Imaging system having storage medium having trained model enabling reconstruction of medical image having preferable feature value of image quality index, and method for producing trained model

Abstract

To provide an image with a preferable feature value of an image quality index.SOLUTION: Provided in one embodiment is a medical imaging system including a storage medium storing a trained model obtained by learning a first medical image in which a first image quality index is prioritized and a second medical image in which a second image quality index is prioritized. Learning data includes, as annotations, at least one of a first algorithm and a first condition for processing data collected to obtain the first medical image, a feature value of the first image quality index, at least one of a second algorithm and a second condition for processing data collected to obtain the second medical image, and a feature value of the second image quality index. The trained model is configured to determine at least one of a first model algorithm and a first model parameter and determine at least one of a second model algorithm and a second model parameter.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a medical imaging system, and more particularly to a technique for analyzing medical images using a trained model. [Background technology]

[0002] When collecting medical images, there is a need to prioritize features of certain types of image quality indexes over features of other types of image quality indexes depending on what is being observed. For example, when observing intracranial bleeding, such as epidural hematoma, subdural hematoma, subarachnoid hemorrhage, or intracerebral hemorrhage, high density resolution and low noise levels are required because the difference in CT values ​​between the intracranial tissue and the bleeding area is slight. On the other hand, high spatial resolution is not required because bleeding generally spreads over a certain area. Also, for example, when observing a skull fracture (e.g., a fracture with a linear crack), high density resolution is not required, and spatial resolution is prioritized.

[0003] For example, to solve these problems when performing X-ray CT scans, radiologists must set appropriate acquisition parameters and reconstruction algorithms under various constraints, such as radiation dose limits. While there are multiple image quality indices, such as spatial resolution, contrast resolution, noise, and artifacts, certain image quality indices must be prioritized for specific clinical purposes, while other image quality indices must be maintained at a relatively low level. It is not possible to improve all image quality indices, and improving one often leads to a decline in other image quality indices. Such settings are not always easy for less-skilled radiologists, and require a great deal of time and effort. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2016-507085 [Patent Document 2] Patent Publication No. 2020-64609 [Patent Document 3] Patent Publication No. 2023-180532 Summary of the Invention [Problem to be solved by the invention]

[0005] Therefore, there is a need for an image generation system that allows easy specification of desirable features for an image quality index, and a system that allows easy and quick specification of desirable features for an image quality index that is suited to the type and location of diagnosis, without requiring a high level of skill.

[0006] Furthermore, a system is desired that allows images having the same image quality as existing images to be easily obtained. [Means for solving the problem]

[0007] The present disclosure provides a medical imaging system including a storage medium storing a trained model that has been trained using, as training data, a first medical image for which a first image quality index is prioritized and a second medical image for which a second image quality index different from the first image quality index is prioritized. The training data includes, as annotations, at least one of a first algorithm and first conditions for processing data collected to obtain the first medical image, feature values ​​of the first image quality index, at least one of a second algorithm and second conditions for processing data collected to obtain the second medical image, and feature values ​​of the second image quality index. The trained model is configured to determine first information consisting of a first model algorithm and at least one of first model parameters for obtaining a first model medical image with a first image quality index prioritized, and model features of the first image quality index, and to determine second information consisting of a second model algorithm and at least one of second model parameters for obtaining a second model medical image with a second image quality index prioritized, and model features of the second image quality index. Furthermore, when medical image data and a selection of either the first image quality index or the second image quality index are input, the trained model processes the input medical image data using the first or second information according to the input information on the user interface, and outputs a medical image having model features of the first or second image quality index.

[0008] In another aspect of the present disclosure, a method for producing a trained model is provided, the trained model being trained using a first medical image for which a first image quality index is prioritized and a second medical image for which a second image quality index different from the first image quality index is prioritized as training data. The method includes generating training data including, as annotations, at least one of a first algorithm and a first condition for processing data collected to obtain the first medical image, features of the first image quality index, at least one of a second algorithm and a second condition for processing data collected to obtain the second medical image, and features of the second image quality index. The trained model includes determining at least one of a first model algorithm and first model parameters for obtaining the first model medical image for which the first image quality index is prioritized, and determining at least one of a second model algorithm and second model parameters for obtaining the second model medical image for which the second image quality index is prioritized. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram illustrating a schematic configuration of an X-ray CT system according to an embodiment of the present invention. [Figure 2] 2 is a diagram showing the configuration of the main parts of an X-ray tube and an X-ray detection unit. FIG. [Figure 3] FIG. 1 illustrates a learning stage for generating a trained model. [Figure 4] FIG. 1 illustrates the learning stage and inference stage using a trained model. [Figure 5] FIG. 1 is a network diagram showing a trained model. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described, but the present invention is not limited thereto.

[0011] FIG. 1 is a block diagram showing the configuration of an X-ray CT device 100 according to this embodiment. In this disclosure, a medical X-ray CT device will be described as an example; however, the present invention can be applied to non-destructive testing devices such as dental CT devices and CT devices for baggage inspection. This CT device can also be implemented by replacing it with an MRI device, a PET device, a SPECT device, a tomosynthesis device, etc. In the example of FIG. 1, the X-ray CT device 100 includes an operation console 1, an imaging table 10, and a scanning gantry 20. In a preferred embodiment of the present invention, a medical X-ray CT device that collects projection data from a subject, such as a human or a non-human animal, and reconstructs an image will be described as an example; however, the present invention can be applied to dental CT devices, baggage inspection CT devices, PET devices, SPECT devices, tomosynthesis devices, etc.

[0012] The operation console 1 has a computer configuration. Specifically, the operation console 1 includes an input device 2 such as a keyboard or a mouse that accepts input from an operator, a central processing unit 3 that executes scan control processing, preprocessing, image generation processing, etc., and a data acquisition buffer 5 that collects X-ray detector data acquired by the scan gantry 20. The operation console 1 also includes a monitor 6 that displays multi-energy images generated by the image generation processing, etc., and a storage device 7 that stores programs, X-ray detector data, X-ray projection data, dual-energy images, etc. The imaging conditions are input from the input device 2 and stored in the storage device 7.

[0013] The imaging table 10 is provided with a cradle 12 on which the subject 71 is placed and which is moved in and out of an opening 20a (described later) of the scan gantry 20. The cradle 12 is moved up and down and in a horizontal linear motion by a motor built in the imaging table 10.

[0014] The scanning gantry 20 has an opening 20a through which a subject 71 to be imaged is carried. The scanner gantry 20 also has an X-ray tube 21, an X-ray control unit 22 that controls the X-ray tube voltage, X-ray emission timing, etc. of the X-ray tube 21, and a collimator 23 having an aperture that shapes the X-rays emitted from the X-ray tube 21 into a fan-shaped X-ray beam 81. The scanner gantry 20 also has a collimator control unit 27 that controls the aperture of the collimator 23, an X-ray detector 24 that detects the X-rays emitted from the X-ray tube 21, and a data acquisition system (DAS) 25 that collects X-ray detector data (also called raw data) from the output of the X-ray detector 24. The DAS 25 samples analog data received from detector elements of the X-ray detector 24 and converts the analog data into a digital signal for subsequent processing.

[0015] The scanner gantry 20 further includes a gantry rotation unit 15 that holds the X-ray tube 21, collimator 23, and X-ray detector 24 and rotates around the body axis of the subject 71, and a rotation control unit 26 that controls the gantry rotation unit 15. The scanner gantry 20 also includes a gantry control unit 29 that exchanges control signals between the operation console 1 and the X-ray control unit 22, rotation control unit 26, and imaging table 10. In practice, the scanner gantry 20 includes a beam-forming X-ray filter that spatially controls the dose of the X-ray beam 81 and an X-ray filter that controls the radiation quality of the X-ray beam 81, and the gantry rotation unit 15 holds these filters between the collimator 23 and the opening 20a, but illustration and detailed description thereof are omitted here.

[0016] 2 is a diagram showing the configuration of the main parts of the X-ray tube 21 and the X-ray detection unit 24. Here, the vertical direction is defined as the y-axis direction, the direction of transport of the imaging table 10 (which usually coincides with the thickness direction of the X-ray beam 81 or the body axis direction of the subject 71) as the z-axis direction, and the direction perpendicular to the y-axis and z-axis directions (channel direction) as the x-axis direction.

[0017] These components are supported on a predetermined base of the gantry rotating unit 15 and maintain the positional relationship as shown in the figure. That is, the X-ray tube 21 and the X-ray detector 24 are arranged opposite each other with the opening 20a in between. Then, X-rays emitted from the X-ray tube 21 pass through a slit formed by the collimator 23 (not shown in Fig. 2), thereby forming a fan-shaped X-ray beam 81 having a predetermined thickness (cone angle) and spread (fan angle).

[0018] The X-ray tube 21 has a structure in which a cathode sleeve 21s incorporating a focusing electrode and a cathode filament, and a rotating target electrode 21t are housed in a housing 21h, and generates X-rays that diverge from an X-ray focal point F.

[0019] The X-ray detection unit 24 is a so-called multi-row X-ray detector, which is configured by arranging a plurality of, for example, 1,000 X-ray detection elements 24a in a channel direction CH (the direction in which the X-ray beam 81 spreads), and arranging a plurality of, for example, 64 detection element rows in the z-axis direction (the thickness direction of the X-ray beam 81). Here, the detection element rows are numbered 1, 2, 3, . . . , 64 from the end. This realizes a so-called 64-row multi-slice X-ray CT. However, the 64-row detection element row here is merely an example, and the present invention is not limited to this. The X-ray detector 24 forms an X-ray detection surface 24s, which detects the X-ray beam 81 transmitted through the subject 71, using the plurality of X-ray detection elements 24a. The X-ray detection elements 24a are configured as a so-called solid-state detector, for example, by combining a scintillator and a photodiode.

[0020] The central processing unit 3 has a scan control unit 32, a pre-processing unit 34, and an image generation unit 35. The central processing unit 3 is, for example, a processor such as a CPU (Central Processing Unit). The central processing unit 3 executes the functions of the scan control unit 32, the pre-processing unit 34, and the image generation unit 35 by reading and executing a program stored in the storage device 7. The program is an example of an embodiment of the control program according to the present invention.

[0021] The scan control unit 32 controls the X-ray control unit 22, the rotation control unit 26, the collimator control unit 27, and the imaging table 10 via the gantry control unit 29 so as to perform multi-energy imaging of the subject 71. Specifically, the scan control unit 32 controls the above-mentioned units to rotate the X-ray tube 21 and the X-ray detector 24 around the subject 71 and collect X-ray projection data.

[0022] In this embodiment, the X-ray tube 21 can irradiate not only monochromatic X-ray beams but also any number of polychromatic X-ray beams equal to or greater than 2. A tube voltage of any value between 50 and 200 kV, such as 80 kV, 85 kV, 100 kV, 120 kV, 130 kV, 140 kV, 150 kV, or 200 kV, is applied to the X-ray tube 21 by switching it for each view to be acquired, and the X-ray tube 21 irradiates an X-ray beam having an energy spectrum corresponding to the applied tube voltage.

[0023] FIG. 3 shows a trained model production system 300. The trained model production system 300 includes an imaging diagnostic device 310, a data store 320, a model trainer 330, a modeler 340, an output processor 350, a feedback unit 360, a selector 370, and an annotation adding unit 380. Each functional block can be executed by one or more processors. In this embodiment, the imaging diagnostic device 310 in FIG. 3 corresponds to the X-ray CT device 100 in FIG. 1, and the data store 320 corresponds to the storage device 7 in FIG. 1. In another embodiment, the imaging diagnostic device 310 in FIG. 3 corresponds to multiple X-ray CT devices including the X-ray CT device 100 in FIG. 1. In a specific example, the data store 320 is located in a database of a server to which multiple X-ray CT scanners are connected via a network, and the data store 320, model trainer 330, modeler 340, output processor 350, feedback unit 360, selector 370, and annotation adding unit 380 are located in a central processing unit 3 that functions as a trained model production terminal. In another specific example, the data store 320, model trainer 330, modeler 340, output processor 350, feedback unit 360, selector 370, and annotation adding unit 380 are located in a server connected to the X-ray CT scanner 100 via a network. In yet another specific example, the data store 320, model trainer 330, modeler 340, output processor 350, feedback unit 360, selector 370, and annotation adding unit 380 are located in one or more central processing units 3 associated with multiple imaging diagnostic scanners connected via a network. The data store 320, model trainer 330, modeler 340, output processor 350, feedback unit 360, selector 370, and annotation unit 380 may be distributed across a network and processed by multiple processing devices. The image data to be analyzed is stored in the data store 320 (e.g., a database, data structure, hard drive, solid-state memory, flash memory, other computer memory, etc.).The data store 320 may also be configured from multiple storage devices distributed across a network.

[0024] The data store 320 stores one or more images acquired by the diagnostic imaging device 310 (such as the X-ray CT device 100 in FIG. 1 ), the acquisition parameters used when acquiring the images, the reconstruction algorithm and image reconstruction parameters used to reconstruct the images, and the examination purpose, in association with each other. The acquisition parameters include the tube voltage, tube current magnitude, slice width, contrast agent protocol, etc. used in dual-energy imaging. If the examination purpose is added by the annotation adding unit 380 (described later), the data store 320 does not store the examination purpose. Images are characterized by one or more of spatial resolution, temporal resolution, contrast resolution, noise, and artifacts. There are various types of artifacts observed in CT images, such as metal objects, Poisson noise, Gaussian noise, streaks, and scatter. These artifacts can also be classified into types such as system design-induced (Gaussian noise, etc.), X-ray tube-induced (defocus, X-ray tube vibration, etc.), detector-induced (non-uniformity of detector response, etc.), patient-induced (patient movement during scanning, metal objects, etc.), and operator-induced (insufficient exposure (insufficient dose), inappropriate slice width setting, etc.) In addition, poor image quality may also be caused by insufficient spatial resolution or contrast resolution.

[0025] The selector 370 receives multiple target images from the data store 320 and provides a function for selecting one or more specific cross sections. For example, when the X-ray CT device 1 scans the chest and lower abdomen, tens to thousands of slice images may be generated depending on the slice width. The selector 370 provides various functions for selecting one or more radiological images to which annotations are to be added from these images. The selector 370 displays an image of the selected organ or region on the monitor 6 in response to an operator's selection of an organ or region. The selected organ or region is automatically or manually enlarged as needed. The selector 370 may also have an image analysis function. The selector 370 applies a trained AI model to the target image to identify the presence of specific lesions, specific noise, or specific artifacts in the image. The selector 370 may also identify the presence of specific lesions, specific noise, or specific artifacts using template matching or the like instead of an AI model. Additionally, the selector 370 can select target images according to the type of acquisition parameters, such as images acquired with a monochromatic x-ray beam, images acquired with a polychromatic x-ray beam, images using contrast agents, images of the head (using head-specific acquisition parameters), etc. The training network 420 and trained model 460, described below, can be generated according to the type of acquisition parameters.

[0026] In a preferred embodiment of the present invention, the selector 370 can automatically extract images from the data store 320 to facilitate annotation by the operator. The operator can also use known image analysis functions such as AI and pattern matching provided in the selector 370 to extract images containing a specific organ, a specific lesion, or a specific type of artifact, and display them on the monitor 6.

[0027] In a preferred embodiment of the present invention, if the operator selects a large number of images that are not suitable as training data, the selector 370 outputs a display and / or a sound prompting the operator to further narrow down the images to be selected as training data.

[0028] When the selector 370 identifies an image to be used as training data in response to an operator's request, the operator can use the functions of the annotation adding unit 380. In Fig. 3, the annotation adding unit 380 is shown as a separate functional block from the selector 370, but the two can also be implemented as a single functional block.

[0029] Deep learning techniques have made it possible to detect objects in images and classify them. However, learning requires a large number of images and accompanying information (annotations). It is difficult to uniformly collect and annotate the images required for learning. The annotation addition unit 380 provides both automatic and manual annotation functions, reducing the annotation workload.

[0030] In a preferred embodiment of the present invention, the annotation adding unit 380 displays the image selected by the selector 370 on the monitor 6. The operator can display the image in various ways according to their preference, for example, by enlarging or reducing the image. By utilizing the functions of the annotation adding unit 380, a first medical image is annotated with at least one of a first algorithm and first conditions for processing data collected to obtain the first medical image, and a feature value of a first image quality index. Furthermore, a second medical image is annotated with at least one of a second algorithm and second conditions for processing data collected to obtain the second medical image, and a feature value of a second image quality index.

[0031] For example, a medical image containing intracranial bleeding is annotated with the features of spatial resolution, density resolution, and noise level, as well as the type of image reconstruction algorithm important for achieving these features, whether a specific image reconstruction algorithm is used, or the purpose of the examination, to become training data. Similarly, a medical image containing a skull fracture is annotated with the features of spatial resolution, density resolution, and noise level, as well as the type of image reconstruction algorithm important for achieving these features, whether a specific image reconstruction algorithm is used, or the purpose of the examination, to become training data.

[0032] The automatic and / or manual annotation can be performed in various ways. In a specific embodiment of the present invention, the operator identifies the location of an organ and / or lesion by surrounding the desired location with a geometric shape, such as a circle, ellipse, polygon including a rectangle, or a complex curve drawn by edge detection. The annotation adding unit 380 identifies the type of organ and / or lesion and displays a list of examination purposes corresponding to the identified organ and / or lesion on the monitor 6. For example, if a skull with a crack is identified by image recognition, examination purposes such as skull fracture and intracerebral hemorrhage are listed. If the examination purpose to be registered is displayed at the top of the list, the operator can register (annotate) the examination purpose by simply pressing the enter key or the confirm button. If the examination purpose to be registered is displayed in a position other than the top of the list, the operator can register (annotate) the examination purpose by selecting it and then pressing the enter key or the confirm button.

[0033] The annotation adding unit 380 also provides the operator with a function for manually adding annotations. The annotation adding unit 380 provides the operator with a text box for inputting the examination purpose and accepts input from the operator. Even if the examination purpose is stored in the data store 320, the operator can adopt or modify the stored examination purpose. Note that the examination purpose can include, in addition to the type of lesion such as bleeding or tumor, the examination site (skull, head brain, lungs, abdomen, liver, etc.), age (low radiation doses are required for young people), gender, race, and attending physician (doctor A and doctor B may have different preferences for image quality), etc.

[0034] Trainer 330 and modeler 340 receive the set of training data via selector 370 and annotation unit 380. If the data store 320 stores an examination purpose, trainer 330 and modeler 340 can receive the set of training data from data store 320 without going through selector 370 and annotation unit 380. Trainer 330 can identify the training network 420 to train based on the type of acquisition parameters, such as images acquired with a monochromatic X-ray beam, images acquired with a polychromatic X-ray beam, images using a contrast agent, or images of the head (using head-specific acquisition parameters).

[0035] In a preferred embodiment of the present application, the trainer 330 inputs the examination objective and images, and repeatedly trains the training network 420 to output a reconstruction algorithm and reconstruction parameters. During training, a first medical image prioritized by a first image quality index and a second medical image prioritized by a second image quality index different from the first image quality index are used as training data. For example, medical images containing intracranial bleeding, such as epidural hematoma, subdural hematoma, subarachnoid hemorrhage, and intracerebral hemorrhage, prioritize high density resolution and low noise level over spatial resolution, and such images are used as training data with a density resolution / noise level priority. Images containing skull fractures (e.g., fractures with linear cracks) prioritize spatial resolution over density resolution, and such images are also used as training data with a spatial resolution priority.

[0036] In another embodiment of the present application, the trainer 330 inputs the inspection objective and image quality index features and repeatedly trains the training network 420 to output a reconstruction algorithm and reconstruction parameters. When image quality index features extracted from images, rather than the images themselves, are input to the network, this can be achieved by providing an image analysis function, either internal or external to the trainer 330, that analyzes the images and outputs the image quality index features. When analyzing the images, the spatial resolution, contrast resolution, temporal resolution, noise, and artifacts of the image are analyzed, and the features of these items are analyzed. The features can be identified using noise values, noise power spectrum (NPS), modulated transfer function (MTF), etc.

[0037] The modeler 340 uses the deployed artificial intelligence model (a trained model, e.g., a neural network, a random forest, etc.) to determine, in response to an input indicating a preference for a first image quality index, a first model algorithm and at least one first model parameter for obtaining a first model medical image in which the first image quality index is preferred, and to determine, in response to an input indicating a preference for a second image quality index, a second model algorithm and at least one second model parameter for obtaining a second model medical image in which the second image quality index is preferred.

[0038] The modeler 340 can use the developed artificial intelligence model to receive the examination purpose and image quality index features input from the input device 2 and identify a reconstruction algorithm and / or reconstruction parameters to be used. Furthermore, in response to input of a sample medical image, the modeler 340 can identify first and / or second image quality index features possessed by the sample medical image and output a reconstruction algorithm and / or reconstruction parameters for outputting a medical image having model features of the first and / or second image quality index. Furthermore, in response to input of medical image data and a selection to prioritize either the first image quality index or the second image quality index, the modeler 340 processes the input medical image data and outputs a reconstruction algorithm and / or reconstruction parameters for outputting a medical image having model features of the prioritized first or second image quality index. In certain embodiments, the modeler 340, in response to input of the examination purpose, can identify model features of an image quality index that match the examination purpose and output a reconstruction algorithm and / or reconstruction parameters for outputting a medical image having these model features. In other words, the model features of the image quality index are indirectly selected by identifying the examination purpose. Identifying model features of an image quality index that match the examination purpose can be performed for each attending physician, each hospital, or each affiliated hospital group having multiple hospitals. In certain embodiments, identifying model features of an image quality index that match the examination purpose can be customized by retraining the trained model 460. Retraining for each attending physician can provide image quality that meets the attending physician's preferences, but the amount of training data may be insufficient. Conversely, retraining for each affiliated hospital group having multiple hospitals increases the likelihood of obtaining a sufficient amount of training data, but may decrease the likelihood of simultaneously satisfying the different preferences of individual radiologists. The deployed trained model 460 can also be shared among affiliated hospitals.If systems with different image reconstruction capabilities are introduced between affiliated hospitals (for example, in a certain image reconstruction algorithm, a higher-level model can set image reconstruction parameters in the range of 1-10, while a lower-level model can only set them in the range of 1-5), a trained model 460 for the higher-level model and a trained model 460 for the lower-level model may be provided. Furthermore, the trained model 460 can undergo continuous learning at each affiliated hospital. Data captured with a system of a different type from the system that provided the images to the trainer 330 can also be used for continuous learning. In this case, the trained model 460 can be customized. Conversely, if continuous learning is performed across affiliated hospitals, the amount of learning data can be increased.

[0039] Reconstruction algorithms include image reconstruction methods such as analytical image reconstruction, filtered back projection, iterative reconstruction, successive iterative reconstruction, model-based iterative reconstruction, and deep learning image reconstruction, as well as the algorithms used by each image reconstruction method. The algorithms used by each image reconstruction method include algorithms for removing various artifacts and noise removal algorithms. Reconstruction parameters include parameters used by the image reconstruction method, flags that specify whether the algorithms used by each image reconstruction method are on or off, and parameters used by the algorithms used by each image reconstruction method.

[0040] In a preferred embodiment of the present application, a simulated image that is expected to be obtained when a specified reconstruction algorithm and / or reconstruction parameters are used is displayed on the monitor 6 along with the reconstruction algorithm and the reconstruction parameters. The monitor 6 also displays numerical values ​​corresponding to characteristic quantities of image quality indexes (including spatial resolution, contrast resolution, noise, artifacts, etc.) of the simulated image. The operator and / or physician of the X-ray CT apparatus 1 determines the appropriateness of the reacquisition algorithm and reconstruction parameters generated by the modeler 340, and if appropriate, the image generator 35 performs image reconstruction using the reacquisition algorithm and reconstruction parameters generated by the modeler 340. The determination of appropriateness may be made before or after the acquisition of the projection data to be reconstructed.

[0041] If the simulation image is not valid, the operator can modify, via the input device 2, the numerical values ​​corresponding to the feature quantities of the image quality index (including spatial resolution, contrast resolution, noise, artifacts, etc.) of the simulation image displayed on the monitor 6. The modeler 340 then outputs the reconstruction algorithm and reconstruction parameters again based on the feature quantities of the modified image quality index. Furthermore, a new simulation image, reconstruction algorithm and / or reconstruction parameters, and the feature quantities of the image quality index corresponding to the modified numerical values ​​are displayed on the monitor 6. The reconstruction algorithm and / or reconstruction parameters generated by the modeler 340 are modified accordingly. If the validity is determined before the projection data to be reconstructed is acquired, the acquisition parameters can also be modified as necessary to obtain ideal image quality.

[0042] If the operator determines that an image reconstructed using a re-acquisition algorithm and reconstruction parameters generated by the modeler 340 that were determined to be appropriate is in fact inappropriate, the operator can modify, via the input device 2, a numerical value corresponding to a feature amount of an image quality index (including spatial resolution, contrast resolution, noise, artifacts, etc.) of the image displayed on the monitor 6. A new simulation image, reconstruction algorithm and / or reconstruction parameters, and image quality index feature amount corresponding to the modified numerical value are displayed on the monitor 6. The reconstruction algorithm and / or reconstruction parameters generated by the modeler 340 are modified accordingly. If the output of the trained model 460 is inappropriate, such as when an image reconstructed using a re-acquisition algorithm and reconstruction parameters generated by the modeler 340 that were determined to be appropriate is significantly different from the original simulation image, the result can be passed to the feedback unit 360 (described later), and the training network 420 can be retrained.

[0043] If feedback from the operator is provided to the trainer 330 by the feedback unit 360, the trainer 330 can use the feedback to improve the artificial intelligence model. For example, the trainer 330 can adjust (lower) the weights of the nodes and change the connections between nodes in the learning model network used to obtain the output that caused the feedback. Conversely, if the feedback is not provided and is accepted by the user, the weights of the nodes in the learning model network used for that output can be strengthened.

[0044] In a preferred embodiment of the present invention, the trained model 460 is transmitted to a central management device that manages multiple X-ray CT scanners and / or other imaging devices. In the X-ray CT scanner 1 and / or other imaging devices, automatic retraining and / or manual retraining of the trained model 460 is performed.

[0045] FIG. 4 illustrates exemplary implementations of trainer 330 and modeler 340. As shown in the example of FIG. 4, trainer 330 includes an input processor 410, a training network 420, an output validator 430, and a feedback processor 440. In this example, modeler 340 includes a preprocessor 450, a trained model 460, which is a deployed model, and a postprocessor 470. In the example of FIG. 4, inputs such as images, acquisition and / or reconstruction parameters, and annotations received directly from data store 320 or via selector 370 are provided to input processor 410, which selects a training network 420 from multiple training networks 420 corresponding to the input and prepares the data to be input to training network 420. For example, the data can be modified in ways such as filtering, supplementing, and / or sanitizing images to make the information input to network 420 more suitable for learning.

[0046] In the illustrated example, the training network 420 analyzes input data from the input processor 410 and generates an output, which is validated by the output validator 430. The output validator 430 can, for example, verify the accuracy of the output from the training network 420. It may determine that the output is incorrect if the reconstruction algorithm or reconstruction parameters are outside of the available range. If the output of the network 420 is inaccurate, the network 420 can be modified (e.g., by adjusting network weights, changing node connections, etc.) to update the network 420 and generate the correct output. Once the output of the training network 420 has been validated by the output validator 430, the training network 420 can be used to generate and deploy a trained model 460 for the modeler 340.

[0047] Feedback can be periodically input to a feedback processor 440, which processes the feedback and evaluates whether to trigger an update or regeneration of the network 420. For example, if the output of the deployed trained model 460 remains accurate, there may be no reason to update. However, for example, if the output of the trained model 460 becomes inaccurate, this can trigger a regeneration or other update of the network 420 and deploy an updated trained model 460 (e.g., based on additional data, new constraints, updated configuration, etc.).

[0048] The developed trained model 460 is used by the modeler 340 to process input image data and identify reconstruction algorithms and reconstruction parameters. Input data, such as from the data store 320, is prepared by the preprocessor 450, which then supplies the input data to a trained model 460 (e.g., a deep learning network model, a machine learning model, another network model, etc.) selected by the preprocessor 450 based on the input image data. The preprocessor 450 can adjust the input image data, such as thinning (e.g., removing even-numbered slices), contrast, brightness, levels, artifacts / noise, etc., before providing the image data to the trained model 460. In certain embodiments, a medical image is input to the trained model 460. In other embodiments, image quality index features extracted from the medical image are input to the trained model 460, rather than the medical image itself. When image quality index features are input to the trained model 460, the preprocessor 450 analyzes the input medical image, identifies features for each analyzed image quality index, and passes the identified features to the trained model 460.

[0049] Trained model 460 processes data from pre-processor 450 and outputs a reconstruction algorithm and reconstruction parameters. Output from trained model 460 is post-processed by post-processor 470, which can clean up, organize, and / or otherwise modify the output data to form a composite 2D image. In certain examples, post-processor 470 can generate simulated images corresponding to the output reconstruction algorithm and reconstruction parameters, and can validate and / or otherwise perform quality checks on the output simulated images, reconstruction algorithm, and / or reconstruction parameters before storing, displaying, transmitting to another system, etc.

[0050] 5 shows a typical deep learning neural network 500. Typical neural network 500 includes layers 520, 540, 560, and 580. Layers 520 and 540 are connected by neural connection 530. Layers 540 and 560 are connected by neural connection 550. Layers 560 and 580 are connected by neural connection 570. Data flows forward from input layer 520 via inputs 512, 514, and 516 to output layer 580 and output 590.

[0051] Layer 520, in the example of Figure 5, is an input layer that includes multiple nodes 522, 524, and 526. Layers 540 and 560 are hidden layers that, in the example of Figure 5, include nodes 542, 544, 546, 548, 562, 564, 566, and 568. Neural network 500 may include more or fewer hidden layers 540 and 560 than are shown. Layer 580 is an output layer that, in the example of Figure 5, includes node 582 having output 590. Each input 512-516 corresponds to a node 522-526 in input layer 520, and each node 522-526 in input layer 520 has a connection 530 to a respective node 542-548 in hidden layer 540. Each node 542-548 in hidden layer 540 has a connection 550 to a respective node 562-568 in hidden layer 560. Each node 562-568 in hidden layer 560 has a connection 570 to output layer 580. Output layer 580 has an output 590 that provides the output from the exemplary neural network 500.

[0052] Of the connections 530, 550, and 570, certain exemplary connections 532, 552, and 572 may be given additional weighting, while other exemplary connections 534, 554, and 574 may be given less weighting in the neural network 500. Input nodes 522-526 are activated, for example, by receiving input data via inputs 512-516. Nodes 542-548 and 562-568 in the hidden layers 540 and 560 are activated by the forward flow of data through the network 500 via connections 530 and 550, respectively. Node 582 in the output layer 580 is activated after data processed in the hidden layers 540 and 560 is sent via connection 570. When output node 582 in the output layer 580 is activated, node 582 outputs an appropriate value based on the processing accomplished in the hidden layers 540 and 560 of the neural network 500. Images, noise and artifacts that appear in specific locations within the image, and their location information can be used to create models for object detection, shape detection (segmentation), and classification. In addition, models can be created based on the spatial resolution, contrast resolution, and temporal resolution of the image, using frequency analysis, etc.

[0053] The above description focuses on the most preferred embodiment of the present invention, but as will be apparent to those skilled in the art, the present invention can be implemented by making various changes and modifications to the embodiment within the technical scope of the present invention.

[0054] Furthermore, a program for causing a computer to function as each of the means for controlling and processing the X-ray CT apparatus is also an example of an embodiment of the invention. [Explanation of symbols]

[0055] 2 Input devices 3. Central Processing Unit 5 Data Collection Buffer 6 monitors 7 Storage device 10 Photographic Table 12 Cradle 15 Gantry rotating part 20 Scanning Gantry 20a opening 21 X-ray tube 21h Housing 21s Cathode Sleeve 21t target electrode 22 X-ray control unit 23 Collimator 24 X-ray detector 24a X-ray detector 24s X-ray detection surface 25 DAS 26 Rotation control section 27 Collimator control section 29 Gantry control unit 32 Scan control section 34 Pretreatment section 35 Image generation unit 71 Subject / Imaging Subject 81 X-ray beam 100 X-ray CT device 300 Trained Model Production System 310 Diagnostic Imaging Devices 320 Data Store 330 Model Trainer 340 Modeler 350 Output Processor 360 Feedback Unit 370 Selector 380 Annotation Addition Section 410 Input Processor 420 Training Network 430 Output Validator 440 Feedback Processor 450 Preprocessor 460 trained models 470 Post Processor

Claims

1. A medical imaging system including a storage medium storing a trained model trained using a first medical image prioritized by a first image quality index and a second medical image prioritized by a second image quality index different from the first image quality index as training data, The learning data is the annotations include at least one of a first algorithm and a first condition for processing data collected to obtain the first medical image, a feature of the first image quality index, at least one of a second algorithm and a second condition for processing data collected to obtain the second medical image, and a feature of the second image quality index; The trained model is determining at least one of a first model algorithm and first model parameters for obtaining a first model medical image with the first image quality index prioritized; A medical imaging system configured to determine at least one of a second model algorithm and second model parameters for obtaining a second model medical image in which the second image quality index is prioritized.

2. The trained model is In response to direct or indirect input of medical image data and a selection of either the first image quality index or the second image quality index, processing the input medical image data and outputting a reconstruction algorithm and / or reconstruction parameters for outputting a medical image having model features of the first or second image quality index; The medical image capturing system according to claim 1 , wherein the first image quality index and / or the second image quality index are indirectly selected by selecting an examination purpose.

3. the first image quality index is one or more of spatial resolution, contrast resolution, temporal resolution, noise, and artifacts; the second image quality index is one or more of spatial resolution, contrast resolution, temporal resolution, noise, and artifacts; 2. The medical imaging system according to claim 1, wherein the feature amount of the first and / or second image quality index includes any one of a noise value, a noise power spectrum (NPS), and a modulation transfer function (MTF).

4. 4. The medical image capturing system according to claim 3, wherein the first and second medical images are reconstructed based on projection signals collected by a radiation imaging device.

5. 5. The medical image capturing system according to claim 4, wherein the radiation imaging device is any one of a CT device, a PET device, a SPECT device, and a tomosynthesis device.

6. the projection signals are signals collected from a subject, which may be a human or a non-human animal; 6. The medical imaging system according to claim 5, wherein the first and / or second conditions include information on a purpose of examination of the subject, a lesion present or suspected to be present in the subject, and / or a specific part of the subject.

7. 7. The medical imaging system of claim 6, wherein the trained models include one or more trained models associated with one or more of the acquisition parameters used when the projection signals were acquired, the examination purpose, the lesion, and / or the site.

8. the learned model is configured to select a reconstruction algorithm to which the reconstruction parameters are applied from among a plurality of reconstruction algorithms based on the first and / or second conditions; the second model parameters include a flag indicating non-use of the first model algorithm; 8. The medical imaging system according to claim 1, wherein the selected reconstruction algorithm is one or more of an analytical image reconstruction method, a filtered back projection method, an iterative reconstruction method, a successive approximation reconstruction method, a model-based iterative reconstruction method, a deep learning image reconstruction method, and an artifact removal algorithm.

9. a user interface including an input device for accepting operator input and a display device for displaying the reconstructed image; the input device is configured to accept input of the first image quality index and / or the second image quality index; the display device is configured to display a numerical value corresponding to a feature of an image quality index of the reconstructed image currently displayed on the display device; an operator can modify the numerical value via the input device; The medical imaging system according to claim 8 , wherein the display device is further configured to display a reconstructed image having the characteristic amount of the image quality index corresponding to the modified numerical value.

10. A method for producing a trained model trained using a first medical image prioritized by a first image quality index and a second medical image prioritized by a second image quality index different from the first image quality index as training data, the method comprising: The learning data is generating training data including, as annotations, at least one of a first algorithm and a first condition for processing data collected to obtain the first medical image, a feature of the first image quality index, at least one of a second algorithm and a second condition for processing data collected to obtain the second medical image, and a feature of the second image quality index; The trained model is determining at least one of a first model algorithm and first model parameters for obtaining a first model medical image with the first image quality index prioritized; The method is configured to determine at least one of a second model algorithm and second model parameters for obtaining a second model medical image in which the second image quality index is prioritized.

11. The trained model is In response to direct or indirect input of medical image data and a selection of either the first image quality index or the second image quality index, processing the input medical image data and outputting a reconstruction algorithm and / or reconstruction parameters for outputting a medical image having model features of the first or second image quality index; The method of claim 1 , wherein the first image quality index and / or the second image quality index are selected indirectly by selecting an inspection purpose.

12. the first image quality index is one or more of spatial resolution, contrast resolution, temporal resolution, noise, and artifacts; the second image quality index is one or more of spatial resolution, contrast resolution, temporal resolution, noise, and artifacts; The method of claim 10 , wherein the first and / or second image quality index features include any one of a noise value, a noise power spectrum (NPS), and a modulation transfer function (MTF).

13. The method according to claim 12 , wherein the first and second medical images are reconstructed based on projection signals collected by a radiation imaging device.

14. The method according to claim 13 , wherein the radiation imaging device is one of a CT device, a PET device, a SPECT device, and a tomosynthesis device.

15. the projection signals are signals collected from a subject, which may be a human or a non-human animal; The method according to claim 14 , wherein the first and / or second conditions include information on the purpose of examination of the subject, a lesion present or suspected to be present in the subject, and / or a specific site of the subject.

16. 16. The method of claim 15, wherein the trained models include one or more trained models associated with one or more of acquisition parameters used when the projection signals were acquired, the examination purpose, the lesion, and / or the region.

17. the learned model is configured to select a reconstruction algorithm to which the reconstruction parameters are applied from among a plurality of reconstruction algorithms based on the first and / or second conditions; the second model parameters include a flag indicating non-use of the first model algorithm; The method of claim 10, wherein the selected reconstruction algorithm is one or more of an analytical image reconstruction method, a filtered backprojection method, an iterative reconstruction method, a successive approximation reconstruction method, a model-based iterative reconstruction method, a deep learning image reconstruction method, and an artifact removal algorithm.

18. receiving an input of the first image quality index and / or the second image quality index from an input device; displaying a numerical value corresponding to the feature of the image quality index of the currently displayed reconstructed image on a display device; Including, the numerical value is modifiable via the input device; The method of claim 17 , wherein the display device is further configured to display the reconstructed image having the image quality index feature corresponding to the modified numerical value.

19. A non-transitory machine-readable storage medium containing executable instructions, The executable instructions, when executed by a processor, A non-transitory machine-readable storage medium on which the method of any of claims 10 to 18 is performed.

Images