Medical-image diagnosis assistance device, method of operating medical-image diagnosis assistance device, program, medical-image capturing device, learning model, and learning method

Abstract

Provided are a medical-image diagnosis assistance device, a method of operating the medical-image diagnosis assistance device, a program, a medical-image capturing device, a learning model, and a learning method, which enable the visual recognition of motion artifact prediction results when capturing medical images of a subject being imaged. The medical-image diagnosis assistance device: acquires body motion information of a subject being imaged; inputs the body motion information of the subject being imaged into a learning model that has learned the relationship between body motion information and artifacts in medical images and outputs prediction information for an artifact in the medical image obtained by imaging the subject being imaged when the body motion information of the subject being imaged is inputted; acquires the prediction information corresponding to the body motion information of the subject being imaged from the trained model; and outputs the prediction information.

Description

Medical Image Diagnosis Support Device, Operating Method of Medical Image Diagnosis Support Device, Program, Medical Image Imaging Device, Learning Model, and Learning Method

[0001] The present disclosure relates to a medical image diagnosis support device, an operating method of the medical image diagnosis support device, a program, a medical image imaging device, a learning model, and a learning method.

[0002] Magnetic resonance imaging devices used for medical image diagnosis can non-invasively acquire information from the whole body of a living body and are widely used in the medical field. Such a magnetic resonance imaging device is likely to be affected by the movement of the subject during the acquisition of imaging data of the subject. Motion artifacts caused by the movement of the subject are one of the artifacts that are likely to occur in the clinical field.

[0003] Since artifacts affect reading of images, the imaging data or the reconstructed image is processed to reduce the artifacts. On the other hand, re-imaging may be required depending on the degree of the artifacts. Re-imaging is a factor in extending the examination time, and there are concerns about an increase in the physical and mental burden on the subject. In addition, there are concerns about the impact on the costs of examination institutions such as hospitals.

[0004] Patent Document 1 describes a magnetic resonance imaging data processing system that uses a deep learning network to remove motion artifacts from a magnetic resonance imaging set.

[0005] The system described in the same document includes a deep learning network that detects and removes motion artifacts in a magnetic resonance imaging set. For the learning of the deep learning network, a plurality of magnetic resonance imaging training data sets with and without motion artifacts are used. The generation of the training data includes introducing artificially generated motion artifacts into a magnetic resonance imaging set without motion artifacts.

[0006] Patent Document 2 describes a system that monitors neurophysiological signals while a patient is being imaged, predicts a part of the patient's movement that has an adverse effect on the imaging data, and prevents the acquisition of image data during this movement period.

[0007] The system described in the document includes a monitoring system that includes sensors such as cameras to measure the patient's physiological state, and a survey feature information module that extracts survey information related to image surveys of body parts, etc. Furthermore, it includes a prediction engine that interprets the signals measured by the monitoring system and determines, based on the extracted survey information, whether there is a high probability of patient movement and at what time there is a high probability of patient movement.

[0008] Special table 2021-501015 publication Special table 2016-514508 publication

[0009] If it is possible to visually predict the degree of artifacts superimposed on the final image due to the subject's body movement during imaging, a decision can be made mid-imaging based on the user's tolerance for the degree of artifacts, such as whether to continue imaging or interrupt imaging and perform re-imaging. If a decision to perform re-imaging can be made mid-imaging, the extension of examination time due to re-imaging can be suppressed.

[0010] The system described in Patent Document 1 uses a deep learning model to detect the presence of motion artifacts in a magnetic imaging dataset and corrects the motion artifacts present in the magnetic imaging dataset. On the other hand, the system described in Patent Document 1 does not take the above prediction into consideration.

[0011] The system described in Patent Document 2 may, even if it detects patient movement that does not cause artifacts, be judged to have a high probability of causing further patient movement, and there is a concern that image data acquisition may be stopped.

[0012] This disclosure is made in view of these circumstances and aims to provide a medical image diagnostic support device, a method for operating the medical image diagnostic support device, a program, a medical image acquisition device, a learning model, and a learning method that enable the visual recognition of motion artifact prediction results in the acquisition of medical images of the subject being imaged.

[0013] A medical image diagnostic support device according to a first aspect of this disclosure comprises a processor and a memory that stores a program to be executed by the processor, wherein the processor acquires subject motion information representing the body movements of the subject being imaged, and is a learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging any subject, and when subject motion information is input to the learning model, it outputs prediction information of artifacts in medical images generated by imaging the subject, and the medical image diagnostic support device acquires prediction information corresponding to the subject motion information from the learning model and outputs the prediction information.

[0014] According to the medical image diagnostic support device of the first embodiment, when acquiring medical images of a subject, predictive information about artifacts in the medical images can be visually recognized. This makes it possible to make decisions such as continuing or interrupting imaging based on the predictive information about artifacts.

[0015] During the imaging period for the subject, motion information of the subject may be acquired. Predictive information may be output during the imaging period for the subject.

[0016] The movement of the subject being imaged may be detected using a sensor that detects the subject's body movement. An example of a sensor that detects the subject's body movement is a camera that images the subject. The camera may image the entire body of the subject, or it may image the area of ​​the subject being examined.

[0017] The medical image diagnostic support device may include, in addition to the first processor used for acquiring motion information of the subject being imaged, a second processor that functions as a learning model, and a second memory. The second memory may store a program that includes one or more instructions executed by the second processor.

[0018] The medical image diagnostic support device according to the second embodiment is a medical image diagnostic support device according to the first embodiment in which the training dataset applied to training the learning model may include, as input training data, body movement information of any subject during the imaging period, and as output training data, medical images generated during imaging performed on any subject during the imaging period.

[0019] In the medical image diagnostic support device according to the third embodiment, the input learning data may include examination information that includes at least one of imaging site information and body position information of an arbitrary subject, in addition to the medical image diagnostic support device according to the second embodiment.

[0020] In the medical image diagnostic support device according to the fourth embodiment, the input learning data may include image-related information, including at least one of imaging parameters, a method for filling k-space, and a k-space filling position, in the medical image diagnostic support device according to the second or third embodiment.

[0021] The medical image diagnostic support device according to the fifth embodiment is a medical image diagnostic support device according to any one embodiment of the second to fourth embodiments, in which the input learning data may include medical images generated by imaging any subject.

[0022] The medical image diagnostic support device according to the sixth embodiment is a medical image diagnostic support device according to any one embodiment of the second to fifth embodiments, wherein the processor may acquire predictive information from a learning model that includes at least one of the following: character information relating to the artifact, the amount of displacement of the artifact at its location from the actual examination site, and a medical image containing the artifact, and output the predictive information acquired from the learning model.

[0023] The medical image diagnostic support device according to the seventh embodiment is a medical image diagnostic support device according to the sixth embodiment in which the training dataset applied to training the learning model includes, as input training data, body movement information of any subject during the imaging period, and as output training data, medical images generated during imaging performed during the imaging period of any subject, and may further include at least one of textual information relating to artifacts and the amount of movement of the artifact location from the actual examination site.

[0024] In the medical image diagnostic support device according to the eighth embodiment, the processor may obtain predictive information indicating whether or not re-imaging is necessary from a learning model and output the obtained predictive information.

[0025] The medical image diagnostic support device according to the ninth embodiment is a medical image diagnostic support device according to the eighth embodiment in which the training dataset applied to training the learning model includes, as input training data, body movement information of any subject during the imaging period, and as output training data, medical images generated during imaging performed during the imaging period of any subject, and may further include a label indicating whether or not re-imaging is necessary.

[0026] In the medical image diagnostic support device according to the 10th embodiment, in the medical image diagnostic support device according to any one embodiment from the 1st to the 9th embodiment, the processor may acquire predictive information from a learning model indicating that it is necessary to confirm the condition of the person being imaged, and output the acquired predictive information.

[0027] The medical image diagnostic support device according to the 11th embodiment is a medical image diagnostic support device according to the 10th embodiment in which the learning dataset applied to the learning model includes, as input learning data, body movement information of any subject during the imaging period, and as output learning data, medical images generated during imaging performed during the imaging period of any subject, and may also include a label indicating that confirmation of the condition of the subject being imaged is necessary.

[0028] In the medical image diagnostic support device according to the 12th embodiment, in the medical image diagnostic support device according to the 8th or 10th embodiment, the processor may output predictive information indicating that re-imaging is necessary, or predictive information indicating that the condition of the subject to be imaged needs to be checked, to the imaging control device that controls imaging of the subject to be imaged.

[0029] The medical image diagnostic support device according to the 13th embodiment is a medical image diagnostic support device according to any one embodiment from the 1st to the 12th embodiment, wherein the processor may acquire examination information that is applied to imaging of the subject and includes the examination area, input the examination information into a learning model, and acquire a medical image that includes artifacts predicted based on the subject's body movement information as prediction information output from the learning model.

[0030] In the medical image diagnostic support device according to the 14th embodiment, in the medical image diagnostic support device according to any one embodiment from the 1st to the 13th embodiment, the processor may acquire predictive information from a learning model to which at least one of text information, icons, sounds, and vibrations is applied.

[0031] A method for operating a medical image diagnostic support device according to a 15th aspect of this disclosure is a method for operating a medical image diagnostic support device in which a computer functioning as a medical image diagnostic support device performs the following steps: acquiring subject motion information representing the body movements of the subject being imaged; inputting subject motion information to a learning model which is a learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging any subject, and which outputs predicted information about artifacts in medical images generated by imaging a subject when subject motion information is input; acquiring predicted information corresponding to the subject motion information from the learning model; and outputting the predicted information.

[0032] According to the operating method of the medical image diagnostic support device according to the 15th aspect of this disclosure, it is possible to obtain the same effects as those of the medical image diagnostic support device according to the first aspect. The constituent elements of the medical image diagnostic support device according to the second to 14th aspects can be applied as constituent elements of the operating method of the medical image diagnostic support device according to the other aspects.

[0033] The program according to the 16th aspect of this disclosure is a program that enables a computer functioning as a medical image diagnostic support device to implement the following functions: a function to acquire subject motion information representing the body movements of the subject being imaged; a learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging any subject; a function to input subject motion information to a learning model that outputs predicted information about artifacts in medical images generated by imaging a subject when subject motion information is input; a function to acquire predicted information corresponding to the subject motion information from the learning model; and a function to output the predicted information.

[0034] According to the program of the sixteenth aspect of this disclosure, it is possible to obtain the same effects and advantages as the medical image diagnostic support device of the first aspect. The constituent elements of the medical image diagnostic support devices of the second to fourteenth aspects can be applied as constituent elements of the programs of the other aspects.

[0035] A medical image acquisition device according to the 17th aspect of this disclosure is a medical image acquisition device comprising: an imaging unit that images a subject to be imaged and generates imaging data of the subject; an image reconstruction unit that generates a reconstruction based on the imaging data; a processor; and a memory that stores a program to be executed by the processor, wherein the processor is a learning model that acquires subject motion information representing the body movements of the subject to be imaged, learns the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging any subject, and when subject motion information is input to the learning model, it outputs prediction information of artifacts in medical images generated by imaging the subject, and the medical image acquisition device inputs subject motion information to the learning model, obtains prediction information corresponding to the subject motion information from the learning model, and outputs the prediction information.

[0036] A medical image acquisition device according to the 17th aspect of this disclosure can achieve the same effects as a medical image diagnostic support device according to the first aspect. The constituent elements of the medical image diagnostic support devices according to the second to 14th aspects can be applied as constituent elements of medical image acquisition devices according to the other aspects.

[0037] The learning model according to the 18th aspect of this disclosure is a learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging any subject. When motion information of an imaging subject representing the body movements of the imaging subject is input, the learning model outputs predictive information about artifacts in medical images generated by imaging the imaging subject.

[0038] According to the learning model of the 18th aspect of this disclosure, it is possible to obtain the same effects and advantages as the medical image diagnostic support device of the first aspect. The constituent elements of the medical image diagnostic support devices of the second to 14th aspects can be applied as constituent elements of the learning models of the other aspects.

[0039] A learning method relating to the 19th aspect of this disclosure is a learning method for a learning model that outputs predictive artifact information in a medical image generated by imaging a subject when subject motion information representing the body movement of the subject is input, and the learning method learns the relationship between motion information representing the body movement of any subject and artifacts in a medical image generated by imaging any subject.

[0040] The learning method according to the 19th aspect of this disclosure makes it possible to obtain the same effects as the medical image diagnostic support device according to the first aspect. The constituent elements of the medical image diagnostic support devices according to the second to 14th aspects can be applied as constituent elements of the learning method according to the other aspects.

[0041] According to this disclosure, predictive information about artifacts in medical images can be visually recognized during the acquisition of medical images of the subject. This makes it possible to make decisions such as continuing or interrupting imaging based on the predictive information about artifacts.

[0042] FIG. 1 is a conceptual diagram of a motion artifact prediction method according to the first embodiment. FIG. 2 is a schematic diagram showing the relationship along the time series between the body movement detection of the subject and the output of the prediction information. FIG. 3 is a block diagram showing a configuration example of a medical image diagnosis support device according to the first embodiment. FIG. 4 is a schematic diagram of a learned AI applied to an AI processing circuit. FIG. 5 is a flowchart showing the procedure of the motion artifact prediction method according to the first embodiment. FIG. 6 is an example of display of prediction information. FIG. 7 is a schematic diagram of a learned AI according to another aspect. FIG. 8 is an example of display of the prediction information illustrated in FIG. 7. FIG. 9 is a schematic diagram of a learned AI applied to the motion artifact prediction method according to the second embodiment. FIG. 10 is a flowchart showing the procedure of the artifact prediction method according to the second embodiment. FIG. 11 is an example of display of prediction information including a determination result indicating the necessity of re-imaging. FIG. 12 is a block diagram showing a configuration example of a medical image diagnosis support device according to the fourth embodiment. FIG. 13 is a flowchart showing the procedure of the artifact prediction method according to the fourth embodiment. FIG. 14 is a block diagram showing the hardware configuration of the electrical configuration of the medical image diagnosis support device. FIG. 15 is an explanatory diagram of learning data. FIG. 16 is a schematic diagram showing a specific example of learning. FIG. 17 is a perspective view showing the appearance of an MRI device. FIG. 18 is a schematic diagram showing the internal configuration of the MRI device.

[0043] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description and the accompanying drawings, the same reference numerals are given to the same components, and duplicate explanations are omitted. Further, when a plurality of components are exemplified in the following embodiments, it can be interpreted as including at least one of the plurality of components.

[0044] [First Embodiment] FIG. 1 is a conceptual diagram of a motion artifact prediction method according to the first embodiment. In the figure, a method for predicting an artifact generated in an MR image generated by imaging an imaging subject P, who is a subject to be imaged, using an MRI device, and a method for predicting a motion artifact caused by the body movement of the imaging subject P are schematically illustrated.

[0045] Here, imaging includes acquisition of imaging data of the imaging subject. Generation of a reconstructed image based on the imaging data may also be included in the imaging. The MR image output from the MRI device may be a reconstructed image reconstructed based on the imaging data.

[0046] The motion artifact caused by the body movement of the imaging subject P may be a motion artifact that occurs in the MR image when the body movement of the imaging subject P occurs during the imaging period. Note that MRI is an abbreviation for Magnetic Resonance Imaging. MR is an abbreviation for Magnetic Resonance. The MR image may be referred to as an MRI image.

[0047] In the motion artifact prediction method, during the imaging period of the imaging subject P, based on the imaging image of the imaging subject P lying on the bed BE using the camera ID, the body movement of the imaging subject P is detected.

[0048] The imaging image of the imaging subject P obtained using the camera ID may be a two-dimensional image or a three-dimensional image. The imaging image of the imaging subject P may be a moving image to which a prescribed frame rate is applied, or may be a plurality of still images obtained at a plurality of different timings.

[0049] The camera ID may be a visible light camera or an infrared camera. Detection of the body movement of the imaging subject P may use inspection information including the imaging site and the imaging position.

[0050] The body movement information BI obtained as a result of detecting the body movement of the imaging subject P is input to the AI to which the learned learning model is applied. The AI may acquire inspection information II including at least one of the imaging site information and the body position information of the imaging subject P. Note that AI is an abbreviation for Artificial Intelligence. The body movement information BI is an example of the imaging subject body movement information of the present disclosure.

[0051] The AI ​​generates predictive information PI. As predictive information PI, the AI ​​generates a predicted image PII, which is an MR image containing motion artifacts caused by the body movements of the subject P being imaged. The AI ​​may also generate text representing the motion artifacts as predictive information PI.

[0052] The AI ​​outputs prediction information PI. The prediction information PI output from the AI ​​is transmitted to the display device DP. The display device DP displays the prediction information PI. The predicted image PII output from the AI ​​may be a standard MR image defined based on the gender, physique, and age of the subject P, with predicted motion artifacts superimposed on it.

[0053] Figure 2 is a schematic diagram illustrating the time-series relationship between the detection of the subject's body movement and the output of predictive information. Figure 2 schematically shows the timing of the detection of the subject P's body movement and the output of predictive information by the AI ​​during the imaging period of the MRI device.

[0054] During the imaging period of the MRI device, the AI ​​may continuously acquire body motion information BI, or it may acquire body motion information BI intermittently by applying a specified sampling period. The AI ​​may also acquire body motion information BI cumulatively.

[0055] When acquiring motion information cumulatively, the AI ​​acquires motion information BI for the period from the start of imaging to each sampling timing. For example, in the first acquisition of motion information BI, the AI ​​acquires motion information BI from the start of imaging to the first sampling timing. In the second acquisition of motion information BI, the AI ​​acquires motion information BI from the start of imaging to the second sampling timing.

[0056] Thus, by acquiring cumulative body motion information (BI), it is possible to suppress AI malfunctions caused by body movements that occur between sampling timings but are not significant enough to cause motion artifacts.

[0057] The AI ​​can output predictive information PI during the imaging period, and the technician can visually review this predictive information PI during the imaging period. This can shorten the time it takes for the technician to decide whether to re-image compared to when re-imaging is performed after the technician has visually reviewed the reconstructed image. The technician may also be referred to as an operator, operator, or user.

[0058] [Example of configuration of a medical image diagnostic support device according to the first embodiment] Figure 3 is a block diagram showing an example of the configuration of a medical image diagnostic support device according to the first embodiment. The medical image diagnostic support device 10 shown in the figure comprises a camera 12, a patient information storage circuit 14, a body motion information processing circuit 16, an image-related information storage circuit 18, an AI processing circuit 20, and an output device 22.

[0059] Here, the circuit may include an integrated form on one or more devices. The integrated circuit may be produced using a semiconductor process. The memory circuit may include devices such as memory elements, memory media, and memory devices.

[0060] Camera 12 uses the subject P, who is lying on the bed BE, as its subject and images the entire body of the subject P. Camera 12 may also use the area being examined and the area surrounding the area being examined as its subject. Camera 12 is positioned above the top plate TB of the bed BE and images the subject P from above. Here, the upper side is a position where the distance from the reference position in the vertical direction is relatively large relative to the position of camera ID, and the lower side is a position where the distance from the reference position in the vertical direction is relatively small relative to the position of camera ID.

[0061] The placement of camera 12 is not limited to a position above the top plate TB of the bed BE, but from the viewpoint of capturing images of the entire body of the subject P lying on the top plate TB of the bed BE, the placement of camera 12 above the top plate TB of the bed BE is preferable. Camera 12 is not limited to one unit, but may be multiple units. Camera 12 may be placed on the ceiling of the shielded room, inside the gantry, etc. The gantry is denoted by reference numeral 120 and is shown in Figure 17.

[0062] The subject information storage circuit 14 stores the examination information II of the subject P. The examination information II includes at least one of the examination area information and the position information of the subject P. The subject information storage circuit 14 may store the examination area information and the like included in the examination order of the subject P as examination information.

[0063] The motion information processing circuit 16 detects the motion of the subject P from the image of the subject P transmitted from the camera 12 and generates motion information BI of the subject P. The motion information processing circuit 16 may acquire the image of the subject P transmitted from the camera 12.

[0064] The motion information processing circuit 16 may acquire examination site information from the subject information storage circuit 14 and generate motion information BI in the examination site of the subject P. The motion information processing circuit 16 may acquire body position information of the subject P from the subject information storage circuit 14 and identify the examination site in the captured image of the subject P. The motion information processing circuit 16 is an example of a circuit implemented using the processor according to this disclosure.

[0065] The image-related information storage circuit 18 stores image-related information from the MRI apparatus. The image-related information includes at least one of the following: imaging parameters such as the sequence, the k-space filling method, and the k-space filling position when body motion occurs. When body motion occurs, the period before and after the timing of the body motion may be included.

[0066] Image-related information may include information about the area being examined. For example, the image-related information storage circuit 18 may store information about the area being examined, etc., included in the examination order of the person being imaged P, as image-related information.

[0067] The AI ​​processing circuit 20 acquires examination information II from the subject information storage circuit 14. The AI ​​processing circuit 20 acquires body movement information BI from the body movement information processing circuit 16. The AI ​​processing circuit 20 acquires image-related information from the image-related information storage circuit 18.

[0068] The AI ​​processing circuit 20 uses the examination information II, body movement information BI, and image-related information of the subject P to predict motion artifacts that may occur in the MR image of the subject P due to the subject P's body movement, and generates prediction information PI. The AI ​​processing circuit 20 is equipped with an AI that generates the prediction information. A deep learning method is applied to the AI. For example, the AI ​​processing circuit 20 shown in Figure 3 is equipped with an AI to which the trained learning model shown in Figure 1 is applied. The AI ​​processing circuit 20 may be equipped with peripheral circuits and the like for the circuit that functions as the AI.

[0069] The AI ​​processing circuit 20 outputs a signal representing prediction information PI to the output device 22. The output device 22 may be the display device DP shown in Figure 1. The output device 22 may be any device that outputs prediction information PI, and may be a printing device that prints prediction information PI, or a device that outputs sound or vibration representing prediction information PI. Note that the AI ​​processing circuit 20 is an example of a circuit implemented using the processor according to this disclosure.

[0070] Figure 3 illustrates a configuration in which the medical image diagnostic support device 10 includes a camera 12 and an output device 22, but the camera 12 and output device 22 may be external devices to the medical image diagnostic support device 10.

[0071] Figure 4 is a schematic diagram of a trained AI applied to an AI processing circuit. The trained AI 21 takes body movement information BI of the subject P during the imaging period of the MRI device as input and outputs prediction information PI which includes the prediction result of motion artifacts during imaging by the MRI device. The trained AI 21 shown in Figure 4 may output a prediction image PII, which is an MR image containing motion artifacts, as the prediction information PI.

[0072] Figure 4 schematically illustrates an example of motion information BI, which includes the amount of movement of the subject P, the waveform representing the motion, the timing at which the motion occurs, and the magnitude of the motion evaluation value. The motion information BI may be any combination of two or more of these elements. Figure 4 illustrates an example of motion evaluation values, which represent the magnitude of the motion evaluation value for each sampling timing along the time series.

[0073] The trained AI21 acquires cumulative information on the magnitude of the body movement evaluation value for each sampling timing along the time series. For example, at the second sampling timing, the total evaluation value obtained by adding the magnitude of the second evaluation value at the second sampling timing to the magnitude of the first evaluation value at the first sampling timing is grasped as body movement information BI.

[0074] The training dataset of the trained AI 21 consists of input training data consisting of body movement information BI of any subject during the imaging period of the MRI device, and output training data consisting of MR images generated by imaging that subject using the MRI device. In other words, the training dataset of the trained AI 21 is a collection of multiple sets of (1) body movement information BI of a subject during the imaging period of the MRI device as input training data, and (2) MR images generated by imaging that subject using the MRI device as output training data. The input training data in (1) and the output training data in (2) described above can also be applied to the description of other training datasets in this specification. The trained AI 21 may be associated with examination sites. That is, the input training data of the trained AI 21 may include information about the examination site. The trained AI 21 may be trained for each examination site.

[0075] The AI ​​processing circuit 20 shown in Figure 3 comprises multiple trained AIs 21, and each of the multiple trained AIs 21 may be trained and generated for each inspection area. The AI ​​processing circuit 20 may select one trained AI 21 from among the multiple trained AIs 21 according to the inspection area.

[0076] The trained AI 21 may be associated with image-related information. For example, sequences included in the image-related information are related to how artifacts are superimposed on MR images. Therefore, the input training data for the trained AI 21 may include sequences. The trained AI 21 may be trained for each sequence.

[0077] The AI ​​processing circuit 20 includes multiple trained AIs 21, and each of the multiple trained AIs 21 may be trained and generated for each sequence. The AI ​​processing circuit 20 may select one trained AI 21 from among the multiple trained AIs 21 according to the sequence.

[0078] The predicted image PII output from the trained AI 21 does not have to be a predicted MR image of the subject P, but may be any MR image for each examination area according to the gender, age, and physique of the subject P.

[0079] [Procedure for Motion Artifact Prediction Method] Figure 5 is a flowchart showing the procedure for the motion artifact prediction method according to the first embodiment. Each step shown in Figure 5 may be executed by a processor in a computer, and some steps may be executed by an external device and / or operator. The steps executed by the processor may be executed in response to the operator's actions.

[0080] The subject P, as shown in Figure 3, enters the room where the MRI machine is installed, and the subject P is laid down on the examination table BE. Imaging of the subject P is started using the MRI machine. In response to the start of imaging of the subject P, the medical image diagnostic support device 10 starts the motion artifact prediction procedure.

[0081] In step S10, the motion information processing circuit 16 shown in Figure 3 acquires an image of the subject P from the camera 12, and also acquires information such as the examination area information of the subject P, as well as the subject P's name, personal ID (identification), age, and gender, as examination information from the subject information storage circuit 14.

[0082] In step S10, the motion information processing circuit 16 detects the motion of the subject P based on the captured image and inspection information transmitted from the camera 12, and generates motion information BI. The motion information processing circuit 16 outputs the motion information BI of the subject P.

[0083] In step S12, the AI ​​processing circuit 20 generates prediction information PI based on the motion information BI of the subject P generated in step S10. In step S14, the AI ​​processing circuit 20 outputs the prediction information PI generated in step S12.

[0084] In step S16, the prediction information PI is communicated using the output device 22. If a display device is used as the output device 22, the operator can confirm the prediction information PI displayed on the display device. After confirming the prediction information PI, the operator may continue imaging with the MRI device. If imaging with the MRI device is to be continued, each step from step S10 to step S16 is repeated.

[0085] On the other hand, an operator who has confirmed the predictive information PI may interrupt the imaging process of the MRI device. When interrupting imaging, the operator inputs a command to the imaging control device of the MRI device indicating the interruption of imaging.

[0086] The imaging control unit of the MRI device may resume imaging if it receives a command to resume imaging while imaging is interrupted. The imaging control unit of the MRI device may terminate imaging if it receives a command to terminate imaging.

[0087] The motion artifact prediction method is an example of how the medical image diagnostic support device described herein operates.

[0088] [Example of displaying prediction information] Figure 6 shows an example of displaying prediction information. When prediction information PI is generated, prediction information PI is displayed. The display device on which prediction information PI is displayed is, for example, the console display, but is not limited to this, and any display device that the operator can see may be used. The display form of prediction information PI may be, for example, a pop-up display on the inspection screen 50. Figure 6 illustrates an example in which a prediction image PII, as an example of the display form of prediction information PI, is displayed as a pop-up on the inspection screen 50 as a prediction result display screen 51. The display form of prediction information PI is not limited to pop-up display. For example, the prediction result display screen 51 may be incorporated into the inspection screen 50.

[0089] Figure 6 illustrates an example in which the prediction information PI is displayed in the center of the inspection screen 50. However, the prediction information PI may be displayed at any position on the inspection screen 50, such as the upper right position in Figure 6. The size of the prediction information PI can also be defined as appropriate.

[0090] The screen on which the prediction information PI is displayed is not limited to the inspection screen 50. The prediction information PI may be displayed on a screen different from the inspection screen 50. For example, the prediction information PI may be displayed on a dedicated screen for displaying the prediction information PI.

[0091] [Other embodiments of prediction information] Figure 7 is a schematic diagram of a trained AI relating to another embodiment. The trained AI 21A shown in the figure outputs character information PIT that explains motion artifacts as prediction information PIA. The character information PIT may include symbols and icons, etc.

[0092] Figure 8 shows an example of the display of the prediction information illustrated in Figure 7. In the inspection screen 50A shown in Figure 8, a prediction result display screen 51A containing the prediction information PIA is displayed as a pop-up. The display format, display position, and size of the prediction information PIA shown in Figure 8 can be defined in the same way as the prediction information PI shown in Figure 6.

[0093] The prediction information PI may be a combination of the prediction image PII shown in Figure 6 and the text information PIT shown in Figure 8. That is, the inspection screen 50 shown in Figure 6 may display prediction information PI that includes the prediction image PII and the text information PIT shown in Figure 8.

[0094] [Effects of the First Embodiment] The medical image diagnostic support device and motion artifact prediction method according to the first embodiment can obtain the following effects.

[0095] [1] Based on the detection results of the subject P's body movement during the imaging period of the MRI device, subject P's body movement information BI is generated. Based on subject P's body movement information BI, predictive information PI of motion artifacts that will occur in the MR image generated by imaging subject P is generated. The predictive information PI is output to the MRI device during the imaging period. This allows the operator, having grasped the predictive information PI, to decide whether to continue imaging or interrupt imaging and perform re-imaging.

[0096] [2] Predictive information PI is output from a trained AI 21, etc., which has been input with the motion information BI of the subject P being imaged. The trained AI 21, etc., is trained using a deep learning method, etc., with the motion information BI for the imaging period of any subject as input training data, and MR images with motion artifacts superimposed as output training data. This makes it possible to predict motion artifacts with high accuracy.

[0097] [3] The trained AI21, etc., receives the examination area information of the subject P as examination information II. This makes it possible to predict motion artifacts according to the examination area of ​​the subject P.

[0098] [4] The trained AI 21, etc., receives the positional information of the subject P as examination information II. This allows the examination area to be identified when predicting motion artifacts for each examination area of ​​the subject P.

[0099] [5] Predictive information PI is applied to predictive image PII, which is an MR image with motion artifacts superimposed. This allows the operator to visually grasp the motion artifacts.

[0100] [6] Predictive information PI is applied to textual information PIT that describes motion artifacts. This allows the operator to visually understand the motion artifacts.

[0101] [Second Embodiment] Figure 9 is a schematic diagram of a trained AI applied to the motion artifact prediction method according to the second embodiment. The trained AI 21B shown in the figure has output prediction information PIB which differs from the prediction information PI etc. output from the trained AI 21 shown in Figure 4.

[0102] In other words, the motion artifact prediction method according to the second embodiment can reduce the burden on the operator in deciding whether or not to re-image by having the system evaluate whether or not re-image is necessary.

[0103] The training dataset for the trained AI21B consists of input training data (BI) of body movement information of any subject during the imaging period of the MRI device, and output training data (MR) of labeled images generated by imaging the same subject using the MRI device.

[0104] Labels are applied to indicate whether or not re-imaging is necessary. Examples of labels include a combination of information indicating that imaging can continue (e.g., re-imaging is not necessary and imaging can continue) and information indicating that imaging is interrupted and re-imaging is performed (e.g., re-imaging is necessary).

[0105] When the trained AI21B receives motion information BI during the imaging period of the MRI device, it outputs prediction information PIB, which includes a predicted image PIBI and a judgment result PIBR indicating whether or not re-imaging is necessary. Figure 9 shows text information indicating that re-imaging is necessary as the judgment result PIBR.

[0106] The body movement detection of the subject P and the body movement information BI of the subject P input to the trained AI 21B, which are applied to the motion artifact prediction method according to the second embodiment, are the same as those in the motion artifact prediction method according to the first embodiment.

[0107] [Procedure for Motion Artifact Prediction Method] Figure 10 is a flowchart showing the procedure for the motion artifact prediction method according to the second embodiment. In the motion artifact prediction method according to the second embodiment, step S12A in Figure 10 is performed instead of step S12 in Figure 5, and step S14A in Figure 10 is performed instead of step S14 in Figure 5.

[0108] In step S12A of Figure 10, the AI ​​processing circuit equipped with the trained AI 21B shown in Figure 9 acquires the motion information BI of the subject P during the imaging period of the MRI device, generates the predicted image PIBI shown in Figure 9, and also generates the determination result PIBR of the need for re-imaging. For example, in step S12A, the predicted information PIB shown in Figure 9 is generated.

[0109] In step S14A, the AI ​​processing circuit equipped with the trained AI 21B outputs the predicted image PIBI and the judgment result PIBR generated in step S12A. In step S16, the operator can recognize the character information representing the judgment result PIBR.

[0110] In step S16, if the operator confirms the PIBR result indicating that re-imaging is necessary, they can instruct the imaging control device of the MRI system to perform re-imaging.

[0111] Figure 11 shows an example of displaying predictive information that includes a judgment result indicating whether re-imaging is necessary. The figure illustrates the judgment result PIBR to which text information indicating that re-imaging is necessary is applied, and the predictive information PIB which includes the predicted image PIBI. The figure also illustrates an example in which the predictive information PIB is displayed as a prediction result display screen 51B, popping up on the examination screen 50B.

[0112] The prediction information PIB shown in Figure 11 has the judgment result PIBR on the left side of the figure and the prediction image PIBI on the right side of the figure, but the arrangement of the judgment result PIBR and the prediction image PIBI is not limited to the configuration shown in Figure 11. The display form, display position, and size of the prediction information PIB shown in Figure 11 can be defined in the same way as the prediction information PI.

[0113] [Effects of the Second Embodiment] The motion artifact prediction method according to the second embodiment can obtain the following effects.

[0114] [1] The AI ​​processing circuit equipped with the trained AI21B acquires the motion information BI of the subject P during the imaging period of the MRI device, and outputs predictive information PIB which includes the judgment result PIBR of whether or not re-imaging is necessary. This reduces the burden on the operator in making decisions when performing re-imaging.

[0115] [2] The trained AI21B is trained using a training dataset in which the input training data is the body movement information BI of any subject during the imaging period of the MRI device, and the output training data is the MR image generated by imaging the said subject and to which a label indicating whether or not re-imaging is necessary is attached. As a result, the trained AI21B can output a PIBR, which is a judgment result of whether or not re-imaging is necessary, based on the body movement information BI of the subject P during the imaging period of the MRI device.

[0116] [Third Embodiment] The motion artifact prediction method according to the third embodiment differs from the motion artifact prediction method according to the second embodiment in that the labeling of the training data used to train the trained AI is different.

[0117] Specifically, the training dataset used to train the trained AI consists of body movement information of any subject during the imaging period of the MRI device, which is generated during imaging by the MRI device. The output image data is a labeled MRI image of that subject. The labels include information indicating whether re-imaging is necessary and information prompting confirmation of the state of the subject P.

[0118] Examples of labels include a set of information indicating the continuation of imaging, such as "re-imaging is not necessary and imaging can continue," information indicating the interruption and re-imaging of imaging, such as "re-imaging is not necessary," and information indicating that the condition of the subject P needs to be checked immediately.

[0119] When the trained AI receives motion information BI of the subject P during the imaging period of the MRI device, it outputs either predictive information PI indicating that re-imaging is not necessary, predictive information PI indicating that re-imaging is necessary, or predictive information PI indicating that the state of the subject P needs to be checked.

[0120] The body movement detection of the subject P and the body movement information BI of the subject P that are input to the trained AI, as applied to the motion artifact prediction method according to the third embodiment, are the same as those of the motion artifact prediction method according to the first embodiment.

[0121] Examples of situations in which predictive information indicating the need to check the condition of the subject P is output include cases where significant body movement of the subject P is detected, such as when the subject P, who is lying on their back, turns to the side, or when the subject P coughs violently.

[0122] The medical image diagnostic support device according to the third embodiment differs from the medical image diagnostic support device 10 according to the first embodiment in that it is equipped with a trained AI whose function differs from the trained AI 21 equipped in the AI ​​processing circuit 20 shown in Figure 3. In the third embodiment, the medical image diagnostic support device may be equipped with the same components as the medical image diagnostic support device 10 according to the first embodiment, except for the trained AI.

[0123] [Effects of the Third Embodiment] The motion artifact prediction method according to the third embodiment can obtain the following effects.

[0124] [1] When the trained AI acquires motion information BI of the subject P during the imaging period of the MRI device, it outputs predictive information PI indicating that re-imaging is not necessary, predictive information PI indicating that re-imaging is necessary, or predictive information PI indicating that the status of the subject P needs to be checked. As a result, the operator who recognizes the predictive information PI can instruct the MRI device to interrupt imaging. In addition, if predictive information PI indicating that the status of the subject P needs to be checked is output, the operator can immediately check the status of the subject P.

[0125] [2] The trained AI is trained using a training dataset in which the input training data is the body movement information of any subject during the imaging period of the MRI device, and the output training data is the MR image generated by imaging the said subject, to which a label is attached indicating that re-imaging is not necessary, a label indicating that re-imaging is necessary, or a label indicating that the subject's condition needs to be checked. As a result, when the trained AI 21 acquires the body movement information of the subject P during the imaging period of the MRI device, it can output predictive information PI which contains information corresponding to the labels applied to the training data.

[0126] [Fourth Embodiment] Figure 12 is a block diagram showing an example configuration of a medical image diagnostic support device according to the fourth embodiment. The medical image diagnostic support device 10C shown in the figure includes an AI processing circuit 20C instead of the AI ​​processing circuit 20 shown in Figure 3.

[0127] When the control circuit (not shown) of the imaging control device 60, which controls imaging by the MRI device, receives a control signal from the AI ​​processing circuit 20C indicating that re-imaging is necessary, or a control signal indicating that the status of the subject P needs to be checked, the control circuit of the imaging control device 60 outputs an imaging control signal to interrupt imaging by the MRI device.

[0128] In other words, the medical image diagnostic support device 10C according to the fourth embodiment instructs the MRI device to interrupt imaging in accordance with the motion artifact prediction information PI. For example, if the AI ​​processing circuit 20C outputs prediction information PI that includes information indicating that re-imaging is necessary, or information indicating that the condition of the subject P needs to be checked, the imaging control device 60 interrupts imaging by the MRI device. Here, interruption of imaging is synonymous with stopping imaging, pausing imaging, ceasing imaging, or stopping imaging.

[0129] [Procedure for Motion Artifact Prediction Method] Figure 13 is a flowchart showing the procedure for the artifact prediction method according to the fourth embodiment. The procedure from the subject entering the room to the start of imaging by the MRI device is the same as the procedure for the artifact prediction method according to the first embodiment.

[0130] In step S30, the motion information processing circuit 16 shown in Figure 12 detects the motion of the subject P based on the captured image and inspection information transmitted from the camera 12, and generates motion information BI for the subject P. The motion information processing circuit 16 outputs the motion information BI for the subject P.

[0131] In step S32, the AI ​​processing circuit 20C generates predictive information PI regarding motion artifacts based on the motion information BI of the subject P generated in step S30. The predictive information PI includes information such as whether re-imaging by the MRI device is not necessary, whether re-imaging is necessary, or whether the state of the subject P needs to be checked.

[0132] In step S34, the AI ​​processing circuit 20C acquires the prediction information PI output in step S32 and outputs the acquired prediction information PI.

[0133] In step S34, if predictive information PI is obtained that does not require re-imaging and does not require confirmation of the condition of the subject P, the process proceeds to step S36 and imaging continues. If imaging continues, each step from S30 to S38 is repeatedly executed until imaging is completed.

[0134] On the other hand, if predictive information PI is obtained in step S34 indicating that re-imaging is necessary or that the status of the subject P needs to be checked, the process proceeds to step S38, where the AI ​​processing circuit 20C transmits a control signal to the imaging control device 60 indicating that re-imaging is necessary or that the status of the subject P needs to be checked.

[0135] In step S38, the imaging control device 60 performs control to interrupt imaging by the MRI device. The imaging control device 60 also decides whether to continue the interruption of imaging, perform re-imaging, or resume imaging. If re-imaging is to be performed, the current imaging is terminated and re-imaging is performed. The medical image diagnostic support device 10C may transmit a command signal to the imaging control device 60 indicating re-imaging to perform re-imaging.

[0136] If imaging by the MRI device is to be resumed, each step from step S30 to step S38 is repeatedly executed during the period until imaging is completed. The medical image diagnostic support device 10C may resume imaging by sending a command signal to the imaging control device 60 indicating the resumption of imaging.

[0137] [Effects of the Fourth Embodiment] The medical image diagnostic support device and motion artifact prediction method according to the fourth embodiment can obtain the following effects.

[0138] [1] The prediction information PI output from the AI ​​processing circuit 20C includes information indicating whether re-imaging is necessary or whether the status of the subject P needs to be checked. When re-imaging is necessary or the status of the subject P needs to be checked, the AI ​​processing circuit 20C transmits a control signal representing the above prediction information PI to the imaging control device 60. This allows the imaging control device 60 to interrupt imaging by the MRI device when re-imaging is necessary or when the status of the subject P needs to be checked.

[0139] [2] When the interrupted MRI imaging is resumed, the medical image diagnostic support device 10C performs a prediction of motion artifacts during the imaging period of the MRI device after the resumption. This allows for the continued prediction of motion artifacts for the subject P being imaged.

[0140] [Modifications of the Embodiment] The information input to the trained AI 21 etc. shown in Figure 4 may include not only the body movement information of the subject P being imaged, but also additional image-related information. For example, different imaging sequences result in different patterns of artifact appearance in MR images. Imaging parameters other than the imaging sequence, the k-space filling method, and the k-space filling position at the time of body movement also affect the patterns of artifact appearance in MR images. The image-related information may include various types of information that affect the patterns of artifact appearance in MR images.

[0141] The predictive information PI output from the trained AI 21, etc., is not limited to a form that includes a predictive image PII, which is an MR image that may be generated by imaging of the subject P, or a form that includes textual information PIT related to artifacts. The predictive information PI may include at least one of the amount of artifact displacement from the actual position of the examination site and the k-space generated by imaging.

[0142] For example, a trained AI 21 may output predictive information PI, which includes the amount of artifact movement from the actual position of the examination site, when it receives input such as motion information BI of the subject P during the imaging period, the imaging sequence, and the method of filling k-space.

[0143] [Example of Hardware Configuration of Medical Image Diagnostic Support Device] Figure 14 is a block diagram showing the hardware configuration of the electrical configuration of a medical image diagnostic support device. The medical image diagnostic support device 10 shown in Figure 3 is configured using a computer.

[0144] The processing functions of the medical image diagnostic support device 10 may be implemented by a computer system including multiple computers. The computer used in the medical image diagnostic support device 10, etc., may be a virtual machine.

[0145] The medical image diagnostic support device 10 comprises a processor 82, a main memory 84, an auxiliary storage device 86, an input / output interface 88, and a bus 90. The processor 82 is connected to the memory 84, the storage 86, the input / output interface 88, the input device 92, and the display device 94 via the bus 90.

[0146] The processor 82 executes instructions stored in memory 84. Memory 84 includes RAM. Memory 84 may also include ROM. For example, storage 86 may include a hard disk drive or a solid-state drive. Storage 86 may be a combination of a hard disk drive and a solid-state drive. Storage 86 may also include an external storage device such as a removable disk.

[0147] RAM is an abbreviation for Random Access Memory, and ROM is an abbreviation for Read Only Memory. Hard disk drives can be referred to as HDDs, using the abbreviation for Hard Disk Drive. Solid state drives can be referred to as SSDs, using the abbreviation for Solid State Drive.

[0148] The storage device, including the memory 84 and storage 86, stores programs and data that enable the various functions of the medical image diagnostic support device 10. The processor 82 executes the programs stored in the memory 84 to enable the various functions of the medical image diagnostic support device 10.

[0149] The input / output interface 88 includes a communication interface that can connect to a network, and a connection interface that can connect to external devices. Examples of connection interfaces that can connect to external devices include the Universal Serial Bus and HDMI (registered trademark). The Universal Serial Bus may be referred to as USB, using the abbreviation Universal Serial Bus. HDMI is an abbreviation for High-Definition Multimedia Interface.

[0150] The processor 82 is freely connected to various devices of the medical image diagnostic support device 10 via the input / output interface 88. This allows the processor 82 to send and receive necessary information.

[0151] Various instructions and information entered by the operator via the input device 92 are input to the medical image diagnostic support device 10. The operator interactively operates the medical image diagnostic support device 10 using the input device 92 and the display device 94.

[0152] The display device 94 displays various information from the medical image diagnostic support device 10. The display device 94 is used as part of the user interface when receiving input from the input device 92. Note that the display device 94 is not limited to one; a multi-display configuration with multiple display devices is also possible.

[0153] The hardware configuration of the medical image diagnostic support device 10 described above is also applicable to the medical image diagnostic support device implementing the motion artifact prediction method according to the second embodiment, the medical image diagnostic support device implementing the motion artifact prediction method according to the third embodiment, the medical image diagnostic support device 10C according to the fourth embodiment, and the medical image diagnostic support device according to a modified example.

[0154] [Regarding the computer and processor applied to each process] In this embodiment, each process is executed on any computer. Furthermore, any computer may execute these processes using a processor, a program, or a combination thereof. Any computer may be a general-purpose computer, a computer designed for a specific purpose, a workstation or other system, or other hardware elements capable of running a program.

[0155] A processor may be composed of one or more hardware components, and the type of hardware is not limited. For example, a processor may be composed of hardware such as a CPU (Central Processing Unit), MPU (Micro Processing Unit), FPGA (Field Programmable Gate Array) or other programmable logic devices, an ASIC (Application Specific Integrated Circuit) or other dedicated circuitry for executing specific processes, a GPU (Graphic Processing Unit), or an NPU (Neural Processing Unit). Furthermore, the processor has various parts (Units) or means (Means) that execute the various processes in this embodiment. The type of hardware may also be a combination of different types of hardware. When multiple hardware components are configured to execute one or more processes of a processor, these multiple hardware components may reside in physically separate devices or in the same device. In addition, in any embodiment, the order of each process performed by the processor is not limited to the order described above and may be changed as appropriate. Hardware is composed of electrical circuits (circuitry) that combine circuit elements such as semiconductor elements.

[0156] Furthermore, this embodiment may be implemented by hardware, software, firmware, microcode, or a combination thereof. The software, firmware, and microcode are composed of a program. The program may also be, for example, a group of program modules, each of which may be implemented by a processor configured to perform its respective function. The program may be program code or multiple code segments stored on one or more non-temporary computer-readable media (e.g., storage media or other storage). The program may be divided and stored on multiple non-temporary computer-readable media located on devices that are physically separated from each other. The program code or code segment may represent any combination of procedures, functions, subprograms, routines, subroutines, modules, software packages, classes, or instructions, data structures, or program statements. The program code or code segment may be connected to other code segments or hardware circuits by sending and receiving information, data, arguments, parameters, or memory contents.

[0157] [Specific Examples of Trained AI] Figure 15 is an explanatory diagram of the training data. The training dataset TDS is used to train the trained AI 21 shown in Figure 4. The training dataset TDS is a set of input training data TDI and output training data TDO.

[0158] The input learning data TDI is information related to body movement acquired during the imaging period of any subject. The input learning data TDI may be the body movement information BI of any subject output from the body movement information processing circuit 16 shown in Figure 3.

[0159] The output training data TDO consists of MR images generated during imaging periods in which information related to the body movements of any given subject is acquired, and MR images containing motion artifacts are applied.

[0160] The input training data TDI may include additional image-related information, which may include at least one of the following: imaging parameters such as the sequence in imaging of any subject, a k-space filling method, and the k-space filling position when body motion occurs. The image-related information may include reconstructed images free of motion artifacts.

[0161] Figure 16 is a schematic diagram illustrating a specific example of learning. The network NW applied to the trained AI 21 etc. shown in Figure 4 is a generative model. An example of a network NW is a GAN. GAN is an abbreviation for Generative Adversarial Networks. The network NW may be a convolutional neural network or the like, and may be implemented by combining multiple networks.

[0162] The input training data TDI includes MR image MI, body motion information BI during the imaging period, and k-space filling position FD. The MR image MI applied to the input training data TDI may be an image in real space or k-space imaging data.

[0163] The output training data TDO is the MR image MIA in which artifacts are present. The artifact image in Figure 16 represents the MR image MIA in which artifacts are present.

[0164] Figure 16 illustrates time-series data of the magnitude of body movement as body movement information BI. In this figure, body movement information BI is illustrated using a graph format in which the horizontal axis is time and the vertical axis is the magnitude of body movement. In addition, the k-space filling position FD is illustrated using a graph format in which the horizontal axis is k-space position and the vertical axis is whether or not measurement was performed.

[0165] Pattern 1 of the input training data TDI and Pattern 1 of the output training data TDO constitute one set of training dataset TDS. Similarly, Pattern 2 of the input training data TDI and Pattern 2 of the output training data TDO constitute one set of training dataset TDS. The network NW performs training using multiple sets of training dataset TDS.

[0166] The MR image MI included in the input training data TDI may be a reconstructed image free of artifacts. The MR image MI included in the input training data TDI may be a reconstructed image generated from some of the imaging data up to the point where the subject's body movement occurs. The MR image MI included in the input training data TDI may have a lower resolution compared to the MR image MIA included in the output training data TDO.

[0167] When training a network NW for each image type, the input training data TDI does not need to include MR image MI. Also, when training a number of network NW equal to the number of image types multiplied by the number of k space filling positions for each image type and k space filling position, the input training data TDI does not need to include k space filling positions. In other words, when training a number of network NW equal to the number of image-related pieces of information for each image-related piece of information, the input training data TDI does not need to include image-related information. Note that the training procedure illustrated in Figure 16 is an example of the training method of this disclosure.

[0168] [Specific Examples of Input Training Data] [Example of Not Using MR Images in Input Training Data TDI] In the first example where MR images are not used in the input training data TDI, a network NW is generated for each imaging area, and a training dataset TDS is prepared, consisting of input training data TDI for each imaging area and output training data TDO for each imaging area.

[0169] The input training data TDI includes body movement information BI. The input training data TDI may also include k-space filling position FD. The output training data TDO may consist of at least one of the following: MR images, textual information representing artifacts, and information indicating whether or not re-imaging is necessary.

[0170] In the second example, where MR images are not used in the input training data TDI, a network NW is generated for each combination of imaging site and imaging sequence, and a training dataset TDS is prepared for each combination of imaging site and imaging sequence. The input training data TDI and output training data TDO are the same as in the first example.

[0171] Here, the imaging protocol includes a combination of imaging area and imaging sequence. The network NW in the second example is generated by learning each imaging protocol set in the imaging process for acquiring MR images.

[0172] In the third example, where MR images are not used in the input training data TDI, the size of the imaging area is further combined with the combination of imaging area and imaging sequence, and a network NW is generated for each combination of imaging area, imaging sequence, and imaging area size. A training dataset TDS is also prepared for each combination of imaging area, imaging sequence, and imaging area size. The input training data TDI and output training data TDO are the same as in the first and second examples.

[0173] For example, prior to performing a series of imaging scans, it is possible to obtain imaging data including three cross-sections or other imaging areas, and then analyze the resulting imaging data to determine the size of the imaging area.

[0174] In the fourth example, where MR images are not used in the input training data TDI, instead of the size of the imaging area as applied in the third example, a network NW is generated for each subject P (the imaging subject), or for each subject P's height and weight, and a training dataset TDS is prepared. From the subject P's weight, or their height and weight, a certain degree of the subject P's physique can be determined, and from the subject P's physique, the size of the imaging area can be estimated.

[0175] [Example of using MR images for input training data TDI] In the example of using MR images for input training data TDI, the input training data TDI will consist of motionless MR images and motion information BI from the same subject, while the output training data TDO will consist of motionless MR images from the same subject. Motionless MR images from the same subject are acquired separately and prepared in advance.

[0176] The input training data TDI may be an MR image from which body motion has been removed by image processing based on body motion information BI applied to the output training data TDO. The output training data TDO may be an MR image from which body motion has been added by image processing based on body motion information BI applied to the input training data TDI.

[0177] In the example where MR images are not used in the input training data TDI, the processing of input to the network NW and the processing of increasing the types of outputs of the network NW, as in the example where MR images are used in the input training data TDI, may be applied as appropriate.

[0178] [Example where MR images are used in the network operation phase but not for training] In examples where MR images are used in the network operation phase but not for training, the input training data TDI is modified with motion information BI, and the output training data TDO is modified with torsion effect, torsion transformer, or difference image. For the generation of torsion effect, torsion transformer, and difference image, MR images without motion and MR images with motion from the same subject are used.

[0179] During the network operation phase, for example, images of the corresponding MR image acquisition area, such as scanogram images (scout images), and body motion information BI are input. A body motion-represented MR image is generated and output by incorporating a twist effect, twist transformer, or difference image into the corresponding MR image acquisition area.

[0180] [Examples of application to programs and program products] The motion artifact prediction method according to the embodiment may be configured such that a processor or a computer equipped with a processor is configured as a program or program product that realizes the functions of each process. For example, the computer may be configured with a program or program product that realizes the function of acquiring body movement information BI of the subject P during the imaging period, the function of generating prediction information PI of artifacts caused by the body movement of the subject P based on the body movement information BI of the subject P, and the function of outputting the prediction information PI.

[0181] [Example of MRI System Configuration] Figure 17 is a perspective view showing the external appearance of an MRI system. The MRI system 100, which is a magnetic resonance imaging system, comprises a gantry 110, which is the main body of the system, and a patient bed 130. The patient bed 130 is equipped with a top plate 130A and is positioned on the front side of the bore 120, which is a cylindrical imaging space provided in the gantry 110. The top plate 130A can be moved into and out of the bore 120 using a top plate drive mechanism provided in the patient bed 130. Note that the top plate drive mechanism is not shown in the diagram.

[0182] The berth 130 may be fixed to the gantry 110, and may be a dockable berth that is a movable berth that can be attached to and detached from the gantry 110.

[0183] Figure 18 is a schematic diagram showing the internal configuration of an MRI apparatus. The MRI apparatus 100 includes a static magnetic field generating magnet 104, a gradient magnetic field coil 106, and an RF transmitting coil 108. RF is an abbreviation for Radio Frequency.

[0184] The MRI apparatus 100 includes a high-frequency magnetic field generator 112, a receiver 114, a gradient magnetic field power supply 116, and a sequencer 118. The MRI apparatus 100 also includes a control unit 140, an operation unit 150, and a display unit 152.

[0185] The subject 102 is placed on the top plate 130A of the bed 130 and positioned in the imaging space. That is, the top plate 130A on which the subject 102 is placed moves to the bore 120. As a result, the area of ​​the subject 102 being examined is positioned at the center of the static magnetic field within the bore 120. Note that the subject 102 shown in Figure 18 includes the imaging subject P shown in Figure 1, etc.

[0186] The static magnetic field generating magnet 104 generates a uniform static magnetic field in the imaging space within the bore 120 where the subject 102 is placed. The static magnetic field generating magnet 104 includes a static magnetic field source of the permanent magnet type, normal conducting type, or superconducting type. The gradient magnetic field coil 106 generates a gradient magnetic field in the imaging space. The gradient magnetic field coil 106 consists of gradient magnetic field coils in the three axes X, Y, and Z, which are real space coordinate systems and stationary coordinate systems. Each gradient magnetic field coil is connected to the gradient magnetic field power supply 116 and supplied with current. This generates gradient magnetic fields in the three axes X, Y, and Z. The Z axis direction is the direction of the static magnetic field, the Y axis direction is the vertical direction, and the X axis direction is perpendicular to the Y axis direction and the Z axis direction, respectively. Note that the X axis direction, Y axis direction, and Z axis direction are omitted in Figure 2.

[0187] The RF transmitting coil 108 is a coil that irradiates the subject 102 with a high-frequency magnetic field pulse. The high-frequency magnetic field pulse may also be referred to as an RF pulse. The RF transmitting coil 108 is connected to a high-frequency magnetic field generator 112, to which a high-frequency pulse current is supplied. The high-frequency magnetic field generator 112 is driven according to commands from the sequencer 118 to amplitude modulate the high-frequency pulse and supply the amplified high-frequency pulse to the RF transmitting coil 108.

[0188] The sequencer 118 sends commands to the high-frequency magnetic field generator 112 and the gradient magnetic field power supply 116 according to the imaging pulse sequence, generating a high-frequency magnetic field and a gradient magnetic field, respectively. The generated high-frequency magnetic field is applied to the subject 102 as a pulsed high-frequency magnetic field through the RF transmitting coil 108. This induces a nuclear magnetic resonance phenomenon in the spins of atoms constituting the biological tissue of the subject 102. Nuclear magnetic resonance can be referred to as NMR, using the abbreviation of Nuclear Magnetic Resonance.

[0189] The MRI apparatus 100 includes a receiving coil unit 200. The receiving coil unit 200 is a multi-channel RF coil unit that includes multiple receiving coils that receive echo signals emitted by the NMR phenomenon of the spins of atoms constituting the biological tissue of the subject 102. The echo signals may be referred to as NMR signals.

[0190] A receiving-side cable, which outputs the NMR signal received by the receiving coil unit 200, is electrically connected to the receiving coil unit 200. A receiving-side connector is electrically connected to the end of the receiving-side cable. The receiving-side connector is electrically connected to the bed-side connector of the bed-side cable. The bed-side cable is electrically connected to the control unit 140. As a result, the receiving coil unit 200 and the control unit 140 are electrically connected in a way that allows for free communication. Note that the receiving-side cable, receiving-side connector, bed-side cable, and bed-side connector are not shown in the diagram.

[0191] Wireless communication may be applied to the communication between the receiving coil unit 200 and the control unit 140. For example, the receiving coil unit 200 may be equipped with a wireless transmitter, and the bed 130 may be equipped with a wireless receiver that is electrically connected to the control unit 140, thereby realizing wireless communication between the receiving coil unit 200 and the control unit 140.

[0192] The receiving coil unit 200 may be a blanket type applicable to imaging of the chest and abdomen, etc. Different receiving coil units 200 may be applied depending on the area being examined. For example, receiving coil units 200 for imaging various parts such as the head, spine, abdomen, legs, and arms can be used. One receiving coil unit 200 may be used for a single imaging scan, or multiple units may be used. Multiple receiving coil units 200 for imaging different parts may be used together. The receiving coil unit 200 may sometimes be simply called a receiving coil. The NMR signal generated from the subject 102 is received using the receiving coil unit 200, amplified by a preamplifier (not shown) in the receiving coil unit 200, and transmitted to the receiver 114.

[0193] The sequencer 118 controls the operation of each component by applying pre-programmed timings and intensities. The program, in particular, which describes the timing and intensity of RF pulses, gradient magnetic fields, and signal reception, is called a pulse sequence. Various types of pulse sequences are known, depending on the purpose.

[0194] The operation unit 150 includes a mouse, keyboard, etc., and functions as part of a GUI that accepts operator input using a display operation window displayed on the display unit 152. In other words, the operation unit 150 and the display unit 152 function as a GUI for the operator to input commands such as starting, stopping, pausing, selecting pulse sequences, imaging conditions, and processing conditions for the MRI device 100. GUI is an abbreviation for Graphical User Interface.

[0195] The control unit 140 controls the operation of the MRI apparatus 100 via the sequencer 118, and receives signals transmitted from the receiver 114, performing various signal processing such as image reconstruction.

[0196] The receiver 114 performs analog-to-digital (AD) conversion and necessary signal processing on the amplified NMR signal to generate data, and transmits this data to the control unit 140. This data is also called the received signal or measurement data.

[0197] The control unit 140 acquires biological information from the subject 102, and the biological information values ​​calculated from the acquired biological information are set as imaging parameters. The control unit 140 uses the biological signal representing the acquired biological information as a trigger to execute a synchronized imaging task synchronized with the biological signal.

[0198] The control unit 140 receives various instruction inputs from the operation unit 150, comprehensively controls each part of the MRI apparatus 100, and performs processing such as inverse Fourier transform on the spatial frequency domain echo signal received via the sequencer 118 to convert it into a real-space image, thereby generating an MR image.

[0199] The control unit 140 can be configured using a computer. The computer used for the control unit 140 may be a personal computer or a workstation. That is, the control unit 140 includes a processor and memory, and the processor executes a program containing instructions stored in the memory to realize various functions of the MRI apparatus 100. The hardware configuration of the control unit 140 may be as shown in Figure 14.

[0200] The MRI apparatus 100 includes a medical image diagnostic support device 160. The medical image diagnostic support device 160 shown in Figure 18 is the same as the medical image diagnostic support device 10 shown in Figure 3, etc. Each part constituting the MRI apparatus 100 may have some or all of the functions of the medical image diagnostic support device 160.

[0201] Various control devices, such as the control unit 140 provided in the MRI apparatus 100, can be configured using the same hardware configuration as the medical image diagnostic support device 10 shown in Figure 14. The imaging control device 60 shown in Figure 12 may include the control unit 140 shown in Figure 18. The imaging control device 60 may include an operation unit 150 and a display unit 152.

[0202] The MRI apparatus 100 is an example of a medical image acquisition apparatus according to this disclosure. The static magnetic field generating magnet 104, gradient magnetic field coil 106, RF transmitting coil 108, high-frequency magnetic field generator 112, receiver 114, gradient magnetic field power supply 116, and sequencer 118 are examples of components of the imaging unit according to this disclosure. The control unit 140 is an example of a component of the image reconstruction unit according to this disclosure.

[0203] This disclosure is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the technical idea of ​​this disclosure. Furthermore, the first embodiment, the second embodiment, the third embodiment, the fourth embodiment, and their variations can be combined as appropriate.

[0204] 10 Medical image diagnostic support device 10C Medical image diagnostic support device 12 Camera 14 Patient information storage circuit 16 Body motion information processing circuit 18 Image-related information storage circuit 20 AI processing circuit 20C AI processing circuit 21 Trained AI 21A Trained AI 21B Trained AI 22 Output device 50 Examination screen 50A Examination screen 50B Examination screen 51 Prediction result display screen 51A Prediction result display screen 51B Prediction result display screen 60 Imaging control device 82 Processor 84 Memory 86 Storage 88 Input / output interface 90 Bus 92 Input device 94 Display device 100 MRI device 102 Patient 104 Static magnetic field generating magnet 106 Gradient magnetic field coil 108 Transmitting coil 110 Gantry 112 High-frequency magnetic field generator 114 Receiver 116 Gradient magnetic field power supply 118 Sequencer 120 Bore 130 Bed 130A Top plate 140 Control unit 150 Operation unit 152 Display unit 200 Receiving coil unit BE Bed BI Body movement information DP Display device FD Space filling position ID Camera II Inspection information MI MR image MIA MR image NW Network P Subject to imaging PI Prediction information PIA Prediction information PIB Prediction information PIBI Prediction image PIBR Judgment result PII Prediction image PIT Text information TB Top plate TDI Input learning data TDO Output learning data TDS Learning dataset S10 to S16 Each step of the motion artifact prediction method S30 to S38 Each step of the motion artifact prediction method

Claims

Processor and A memory in which a program to be executed by the aforementioned processor is stored, Equipped with, The aforementioned processor, Acquire subject movement information that represents the body movements of the subject being imaged. A learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging the subject, and when the motion information of the subject being imaged is input to the learning model, it outputs predicted information about artifacts in medical images generated by imaging the subject, and the motion information of the subject being imaged is input to the learning model. The prediction information corresponding to the motion information of the subject being imaged is obtained from the learning model. Outputting the aforementioned prediction information, Medical image diagnostic support device. 　 The training dataset applied to the training of the aforementioned learning model includes, as input training data, the body movement information of the arbitrary subject during the imaging period, and as output training data, the medical images generated during imaging performed on the arbitrary subject during the imaging period. A medical image diagnostic support device according to claim 1. 　 The input learning data includes examination information that includes at least one of imaging site information and the positional information of any subject. The medical image diagnostic support device according to claim 2. 　 The input learning data includes image-related information, including at least one of imaging parameters, a method for filling k-space, and the filling positions of k-space. The medical image diagnostic support device according to claim 2. 　 The aforementioned input training data includes medical images generated by imaging any subject. The medical image diagnostic support device according to claim 2. 　 The aforementioned processor, The learning model obtains the prediction information which includes at least one of the following: textual information relating to the artifact, the amount of displacement of the artifact from the actual examination site at the artifact's location, and a medical image containing the artifact. Outputs the prediction information obtained from the aforementioned learning model. The medical image diagnostic support device according to claim 2. 　 The training dataset applied to the training of the aforementioned learning model includes, as input training data, the body movement information of the arbitrary subject during the imaging period, and as output training data, medical images generated during imaging performed on the arbitrary subject during the imaging period, and further includes at least one of the following: textual information relating to the artifact and the amount of movement of the artifact from the actual examination site. The medical image diagnostic support device according to claim 6. 　 The aforementioned processor, The predictive information indicating whether or not re-imaging is necessary is obtained from the learning model. Output the acquired prediction information. A medical image diagnostic support device according to claim 1. 　 The training dataset applied to the training of the aforementioned learning model includes, as input training data, the body movement information of the arbitrary subject during the imaging period, and as output training data, medical images generated during imaging performed on the arbitrary subject during the imaging period, and further includes a label indicating whether or not re-imaging is necessary. The medical image diagnostic support device according to claim 8. 　 The processor obtains the predictive information from the learning model that indicates that it is necessary to confirm the state of the person being imaged. Output the acquired prediction information. A medical image diagnostic support device according to claim 1. 　 The training dataset applied to the training of the aforementioned learning model includes, as input training data, the body movement information of the arbitrary subject during the imaging period, and as output training data, medical images generated during imaging performed on the arbitrary subject during the imaging period, and further includes a label indicating that confirmation of the subject's condition is necessary. A medical image diagnostic support device according to claim 10. 　 The aforementioned processor, The predictive information indicating that re-imaging is necessary, or the predictive information indicating that the condition of the subject being imaged needs to be checked, is output to the imaging control device that controls imaging of the subject being imaged. A medical image diagnostic support device according to claim 8 or 10. 　 The aforementioned processor, Applied to imaging of the subject being imaged, and obtaining examination information that includes the examination area, The inspection information is input into the learning model. The learning model outputs the predicted information, which includes a medical image containing artifacts predicted based on the motion information of the subject being imaged. A medical image diagnostic support device according to claim 1. 　 The processor obtains the prediction information from the learning model to which at least one of character information, icons, sounds, and vibrations is applied. A medical image diagnostic support device according to claim 1. 　 A computer that functions as a medical image diagnostic support device, A step to acquire subject movement information representing the body movements of the subject being imaged, A learning model that has learned the relationship between motion information representing the body movements of an arbitrary subject and artifacts in medical images generated by imaging the arbitrary subject, and which outputs predicted artifacts in medical images generated by imaging the subject when the motion information of the subject being imaged is input to the learning model, the step of inputting the motion information of the subject being imaged, The steps include: obtaining the prediction information corresponding to the motion information of the subject being imaged from the learning model, The steps include outputting the aforementioned prediction information, Execute How to operate a medical image diagnostic support device. 　 A computer that functions as a medical image diagnostic support device, A function to acquire subject movement information representing the body movements of the subject being imaged. A learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging the subject, and a function to input the motion information of the subject into the learning model that outputs predicted artifacts in the medical images generated by imaging the subject when the motion information of the subject is input. A function to acquire the predictive information corresponding to the motion information of the subject being imaged from the learning model, and To realize the function of outputting the aforementioned prediction information, program. 　 A non-temporary and computer-readable recording medium on which the program described in claim 16 is recorded. 　 An imaging unit that images a subject to be imaged and generates imaging data of the subject, An image reconstruction unit that generates a reconstruction based on the aforementioned imaging data, Processor and A memory in which a program to be executed by the aforementioned processor is stored, A medical imaging device comprising: The aforementioned processor, Acquire subject movement information that represents the body movements of the subject being imaged. A learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging the subject, and when the motion information of the subject being imaged is input to the learning model, it outputs predicted information about artifacts in medical images generated by imaging the subject, and the motion information of the subject being imaged is input to the learning model. The prediction information corresponding to the motion information of the subject being imaged is obtained from the learning model. Outputting the aforementioned prediction information, Medical imaging device. 　 This is a learning model that has learned the relationship between motion information representing the body movements of any subject and artifacts in medical images generated by imaging the subject. When motion information representing the body movements of the subject being imaged is input, it outputs predictive information about artifacts in medical images generated by imaging the subject. A learning model. 　 A method for training a learning model that, when subject movement information representing the subject's body movement is input, outputs predictive artifact information in a medical image generated by imaging the subject, The system learns the relationship between motion information representing the body movements of an arbitrary subject and artifacts in medical images generated by imaging the said arbitrary subject. Learning methods.

Images