X-ray imaging device, method for generating learned model, and image processing method

Abstract

The X-ray imaging apparatus (100) comprises: an X-ray irradiation unit (2); an X-ray detection unit (3); a pulse width setting unit (72) for setting the pulse width of the irradiated X-rays; an X-ray irradiation control unit (71) for causing the X-ray irradiation unit (2) to irradiate X-rays; an X-ray image generation unit (73) for generating an X-ray image (10); and a device detection unit (74) for detecting a device (200) in the X-ray image (10) based on a learned model (80) generated by machine learning, wherein the pulse width is a jitter suppression pulse width that is less than or equal to the pulse width at which the detection accuracy of the device (200) based on the learned model (80) is maximized.

Description

Technical Field

[0001] The present invention relates to an X-ray imaging device, a method for generating a learned model, and an image processing method. Background Art

[0002] Conventionally, radiographic devices that use machine learning to detect markers from radiographic images are known. For example, Japanese Patent Application Laid-Open No. 2017-185007 discloses such a radiographic device.

[0003] The radiographic apparatus described in Japanese Patent Application Laid-Open No. 2017-185007 is an apparatus that captures radiographic images for imaging the inside of a subject by irradiating radiation during coronary interventional therapy. The radiographic apparatus detects the position and range of a marker for the radiographic image from the radiographic image by image recognition based on learning result data. The marker is set as a mark of a stent inserted into the subject by a catheter. In addition, the learning result data is acquired in advance by machine learning using multiple rotated images obtained by rotating the image including the marker at multiple angles. Moreover, the radiographic apparatus described in Japanese Patent Application Laid-Open No. 2017-185007 enhances the display of the stent in the radiographic image based on the position of the detected marker.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2017-185007 Summary of the Invention

[0007] Problems to be solved by the invention

[0008] Although not described in Japanese Patent Application Laid-Open No. 2017-185007, it is considered to detect medical devices such as catheters and guidewires that are not marked in radiographic images (X-ray images). In this case, it is considered to detect the devices in the X-ray images based on a learned model generated through machine learning.

[0009] However, medical devices such as catheters and guidewires that are placed in the body may move irregularly due to movements such as the beating of the heart in the subject's body. For example, when multiple X-ray images of medical devices such as catheters and guidewires placed in the coronary arteries are taken, since the coronary arteries move (deform) irregularly with the beating of the heart, the positions and shapes of medical devices such as catheters and guidewires also change irregularly with the beating of the heart. Specifically, at the time point when the heart is most dilated and at the time point when the heart is least contracted, the movement speed of the coronary arteries is small (approximately zero), so the change in the position and shape of the device is small. In addition, between the time point when the heart is most dilated and the time point when the heart is least contracted, the movement speed of the coronary arteries increases according to the rate of change of the heart's volume, so the change in the position and shape of the device increases.

[0010] Because the movement (deformation) of devices within the body is not constant (it is irregular), the devices in captured X-ray images exhibit varying degrees of motion. Therefore, in order to detect devices in X-ray images based on a learned model, the learned model must be trained to account for various types of motion. In other words, machine learning must be performed using a large number of training images to account for various types of motion. In this case, machine learning may not converge, and even if it does converge, device detection accuracy based on the learned model generated through machine learning is low.

[0011] The present invention is completed to solve the above-mentioned problems. One of the purposes of the present invention is to provide an X-ray imaging device, a method for generating a learned model, and an image processing method that can detect medical equipment retained in the body of a subject in an X-ray image with high precision based on a learned model generated by machine learning.

[0012] Solutions for solving problems

[0013] In order to achieve the above-mentioned purpose, the first aspect of the present invention provides an X-ray imaging apparatus comprising: an X-ray irradiation unit that irradiates an object having a medical device implanted therein with X-rays; an X-ray detection unit that detects the X-rays after passing through the object; a pulse width setting unit that sets the pulse width of the X-rays irradiated from the X-ray irradiation unit; an X-ray irradiation control unit that causes the X-ray irradiation unit to irradiate X-rays of the pulse width set by the pulse width setting unit; an X-ray image generation unit that generates an X-ray image based on the X-rays of the pulse width detected by the X-ray detection unit; and a device detection unit that detects the device in the X-ray image obtained from the X-ray image based on the pulse width generated by the X-ray image generation unit based on a learned model generated by machine learning, wherein the pulse width is a jitter-suppressed pulse width that is less than or equal to the pulse width that maximizes the detection accuracy of the device based on the learned model.

[0014] The second aspect of the present invention provides a method for generating a learned model, comprising the following steps: obtaining a training input X-ray image in a manner corresponding to an X-ray image obtained based on a jitter-suppressed pulse width of X-rays, wherein the jitter-suppressed pulse width is a pulse width that is less than or equal to the maximum detection accuracy of a medical device retained in the body of the subject, and wherein the training input X-ray image is an image generated in a manner that simulates a device in the body of the subject in an X-ray image obtained based on a jitter-suppressed pulse width; obtaining training output information representing the position or shape of the device in the training input X-ray image; and generating a learned model through machine learning based on the training input X-ray image and the training output information.

[0015] The image processing method of the third aspect of the present invention includes the following steps: based on a learned model generated by machine learning, setting the pulse width of the X-rays irradiated to generate the X-ray image to a jitter-suppressed pulse width, wherein the jitter-suppressed pulse width is less than the pulse width that maximizes the detection accuracy of the medical device retained in the body of the subject from the generated X-ray image; irradiating the subject with the device retained in the body with X-rays of the set jitter-suppressed pulse width; detecting the X-rays after passing through the subject; generating an X-ray image based on the X-rays with the detected jitter-suppressed pulse width; and detecting the device in the X-ray image from the X-ray image obtained based on the generated X-rays with the jitter-suppressed pulse width based on the learned model.

[0016] Effects of the Invention

[0017] According to the X-ray imaging apparatus of the first aspect and the image processing method of the third aspect, the pulse width of the X-rays emitted to generate X-ray images is set to a jitter-suppressed pulse width that maximizes the detection accuracy of medical devices placed within the subject's body from the generated X-ray images. Therefore, it is possible to detect devices from X-ray images obtained based on X-rays with the jitter-suppressed pulse width based on machine learning. Therefore, since X-ray images are captured using X-rays with a relatively small pulse width (i.e., the jitter-suppressed pulse width) that maximizes the detection accuracy of devices, even if a device within the subject's body moves irregularly due to irregular in-vivo motion, the types of device jitter in the captured X-ray images can be suppressed from varying. Thus, even if a device within the subject's body moves irregularly, the device jitter in the X-ray image is suppressed to a smaller than fixed value, thereby suppressing a decrease in device detection accuracy based on the learned model. As a result, medical devices placed within the subject's body can be detected in X-ray images with high accuracy based on the learned model generated through machine learning.

[0018] Furthermore, when the pulse width of the X-rays used to generate the X-ray image is set to be larger than the pulse width that maximizes the detection accuracy of the device, the detection accuracy of the device in the X-ray image becomes lower. Furthermore, when the pulse width is increased, the dose of X-rays irradiated to the subject also increases. In contrast, in the present invention, X-ray images are generated by X-rays with a jitter-suppressed pulse width, where the jitter-suppressed pulse width is a pulse width that is less than the pulse width that maximizes the detection accuracy of the device. Therefore, since X-ray images are generated by X-rays with a relatively small jitter-suppressed pulse width, it is possible to effectively suppress a decrease in the detection accuracy of the device and to suppress an increase in the dose of X-rays irradiated to the subject. As a result, the device can be detected with further high precision, and the dose of X-rays irradiated to the subject can be reduced.

[0019] Furthermore, according to the method for generating a learned model according to the second aspect, the types of jitter can be suppressed by acquiring training input X-ray images generated to simulate the internal devices of the subject in X-ray images obtained with X-rays using a jitter-suppressed pulse width. This reduces the need to acquire a large number of training input X-ray images to account for various types of device jitter. However, using a large number of training input X-ray images for machine learning to account for X-ray images with various types of jitter makes it difficult to achieve convergence, and even if the learning does converge, the detection accuracy of the learned model decreases. In contrast, in the present invention, training input X-ray images are acquired to simulate the internal devices of the subject in X-ray images obtained with X-rays using a jitter-suppressed pulse width. This configuration reduces the types of jitter in the X-ray images obtained with X-rays using a jitter-suppressed pulse width, compared to a case where the pulse width is not restricted and the device jitter in the image varies, thereby reducing the types of jitter in the images acquired as training images. This reduces the number of X-ray image types used for training input, making learning more likely to converge compared to situations with multiple types of device vibration. Furthermore, a learned model capable of highly accurate device detection can be generated. Consequently, a method for generating a learned model capable of highly accurate detection of medical devices placed within a subject in X-ray images can be provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram showing the structure of an X-ray imaging apparatus according to one embodiment.

[0021] Figure 2 This is a block diagram showing the configuration of an X-ray imaging apparatus according to one embodiment.

[0022] Figure 3 This is a diagram for explaining the functional structure of a control unit according to one embodiment.

[0023] Figure 4 This is a diagram for explaining an X-ray image with a large pulse width.

[0024] Figure 5 This is a diagram for explaining an X-ray image with a small pulse width.

[0025] Figure 6 is a diagram for explaining the generation of enhanced images.

[0026] Figure 7 It is a figure for demonstrating the display of a display part.

[0027] Figure 8 This is a diagram for explaining the generation of a learned model.

[0028] Figure 9 This is a diagram for explaining the generation of training input X-ray images.

[0029] Figure 10 This is a diagram for explaining the relationship between the pulse width of X-rays and the detection accuracy of catheters.

[0030] Figure 11 This is a diagram for explaining the relationship between the pulse width of X-rays and the detection accuracy of the guide wire.

[0031] Figure 12 This is a flowchart for explaining a method for generating a learned model according to one embodiment.

[0032] Figure 13 This is a flowchart for explaining an image processing method according to one embodiment. DETAILED DESCRIPTION

[0033] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0034] Reference Figures 1 to 11 The configuration of the X-ray imaging apparatus 100 according to one embodiment, an image processing method according to one embodiment, and a method for generating a learned model according to one embodiment will be described.

[0035] (Structure of X-ray Radiography Apparatus)

[0036] First, refer to Figure 1 and Figure 2 The structure of the X-ray imaging apparatus 100 will be described.

[0037] like Figure 1 As shown, an X-ray imaging device 100 irradiates a subject 101 having a medical device 200 inserted therein with X-rays. Furthermore, the X-ray imaging device 100 performs X-ray photography by detecting the X-rays that have passed through the subject 101. For example, during percutaneous coronary intervention (PCI), the X-ray imaging device 100 generates images for confirming the internal conditions of the subject 101. Percutaneous coronary intervention is a treatment for diseases such as angina pectoris and myocardial infarction caused by narrowing and blockage of the coronary arteries of the heart, using the device 200 to eliminate the narrowing and blockage of the blood vessels.

[0038] <About the device>

[0039] The device 200 is placed in the body of the subject 101. The device 200 includes, for example, a catheter or a guidewire placed in a blood vessel near the heart of the subject 101. The device 200 is made of a flexible material that can move in a manner that allows it to be inserted into a blood vessel of the human body. That is, the device 200 is inserted into the blood vessel of the subject 101 while changing its shape according to the shape of the blood vessel of the subject 101. Moreover, the device 200 is deformed due to the movement of the blood vessel caused by the movement in the body of the subject 101. The movement in the body of the subject 101 includes, for example, the beating of the heart and the deformation of the blood vessel caused by the blood flow. That is, due to the beating of the heart, the device 200 moves three-dimensionally, greatly and irregularly in the blood vessels near the heart. Here, the "blood vessels near the heart" include not only the blood vessels next to the heart, but also the blood vessels of the heart itself (coronary arteries, etc.).

[0040] Furthermore, during percutaneous coronary intervention, the device 200 is used to position a therapeutic device, such as a stent, placed in a narrowed area of ​​a blood vessel (coronary artery) of the subject 101 at a target location within the blood vessel. The device 200 is inserted into a blood vessel (radial artery or femoral artery, etc.) in the wrist or thigh into the narrowed area of ​​the coronary artery. During percutaneous coronary intervention, the stent is positioned in the narrowed area of ​​the coronary artery using the device 200 inserted into the blood vessel. Furthermore, the stent is expanded to treat the narrowed blood vessel.

[0041] <About X-ray Equipment>

[0042] like Figure 2 As shown, the X-ray imaging apparatus 100 includes a top plate 1 , an X-ray irradiation unit 2 , an X-ray detection unit 3 , a moving unit 4 , a display unit 5 , an operation unit 6 , a control unit 7 , and a storage unit 8 .

[0043] A subject 101 to be irradiated with X-rays is placed on the table 1. While placed on the table 1, the subject 101 is inserted into the device 200 and X-rays are taken. The table 1 is configured to be movable by a table moving unit (not shown) under the control of the controller 7.

[0044] The X-ray irradiation unit 2 irradiates (radiates) X-rays toward a subject 101 having a medical device 200 placed within the subject. The X-ray irradiation unit 2 includes an X-ray tube 21 that irradiates X-rays by applying a voltage. The X-ray tube 21 is configured such that the voltage applied to the tube 21 is controlled by the controller 7, thereby controlling the X-rays irradiated by the tube 21. The X-ray irradiation unit 2 irradiates the device 200 within the subject 101 with X-rays one or more times.

[0045] The X-ray detector 3 detects X-rays transmitted through the subject 101. The X-ray detector 3 outputs a detection signal based on the detected X-rays. The X-ray detector 3 includes, for example, an FPD (Flat Panel Detector).

[0046] The moving part 4 holds the X-ray irradiation part 2 and the X-ray detection part 3 in a movable manner. Specifically, the moving part 4 supports the X-ray irradiation part 2 and the X-ray detection part 3 so that they face each other with the top plate 1 for placing the subject 101 therebetween. Moreover, the moving part 4 supports the X-ray irradiation part 2 and the X-ray detection part 3 in a manner that enables the position and angle of the X-ray irradiation part 2 and the X-ray detection part 3 relative to the subject 101 to be changed. In addition, the moving part 4 supports the X-ray irradiation part 2 and the X-ray detection part 3 in a manner that enables the distance between the X-ray irradiation part 2 and the X-ray detection part 3 to be changed. That is, the moving part 4 moves the X-ray irradiation part 2 and the X-ray detection part 3 to perform X-ray photography of the subject 101 from various positions and various angles.

[0047] The display unit 5 is, for example, a monitor such as a liquid crystal display, and is used to display images (still images and moving images) generated by the control unit 7 .

[0048] The operating unit 6 is configured to receive input operations for operating the X-ray imaging apparatus 100. For example, the operating unit 6 receives operations for moving the top plate 1 and the movable unit 4. Furthermore, when performing X-ray imaging on the subject 101, the operating unit 6 receives operations for irradiating the subject 101 with X-rays. Furthermore, the operating unit 6 receives input operations for executing controls of the control unit 7.

[0049] The control unit 7 is a computer including a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The control unit 7 controls each unit of the X-ray imaging apparatus 100, generates the X-ray image 10, and processes the generated X-ray image 10 by executing a predetermined control program on the CPU. Specifically, Figure 3As shown, the control unit 7 includes, as a functional configuration, an X-ray irradiation control unit 71, a pulse width setting unit 72, an X-ray image generator 73, a device detector 74, an image processor 75, and a display controller 76. Specifically, the control unit 7 functions as the X-ray irradiation control unit 71, the pulse width setting unit 72, the X-ray image generator 73, the device detector 74, the image processor 75, and the display controller 76 by having the CPU execute a predetermined control program. Details of the control performed by the control unit 7 will be described later.

[0050] The storage unit 8 is composed of a storage device such as a hard disk drive. The storage unit 8 is configured to store image data, imaging conditions, and various setting values. In addition, the storage unit 8 stores a program for making the control unit 7 function. In addition, the storage unit 8 stores a learning model 80 (see Figure 8 ).

[0051] (Regarding Control of the X-ray Imaging Apparatus by the Control Unit)

[0052] The X-ray irradiation control unit 71 of the control unit 7 controls the X-ray irradiation unit 2 and the X-ray detection unit 3 to perform X-ray photography. The X-ray irradiation control unit 71 controls the irradiation of X-rays by the X-ray irradiation unit 2. Specifically, the X-ray irradiation control unit 71 causes the X-ray irradiation unit 2 to irradiate X-rays having a pulse width set by the pulse width setting unit 72 described later. In detail, the X-ray irradiation control unit 71 controls the pulse width of the X-rays irradiated from the X-ray tube 21 by controlling the voltage applied to the X-ray tube 21. The pulse width refers to the irradiation time (exposure time) for irradiating the subject 101 with X-rays in order to capture one X-ray image 10. In addition, the X-ray irradiation control unit 71 controls the movement of the moving unit 4. In addition, the X-ray irradiation control unit 71 obtains an operation signal based on the input operation received by the operation unit 6, and controls each part of the X-ray imaging device 100 based on the obtained operation signal. For example, the X-ray irradiation control unit 71 controls the moving unit 4 based on an input operation to the operation unit 6 , thereby moving the positions of the X-ray irradiation unit 2 and the X-ray detection unit 3 for performing X-ray imaging.

[0053] The pulse width setting unit 72 sets the pulse width of the X-rays irradiated from the X-ray irradiation unit 2 to the subject 101. In this embodiment, the pulse width setting unit 72 sets the pulse width of the irradiated X-rays to a jitter-suppressed pulse width equal to or less than the pulse width that maximizes the detection accuracy of the detection device 200 based on the learned model 80 described later. The pulse width setting unit 72 can set the pulse width based on input operations input to the operation unit 6 or based on a pulse width setting value pre-stored in the storage unit 8. Details of the jitter-suppressed pulse width will be described later.

[0054] like Figure 4 and Figure 5 As shown, the X-ray image generator 73 of the control unit 7 generates an X-ray image 10 by performing X-ray imaging. In this embodiment, the X-ray image generator 73 generates the X-ray image 10 based on the X-rays detected by the X-ray detector 3. Specifically, the X-ray image generator 73 generates the X-ray image 10 based on the detection signal from the X-ray detector 3 using X-rays having a jitter-suppressed pulse width.

[0055] Here, Figure 4 The X-ray image 10 shown is an X-ray image 10 captured by X-rays with a relatively large pulse width. Figure 5 The X-ray image 10 shown is an X-ray image 10 captured by X-rays having a relatively small pulse width. Thus, in the X-ray image 10 obtained based on X-rays having a relatively large pulse width, the device 200 included in the X-ray image 10 contains large jitter when the movement inside the body of the subject 101 is large, and contains small jitter when the movement inside the body of the subject 101 is small. That is, in the X-ray image 10 obtained based on X-rays having a relatively large pulse width, the device 200 contains multiple jitters. On the other hand, in the X-ray image 10 obtained based on X-rays having a relatively small pulse width, the jitter included in the device 200 included in the X-ray image 10 is small in both cases where the movement inside the body of the subject 101 is large and where the movement inside the body of the subject 101 is small. That is, regardless of the amount of movement (movement speed) of the movement inside the body of the subject 101, the magnitude of the jitter becomes small (constant). For example, when a catheter or a guidewire placed in a blood vessel near the heart of the subject 101 is imaged, the catheter or the guidewire in the generated X-ray image 10 moves (deforms) irregularly due to the beating of the heart of the subject 101. Figure 4 As shown in FIG, since the movement of the heart and other organs in the subject 101 is not constant, the X-ray image 10 captured by the X-ray with a relatively large pulse width becomes an image containing various jitters. Figure 5As shown, in the X-ray image 10 captured by X-rays having a relatively small pulse width, even if the inside of the subject 101 moves (deforms) irregularly like the beating of the heart, the types (sizes) of jitter are reduced.

[0056] In this embodiment, in order to suppress the jitter of the device 200 including the catheter or guidewire in the generated X-ray image 10 from becoming multiple, the X-ray irradiation control unit 71 causes the X-ray irradiation unit 2 to irradiate X-rays with a pulse width below a predetermined threshold, i.e., a jitter-suppressed pulse width. That is, one X-ray irradiation for generating one X-ray image 10 is an irradiation of X-rays with a pulse width (exposure time) below a predetermined threshold. In addition, the X-ray image 10 obtained based on the X-rays with the jitter-suppressed pulse width becomes an X-ray image 10 with a small dose due to its small pulse width. In addition, the X-ray image 10 obtained based on the X-rays with the jitter-suppressed pulse width becomes an image with fewer types of jitter. In addition, the details of the predetermined threshold and the jitter-suppressed pulse width will be described later.

[0057] (Regarding Generation of Enhanced Image by Control Unit)

[0058] The device detection unit 74 of the control unit 7 is configured to detect the device 200 from the X-ray image 10 obtained based on the X-ray of the jitter suppression pulse width based on the learned model 80. The learned model 80 is a model generated by machine learning using images generated in a manner corresponding to the jitter suppression pulse width as training images. Furthermore, the image processing unit 75 of the control unit 7 performs image processing for enhancing the device 200 detected in the X-ray image 10, thereby generating an enhanced image 11 in which the device 200 is enhanced. That is, as Figure 6 As shown, the control unit 7 (the device detection unit 74 and the image processing unit 75 ) generates an enhanced image 11 in which the device 200 has a high resolution based on the X-ray image 10 in which the device 200 has a low resolution.

[0059] The learned model 80 is pre-stored in the storage unit 8. The learned model 80 is pre-generated by machine learning that learns the process of detecting the position of the device 200 from the input X-ray image 10. Details of the generation of the learned model 80 will be described later.

[0060] The device detection unit 74 of the control unit 7 detects the position of the device 200 in the X-ray image 10 from the generated X-ray image 10 based on the learned model 80, thereby obtaining the position information (coordinates) of the device 200 included in the X-ray image 10. Furthermore, the device detection unit 74 distinguishes the area of ​​the device 200 and the area of ​​the background that is not the device 200 from the X-ray image 10 based on the obtained position information. Furthermore, the image processing unit 75 of the control unit 7 performs image processing to increase the concentration of the area in the X-ray image 10 that is determined to be the device 200. Furthermore, the image processing unit 75 performs image processing to decrease the concentration of the area in the X-ray image 10 that is determined to be the background that is not the device 200. In this way, the control unit 7 (device detection unit 74 and image processing unit 75) obtains the position information of the device 200 based on the learned model 80, and generates the enhanced image 11 in which the device 200 is enhanced based on the obtained position information.

[0061] like Figure 7 As shown, the display control unit 76 of the control unit 7 controls the display of the display unit 5. Specifically, the display control unit 76 causes the display unit 5 to display the generated enhanced image 11. For example, the control unit 7 causes the X-ray irradiation unit 2 to irradiate X-rays 15 times in 1 second through the X-ray irradiation control unit 71, thereby generating 15 X-ray images 10 in 1 second through the X-ray image generation unit 73. Furthermore, the image processing unit 75 of the control unit 7 generates 15 enhanced images 11 based on the generated X-ray images 10 in 1 second. Furthermore, the display control unit 76 causes the display unit 5 to display the enhanced image 11 generated in real time as a moving image with a speed of 15 FPS (frames per second). Alternatively, the control unit 7 (display control unit 76) may cause the display unit 5 to display one enhanced image 11 as a still image.

[0062] (About the generation of the learning model)

[0063] like Figure 8As shown, a learned model 80 is generated through machine learning using training images corresponding to X-ray images 10 obtained with X-rays using a jitter-suppressed pulse width. Specifically, the learned model 80 is generated through machine learning based on a training input X-ray image 81, which is generated to simulate a device 200 within the body of the subject 101 as shown in the X-ray image 10 obtained with X-rays using a jitter-suppressed pulse width, and training output information 82, which indicates the position (coordinates) of the device 200 included in the training input X-ray image 81. Furthermore, the training input X-ray image 81 is generated to simulate the device 200 within the body of the subject 101 based on a simulated X-ray image generated with X-rays using a jitter-suppressed pulse width. The learned model 80 is pre-generated by a learning device 110 independent of the X-ray imaging apparatus 100.

[0064] The learning device 110 is, for example, a computer for machine learning including a CPU, a GPU, a ROM, a RAM, and the like.

[0065] like Figure 9 As shown, the learning device 110 acquires simulated X-ray images obtained by simulatedly imaging the subject 101 and the device 200 placed in the body of the subject 101 using X-rays having a pulse width, i.e., a jitter-suppressed pulse width, equivalent to the pulse width used when the X-ray image 10 is generated by the X-ray imaging apparatus 100. The simulated X-ray images include, for example, a simulated human body image 81a obtained by imaging a human body model simulating the subject 101 using X-rays having the jitter-suppressed pulse width, and a simulated device image 81b obtained by imaging the device 200 placed in the body of the subject 101 using X-rays having the jitter-suppressed pulse width.

[0066] The learning device 110 simulates the movement of the device 200 within the subject 101 by performing image processing on the acquired simulated X-ray images (simulated human body image 81a and simulated device image 81b). Specifically, the learning device 110 generates multiple training input X-ray images 81 by combining multiple simulated device images 81b with the acquired simulated human body image 81a to simulate the movement of the device 200 within the subject 101. These multiple simulated device images 81b are images processed to simulate the movement of the device 200 within the subject 101. Specifically, the learning device 110 simulates the movement of the device 200 within the subject 101 by performing image processing on one simulated device image 81b while varying multiple parameters, thereby generating multiple processed simulated device images 81b. Furthermore, the learning device 110 generates multiple training input X-ray images 81 by combining the simulated human body image 81a with the multiple simulated device images 81b obtained by performing image processing while varying multiple parameters. The multiple parameters include the body motion (e.g., heartbeat) of the subject 101 during a period below a predetermined threshold (a period corresponding to the duration of X-ray irradiation with the jitter-suppressed pulse width), the part of the human body to be irradiated with X-rays, and the angle at which the X-rays are irradiated. Furthermore, the learning device 110 acquires multiple simulated device images 81b and multiple simulated human body images 81a while varying the X-ray pulse width within a range below the predetermined threshold (the range of the jitter-suppressed pulse width). The learning device 110 simulates the acquired simulated device images 81b and multiple simulated human body images 81a by performing image processing while varying the multiple parameters. This generates multiple training input X-ray images 81 corresponding to various parts of the subject 101 and various imaging conditions. Specifically, images obtained using X-rays with pulse widths greater than the predetermined threshold are not included in the training input X-ray images 81.

[0067] Furthermore, the learning device 110 detects the position of the device 200 (the area where the device 200 is located) from the acquired simulated device image 81b as position information, thereby acquiring the position information of the device 200 in the corresponding training input X-ray image 81 as training output information 82. The position information includes, for example, the coordinates of the pixels corresponding to the device 200 among the pixels constituting the training input X-ray image 81.

[0068] The learning device 110 uses the training input X-ray images 81 as input and the training output information 82 as output, and generates a learned model 80 through machine learning. Specifically, the learning device 110 uses the training input X-ray images 81 and the training output information 82 as training data (training set) and learns the learned model 80 through machine learning. The learning device 110 generates the learned model 80 using the training input X-ray images 81 generated from a variety of simulated X-ray images. The machine learning method used is deep learning, which utilizes a multilayer neural network. For example, deep learning utilizes fully convolutional neural networks (FCNs).

[0069] The created learned model 80 is provided to the X-ray imaging apparatus 100 via a network or in a form recorded in a recording medium such as a flash memory.

[0070] (About jitter suppression pulse width)

[0071] Here, if Figure 10 and Figure 11 As shown, the detection accuracy (IoU: Intersection over Union) of the device 200 obtained based on the learned model 80 is a value that varies depending on the pulse width of the X-ray used to capture the X-ray image 10. That is, as the pulse width increases, the types of jitters of the image of the device 200 in the captured image increase due to the internal movement of the subject 101 (heartbeat, etc.). On the other hand, as the pulse width of the X-ray decreases, the dose of the irradiated X-ray decreases, so the captured image becomes unclear. Therefore, in both cases where the pulse width of the X-ray is too large and too small, the detection accuracy of the device 200 detected using the learned model 80 decreases. Therefore, the detection accuracy of the device 200 detected based on the learned model 80 has an upward convex change as the pulse width increases. For example, in the learned model 80, the detection accuracy of the device 200 is the highest in the X-ray image 10 obtained based on the X-ray with a pulse width of 8 milliseconds (ms).

[0072] For example, Figure 10 As shown, when the catheter as the device 200 included in the X-ray image 10 is detected using the learned model 80, the detection accuracy of the catheter (device 200) is the highest when the pulse width is 8 milliseconds (ms). Figure 11 As shown, when the guidewire as the device 200 included in the X-ray image 10 is detected using the learned model 80 , the detection accuracy of the guidewire (device 200 ) is highest when the pulse width is 8 milliseconds (ms).

[0073] In addition, the detection accuracy (IoU: Intersection over Union) of the device 200 based on the learned model 80 is expressed as a ratio, which represents the consistency between the position information of the device 200 output when the training input X-ray image 81 is input to the learned model 80 generated in the learning device 110 and the corresponding training output information 82.

[0074] The X-ray jitter suppression pulse width when capturing an X-ray image 10 by the X-ray imaging apparatus 100 is determined based on the detection accuracy of the device 200, as determined by the learned model 80. In this embodiment, the jitter suppression pulse width is a pulse width that maximizes the detection accuracy of the device 200 in the X-ray image 10, as determined by the learned model 80. For example, the predetermined threshold is 8 milliseconds. In other words, the jitter suppression pulse width is a pulse width of 8 milliseconds or less.

[0075] For example, when a pulse width of 5 milliseconds is determined as the jitter-suppressed pulse width for X-ray imaging of a subject 101, the X-ray imaging apparatus 100 generates an X-ray image 10 by irradiating the subject 101 with X-rays having a pulse width of 5 milliseconds. Furthermore, the X-ray imaging apparatus 100 generates an enhanced image 11 from the generated X-ray image 10 based on a learned model 80 generated using a plurality of simulated X-ray images (simulated human body image 81a and simulated device image 81b) obtained with X-rays having a pulse width within a range of 8 milliseconds or less. Furthermore, for X-rays having a jitter-suppressed pulse width, the irradiation time (pulse width) can be appropriately varied within a range of 8 milliseconds or less (or less than a predetermined value).

[0076] As described above, the X-ray imaging apparatus 100 in this embodiment sets a pulse width that maximizes the detection accuracy of the device 200 in the X-ray image 10 based on the learned model 80 as a predetermined threshold, and irradiates X-rays having a pulse width less than the predetermined threshold, i.e., a jitter-suppressed pulse width, thereby generating an X-ray image 10 based on X-rays with the jitter-suppressed pulse width. Furthermore, the X-ray imaging apparatus 100 acquires position information (region) of the device 200 from the generated X-ray image 10 based on the learned model 80, which is a model generated in a manner corresponding to the jitter-suppressed pulse width through machine learning using a simulated human body image 81a, which is an image captured using X-rays with the jitter-suppressed pulse width, and a simulated device image 81b, which is an image captured using X-rays with the jitter-suppressed pulse width.

[0077] (Regarding the method for generating a learned model according to the present embodiment)

[0078] Next, refer to Figure 12 The method for generating a learned model according to this embodiment will be described. The method for generating a learned model is implemented by the learning device 110 .

[0079] First, in step 301, a simulated X-ray image generated using X-rays with a jitter-suppressed pulse width is acquired. Specifically, a simulated X-ray image is acquired that includes a simulated human body image 81a, which is an image obtained by performing X-ray imaging of a human body phantom simulating the subject 101 using X-rays with a jitter-suppressed pulse width, and a simulated device image 81b, which is an image obtained by imaging the device 200 placed within the subject 101 using X-rays with a jitter-suppressed pulse width.

[0080] Next, in step 302, the acquired simulated device image 81b is image processed. Furthermore, the processed simulated device image 81b is synthesized with the acquired simulated human body image 81a. Specifically, by performing image processing on the acquired simulated X-ray images, multiple training input X-ray images 81 are acquired. These multiple training input X-ray images 81 are generated to simulate the device 200 within the body of the subject 101 in the X-ray image 10 obtained by X-rays using a jitter-suppressed pulse width. Specifically, based on the acquired simulated X-ray images (simulated human body image 81a and simulated device image 81b), multiple training input X-ray images 81 are generated to correspond to the X-ray image 10 obtained by X-rays using a jitter-suppressed pulse width. The jitter-suppressed pulse width is equal to or smaller than the pulse width that maximizes the detection accuracy of the medical device 200 placed within the body of the subject 101.

[0081] Next, in step 303 , based on the acquired simulated X-ray image, training output information 82 indicating the position of the device 200 in the training input X-ray image 81 is acquired.

[0082] Next, in step 304 , the training input X-ray image 81 is used as input and the training output information 82 is used as output, and a learned model 80 is generated (learned) through machine learning.

[0083] In addition, either the step of acquiring the training input X-ray image 81 in step 302 or the step of acquiring the training output information 82 in step 303 may be performed first.

[0084] (Regarding the Image Processing Method of the Present Embodiment)

[0085] Next, refer to Figure 13 The image processing method of this embodiment is described below. The image processing method is performed by the control unit 7 of the X-ray imaging apparatus 100.

[0086] First, in step 401, based on the learned model 80 generated through machine learning, the pulse width of the X-rays to be irradiated to generate the X-ray image 10 is set to a jitter-suppressed pulse width. This jitter-suppressed pulse width is a pulse width that maximizes the detection accuracy of the medical device 200 placed within the subject 101 from the generated X-ray image 10. Specifically, the pulse width setting unit 72 sets the pulse width of the X-rays irradiated from the X-ray irradiator 2 to the jitter-suppressed pulse width, thereby irradiating X-rays of the jitter-suppressed pulse width.

[0087] Next, in step 402 , the X-ray irradiation unit 2 irradiates the subject 101 in which the medical device 200 is indwelled with X-rays having the set jitter suppression pulse width.

[0088] Next, in step 403 , the X-ray detector 3 detects the X-rays of the jitter-suppressed pulse width that have passed through the subject 101 .

[0089] Next, in step 404 , an X-ray image 10 is generated based on the X-rays of the detected jitter-suppressed pulse width.

[0090] Next, in step 405, the device 200 is detected from the X-ray image 10 generated based on the jitter-suppressed pulse width X-rays based on the learned model 80. Specifically, the coordinates (region) indicating the position of the device 200 are detected from the X-ray image 10 based on the learned model 80.

[0091] Next, in step 406 , image processing is performed on the X-ray image 10 based on the detected position (coordinates) of the device 200 , thereby generating an enhanced image 11 in which the device 200 is enhanced.

[0092] Next, in step 407 , the generated enhanced image 11 is displayed on the display unit 5 .

[0093] (Effects of this embodiment)

[0094] In this embodiment, the following effects can be obtained.

[0095] According to the X-ray imaging apparatus 100 and image processing method of this embodiment, the pulse width of the X-rays irradiated to generate the X-ray image 10 is set to a jitter-suppressed pulse width, which is equal to or smaller than the pulse width that maximizes the detection accuracy of the medical device 200 placed within the body of the subject 101 from the generated X-ray image 10. Therefore, machine learning can be used to detect the device 200 from the X-ray image 10 obtained based on X-rays with the jitter-suppressed pulse width. Therefore, by capturing the X-ray image 10 using X-rays having a relatively small pulse width, i.e., the jitter-suppressed pulse width, which maximizes the detection accuracy of the device 200, even when the in-vivo device 200 is moving erratically due to irregular movements within the body of the subject 101, it is possible to suppress multiple types of jitter in the captured X-ray image 10. Thus, even when the device 200 inside the body of the subject 101 moves irregularly, the jitter of the device 200 in the X-ray image 10 is suppressed to a smaller than fixed value, thereby suppressing a decrease in the detection accuracy of the device 200 based on the learned model 80. As a result, the medical device 200 placed inside the body of the subject 101 in the X-ray image 10 can be detected with high accuracy based on the learned model 80 generated by machine learning.

[0096] Furthermore, if the pulse width of the X-rays used to generate the X-ray image 10 is greater than the pulse width that maximizes the detection accuracy of the device 200, the detection accuracy of the device 200 in the X-ray image 10 decreases. Furthermore, when the pulse width is increased, the dose of X-rays irradiated to the subject 101 also increases. In contrast, in this embodiment, the X-ray image 10 is generated using X-rays having a pulse width that is less than or equal to the pulse width that maximizes the detection accuracy of the device 200, i.e., a jitter-suppressed pulse width. Therefore, since the X-ray image 10 is generated using X-rays having a relatively small pulse width, i.e., a jitter-suppressed pulse width, the detection accuracy of the device 200 can be effectively suppressed, and the dose of X-rays irradiated to the subject 101 can be suppressed. As a result, the device 200 can be detected with even greater accuracy, and the dose of X-rays irradiated to the subject 101 can be reduced.

[0097] In addition, in the above-described embodiment example, further effects can be obtained by configuring as follows.

[0098] Specifically, in this embodiment, the device detection unit 74 is configured to detect the device 200 from the X-ray image 10 obtained by X-rays with jitter-suppressed pulse widths based on a learned model 80 generated through machine learning using training images corresponding to the X-ray image 10 obtained by X-rays with jitter-suppressed pulse widths. With this configuration, the device detection unit 74 utilizes the learned model 80 generated through machine learning using training images corresponding to the X-ray image 10 obtained by X-rays with jitter-suppressed pulse widths, thereby enabling even higher-precision detection of the device 200 in the X-ray image 10 obtained by X-rays with jitter-suppressed pulse widths.

[0099] Furthermore, in this embodiment, the device detection unit 74 is configured to detect the device 200 from the X-ray image 10 obtained by X-rays with jitter-suppressed pulse width based on a learned model 80 generated through machine learning from a training input X-ray image 81, which is an image generated to simulate the device 200 within the subject 101 in the X-ray image 10 obtained by X-rays with jitter-suppressed pulse width, and training output information 82, which indicates the position or shape of the device 200 included in the training input X-ray image 81. With this configuration, the learned model 80, which is learned using the training input X-ray image 81 that simulates the device 200 in the X-ray image 10 obtained by X-rays with jitter-suppressed pulse width, is utilized. This further improves the detection accuracy of the device 200 based on the learned model 80. Consequently, the device 200 in the X-ray image 10 obtained by X-rays with jitter-suppressed pulse width can be detected with even higher accuracy.

[0100] Furthermore, in this embodiment, the device detection unit 74 is configured to detect the device 200 in the X-ray image 10 obtained by X-rays with jitter-suppressed pulse width based on a learned model 80 generated through machine learning based on a training input X-ray image 81, which is an image generated based on a simulated X-ray image generated by X-rays with jitter-suppressed pulse width to simulate the device 200 within the body of the subject 101 in the X-ray image 10, and training output information 82 indicating the position or shape of the device 200 included in the training input X-ray image 81. With this configuration, the device 200 can be detected in the X-ray image 10 obtained by X-rays with jitter-suppressed pulse width based on the learned model 80, which is a model learned using the training input X-ray image 81 based on the simulated X-ray image generated by X-rays with jitter-suppressed pulse width. That is, both the training input X-ray image 81 used as input for learning to generate the learned model 80 and the X-ray image 10 used as input for inference based on the learned model 80 can be images generated using X-rays with a jitter-suppressed pulse width. Therefore, since both the input for learning the learned model 80 and the input for inference based on the learned model 80 can be images obtained based on X-rays under the same conditions, the accuracy of device 200 detection based on the learned model 80 can be further improved.

[0101] Furthermore, in this embodiment, the device 200 includes a catheter or guidewire placed in a blood vessel near the heart of the subject 101. The device detection unit 74 is configured to detect the catheter or guidewire from an X-ray image 10 obtained based on X-rays with jitter-suppressed pulse widths, based on a learned model 80 generated through machine learning from a training input X-ray image 81 and training output information 82. The training input X-ray image 81 is generated to simulate a catheter or guidewire placed in a blood vessel near the heart of the subject 101 in the X-ray image 10, and the training output information 82 indicates the position or shape of the catheter or guidewire included in the training input X-ray image 81. Here, the blood vessels near the heart move significantly and irregularly in three dimensions due to the beating of the heart. Therefore, the catheter or guidewire placed in the blood vessels near the heart also moves irregularly in three dimensions due to the beating of the heart, resulting in the X-ray image 10 being an image with various jitters present irregularly. Therefore, as in the above-described embodiment, by causing the X-ray irradiation unit 2 to irradiate X-rays having a pulse width equal to or less than a predetermined threshold value, i.e., a jitter-suppressed pulse width, to acquire an X-ray image 10, it is possible to reduce the types of jitter in the X-ray image 10 even for a catheter or guidewire placed in a blood vessel that moves significantly and erratically due to the heartbeat. Furthermore, since the catheter or guidewire can be detected from the X-ray image 10 obtained based on X-rays having the jitter-suppressed pulse width, it is possible to improve the visibility of the catheter or guidewire even when the catheter or guidewire is placed in an organ that moves significantly and erratically, such as a blood vessel near the heart.

[0102] Furthermore, in this embodiment, the jitter suppression pulse width is a pulse width of 8 milliseconds or less, and the device detection unit 74 is configured to detect the device 200 from X-ray images 10 obtained using X-rays having a pulse width of 8 milliseconds or less, based on the learned model 80. This configuration allows the device 200 to be detected from X-ray images 10 in which the jitter of the device 200 is suppressed by X-rays having a pulse width of 8 milliseconds or less. Therefore, since the device 200 is detected from X-ray images 10 obtained using X-rays having a pulse width of 8 milliseconds or less, the detection accuracy of the device 200 can be improved compared to the case where the pulse width is greater than 8 milliseconds.

[0103] Furthermore, in this embodiment, the device detection unit 74 is configured to detect the device 200 from the X-ray image 10 obtained based on the jitter-suppressed pulse width of X-rays, based on a learned model 80 generated through deep learning, which is machine learning using a multi-layer neural network. This configuration enables the detection of the device 200 from the X-ray image 10 based on the learned model 80, which automatically constructs appropriate feature quantities and an algorithm for acquiring the feature quantities through deep learning. This further improves the accuracy of device 200 detection.

[0104] In addition, this embodiment further includes a control unit 7, which includes a pulse width setting unit 72 for setting the pulse width of X-rays emitted from the X-ray irradiator 2; an X-ray irradiation control unit 71 for causing the X-ray irradiator 2 to irradiate X-rays having the set pulse width (jitter suppression pulse width); an X-ray image generator 73 for generating an X-ray image 10 based on the X-rays detected by the X-ray detector 3; and a device detector 74 for detecting devices 200 in the X-ray image 10 generated by the X-ray image generator 73 based on a learned model 80 generated through machine learning. With this configuration, pulse width setting, X-ray irradiation, generation of the X-ray image 10, and detection of the devices 200 can be easily performed through software control by the control unit 7.

[0105] (Effects of the Method for Generating a Learned Model According to the Present Embodiment)

[0106] The method for generating a learned model according to this embodiment can achieve the following effects.

[0107] In the method for generating a learned model according to this embodiment, as described above, multiple types of jitter can be suppressed by acquiring training input X-ray images 81 generated to simulate the device 200 within the body of the subject 101 in the X-ray image 10 obtained with X-rays using a jitter-suppressed pulse width. This prevents the need to acquire a large number of training input X-ray images 81 to address various types of jitter in the device 200. However, using a large number of training input X-ray images 81 to perform machine learning to address X-ray images 10 with various types of jitter makes it difficult to achieve learning convergence. Furthermore, even if learning converges, the detection accuracy of the learned model 80 is low. In contrast, in this embodiment, training input X-ray images 81 generated to simulate the device 200 within the body of the subject 101 in the X-ray image 10 obtained with X-rays using a jitter-suppressed pulse width are acquired. With this configuration, compared to a case where the pulse width is not restricted and the device 200 in the image exhibits a variety of jitter, the types of jitter in the X-ray image 10 are reduced by jittering the X-rays with suppressed pulse widths. This reduces the types of jitter in the images acquired as training images. This reduces the number of types of X-ray images 81 used for training input, making learning more likely to converge than when the device 200 exhibits a variety of jitter, and allows the generation of a learned model 80 capable of highly accurate detection of the device 200. Consequently, a method for generating a learned model capable of highly accurate detection of a medical device 200 indwelling in a subject 101 in an X-ray image 10 can be provided.

[0108] Furthermore, in this embodiment, as described above, the step of acquiring training input X-ray images 81 includes the steps of acquiring simulated X-ray images (simulated human body images 81a and simulated device images 81b) generated using X-rays with a jitter-suppressed pulse width; and performing image processing on the acquired simulated X-ray images to generate training input X-ray images 81 that simulate the device 200 within the body of the subject 101 in the X-ray image 10. With this configuration, the device 200 in the X-ray image 10 obtained using X-rays with a jitter-suppressed pulse width can be detected based on the learned model 80, which is learned using the training input X-ray images 81 obtained using simulated X-ray images generated using X-rays with a jitter-suppressed pulse width. In other words, both the training input X-ray images 81 used as input for learning to generate the learned model 80 and the X-ray images 10 used as input for inference based on the learned model 80 can be images generated using X-rays with a jitter-suppressed pulse width. Therefore, the input in learning the learned model 80 and the input in inferring the learned model 80 can be set to images obtained based on X-rays under the same conditions, thereby further improving the accuracy of detection of the device 200 based on the learned model 80.

[0109] (Variation)

[0110] The embodiments disclosed herein are to be considered in all respects as illustrative and non-restrictive. The scope of the present invention is not indicated by the description of the embodiments described above, but by the claims, and includes all modifications (variations) within the meaning and scope equivalent to the claims.

[0111] For example, in the above embodiment, an enhanced image in which the device is enhanced is generated by performing image processing to increase the density of the detected device area, but the present invention is not limited to this. In the present invention, the device can also be enhanced by coloring the detected device area. In addition, the device can be enhanced by displaying the outline of the detected device area in an enhanced manner (increasing the intensity of the outline).

[0112] In addition, in the above embodiment, an example is shown in which the jitter suppression pulse width can be appropriately changed within a range below 8 milliseconds (a predetermined threshold value), but the present invention is not limited to this. In the present invention, the jitter suppression pulse width can also be a fixed value below the predetermined threshold value. For example, when the jitter suppression pulse width is determined to be a fixed pulse width of 5 milliseconds, the X-ray imaging device can be caused to capture X-ray images using X-rays with a fixed pulse width of 5 milliseconds. In addition, when capturing X-ray images using X-rays with a pulse width of 5 milliseconds, the device can be detected from the captured X-ray image based on a learned model generated using a training input X-ray image and training output information, the training input X-ray image being an image generated based on a simulated X-ray image obtained based on X-rays with a pulse width of 5 milliseconds. In addition, in the case of taking an X-ray image using X-rays with a pulse width of 5 milliseconds, for example, a device can be detected from the taken X-ray image based on a learned model generated by training input X-ray images and training output information, wherein the training input X-ray image is an image generated based on simulated X-ray images obtained based on multiple X-rays with a pulse width of less than 8 milliseconds.

[0113] In the above embodiment, an example is shown in which a control unit is further provided. This control unit uses software to control: causing the X-ray irradiation unit to irradiate X-rays with a jitter-suppressed pulse width; generating an X-ray image based on the detection signal from the X-ray detection unit; and detecting a device from an X-ray image obtained based on the jitter-suppressed pulse width, based on a learned model generated through machine learning using images generated in a manner corresponding to the jitter-suppressed pulse width as teacher images. However, the present invention is not limited to this. In the present invention, the control unit for controlling the X-ray irradiation unit to irradiate X-rays and the control unit for controlling the device detection from the X-ray image may be configured separately. For example, an image generation unit may be configured separately from the control unit, and this image generation unit may include an image processing circuit or the like as hardware that generates an X-ray image based on the detected X-rays. Alternatively, an image processing module may be provided separately from the control unit, and this image processing module may generate an enhanced image from the X-ray image based on the position or shape of the device acquired through the use of the learned model. In other words, the pulse width setting unit, X-ray irradiation control unit, X-ray image generation unit, and device detection unit may be configured separately.

[0114] In addition, the above embodiment illustrates an example in which the control unit (device detection unit) is configured to detect devices from X-ray images based on a learned model generated through machine learning from training input X-ray images, where the training input X-ray images are images generated by simulating devices within the body of a subject in an X-ray image obtained using X-rays with a jitter-suppressed pulse width. However, the present invention is not limited to this. In the present invention, a learned model generated by learning using actual X-ray images as training input data, rather than training input X-ray images generated by simulating a device, may also be used. Furthermore, images generated by simulation without using X-rays may also be used as training input X-ray images.

[0115] Furthermore, in the above embodiment, an example is shown in which the control unit (device detection unit) is configured to detect devices from X-ray images based on a learned model generated through machine learning from training input X-ray images, where the training input X-ray images are images generated by simulated X-ray images generated using X-rays with a jitter-suppressed pulse width to simulate devices within the subject's body within the X-ray images. However, the present invention is not limited thereto. In the present invention, a learned model generated through machine learning from training input X-ray images generated based on simulated X-ray images generated using X-rays with a pulse width different from the jitter-suppressed pulse width may be used.

[0116] Furthermore, in the above embodiment, an example is shown in which the device includes a catheter or guidewire placed in a blood vessel near the heart of the subject, but the present invention is not limited to this. In the present invention, the device may also be a stent or artificial valve placed in a blood vessel near the heart of the subject. Furthermore, the device may also be placed in a blood vessel in the head.

[0117] In addition, in the above embodiment, an example is shown in which the pulse width below the predetermined threshold is a pulse width of 8 milliseconds or less, but the present invention is not limited to this. In the present invention, the pulse width below the predetermined threshold may also be a pulse width of 5 milliseconds or less. In other words, a pulse width smaller than the pulse width that maximizes the detection accuracy of the device may be determined as the predetermined threshold. Thus, by determining a pulse width smaller than the pulse width that maximizes the detection accuracy of the device as the predetermined threshold, the X-ray dose irradiated to the subject can be further reduced.

[0118] Furthermore, in the above embodiment, an example is shown in which a device is detected from X-ray images obtained based on the jitter-suppressed pulse width of X-rays using a learned model generated through machine learning (deep learning) using a multi-layer neural network. However, the present invention is not limited to this. Machine learning methods other than deep learning may also be used in the present invention. For example, a support vector machine (SVM) may be used to detect devices from X-ray images.

[0119] In addition, in the above embodiment, an example of detecting a device from an X-ray image based on a single learned model is shown, but the present invention is not limited thereto. In the present invention, a device may be detected from an X-ray image based on a plurality of learned models.

[0120] In addition, in the above-mentioned embodiment, an example of using a learned model generated by detecting the position (coordinates) of a device in an X-ray image is shown, but the present invention is not limited to this. The present invention may also use a learned model generated by detecting the shape (contour) of a device included in an X-ray image. That is, a learned model generated by machine learning based on training output information representing the shape (contour) of a device included in a training input X-ray image may also be used. For example, a learned model obtained by learning by machine learning using a training input X-ray image and training output information as training data (training set) may be used, wherein the training input X-ray image is an image of an object having a shape or other form corresponding to the device in the X-ray image, and the training output information is a shape (contour) specified by specifying the shape of the object corresponding to the device in the training input X-ray image.

[0121] [Way]

[0122] It should be understood by those skilled in the art that the above-described exemplary embodiments are specific examples of the following aspects.

[0123] (Item 1)

[0124] An X-ray imaging device comprising:

[0125] An X-ray irradiation unit that irradiates X-rays toward a subject having a medical device placed in the body;

[0126] an X-ray detection unit for detecting X-rays that have passed through the subject;

[0127] a pulse width setting unit that sets a pulse width of X-rays irradiated from the X-ray irradiation unit;

[0128] an X-ray irradiation control unit that causes the X-ray irradiation unit to irradiate X-rays of the pulse width set by the pulse width setting unit;

[0129] an X-ray image generating unit that generates an X-ray image based on the X-rays of the pulse width detected by the X-ray detecting unit; and

[0130] a device detection unit that detects the device in the X-ray image, based on a learned model generated by machine learning, from the X-ray image generated by the X-ray image generation unit and obtained based on the pulse width of the X-rays;

[0131] The pulse width is a jitter suppression pulse width that is equal to or smaller than a pulse width that maximizes detection accuracy of the device based on the learned model.

[0132] (Item 2)

[0133] According to the X-ray imaging apparatus described in Item 1, the device detection unit is configured to detect the device from the X-ray image obtained based on the X-ray with the jitter-suppressed pulse width based on the learned model generated by machine learning using a training image, wherein the training image is an image corresponding to the X-ray image obtained based on the X-ray with the jitter-suppressed pulse width.

[0134] (Item 3)

[0135] According to the X-ray imaging device described in Item 2, the device detection unit is configured to detect the device from the X-ray image obtained based on the X-ray of the jitter-suppressed pulse width based on the learned model generated by machine learning based on the training input X-ray image and the training output information, wherein the training input X-ray image is an image generated in a manner that simulates the device inside the body of the subject in the X-ray image obtained based on the jitter-suppressed pulse width, and the training output information represents the position or shape of the device included in the training input X-ray image.

[0136] (Item 4)

[0137] According to the X-ray imaging device described in Item 3, the device detection unit is configured to detect the device from the X-ray image obtained based on the X-ray with the jitter-suppressed pulse width based on the learned model generated by machine learning based on the training input X-ray image and the training output information, wherein the training input X-ray image is an image generated in a manner that simulates the device inside the body of the subject in the X-ray image based on a simulated X-ray image generated by the X-ray with the jitter-suppressed pulse width, and the training output information represents the position or shape of the device included in the training input X-ray image.

[0138] (Item 5)

[0139] The X-ray imaging apparatus according to item 3, wherein the device includes a catheter or a guide wire placed in a blood vessel near the heart of the subject.

[0140] The device detection unit is configured to detect the catheter or the guidewire from the X-ray image obtained based on the X-ray of the jitter-suppressed pulse width based on the learned model generated by machine learning according to the training input X-ray image and the training output information, wherein the training input X-ray image is an image generated in a manner simulating the catheter or the guidewire in the blood vessel near the heart of the subject in the X-ray image, and the training output information represents the position or shape of the catheter or the guidewire included in the training input X-ray image.

[0141] (Item 6)

[0142] According to the X-ray imaging apparatus of item 1, the jitter suppression pulse width is a pulse width of 8 milliseconds or less.

[0143] The device detection unit is configured to detect the device from the X-ray image obtained by X-rays having a pulse width of 8 milliseconds or less based on the learned model.

[0144] (Item 7)

[0145] According to the X-ray imaging device described in Item 1, the device detection unit is configured to detect the device from the X-ray image obtained based on the X-ray of the jitter suppression pulse width based on the learned model generated by deep learning, and the deep learning is machine learning using a multi-layer neural network.

[0146] (Item 8)

[0147] The X-ray imaging apparatus according to Item 1 further includes a control unit, which includes: the pulse width setting unit, which sets the pulse width of the X-rays irradiated from the X-ray irradiation unit; the X-ray irradiation control unit, which causes the X-ray irradiation unit to irradiate X-rays of the set pulse width; the X-ray image generation unit, which generates the X-ray image based on the X-rays detected by the X-ray detection unit; and the device detection unit, which detects the device in the X-ray image generated by the X-ray image generation unit based on the learned model generated by machine learning.

[0148] (Item 9)

[0149] A method for generating a learned model comprises the following steps:

[0150] Acquiring a training input X-ray image in a manner corresponding to an X-ray image obtained by X-rays using a jitter-suppressed pulse width, wherein the jitter-suppressed pulse width is a pulse width that maximizes detection accuracy of a medical device placed in the body of a subject, the training input X-ray image being an image generated in a manner that simulates the device in the body of the subject in the X-ray image obtained by the X-rays using the jitter-suppressed pulse width;

[0151] acquiring training output information indicating the position or shape of the device in the training input X-ray image; and

[0152] A learned model is generated by machine learning based on the training input X-ray image and the training output information.

[0153] (Item 10)

[0154] According to the method for generating a learned model described in item 9,

[0155] The step of obtaining the training input X-ray image comprises the following steps:

[0156] acquiring a simulated X-ray image generated by X-rays having the jitter-suppressed pulse width; and

[0157] The training input X-ray image is generated by performing image processing on the acquired simulated X-ray image so as to simulate the device inside the body of the subject in the X-ray image.

[0158] (Item 11)

[0159] An image processing method comprises the following steps:

[0160] Based on a learned model generated by machine learning, the pulse width of X-rays irradiated to generate an X-ray image is set to a jitter-suppressed pulse width equal to or less than a pulse width that maximizes the accuracy of detecting a medical device placed in a subject's body from the generated X-ray image;

[0161] irradiating the subject in which the device is implanted with X-rays having the set jitter suppression pulse width;

[0162] detecting the X-rays after passing through the subject;

[0163] generating the X-ray image based on the detected X-rays of the jitter-suppressed pulse width; and

[0164] The device in the X-ray image is detected based on the learned model from the X-ray image generated based on the jitter-suppressed pulse width.

[0165] (Other methods)

[0166] In addition, the above-mentioned embodiment can also be a specific example of the following aspects.

[0167] (Item 12)

[0168] A method for improving the image quality of a movable object (device) in a human body by short-time exposure to X-rays, the method comprising:

[0169] radiating one or more pulses of X-rays toward a movable object within a human body (irradiation), wherein the one or more pulses of X-rays include at least one pulse of short exposure (pulse width) X-rays having an exposure period (pulse width) shorter than a predetermined threshold;

[0170] Detecting more than one pulse of X-rays;

[0171] generating a low-resolution X-ray image of a movable object within the human body based on one or more pulses of X-rays;

[0172] Generating a contour of a movable object within the low-resolution X-ray image by inputting the low-resolution X-ray image into one or more machine learning models (learned models), wherein the one or more machine learning models are respectively trained (learned) according to a training set (training data), the training set including an X-ray image of an object having a shape and material corresponding to the movable object within the low-resolution X-ray image (training input X-ray image) and a specified contour of the object within the X-ray image (training input X-ray image); and

[0173] By increasing the intensity of the generated contour of the movable object, a new X-ray image (enhanced image) is generated based on the low-resolution X-ray image and output, wherein the new X-ray image has a higher resolution of the movable object than the low-resolution X-ray image.

[0174] (Item 13)

[0175] A device for improving the image quality of short-time exposure X-rays of movable objects inside the human body,

[0176] The device includes an X-ray generator (X-ray irradiation unit),

[0177] The X-ray generator is configured to generate one or more X-ray pulses and radiate the one or more X-ray pulses toward a movable object in the human body.

[0178] wherein the one or more pulses of X-rays include at least one pulse of short exposure (pulse width) X-rays having an exposure period shorter than a predetermined threshold exposure period (pulse width),

[0179] The device comprises:

[0180] an X-ray detector (X-ray detection unit) configured to detect one or more pulses of X-rays;

[0181] an image generating unit configured to generate a low-resolution X-ray image of a movable object in a human body based on one or more pulses of X-rays; and

[0182] One or more learned machine learning systems are configured to generate contours of movable objects within a low-resolution X-ray image by inputting a low-resolution X-ray image into one or more machine learning models (learned models).

[0183] wherein one or more machine learning models are trained (learned) according to a training set (training data), wherein the training set includes an X-ray image of an object having a shape and material corresponding to a movable object in a low-resolution X-ray image (training input X-ray image) and a specified contour of the object in the X-ray image (training input X-ray image) (training output information),

[0184] The apparatus further includes an image processing module configured to generate and output a new X-ray image (enhanced image) based on the low-resolution X-ray image by increasing the intensity of the generated contour of the movable object.

[0185] Among them, the new X-ray image has a higher resolution ratio of the movable object than the low-resolution X-ray image.

[0186] (Item 14)

[0187] A method for training a machine learning model for improving the image quality of short-time exposure X-rays of a movable object (equipment) is provided.

[0188] Acquire a plurality of X-ray images (training input X-ray images) depicted with the outline of an object (training output information) having a shape and material corresponding to a movable object (device),

[0189] wherein the plurality of X-ray images include at least one X-ray image including a contour depicted at a low resolution,

[0190] One or more machine learning models are trained using multiple X-ray images (training input X-ray images) and contour lines (training output information) so that the X-ray images (training input X-ray images) are correlated with the contours of an object having a shape and material corresponding to a movable object (training output information).

[0191] Description of Reference Numerals

[0192] 2: X-ray irradiation unit; 3: X-ray detection unit; 7: Control unit; 10: X-ray image; 71: X-ray irradiation control unit; 72: Pulse width setting unit; 73: X-ray image generation unit; 74: Equipment detection unit; 80: Learned model; 81: X-ray image for training input; 82: Information for training output; 100: X-ray imaging device; 101: Subject; 200: Equipment.

Claims

1. An X-ray imaging device comprising: An X-ray irradiation unit that irradiates X-rays toward a subject having a medical device placed in the body; an X-ray detection unit for detecting X-rays that have passed through the subject; a pulse width setting unit that sets a pulse width of X-rays irradiated from the X-ray irradiation unit; an X-ray irradiation control unit that causes the X-ray irradiation unit to irradiate X-rays of the pulse width set by the pulse width setting unit; an X-ray image generating unit that generates an X-ray image based on the X-rays of the pulse width detected by the X-ray detecting unit; as well as a device detection unit that detects the device in the X-ray image, based on a learned model generated by machine learning, from the X-ray image generated by the X-ray image generation unit and obtained based on the pulse width of the X-rays; The pulse width is a jitter suppression pulse width that is equal to or smaller than a pulse width that maximizes detection accuracy of the device based on the learned model.

2. The X-ray imaging device according to claim 1, wherein The device detection unit is configured to detect the device from the X-ray image obtained based on the X-rays with the jitter-suppressed pulse width based on the learned model generated by machine learning using a training image, wherein the training image is an image corresponding to the X-ray image obtained based on the X-rays with the jitter-suppressed pulse width.

3. The X-ray imaging device according to claim 2, wherein: The device detection unit is configured to detect the device from the X-ray image obtained based on the X-ray of the jitter-suppressed pulse width based on the learned model generated by machine learning according to the training input X-ray image and the training output information, wherein the training input X-ray image is an image generated in a manner that simulates the device inside the body of the subject in the X-ray image obtained based on the jitter-suppressed pulse width, and the training output information represents the position or shape of the device included in the training input X-ray image.

4. The X-ray imaging device according to claim 3, wherein The device detection unit is configured to detect the device from the X-ray image obtained based on the X-ray of the jitter-suppressed pulse width based on the learned model generated by machine learning according to the training input X-ray image and the training output information, wherein the training input X-ray image is an image generated in a manner that simulates the device inside the body of the subject in the X-ray image based on a simulated X-ray image generated by the X-ray of the jitter-suppressed pulse width, and the training output information represents the position or shape of the device included in the training input X-ray image.

5. The X-ray imaging device according to claim 3, wherein The device includes a catheter or a guidewire placed in a blood vessel near the heart of the subject. The device detection unit is configured to detect the catheter or the guidewire from the X-ray image obtained based on the X-ray of the jitter-suppressed pulse width based on the learned model generated by machine learning according to the training input X-ray image and the training output information, wherein the training input X-ray image is an image generated in a manner simulating the catheter or the guidewire in the blood vessel near the heart of the subject in the X-ray image, and the training output information represents the position or shape of the catheter or the guidewire included in the training input X-ray image.

6. The X-ray imaging device according to claim 1, wherein The jitter suppression pulse width is a pulse width of less than 8 milliseconds, The device detection unit is configured to detect the device from the X-ray image obtained by X-rays having a pulse width of 8 milliseconds or less based on the learned model.

7. The X-ray imaging device according to claim 1, wherein The device detection unit is configured to detect the device from the X-ray image obtained based on the X-rays of the jitter suppression pulse width based on the learned model generated by deep learning, which is machine learning using a multilayer neural network.

8. The X-ray imaging device according to claim 1, wherein A control unit is provided, the control unit comprising: the pulse width setting unit for setting the pulse width of the X-ray irradiated from the X-ray irradiation unit; the X-ray irradiation control unit causing the X-ray irradiation unit to irradiate X-rays of the set pulse width; and the X-ray image generation unit generating the X-ray image based on the X-rays detected by the X-ray detection unit; and the device detection unit detects the device in the X-ray image generated by the X-ray image generation unit based on the learned model generated by machine learning.

9. A method for generating a learned model, comprising the following steps: Acquiring a training input X-ray image in a manner corresponding to an X-ray image obtained by X-rays using a jitter-suppressed pulse width, wherein the jitter-suppressed pulse width is a pulse width that maximizes detection accuracy for a medical device placed inside a subject, the training input X-ray image being an image generated to simulate the device inside the subject in the X-ray image obtained by the X-rays using the jitter-suppressed pulse width; acquiring training output information indicating the position or shape of the device in the training input X-ray image; as well as A learned model is generated by machine learning based on the training input X-ray image and the training output information.

10. The method for generating a learned model according to claim 9, wherein: The step of obtaining the training input X-ray image comprises the following steps: acquiring a simulated X-ray image generated by X-rays having the jitter-suppressed pulse width; and The training input X-ray image is generated by performing image processing on the acquired simulated X-ray image so as to simulate the device inside the body of the subject in the X-ray image.

11. An image processing method comprising the following steps: Based on a learned model generated by machine learning, the pulse width of X-rays irradiated to generate an X-ray image is set to a jitter-suppressed pulse width equal to or less than a pulse width that maximizes the accuracy of detecting a medical device placed in a subject's body from the generated X-ray image; irradiating the subject in which the device is implanted with X-rays having the set jitter suppression pulse width; detecting the X-rays after passing through the subject; generating the X-ray image based on the detected X-rays of the jitter-suppressed pulse width; as well as The device in the X-ray image is detected based on the learned model from the X-ray image generated based on the jitter-suppressed pulse width.

Images