponding to the type of the bone.

[0108] As described above, the estimation device 1 and the bone density estimation system 600 are described in detail, but the above description is illustrative in all aspects, and the present disclosure is not limited to this. In addition, the above various examples can be combined and applied as long as they are not contradictory. Moreover, the countless examples not illustrated can be interpreted as examples obtained by conceiving without departing from the scope of the present disclosure.

[0109] Implementation method 2.

[0110] Figure 7 is a diagram showing an example of the structure of the estimation device 1A involved in the present embodiment. In the estimation device 1A, the approximator 280 also has a second neural network 900. The second neural network 900 can sense fractures based on the learned parameters 910. In addition, the estimation device 1A involved in the present embodiment has a structure equivalent to the estimation device 1 involved in the first embodiment, and the description of the equivalent structure is omitted. In addition, for the convenience of description, the neural network 200 described in the above example is referred to as the first neural network 200. The second neural network 900, for example, has a structure equivalent to the first neural network 200.

[0111] The second neural network 900 can sense fracture based on the same estimation image data as the estimation image data included in the estimation data 120 input to the first neural network 200. That is, based on one estimation image data, the bone density can be estimated in the first neural network 200, and the fracture can be sensed in the second neural network 900. In addition, the perception result 920 of the second neural network 900 can be output from the output layer 230 of the second neural network 900 in the same manner as in the above example.

[0112] In the learning of the second neural network 900, the parameters are learned using learning image data of bones that are not fractured and learning image data of bones that are fractured. In addition, in the training data, for each learning image data, a correspondence is established between information indicating whether there is a current fracture and information indicating the fracture site for the bone that has been photographed with the learning image data. In addition, the training data may also include information indicating past fracture history and information indicating past fracture sites. As a result, the second neural network 900 can sense the presence and location of a fracture for the bone that has been photographed with the estimation image data based on the estimation image data, and output the perception result 920.

[0113] In addition, the estimation device 1A according to the present embodiment may also be configured as follows: Figure 8As shown, there is a determination unit 930 for determining whether the subject has osteoporosis. The determination unit 930 can compare the estimation result 300 of the first neural network 200 and the perception result 920 of the second neural network 900 to determine whether the subject has osteoporosis.

[0114] The determination unit 900 may also determine osteoporosis based on a specific benchmark or a known criterion, for example. Specifically, the determination unit 900 may also determine that it is osteoporosis when the perception result 920 indicates a fracture at the vertebral body or the proximal part of the femur. In addition, when the estimated bone density value 300 output by the first neural network 200 indicates YAM, the determination unit 900 may also determine that it is osteoporosis when YAM is a value less than 80% and the perception result 920 indicates a fracture other than the vertebral body and the proximal part of the femur. In addition, the determination unit 900 may also determine that it is osteoporosis when the YAM indicated by the estimated bone density value 300 indicates a value below 70%.

[0115] In the estimation device 1A according to the present embodiment, the approximator 280 may also be configured as follows: Fig. 9 As shown, the third neural network 950 is further provided. The third neural network 950 can segment bones of the subject from the estimation image data included in the estimation data 120 based on the learned parameters 960.

[0116] The third neural network 950 outputs the part information indicating the part of the bone indicated by the pixel data for each pixel data of the input estimation image data. Thus, the bone photographed in the X-ray image indicated by the estimation image data can be segmented. The part information is sometimes referred to as segment data.

[0117] For example, when the lumbar spine is captured in the X-ray image represented by the input estimation image data, the third neural network 950 outputs, for each pixel data of the estimation image data, location information indicating which location among L1 to L5 of the lumbar spine the pixel data indicates. For example, when certain pixel data of the estimation image data indicates L1 of the lumbar spine, the third neural network 950 outputs location information indicating L1 as location information corresponding to the pixel data.

[0118] The third neural network 950 uses learned parameters 960 corresponding to the type of bones captured in the X-ray image represented by the estimation image data. The third neural network 950 has a plurality of learned parameters 960 corresponding to the plurality of bones. The third neural network 950 uses the learned parameters 960 corresponding to the type of bones captured in the X-ray image represented by the input estimation image data to divide the bones captured in the X-ray image represented by the estimation image data. For example, when the lumbar vertebrae are captured in the X-ray image represented by the input estimation image data, the third neural network 950 divides the lumbar vertebrae into L1 to L5 using the learned parameters 960 corresponding to the lumbar vertebrae. The third neural network 950 uses, for example, learned parameters 960 indicated by the user through the input unit 50, among the plurality of learned parameters 960. In this case, the user instructs the third neural network 950 to use learned parameters 960 based on the type of bone captured in the X-ray image represented by the estimation image data input to the third neural network 950.

[0119] The third neural network 950 may also classify the bone captured in the X-ray image represented by the input estimation image data into a first site where an implant is embedded, a second site where a tumor is present, and a third site where a fracture is present. In this case, the third neural network 950 outputs site information indicating which site among the first site, the second site, and the third site the pixel data indicates for each pixel data of the estimation image data. The third neural network 950 outputs site information indicating that the pixel data indicates a site other than the first site, the second site, and the third site when the pixel data indicates a site other than the first site, the second site, and the third site. When the third neural network 950 classifies the bone captured in the X-ray image represented by the estimation image data into the first site, the second site, and the third site, it can also be said that the third neural network 950 perceives the implant embedded in the bone captured in the X-ray image represented by the estimation image data, the fracture of the bone, and the tumor of the bone.

[0120] In the learning of the third neural network 950, a plurality of learning image data representing a plurality of X-ray images of the same type of bone are used to generate a learning-completed parameter 960 corresponding to the type of the bone. In the case where the third neural network 950 divides the bone captured in the X-ray image represented by the input estimation image data into the first part, the second part, and the third part, the plurality of learning image data includes learning image data representing X-ray images of cases in which implants are embedded, learning image data representing X-ray images of cases in which tumors are captured in the bones, and learning image data representing X-ray images of cases in which fractures are captured. In addition, in the training data, for each learning image data, annotation information for the division of the bone represented by the learning image data is included. In the annotation information, for each pixel data of the corresponding learning image data, part information indicating the part of the bone represented by the pixel data is included.

[0121] Alternatively, the first neural network 200 may estimate the bone density for each of the parts divided by the third neural network 950. In this case, Fig.10 As shown, the first neural network 200 is input with the estimation image data 121 and the part information 965 corresponding to each pixel data of the estimation image data 121 outputted by the third neural network 950 based on the estimation image data 121. The first neural network 200 outputs the bone density estimated value 300 of the part for each part divided by the third neural network based on the learned parameter 110 corresponding to the type of the bone photographed in the X-ray image represented by the estimation image data 121. For example, when the third neural network 950 divides the cervical vertebrae photographed in the X-ray image represented by the estimation image data 121 into L1 to L5, the first neural network 200 separately outputs the bone density estimated value 300 of L1, the bone density estimated value 300 of L2, the bone density estimated value 300 of L3, the bone density estimated value 300 of L4, and the bone density estimated value 300 of L5.

[0122] In the learning of the first neural network 200, a plurality of learning image data representing a plurality of X-ray images of the same type of bone are used to generate the learned parameters 110 corresponding to the type of the bone. In addition, the training data includes the reference bone density of each part of the bone represented by the learning image data for each learning image data.

[0123] The first neural network 200 uses, for example, learned parameters 110 instructed by the user through the input unit 50, among the plurality of learned parameters 110. In this case, the user instructs the learned parameters 110 to be used by the first neural network 200, based on the type of bone captured in the X-ray image represented by the estimation image data input to the first neural network 200.

[0124] When the third neural network 950 divides the bone captured in the X-ray image represented by the estimated image data into a first part where an implant is embedded, a second part where a tumor is present, and a third part where a fracture is present, the brightness of the first part of the image data representing the first part, the second part of the image data representing the second part where the tumor is present, and the third part of the image data representing the third part in the estimated image data can also be adjusted. Fig.11 It is a figure which shows the structural example in this case.

[0125] like Fig.11 As shown in FIG. 1 , the estimation image data 121 and the part information 965 corresponding to each pixel data of the estimation image data 121 outputted by the third neural network 950 based on the estimation image data 121 are inputted to the adjustment unit 968. The adjustment unit 968 specifies the first partial image data, the second partial image data, and the third partial image data included in the estimation image data 121 based on the part information 965. Then, the adjustment unit 968 adjusts the brightness of the specified first partial image data, the second partial image data, and the third partial image data.

[0126] The adjustment unit 968 stores, for example, the brightness of the first part captured in the general X-ray image as the first reference brightness. In addition, the adjustment unit 968 stores the brightness of the second part captured in the general X-ray image as the second reference brightness. Furthermore, the adjustment unit 968 stores the brightness of the third part captured in the general X-ray image as the third reference brightness. The adjustment unit 968 subtracts the first reference brightness from the brightness of the first partial image data, thereby adjusting the brightness of the first partial image data. Furthermore, the adjustment unit 968 subtracts the second reference brightness from the brightness of the second partial image data, thereby adjusting the brightness of the second partial image data. Furthermore, the adjustment unit 968 subtracts the third reference brightness from the brightness of the third partial image data, thereby adjusting the brightness of the third partial image data. The adjustment unit 968 inputs the estimated image data in which the brightness of the first partial image data, the second partial image data, and the third partial image data are adjusted in the estimated image data as the estimated image data after brightness adjustment to the first neural network 200. The first neural network 200 estimates the bone density of the bone captured in the X-ray image represented by the estimation image data based on the estimation image data after brightness adjustment.

[0127] Here, it is not easy to accurately estimate the bone density based on the first site where the implant is embedded, the second site where the tumor is located, and the third site where the fracture is located. By adjusting the brightness of the first partial image data representing the first site, the second partial image data representing the second site, and the third partial image data representing the third site to be smaller as described above, the bone density of the bone captured in the X-ray image represented by the estimation image data can be accurately estimated.

[0128] In addition, the adjustment unit 968 may input the estimation image data in which the brightness of the first partial image data, the second partial image data, and the third partial image data is forcibly set to zero as the estimation image data after brightness adjustment into the first neural network 200.

[0129] In addition, the third neural network 950 may also sense only one of the implant, fracture, and tumor. In addition, the third neural network 950 may also sense only two of the implant, fracture, and tumor. That is, the third neural network 950 may also sense at least one of the implant, fracture, and tumor.

[0130] In addition, the estimation device 1A may include the first neural network 200 and the third neural network 950 instead of the second neural network 900. In addition, the estimation device 1A may include at least one of the second neural network 900 and the third neural network 950 instead of the first neural network 200.

[0131] As described above, the estimation device 1A is described in detail, but the above description is illustrative in all aspects, and the present disclosure is not limited to this. In addition, the above various examples can be combined and applied as long as they are not contradictory. Moreover, countless examples that are not illustrated can be interpreted as examples obtained by assumption without departing from the scope of the present disclosure.

[0132] Implementation method 3.

[0133] Fig.12 1 is a diagram showing an example of the structure of the estimation device 1B involved in the present embodiment. The estimation device 1B has a fracture prediction unit 980. The fracture prediction unit 980 can predict the probability of fracture based on the estimation result 300 of the neural network 200 of the estimation device 1 involved in embodiment 1, for example. Specifically, an operation formula 990 representing the relationship between the estimation result associated with bone density (such as bone density, etc.) and the probability of fracture is obtained based on past literature, for example. The fracture prediction unit 980 stores the operation formula 990. The fracture prediction unit 980 can predict the probability of fracture based on the input estimation result 300 and the stored operation formula 990.

[0134] Furthermore, the calculation formula 990 may be a calculation formula that represents the relationship between the estimated result related to bone density and the probability of fracture after bone screw implantation. As a result, it is possible to study whether to implant a bone screw and a treatment plan including drug administration.

[0135] In addition, the estimation device 1B may include a second neural network 900 . In addition, the estimation device 1B may include a third neural network 950 .

[0136] As described above, the estimation device 1B is described in detail, but the above description is illustrative in all aspects, and the present disclosure is not limited to this. In addition, the above various examples can be combined and applied as long as they are not contradictory. Moreover, countless examples that are not illustrated can be interpreted as examples obtained by assumption without departing from the scope of the present disclosure.

[0137] Implementation method 4.

[0138] exist Fig.13 8 shows the concept of the structure of the estimation system 801 of this embodiment.

[0139] The estimation system 801 of the present disclosure can estimate the future bone mass of the subject based on an image of the subject's bones captured by an X-ray image, etc. The estimation system 801 of the present disclosure has a terminal device 802 and an estimation device 803. In addition, the so-called bone mass is an indicator related to bone density and is a concept that includes bone density.

[0140] The terminal device 802 can obtain input information I for input to the estimation device 803. The input information I can be, for example, an X-ray image. In this case, the terminal device 802 can be a device used by a doctor or the like to take an X-ray image of a subject. For example, the terminal device 802 can be a simple X-ray imaging device (in other words, a general X-ray imaging device or a x-ray imaging device).

[0141] In addition, the terminal device 802 is not limited to a simple X-ray imaging device. The terminal device 802 may be, for example, an X-ray fluoroscopic imaging device, CT (Computed Tomography), MRI (Magnetic Resonance Imaging), SPECT (Single Photon Emission Computed Tomography)-CT, or tomosynthesis. In this case, the input information I may be, for example, an X-ray fluoroscopic image, a CT (Computed Tomography) image, an MRI (Magnetic Resonance Imaging) image, a bone scintigraphy image, or a tomosynthesis image.

[0142] The estimation system 801 can be used, for example, for diagnosing osteoporosis in patients who go to the hospital. The estimation system 801 of the present disclosure uses, for example, a terminal device 802 set in an x-ray room to take x-ray images of the patient. Then, the image data is transmitted from the terminal device 802 to the estimation device 803, and through the estimation device 803, it is possible to estimate not only the bone mass or bone density of the patient at the current time point, but also the bone mass or bone density of the patient in the future than when the image was taken.

[0143] In addition, the terminal device 802 may not directly transmit the input information I to the estimation device 803. In this case, for example, the input information I acquired by the terminal device 802 may be stored in a storage medium, and the input information I may be input to the estimation device 803 via the storage medium.

[0144] exist Fig.14 , the concept of the structure of the estimation device 803 involved in this embodiment is shown.

[0145] The estimation device 803 can estimate the future bone mass or bone density of the subject based on the input information I acquired by the terminal device 802. In the estimation device 803, the future bone mass or bone density of the subject can be estimated based on the image data acquired by the terminal device 802, and the estimation result O can be output.

[0146] The estimation device 803 includes an input unit 831, an approximator 832, and an output unit 833. The input unit 831 inputs input information I from the terminal device 802. The approximator 832 can estimate the future bone mass or bone density based on the input information I. The output unit 833 can output the estimation result O predicted by the approximator 832.

[0147] The estimation device 803 has various electronic components and circuits. As a result, the estimation device 803 can form various structural elements. For example, the estimation device 803 can integrate multiple semiconductor elements to form at least one integrated circuit (such as IC: Integrated Circuit or LSI: Large Scale Integration), etc., or further integrate multiple integrated circuits to form at least one unit, etc., thereby forming various functional parts of the estimation device 803.

[0148] The plurality of electronic components may be active components such as transistors and diodes, or passive components such as capacitors. In addition, the plurality of electronic components and an integrated circuit formed by integrating them can be formed by a conventionally known method.

[0149] The input unit 831 inputs information used in the estimation device 803. For example, input information I having an X-ray image acquired by the terminal device 802 is input to the input unit 831. The input unit 831 has a communication unit, and the input information I acquired by the terminal device 802 is directly input from the terminal device 802. In addition, the input unit 831 may also include an input device capable of inputting the input information I or other information. The input device may be, for example, a keyboard, a touch screen, or a mouse.

[0150] The approximator 832 estimates the future bone mass or bone density of the subject based on the information input to the input unit 831. The approximator 832 has AI (Artificial Intelligence). The approximator 832 has a program that functions as AI and various electronic components and circuits for executing the program. The approximator 832 has a neural network.

[0151] The approximator 832 performs learning processing on the relationship between input and output in advance. That is, machine learning can be applied to the approximator 832 using learning data and training data, so that the approximator 832 can calculate the estimation result O based on the input information I. In addition, the learning data or training data only needs to be the input information I input to the estimation device 803 and the data corresponding to the estimation result O output from the estimation device 803.

[0152] Fig.15 The concept of the structure of the approximator 832 of the present disclosure is shown.

[0153] The approximator 832 has a first neural network 8321 and a second neural network 8322. The first neural network 8321 can be any neural network suitable for processing time series information. For example, the first neural network 8321 can be a ConvLSTM network that combines LSTM (Long short-term memory) and CNN (Convolutional Neural Network). The second neural network 8322 can be, for example, a convolutional neural network constructed using CNN.

[0154] The first neural network 8321 includes an encoder E and a decoder D. The encoder E can extract the time variation of the input information I and the feature quantity of the position information. The decoder D can calculate a new feature quantity based on the feature quantity extracted by the encoder E, the time variation of the input information I, and the initial value.

[0155] Fig.16 The concept of the structure of the first neural network 8321 of the present disclosure is shown.

[0156] The encoding unit E has a plurality of ConvLSTM (Convolutional Long short-term memory) layers E1. The decoding unit D has a plurality of ConvLSTM (Convolutional Long short-term memory) layers D1. Each of the encoding unit E and the decoding unit D may also have more than three ConvLSTM layers E1 and D1. In addition, the number of the plurality of ConvLSTM layers E1 and the number of the plurality of ConvLSTM layers D1 may also be the same.

[0157] In addition, the contents learned by the multiple ConvLSTM layers E1 may be different. The contents learned by the multiple ConvLSTM layers D1 may be different. For example, a certain ConvLSTM layer learns subtle contents such as the change of one pixel at a time, while another ConvLSTM layer learns coarse contents such as the change of the entire image.

[0158] Fig.17 The concept of the structure of the second neural network 8322 is shown.

[0159] The second neural network 8322 has a conversion unit C. The conversion unit C can convert the feature amount calculated by the decoding unit D of the first neural network 8321 into bone mass or bone density. The conversion unit C has a plurality of convolutional layers C1, a plurality of pooling layers C2, and a fully connected layer C3. The fully connected layer C3 is located in the front stage of the output unit 33. Moreover, in the conversion unit C, the convolutional layer C1 and the pooling layer C2 are alternately configured between the first neural network 8321 and the fully connected layer C3.

[0160] In addition, the learning data is input to the encoding unit E of the approximator 832 when the approximator 832 is learning. The training data is compared with the output data output from the conversion unit C of the approximator 832 when the approximator 832 is learning. The training data is data indicating the values ​​measured using a conventional bone density measuring device.

[0161] The output unit 833 can display the estimation result O. The output unit 833 is, for example, a liquid crystal display or an organic EL display. The output unit 833 can display various information such as text, symbols, and graphics. The output unit 833 can display numbers or images, for example.

[0162] The estimation device 803 of the present disclosure further includes a control unit 834 and a storage unit 835. The control unit 834 can centrally manage the operation of the estimation device 803 by controlling other components of the estimation device 803.

[0163] The control unit 834 includes, for example, a processor. The processor may include, for example, one or more processors, controllers, microprocessors, microcontrollers, application-specific integrated circuits (ASICs), digital signal processing devices, programmable logic devices, or any combination of these devices or calibrations, or other basic devices or calibrations. The control unit 834 includes, for example, a CPU.

[0164] The storage unit 835 includes, for example, a non-transitory storage medium such as a RAM (Random Access Memory) or a ROM (Read-Only Memory) that can be read by the CPU of the control unit 834. The storage unit 835 stores a control program such as firmware for controlling the estimation device 803. In addition, the storage unit 835 may store input information I, learning data, and training data.

[0165] The processor of the control unit 834 can execute one or more data calculation programs or processes according to the control program of the storage unit 835. Various functions of the control unit 834 are realized by the CPU of the control unit 834 executing the control program in the storage unit 835.

[0166] In addition, the control unit 834 may perform other processing as pre-processing of the calculation processing as necessary.

[0167] <An example of input information, learning data, and training data>

[0168] The input information (hereinafter also referred to as the first input information I1) has image data of a bone of an object to be estimated for bone mass or bone density. The image data may be, for example, a simple X-ray image. The object to be estimated for bone mass or bone density is, for example, a person. In this case, it can be said that the first input information I1 is image data of a simple X-ray image of a person's bones. A simple X-ray image is a two-dimensional image, also called a general X-ray image or a Roentgen image.

[0169] The first input information I1 is preferably a simple X-ray image that is relatively easy to obtain, but is not limited thereto. For example, there is a case where the bone mass or bone density can be estimated more accurately by using an X-ray fluoroscopic image, a CT (Computed Tomography) image, an MRI (Magnetic Resonance Imaging) image, a bone scintigraphy image, or a tomosynthesis image as input information.

[0170] In addition, the estimated object of bone mass or bone density can also be other than human. For example, the estimated object of bone mass or bone density can also be animals such as dogs, cats or horses. In addition, the bone set as the object is mainly cortical bone and spongy bone of biological origin, but in the bone set as the object, artificial bone with calcium phosphate as the main component or regenerated bone artificially manufactured by regenerative medicine, etc. can also be included.

[0171] The X-ray image may be taken from the neck, chest, waist, proximal part of the femur, knee joint, ankle joint, shoulder joint, elbow joint, wrist joint, finger joint or jaw joint. In addition, parts other than bones may be taken from the X-ray image. For example, in the case of a simple chest X-ray image, an image of the lungs and an image of the thoracic spine may be included. The X-ray image may be a frontal image of the object part taken from the front or a side image of the object part taken from the side.

[0172] The learning data or training data may be any data corresponding to the input information I input to the estimation device 803 and the estimation result O output from the estimation device 803 .

[0173] The learning data has the same type of information as the first input information I1. For example, if the first input information I1 is a simple X-ray image, the learning data only needs to have the simple X-ray image. Furthermore, if the first input information I1 is a simple chest X-ray image, the learning data only needs to have the simple chest X-ray image.

[0174] The learning data includes learning image data of multiple simple X-ray images of bones. The shooting parts of the multiple learning image data include, for example, at least one of the neck, chest, waist, proximal part of the femur, knee joint, ankle joint, shoulder joint, elbow joint, wrist joint, finger joint and jaw joint. The learning data may include image data of some of the 11 types of image data or all types of image data. In addition, the multiple learning image data may include both frontal images and side images.

[0175] In the learning data, bones of multiple different people are photographed. For each of the multiple learning image data, the measured value of the bone mass or bone density of the subject of each learning image data is established as training data. The measured value of the bone mass or bone density is the measured value measured at a time period that is almost the same as the time period when the learning image data is photographed.

[0176] In addition, the learning image data of the learning data may be a series of data captured on different time axes of the same person. That is, the learning image data may include the first learning data having an X-ray image of bones and an image of the same person as the first learning data, and may also include the second learning data having an X-ray image captured after the first learning data.

[0177] The learning image data may be a data set obtained by photographing the same part of another person but with different ages, etc. The learning image data may be a series of data obtained by photographing the same person and the same part but with different time axes.

[0178] The learning data and the first input information I1 may also use input information obtained by compressing image data representing grayscale of a simple X-ray image captured by a simple X-ray imaging device (in other words, a general X-ray imaging device or a x-ray imaging device) and reducing the number of grayscales. For example, consider a case where the number of pixel data of the image data is greater than (1024×640) and the number of bits of the pixel data is 16 bits. In this case, input information obtained by compressing the number of pixel data to, for example, (256×256), (1024×512), or (1024×640) and reducing the number of bits of the pixel data to 8 bits is used as the first input information I1 and the learning data.

[0179] The training data includes, for each of the plurality of learning image data included in the learning data, a measured value of the bone mass or bone density of the bone captured in the simple learning X-ray image represented by the learning image data. The bone mass or bone density may be measured, for example, by a DEXA (dual-energy X-ray absorptiometry) method or an ultrasonic method.

[0180] <Neural network learning example>

[0181] The control unit 834 performs machine learning using the learning data and the training data on the approximator 832, so that the approximator 832 can calculate the estimated result O related to the bone mass or bone density based on the input information I. The approximator 832 is optimized by the well-known machine learning using the training data. The approximator 832 performs calculations based on the learning data input to the encoding unit E, and adjusts the variable parameters in the approximator 832 so that the difference between the pseudo estimated result output from the conversion unit C and the training data is small.

[0182] Specifically, the control unit 834 inputs the learning data in the storage unit 835 to the encoding unit E. When the learning data is input to the encoding unit E, the control unit 834 inputs the plurality of pixel data constituting the learning image data to the plurality of artificial neurons constituting the encoding unit E. Then, when the learning image data is input to the encoding unit E, the control unit 834 adjusts the parameters so that the error of the estimated result O output from the conversion unit C relative to the bone mass or bone density actual measurement value corresponding to the learning image data becomes smaller. The adjusted parameters become the parameters after learning and are stored in the storage unit 835.

[0183] As a parameter adjustment method, for example, an error back propagation method may be used. The parameters include, for example, parameters used by the encoding unit E, the decoding unit D, and the transformation unit C. Specifically, the parameters include weighting coefficients used by the ConvLSTM layers of the encoding unit E and the decoding unit D, and the convolutional layer and the fully connected layer of the transformation unit C.

[0184] As a result, the approximator 832 performs operations based on the learned parameters for the input information I input to the encoder E, and outputs the estimated result O from the transform unit C. When the X-ray image data as the input information I is input to the encoder E, the plurality of pixel data constituting the image data are respectively input to the plurality of artificial neurons constituting the input unit 831. Then, the ConvLSTM layer, the convolution layer, and the fully connected layer can perform operations using the weighting coefficients included in the learned parameters, thereby outputting the estimated result O.

[0185] As described above, in the estimation system 801, the image data of the simple X-ray image is used to learn the approximator 832 and estimate the bone mass or bone density in the approximator 832. Therefore, input information I can be input to the estimation system 801, and the future bone mass or bone density can be output as an estimation result O.

[0186] The estimation result O of the estimation system 801 may be any estimation result for a date in the future relative to the acquisition date of the input information I. For example, the estimation system 1 can estimate the bone mass or bone density from 3 months to 50 years after the photographing, more preferably from 6 months to 10 years after the photographing.

[0187] The estimated result O can be output as a value. For example, it can be represented by at least one of YAM (Young Adult Mean), T score, and Z score. For example, the output unit 833 can output an estimated value represented by YAM, an estimated value represented by YAM, an estimated value represented by a T score, and an estimated value represented by a Z score.

[0188] In addition, the estimation result O can also be output as an image. In the case where the estimation result O is an image, for example, an image that simulates an X-ray image can be displayed. In addition, the so-called image that simulates an X-ray image is an image that imitates an X-ray image. In addition, when ConvLSTM is used to learn a series of data that is taken by the same person and the same part and has different time axes, it is possible to predict the time change of the image. As a result, it is possible to generate a future image based on an X-ray image of another patient at one time point.

[0189] In the learning data and the input information I, internal organs, muscles, fat, or blood vessels may be captured in addition to bones. Even in this case, high-precision estimation can be performed.

[0190] The first input information I1 may include individual data of the subject (first individual data). The first individual data may be, for example, age information, gender information, height information, weight information, or fracture history, etc. As a result, a highly accurate estimation can be performed.

[0191] The first input information I1 may include second individual data of the subject. The second individual data may include information on blood pressure, lipids, cholesterol, neutral fat, and blood sugar levels, for example. As a result, high-precision estimation can be performed.

[0192] The first input information I1 may include the subject's lifestyle information. The lifestyle information may be any information on drinking habits, smoking habits, exercise habits, or eating habits. As a result, a high-precision estimation can be performed.

[0193] The first input information I1 may also include bone metabolism information of the subject. The so-called bone metabolism information may be, for example, bone absorption capacity or bone formation capacity. For example, they can be measured by at least one of type I collagen cross-linked N-terminal peptide (NTX), type I collagen cross-linked C-terminal peptide (CTX), tartrate-resistant acid phosphatase (TRACP-5b), deoxypyridinoline (DPD), bone alkaline phosphatase (BAP), type I collagen cross-linked N-propeptide (P1NP), and low carboxyl osteocalcin (ucOC) as a bone association matrix marker as a bone absorption marker. The bone absorption marker can be measured by serum or urine as a test body.

[0194] In the estimation system 801, second input information I2 related to the future scheduled action of the subject can also be further input as input information I. The so-called second input information I2, for example, can be information related to individual data that is scheduled to be improved or improved, or living habits, exercise habits, and eating habits that are scheduled to be improved or improved. Specifically, the second input information I2 can be information such as weight data, drinking habits, smoking habits, sunbathing time, number of steps or walking distance per day, intake of dairy products, or improved intake of foods containing more vitamin D such as fish and mushrooms. As a result, the estimation system 801 can indicate an estimated result O that the future bone mass or bone density has improved.

[0195] In addition, the second input information I2 may be, for example, information related to a lifestyle that is expected to deteriorate. As a result, the estimation system can indicate an estimation result O of a future deterioration in bone mass or bone density.

[0196] In the estimation system 801, as input information I, third input information I3 related to a therapy for the subject may be further input. The third input information I3 is, for example, information related to a physical therapy or a drug therapy. Specifically, the third input information I3 only needs to be at least one of calcium tablets, female hormone drugs, vitamin drugs, bisphosphonate drugs, SERM (Selective Estrogen Receptor Modulator) drugs, calcitonin drugs, thyroid hormone drugs, and denosumab drugs.

[0197] In the estimation system 801, the estimation result O may output a first result O1 based only on the first input information I1 and a second result O2 based on the first input information I1 and at least one of the second and third input information I2 and I3. As a result, the effects of planned actions in the future can be compared.

[0198] In the estimation system 801, the estimation result O can output not only the future bone mass or bone density but also the current result. As a result, the changes in bone mass or bone density over time can be compared.

[0199] Fig.18 The concept of the structure of the approximator 832 of another embodiment of the estimation system 1 is shown.

[0200] The estimation device 803 of the estimation system 801 may include a first approximator 832a and a second approximator 832b. That is, in addition to the above-mentioned approximator 832 (first approximator 832a), the second approximator 832b may be included. The second approximator 832b may be, for example, a CNN.

[0201] In this case, in the estimation system 801, the first approximator 832a outputs the first image and the first value to the first output unit 833a as the first estimation result O1. Furthermore, the second approximator 832b outputs the second value to the second output unit 833b as the second estimation result O2 based on the first image from the first output unit 833a. As a result, the first value and the second value can be compared as the estimation result O of the future bone mass or bone density.

[0202] The estimation system 801 may output a third value based on the first value and the second value as the estimation result O. As a result, for example, the estimation result O can be a result (third value) obtained by calibrating the first value based on the second value.

[0203] As described above, the estimation system 801 is described in detail, but the above description is illustrative in all aspects, and the present disclosure is not limited to this. In addition, the above various examples can be combined and applied as long as they are not contradictory. Moreover, the countless examples not illustrated can be interpreted as examples obtained by conceiving without departing from the scope of the present disclosure.

[0204] Explanation of symbols

[0205] 1: Computer device (presumption device);

[0206] 20: storage unit;

[0207] 100: control program;

[0208] 110, 910, 960: learning completed parameters;

[0209] 120: Presumptive data;

[0210] 130: Learning data;

[0211] 140: training data;

[0212] 200: Neural Networks;

[0213] 210: input layer;

[0214] 230: output layer;

[0215] 280, 832: approximator;

[0216] 500: processing device;

[0217] 600: Bone density estimation system;

[0218] 801: Presumption system;

[0219] 802: terminal device;

[0220] 803: Presumption device;

[0221] 831: Input unit;

[0222] 833: output unit;

[0223] 834: Control Department;

[0224] 835: storage unit;

[0225] 900, 8322: 2nd neural network;

[0226] 930: Judgment Department;

[0227] 950: 3rd neural network;

[0228] 980: Fracture Prediction Department;

[0229] 8321: 1st neural network;

[0230] O: Presumed result;

[0231] I: input information;

[0232] E: Encoding department;

[0233] D: decoding unit;

[0234] C: Transformation unit.

Claims

1. An estimation device comprising: An input unit for inputting input information including a Roentgen image of a bone; and an approximator capable of estimating an estimation result associated with the bone density of the bone based on the input information input into the input unit, The approximator has learned parameters for obtaining an estimated result related to the bone density of the bone based on the input information. The approximator has a neural network, The neural network receives the input information, estimates the estimation result based on the input information and the learned parameter, and outputs the estimated estimation result. The learned parameters are obtained by learning the relationship between learning data and training data, wherein the learning data includes image data of multiple simple X-ray images of bones, and the training data includes, for each of the multiple learning image data included in the learning data, a measured value of bone density of a person having a bone captured in the learning simple X-ray image represented by the learning image data.

2. The estimation device according to claim 1, wherein: The input information includes a first simple X-ray image.

3. The estimation device according to claim 1 or 2, wherein: The estimated results are calculated by bone mineral density per unit area (g / cm 2 ), bone mineral density per unit volume (g / cm 3 ), YAM, T score, or Z score.

4. The estimation device according to claim 1 or 2, wherein: The approximator is able to estimate future estimated results associated with bone density.

5. The estimation device according to claim 4, wherein: The approximator comprises: an encoding unit for extracting a feature quantity of a temporal change of the input information and position information; a decoding unit for calculating a new feature quantity based on the feature quantity, the temporal change of the input information and an initial value; and a conversion unit that converts the new feature value into bone density.

6. The estimation device according to claim 1 or 2, wherein: The estimation result has an image.

7. The estimation device according to claim 6, wherein: The image has an image resembling an X-ray image.

8. The estimation device according to claim 6, wherein: The approximator includes: a first approximator capable of estimating the image and a first value as the estimation result; and a second approximator capable of estimating a second value from the image as the estimation result.

9. The estimation device according to claim 8, wherein: A third value is output as the estimation result based on the first value and the second value.

10. The estimation device according to claim 1 or 2, wherein: The input information also includes individual data of the subject.

11. The estimation device according to claim 4, wherein: When the input information is set as the first input information, second input information related to the future planned behavior of the subject is further input to the input unit.

12. The estimation device according to claim 4, wherein: Furthermore, third input information related to a therapy for the subject is input to the input unit.

13. The estimation device according to claim 11, wherein: The estimation result includes: a first result based only on the first input information; and a second result based on the first input information and the second input information.

14. The estimation device according to claim 4, wherein: The approximator performs a learning process using first learning data including the X-ray image and second learning data including the X-ray image which is an image of the same person as the first learning data and is captured after the first learning data.

15. The estimation device according to claim 1 or 2, wherein: The approximator further comprises: a third approximator capable of sensing at least one of the bone fracture, tumor, and implant based on the input information input to the input unit; The third approximator has parameters for obtaining learning completion of at least one of the perceptions of bone fracture, tumor, and implant based on the input information.

16. The estimation device according to claim 1 or 2, wherein: The estimation device further includes a determination unit that determines whether the subject suffers from osteoporosis based on the estimation result of the approximator.

17. The estimation device according to claim 1 or 2, wherein: The estimation device further includes a fracture prediction unit configured to predict a fracture of the bone based on an estimation result obtained by the approximator.

18. The estimation device according to claim 1 or 2, wherein: The approximator divides the bones captured in the Roentgen image and estimates the bone density for each divided bone.

19. The estimation device according to claim 1, wherein: The Roentgen image of bones is a Roentgen image of bones used in medical examinations in hospitals.

20. A presumption system comprising: An input unit for inputting input information including a Roentgen image of a bone; and an approximator having learned parameters for obtaining an estimated result associated with the bone density of the bone based on the input information, and capable of estimating the estimated result associated with the bone density of the bone based on the input information input to the input unit, When input information is input to the input unit, the input information is processed by the approximator. The approximator has a neural network, The neural network receives the input information, estimates the estimation result based on the input information and the learned parameter, and outputs the estimated estimation result. The learned parameters are obtained by learning the relationship between learning data and training data, wherein the learning data includes image data of multiple simple X-ray images of bones, and the training data includes, for each of the multiple learning image data included in the learning data, a measured value of bone density of a person having a bone captured in the learning simple X-ray image represented by the learning image data.

21. A computer-readable non-transitory recording medium recording an estimation program, the estimation program causing a device to function as a neural network, The neural network is input with input information having an X-ray image of a bone, performs calculations based on learned parameters, and outputs an estimated value of the bone density of the bone captured in the image, wherein the learned parameters are used to obtain an estimated result associated with the bone density of the bone based on the input information. The learned parameters are obtained by learning the relationship between learning data and training data, wherein the learning data includes image data of multiple simple X-ray images of bones, and the training data includes, for each of the multiple learning image data included in the learning data, a measured value of bone density of a person having a bone captured in the learning simple X-ray image represented by the learning image data.

Images