Medical information processing method, medical information processing apparatus, and program

Abstract

To provide a medical information processing method, a medical information processing apparatus, and a program capable of improving the accuracy of noise reduction in spectral imaging.SOLUTION: The medical information processing method according to the embodiment includes acquiring X-ray projection data, generating a medical image from the projection data by reconstruction processing, separating the medical image into a signal image and a noise image by whitening conversion, de-noising at least one of the signal image and the noise image using a first learned model, generating a de-noised medical image by combining both de-noised images by inverse whitening conversion, and generating the de-noised medical image by combining one de-noised image and the other non-de-noised image.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The embodiments disclosed in this specification and the drawings relate to a medical information processing method, a medical information processing device, and a program. [Background technology]

[0002] In recent years, noise reduction technology using deep learning has been actively proposed for X-ray CT (Computed Tomography) equipment, which has resulted in significant contributions to reducing radiation exposure. Furthermore, radiation exposure reduction must also be constantly considered in spectral imaging such as dual energy (DE) and photon counting CT (PCCT), making noise reduction a necessary technology.

[0003] Recently, noise reduction methods using deep learning for spectral imaging have been proposed. However, conventional techniques still have room for improvement in the accuracy of noise reduction. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-184428 Summary of the Invention [Problem to be solved by the invention]

[0005] The problem to be solved by the embodiments disclosed in this specification and the drawings is to improve the accuracy of noise reduction in spectral imaging. However, the problem to be solved by the embodiments disclosed in this specification and the drawings is not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems. [Means for solving the problem]

[0006] a first input image corresponding to the first output image having less noise than the first input image; a first trained model that is a machine learning model trained based on a training dataset in which a first output image having less noise than the first input image is associated with a first input image as a target; when both the signal image and the noise image are denoised, an inverse whitening transformation is used to combine the denoised signal image and the denoised noise image to generate a denoised medical image; and when one of the signal image and the noise image is denoised, an inverse whitening transformation is used to combine one of the denoised signal image and the denoised noise image with the other un-denoised signal image and the noise image to generate the denoised medical image. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a medical information processing system 1 according to a first embodiment. [Figure 2] 1 is a diagram showing an example of the arrangement of an X-ray CT apparatus 100 according to a first embodiment. [Figure 3] 4 is a flowchart showing an example of a series of processing steps of the X-ray CT apparatus 100 in the first embodiment. [Figure 4] 2 is a diagram schematically showing a series of processing steps of the X-ray CT apparatus 100 according to the first embodiment. [Figure 5] FIG. 2 is a diagram illustrating an example of the configuration of a learning device 200 according to the first embodiment. [Figure 6] 3 is a flowchart showing the flow of a series of processes of the learning device 200 according to the first embodiment. [Figure 7]FIG. 2 is a diagram schematically illustrating the flow of a series of processes of the learning device 200 according to the first embodiment. [Figure 8] FIG. 10 is a diagram showing an example of the arrangement of a DAS 116 of an X-ray CT apparatus 100 according to a second embodiment. [Figure 9] FIG. 10 is a diagram showing an example of the functional configuration of a reconstruction function 153 according to the second embodiment. [Figure 10] FIG. 11 is a diagram schematically showing a flow of a series of processes in the X-ray CT apparatus 100 according to the third embodiment. [Figure 11] FIG. 11 is a diagram schematically illustrating the flow of a series of processes of the learning device 200 according to the third embodiment. [Figure 12] FIG. 10 is a diagram schematically showing a series of processing steps of the X-ray CT apparatus 100 according to the fourth embodiment. [Figure 13] FIG. 10 is a diagram schematically illustrating the flow of a series of processes of a learning device 200 according to a fourth embodiment. [Figure 14] 13 is a flowchart showing an example of a series of processing steps of the X-ray CT apparatus 100 according to the fifth embodiment. [Figure 15] FIG. 12 is a diagram schematically showing a series of processing steps of the X-ray CT apparatus 100 according to the fifth embodiment. [Figure 16] 10 is a flowchart showing the flow of a series of processes of the learning device 200 according to the fifth embodiment. [Figure 17] FIG. 12 is a diagram schematically showing the flow of a series of processes of the learning device 200 according to the fifth embodiment. [Figure 18] FIG. 13 is a diagram schematically showing the flow of a series of processes of the learning device 200 according to the sixth embodiment. [Figure 19] FIG. 13 is a diagram schematically showing the flow of a series of processes of the learning device 200 according to the seventh embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, a medical information processing method, a medical information processing apparatus, and a program according to an embodiment will be described with reference to the drawings.

[0009] (First embodiment) [Configuration of medical information processing system] 1 is a diagram illustrating an example of the configuration of a medical information processing system 1 according to the first embodiment. The medical information processing system 1 includes, for example, a plurality of medical image diagnostic devices 100 and a learning device 200. The medical image diagnostic devices 100 and the learning device 200 are communicably connected via a communication network NW.

[0010] The communication network NW may refer to any information and communication network that uses telecommunications technology. For example, the communication network NW may include wireless / wired LANs such as hospital backbone LANs (Local Area Networks), the Internet, telephone communication networks, optical fiber communication networks, cable communication networks, satellite communication networks, etc.

[0011] The medical image diagnostic apparatus 100 is, for example, an X-ray CT (Computed Tomography) apparatus, and is typically installed in a medical institution, a research facility, or the like.

[0012] The learning device 200 receives information from each X-ray CT apparatus (medical image diagnostic apparatus) 100 via the communication network NW, and uses the received information to learn a machine learning model MDL for removing noise from medical images. Then, the learning device 200 transmits the learned machine learning model MDL to each X-ray CT apparatus (medical image diagnostic apparatus) 100 via the communication network NW.

[0013] The learning device 200 may be a single device, or may be a system in which multiple devices connected via a communication network NW operate in cooperation with each other. That is, the learning device 200 may be realized by multiple computers (processors) included in a distributed computing system or a cloud computing system. Furthermore, the learning device 200 does not necessarily have to be a separate device from the X-ray CT device (medical image diagnostic device) 100, but may be a device integrated with the X-ray CT device (medical image diagnostic device) 100.

[0014] [X-ray CT system configuration] 2 is a diagram showing an example of the configuration of an X-ray CT device 100 in the first embodiment. The X-ray CT device 100 is a device that generates a medical image (hereinafter referred to as a CT image) of a subject P by scanning the subject P with X-rays, and diagnoses the subject P based on the CT image. The subject P is typically a human, but is not limited to this, and may be other animals such as a dog or a cat, or may be a plant. In the following description, the subject P is assumed to be a human as an example.

[0015] The X-ray CT apparatus 100 may be, for example, a CT apparatus using DE (Dual Energy) or PCCT (Photon Counting CT). In the first embodiment, the X-ray CT apparatus 100 will be described as a CT apparatus using DE (Dual Energy).

[0016] As shown in the figure, the X-ray CT apparatus 100 includes, for example, a gantry 110, a bed 130, and a console 140. In the illustrated example, for convenience of explanation, both a view of the gantry 110 from the Z-axis direction and a view of the gantry 110 from the X-axis direction are shown, but in reality, there is only one gantry 110. In this embodiment, the rotation axis of the rotating frame 117 in a non-tilted state or the longitudinal direction of the tabletop 133 of the bed 130 is defined as the Z-axis direction, an axis perpendicular to the Z-axis direction and horizontal to the floor surface is defined as the X-axis direction, and a direction perpendicular to the Z-axis direction and perpendicular to the floor surface is defined as the Y-axis direction.

[0017] The gantry device 110 includes, for example, an X-ray tube 111, a wedge 112, a collimator 113, an X-ray high voltage device 114, an X-ray detector 115, a data acquisition system (hereinafter referred to as DAS: Data Acquisition System) 116, a rotating frame 117, and a control device 118.

[0018] The X-ray tube 111 generates X-rays by irradiating thermoelectrons from a cathode (filament) to an anode (target) when a high voltage is applied from the X-ray high voltage device 114. The X-ray tube 111 includes a vacuum tube. For example, the X-ray tube 111 is a rotating anode type X-ray tube that generates X-rays by irradiating a rotating anode with thermoelectrons.

[0019] The wedge 112 is a filter for adjusting the amount of X-rays irradiated from the X-ray tube 111 to the subject P. The wedge 112 attenuates the X-rays passing through it so that the distribution of the amount of X-rays irradiated from the X-ray tube 111 to the subject P becomes a predetermined distribution. The wedge 112 is also called a wedge filter or a bow-tie filter. The wedge 112 is made by processing aluminum so as to have a predetermined target angle and a predetermined thickness, for example.

[0020] The collimator 113 is a mechanism for narrowing down the irradiation range of the X-rays that have passed through the wedge 112. The collimator 113 narrows down the irradiation range of the X-rays, for example, by forming a slit by combining multiple lead plates. The collimator 113 is also called an X-ray aperture.

[0021] The X-ray high voltage device 114 includes, for example, a high voltage generator and an X-ray control device. The high voltage generator has an electric circuit including a transformer, a rectifier, etc., and generates a high voltage to be applied to the X-ray tube 111. The X-ray control device controls the output voltage of the high voltage generator according to the X-ray dose to be generated by the X-ray tube 111. The high voltage generator may be one that performs voltage boosting using the above-mentioned transformer, or one that performs voltage boosting using an inverter. The X-ray high voltage device 114 may be provided on the rotating frame 117, or may be provided on a fixed frame (not shown) of the gantry device 110.

[0022] The X-ray detector 115 detects the intensity of X-rays generated by the X-ray tube 111 and incident upon the subject P. The X-ray detector 115 outputs an electrical signal (which may be an optical signal, etc.) corresponding to the intensity of the detected X-rays to the DAS 116. The X-ray detector 115 has, for example, multiple X-ray detection element rows. Each of the multiple X-ray detection element rows has multiple X-ray detection elements arranged in the channel direction along an arc centered on the focal point of the X-ray tube 111. The multiple X-ray detection element rows are arranged in the slice direction (column direction, row direction).

[0023] The X-ray detector 115 is an indirect detector having, for example, a grid, a scintillator array, and a photosensor array. The scintillator array has multiple scintillators. Each scintillator has scintillator crystals. The scintillator crystals emit light with an amount of light corresponding to the intensity of the incident X-rays. The grid is arranged on the surface of the scintillator array on which the X-rays are incident and has an X-ray shielding plate that has the function of absorbing scattered X-rays. The grid is sometimes called a collimator (one-dimensional collimator or two-dimensional collimator). The photosensor array has, for example, a photosensor such as a photomultiplier tube (PMT). The photosensor array outputs an electrical signal corresponding to the amount of light emitted by the scintillator. The X-ray detector 115 may also be a direct conversion detector having a semiconductor element that converts incident X-rays into an electrical signal.

[0024] The DAS 116 includes, for example, an amplifier, an integrator, and an A / D converter. The amplifier amplifies the electrical signal output by each X-ray detection element of the X-ray detector 115. The integrator integrates the amplified electrical signal over a view period (described below). The A / D converter converts the electrical signal indicating the integration result into a digital signal. The DAS 116 outputs detection data based on the digital signal to the console device 140. The detection data is a digital value of X-ray intensity identified by the channel number and column number of the X-ray detection element that generated the data, and a view number indicating the acquired view. The view number is a number that changes according to the rotation of the rotating frame 117, and is, for example, a number that is incremented according to the rotation of the rotating frame 117. Therefore, the view number is information that indicates the rotation angle of the X-ray tube 111. The view period is the period from the rotation angle corresponding to a certain view number to the rotation angle corresponding to the next view number. The DAS 116 may detect the view switching by a timing signal input from the control device 118, by an internal timer, or by a signal acquired from a sensor (not shown). When a full scan is performed and X-rays are continuously emitted by the X-ray tube 111, the DAS 116 collects a group of detection data for the entire circumference (360 degrees). When a half scan is performed and X-rays are continuously emitted by the X-ray tube 111, the DAS 116 collects detection data for half the circumference (180 degrees).

[0025] The rotating frame 117 is an annular rotating member that rotates the X-ray tube 111, wedge 112, collimator 113, and X-ray detector 115 while holding them facing each other. The rotating frame 117 is supported by a fixed frame so as to be rotatable around the subject P introduced inside. The rotating frame 117 also supports the DAS 116. Detection data output by the DAS 116 is transmitted by optical communication from a transmitter having a light-emitting diode (LED) provided on the rotating frame 117 to a receiver having a photodiode provided on a non-rotating portion of the gantry device 110 (e.g., the fixed frame), and then transferred to the console device 140 by the receiver. Note that the method of transmitting the detection data from the rotating frame 117 to the non-rotating portion is not limited to the above-mentioned method using optical communication, and any non-contact transmission method may be adopted. The rotating frame 117 is not limited to an annular member, and may be an arm-like member as long as it can support and rotate the X-ray tube 111 and the like.

[0026] The control device 118 includes, for example, a processing circuit having a processor such as a CPU (Central Processing Unit), and a drive mechanism including a motor, an actuator, etc. The control device 118 receives input signals from an input interface 143 attached to the console device 140 or the gantry device 110, and controls the operations of the gantry device 110 and the bed device 130.

[0027] The control device 118, for example, rotates the rotating frame 117, tilts the gantry 110, or moves the top board 133 of the bed device 130. When tilting the gantry 110, the control device 118 rotates the rotating frame 117 around an axis parallel to the Z-axis direction based on the inclination angle (tilt angle) input to the input interface 143. The control device 118 acquires the rotation angle of the rotating frame 117 from the output of a sensor (not shown), etc. The control device 118 also outputs the rotation angle of the rotating frame 117 to the processing circuit 150 as needed. The control device 118 may be provided in the gantry 110 or in the console device 140.

[0028] The control device 118 causes the gantry device 110 to perform a main scan and a scanogram, which is a positioning image taken before the main scan is performed.

[0029] The bed device 130 is a device on which the subject P to be scanned is placed and introduced into the rotating frame 117 of the gantry device 110. The bed device 130 has, for example, a base 131, a bed driving device 132, a top plate 133, and a support frame 134. The base 131 includes a housing that supports the support frame 134 so that the support frame 134 can move in the vertical direction (Y-axis direction). The bed driving device 132 includes a motor and an actuator. The bed driving device 132 moves the top plate 133, on which the subject P is placed, along the support frame 134 in the longitudinal direction of the top plate 133 (Z-axis direction). The top plate 133 is a plate-shaped member on which the subject P is placed.

[0030] The console device 140 includes, for example, a memory 141 (storage circuit), a display 142, an input interface 143, a communication interface 144, a speaker 145, and a processing circuit 150. In this embodiment, the console device 140 is described as being separate from the gantry device 110, but the gantry device 110 may include some or all of the components of the console device 140.

[0031] The memory 141 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory), a flash memory, a hard disk, an optical disk, etc. The memory 141 may also include a storage medium such as a ROM (Read Only Memory) or a register.

[0032] The memory 141 stores, for example, detection data, projection data, reconstructed images, etc. These data may be stored in an external memory (for example, a NAS (Network Attached Storage)) with which the X-ray CT apparatus 100 can communicate, instead of (or in addition to) the memory 141.

[0033] The memory 141 also stores model definition data. The model definition data is data such as a program or algorithm that defines several trained models MDLα, MDLβ, MDLκ, and MDLλ, which will be described later. The trained models MDLα and MDLβ are examples of a "first trained model," and the trained models MDLκ and MDLλ are examples of a "second trained model."

[0034] The display 142 displays various types of information. For example, the display 142 displays a CT image generated by the processing circuitry 150, a GUI (Graphical User Interface) that accepts various operations by an operator, and the like. The operator is, for example, a medical professional such as a doctor, technician, or nurse. The display 142 is, for example, a liquid crystal display, a CRT, an organic EL (Electro Luminescence) display, or the like.

[0035] The input interface 143 receives various input operations from the operator and outputs an electrical signal indicating the content of the received input operation to the processing circuitry 150 .

[0036] For example, the input interface 143 is realized by a pointing device (such as a mouse, touch panel, trackball, joystick, pen tablet, or stylus), a keyboard, a switch, a button, a foot pedal, a camera, an infrared sensor, or a microphone. In this specification, the input interface 143 is not limited to an interface having physical operating parts such as a mouse or a keyboard. For example, an example of the input interface 143 also includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs this electrical signal to a control circuit.

[0037] The communication interface 144 includes, for example, a network interface card (NIC), a wireless communication module, etc. The communication interface 144 communicates with external devices such as the learning device 200 via the communication network NW.

[0038] The speaker 145 outputs sound based on the information output by the processing circuit 150 .

[0039] The processing circuitry 150 controls the overall operation of the X-ray CT apparatus 100. The processing circuitry 150 executes, for example, a system control function 151, a preprocessing function 152, a reconstruction function 153, an image processing function 154, a conversion function 155, a denoising function 156, and an output control function 157. The processing circuitry 150 realizes these functions by, for example, causing a hardware processor to execute a program stored in the memory 141.

[0040] The DAS 116 and the preprocessing function 152 are an example of an "acquisition unit." The reconstruction function 153 is an example of a "reconstruction unit." The conversion function 155 is an example of a "conversion unit." The denoising function 156 is an example of a "denoising processing unit."

[0041] A hardware processor refers to a circuit such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an Application Specific Integrated Circuit (ASIC), a programmable logic device (e.g., a Simple Programmable Logic Device (SPLD) or a Complex Programmable Logic Device (CPLD), or a Field Programmable Gate Array (FPGA)).

[0042] The program may be directly embedded in the circuit of the hardware processor instead of storing the program in memory 141. In this case, the hardware processor realizes its function by reading and executing the program embedded in the circuit.

[0043] The above program may be stored in memory 141 in advance, or may be stored in a non-transitory storage medium such as a DVD or CD-ROM, and installed into memory 141 from the non-transitory storage medium by inserting the non-transitory storage medium into a drive device (not shown) of console device 140.

[0044] The hardware processor is not limited to being configured as a single circuit, but may be configured as a single hardware processor by combining multiple independent circuits to realize each function, or multiple components may be integrated into a single hardware processor to realize each function.

[0045] Each component of the console device 140 or the processing circuitry 150 may be distributed and realized by multiple pieces of hardware. The processing circuitry 150 may not be a component of the console device 140, but may be realized by an external device (e.g., the learning device 200) that can communicate with the console device 140. The external device may be, for example, a workstation connected to one X-ray CT device 100, or may be a device (e.g., a cloud server) connected to multiple X-ray CT devices 100 that collectively executes processing equivalent to that of the processing circuitry 150.

[0046] The system control function 151 controls various functions of the processing circuit 150 based on operations input to the input interface 143. The system control function 151 also acquires information from external devices such as the learning device 200 via the communication interface 144.

[0047] The preprocessing function 152 performs preprocessing on the detection data output by the DAS 116 to generate projection data, and stores the generated projection data in the memory 141. The preprocessing includes various processes such as logarithmic conversion, offset correction, inter-channel sensitivity correction, and beam hardening correction.

[0048] The reconstruction function 153 performs reconstruction processing on the projection data generated by the preprocessing function 152 using a filtered back projection method, an iterative reconstruction method, or the like to generate a CT image (also called a reconstructed image), and stores the generated CT image (reconstructed image) in the memory 141.

[0049] The image processing function 154 converts the CT image (including a denoised CT image, which will be described later) into a three-dimensional image or cross-sectional image data of an arbitrary cross section by a known method, based on an operation input to the input interface 143. The conversion into a three-dimensional image may be performed by the pre-processing function 152.

[0050] The transformation function 155 separates the reconstructed CT image into a signal image and a noise image by a whitening transform. The whitening transform is a linear transformation, and is a method for eliminating the correlation between signal components and noise components by using covariance analysis, eigenvalue decomposition, principal component analysis, independent principal component analysis, luma-chromatography, etc. A signal image is an image in which, of the signal components and noise components contained in the CT image, the signal components are significantly larger than the noise components. A noise image is an image in which, of the signal components and noise components contained in the CT image, the noise components are significantly larger than the signal components.

[0051] Furthermore, the transform function 155 generates a denoised CT image by combining the denoised signal image and the noise image, which will be described later, using an inverse whitening transform.

[0052] The denoising function 156 denoises (removes noise from) at least one of the signal image and the noise image separated from the CT image by the conversion function 155. For example, the denoising function 156 may denoise noise from each image using a pre-trained machine learning model MDL (hereinafter also referred to as a trained model MDL). MDL is simply a symbol meaning a model. Details of the denoising process using the trained model MDL will be described later.

[0053] The output control function 157 controls the display 142, the input interface 143, the communication interface 144, and the speaker 145 to output various types of information.

[0054] For example, the output control function 157 may cause the display 142 to display a denoised CT image, or cause the display 142 to display a CT image before denoising. Furthermore, the output control function 157 may cause the display 142 to display a GUI (Graphical User Interface) that accepts various operations from an operator such as a doctor or a technician.

[0055] Furthermore, for example, the output control function 157 may transmit the denoised CT image or the CT image before denoising to an external device via the communication interface 144.

[0056] [X-ray CT system processing flow and inference (runtime)] A series of processing flows by the X-ray CT apparatus 100 will be described below with reference to Fig. 3 and Fig. 4. Fig. 3 is a flowchart showing an example of a series of processing flows by the X-ray CT apparatus 100 in the first embodiment. Fig. 4 is a diagram schematically showing a series of processing flows by the X-ray CT apparatus 100 in the first embodiment. The processing of this flowchart is executed after learning of the machine learning model MDL is completed.

[0057] First, the preprocessing function 152 performs preprocessing on the detection data output by the DAS 116 to obtain projection data (step S100).

[0058] The X-ray CT device 100 according to the first embodiment is a DECT device, and images the subject P using two types of X-rays with different tube voltages, which are voltages of the X-ray tube 111. Therefore, the projection data includes projection data with a high tube voltage (kVp equal to or higher than a threshold) and projection data with a low tube voltage (kVp less than a threshold). The threshold is, for example, about 120 kVp.

[0059] Next, the reconstruction function 153 performs material decomposition based on the acquired projection data at high tube voltage (high kVp) and projection data at low tube voltage (low kVp) (step S102).

[0060] For example, the reconstruction function 153 distinguishes between two different types of reference materials (also called basis materials) a and b from two types of projection data using the attenuation coefficients of the two different types of reference materials a and b, such as iodine and water. Note that the number of reference materials is not limited to two, and may be one, or three or more.

[0061] Next, the reconstruction function 153 performs reconstruction processing on the material-decomposed projection data to generate a CT image (also called a reference material image) in which each reference material is enhanced or suppressed (step S104). The reference material image is also called a spectral image.

[0062] For example, the reconstruction function 153 performs reconstruction processing on projection data in which reference material a has been discriminated, to generate a CT image in which reference material a (e.g., iodine) has been emphasized. Similarly, the reconstruction function 153 performs reconstruction processing on projection data in which reference material b has been discriminated, to generate a CT image in which reference material b (e.g., water) has been emphasized. Hereinafter, the CT image in which reference material a has been emphasized will be referred to as "reference material image a," and the CT image in which reference material b has been emphasized will be referred to as "reference material image b." The projection data in which reference material a has been discriminated and the projection data in which reference material b has been discriminated are examples of "reference material projection data."

[0063] Next, the transformation function 155 separates the reference material image into a signal image s and a noise image n by whitening transformation (step S106).

[0064] For example, when reference material image a and reference material image b are generated, the conversion function 155 separates reference material image a into signal image s and noise image n, while separating reference material image b into signal image s and noise image n, using a whitening conversion.

[0065] Next, the denoising function 156 denoises each of the signal image s and the noise image n using the learned model MDL (step S108).

[0066] As described above, the signal image s separated by the whitening transformation is an image in which the signal components are significantly larger than the noise components, and the noise image n separated by the whitening transformation is an image in which the noise components are significantly larger than the signal components. In other words, the whitening transformation cannot completely separate the signal and noise, and the signal image s is contaminated with noise components, and the noise image n is contaminated with signal components.

[0067] Therefore, the denoising function 156 uses the learned model MDL to remove noise components from the signal image s and signal components from the noise image n.

[0068] For example, the denoising processing function 156 reads out the first trained model MDLα and the second trained model MDLβ defined by the model definition data stored in the memory 141, inputs a signal image s to the first trained model MDLα, and inputs a noise image n to the second trained model MDLβ.

[0069] The first trained model MDLα and the second trained model MDLβ may be implemented by a deep neural network such as a convolutional neural network (CNN). Instead of a neural network, the first trained model MDLα and the second trained model MDLβ may be implemented using other machine learning models such as a support vector machine, a decision tree, a random forest, or a logistic regression.

[0070] When the first trained model MDLα and the second trained model MDLβ are implemented by a neural network, the model definition data includes, for example, connection information regarding how the units contained in each of the layers that make up the neural network, such as the input layer, one or more hidden layers (intermediate layers), and output layer, are connected to each other, and weight information regarding the connection coefficients assigned to the data input and output between the connected units.

[0071] The connection information includes, for example, the number of units included in each layer, information specifying the type of unit to which each unit is connected, the activation function that realizes each unit, and gates provided between units in the hidden layer.

[0072] The activation functions that realize the units may be, for example, ReLU (Rectified Linear Unit) functions, ELU (Exponential Linear Units) functions, clipping functions, sigmoid functions, step functions, hyperbolic tangent functions, identity functions, etc. The gates selectively pass or weight data transmitted between units depending on, for example, the value (e.g., 1 or 0) returned by the activation functions.

[0073] The connection coefficients include, for example, weights assigned to output data when data is output from a unit in a hidden layer of a neural network to a unit in a deeper layer, and may also include bias components specific to each layer.

[0074] The first trained model MDLα and the second trained model MDLβ are trained to output, in response to an input image, an image in which noise has been reduced from the input image (i.e., a denoised image).

[0075] Therefore, the denoising processing function 156 inputs the signal image s separated from the reference material image a to the first trained model MDLα, thereby obtaining a denoised image (hereinafter referred to as the denoised signal image s) of the signal image s separated from the reference material image a from the first trained model MDLα.

[0076] Similarly, the denoising processing function 156 inputs the signal image s separated from the reference material image b to the first trained model MDLα, thereby obtaining a denoising signal image s of the signal image s separated from the reference material image b from the first trained model MDLα.

[0077] Furthermore, the denoising processing function 156 inputs the noise image n separated from the reference material image a to the second trained model MDLβ, thereby obtaining a denoised image of the noise image n separated from the reference material image a (hereinafter referred to as the denoised noise image n) from the second trained model MDLβ.

[0078] Similarly, the denoising processing function 156 inputs the noise image n separated from the reference material image b to the second trained model MDLβ, thereby obtaining a denoised noise image n of the noise image n separated from the reference material image b from the second trained model MDLβ.

[0079] Next, the transformation function 155 generates a denoised reference material image (hereinafter referred to as the denoised reference material image) by combining the denoised signal image s and the denoised noise image n for each reference material image using an inverse whitening transformation (step S110).

[0080] For example, the transform function 155 generates the denoised reference material image a' by combining the denoised signal image s and the denoised noise image n derived from the reference material image a using an inverse whitening transform. Similarly, the transform function 155 generates the denoised reference material image b' by combining the denoised signal image s and the denoised noise image n derived from the reference material image b using an inverse whitening transform.

[0081] Next, the output control function 157 outputs the denoised images (step S112). For example, the output control function 157 may cause the display 142 to display the denoised reference material image a' and the denoised reference material image b'. The output control function 157 may also transmit the denoised reference material image a' and the denoised reference material image b' to an external device (e.g., a computer used by a doctor, technician, etc.) via the communication interface 144. This completes the processing of this flowchart.

[0082] [Learning device configuration] The following describes the configuration of a learning device 200 that learns the machine learning model MDL. Fig. 5 is a diagram showing an example configuration of the learning device 200 in the first embodiment. The learning device 200 includes, for example, a communication interface 202, an input interface 204, an output interface 206, a memory 208, and a processing circuit 210.

[0083] The communication interface 202 communicates with external devices via the communication network NW. The communication interface 202 includes, for example, a NIC and an antenna for wireless communication.

[0084] The input interface 204 receives various input operations from an operator, converts the received input operations into electrical signals, and outputs the electrical signals to the processing circuitry 210 .

[0085] For example, the input interface 204 includes a mouse, a keyboard, a trackball, a switch, a button, a joystick, a touch panel, etc. The input interface 204 may be, for example, a user interface that accepts audio input from a microphone, etc. If the input interface 204 is a touch panel, the input interface 204 may also have the display function of a display 213a included in the output interface 206, which will be described later.

[0086] In this specification, the input interface 204 is not limited to an interface having physical operation parts such as a mouse, keyboard, etc. For example, an example of the input interface 204 also includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs this electrical signal to a control circuit.

[0087] The output interface 206 includes, for example, a display 213a and a speaker 213b. The display 213a displays various types of information. For example, the display 213a displays images generated by the processing circuit 210 and a GUI for receiving various input operations from an operator. For example, the display 213a is an LCD, a CRT display, an organic EL display, or the like. The speaker 213b outputs information input from the processing circuit 210 as sound.

[0088] The memory 208 is realized by, for example, a semiconductor memory element such as RAM or flash memory, a hard disk, or an optical disk. These non-transitory storage media may also be realized by other storage devices connected via a communication network NW, such as a NAS or an external storage server device.

[0089] The memory 208 may also include a non-transitory storage medium such as a ROM, a register, etc. The memory 208 stores programs executed by the hardware processor of the processing circuitry 210, various calculation results by the processing circuitry 210, model definition data, etc.

[0090] The processing circuit 210 includes, for example, an acquisition function 212 , a generation function 214 , a reconstruction function 216 , a conversion function 218 , a machine learning function 220 , and an output control function 222 .

[0091] The processing circuitry 210 realizes these functions by, for example, a hardware processor (computer) executing a program stored in a memory 208 (storage circuitry).

[0092] The hardware processor in processing circuit 210 refers to circuitry such as a CPU, GPU, application specific integrated circuit, programmable logic device (e.g., simple programmable logic device or complex programmable logic device, field programmable gate array), etc.

[0093] Instead of storing the program in memory 208, the program may be directly embedded in the circuit of the hardware processor. In this case, the hardware processor realizes its function by reading and executing the program embedded in the circuit. The program may be stored in memory 208 in advance, or may be stored on a non-transitory storage medium such as a DVD or CD-ROM, and installed into memory 208 from the non-transitory storage medium when the non-transitory storage medium is inserted into a drive device (not shown) of learning device 200.

[0094] The hardware processor is not limited to being configured as a single circuit, but may be configured as a single hardware processor by combining multiple independent circuits to realize each function, or multiple components may be integrated into a single hardware processor to realize each function.

[0095] [Learning device processing flow training] The training process by the learning device 200 will be described below with reference to Figures 6 and 7. Figure 6 is a flowchart showing the flow of a series of processes by the learning device 200 according to the first embodiment. Figure 7 is a diagram schematically showing the flow of a series of processes by the learning device 200 according to the first embodiment. The process of this flowchart is executed when learning the machine learning model MDL.

[0096] First, the acquisition function 212 acquires projection data in which the X-ray dose is equal to or greater than a threshold (that is, high-dose projection data) (step S200).

[0097] The acquisition function 212 acquires projection data at a high tube voltage (high kVp) and projection data at a low tube voltage (low kVp) as high-dose projection data.

[0098] For example, the acquisition function 212 may access the X-ray CT apparatus 100 via the communication interface 202 and acquire high-dose projection data from the X-ray CT apparatus 100. Furthermore, if a user inputs high-dose projection data into the input interface 204, the acquisition function 212 may acquire the high-dose projection data from the input interface 204. Furthermore, if high-dose projection data is stored in the memory 208, the acquisition function 212 may acquire the high-dose projection data from the memory 208.

[0099] Next, the generating function 214 generates projection data in which the X-ray dose is less than a threshold (that is, low-dose projection data) from the high-dose projection data acquired by the acquiring function 212 (step S202).

[0100] For example, the generation function 214 may generate low-dose, high-tube voltage (high kVp) projection data from high-dose, high-tube voltage (high kVp) projection data by noise simulation, and may generate low-dose, low-tube voltage (low kVp) projection data from high-dose, low-tube voltage (low kVp) projection data.

[0101] Next, the reconstruction function 216 performs material decomposition based on a combination of high-dose, high-tube voltage (high kVp) projection data and high-dose, low-tube voltage (low kVp) projection data, and also performs material decomposition based on a combination of low-dose, high-tube voltage (high kVp) projection data and low-dose, low-tube voltage (low kVp) projection data (step S204).

[0102] For example, the reconstruction function 216 uses the attenuation coefficients of two different types of reference materials a and b to distinguish between reference materials a and b from a combination of high-dose, high-tube voltage (high kVp) projection data and high-dose, low-tube voltage (low kVp) projection data, and to distinguish between reference materials a and b from a combination of low-dose, high-tube voltage (high kVp) projection data and low-dose, low-tube voltage (low kVp) projection data.

[0103] Next, the reconstruction function 216 performs reconstruction processing on the material-decomposed projection data to generate reference material images, which are CT images in which each reference material is enhanced or suppressed (step S206).

[0104] For example, the reconstruction function 216 performs reconstruction processing on high-dose projection data in which the reference materials a and b have been differentiated to generate a reference material image a_high and a reference material image b_high. Similarly, the reconstruction function 216 performs reconstruction processing on low-dose projection data in which the reference materials a and b have been differentiated to generate a reference material image a_low and a reference material image b_low.

[0105] Next, the transformation function 218 separates the four types of reference material images into a signal image s and a noise image n by whitening transformation (step S208).

[0106] For example, the transformation function 218 separates the reference material image a_high into a signal image s_high and a noise image n_high by whitening transformation, while separating the reference material image b_high into a signal image s_high and a noise image n_high. Similarly, the transformation function 155 separates the reference material image a_low into a signal image s_low and a noise image n_low by whitening transformation, while separating the reference material image b_low into a signal image s_low and a noise image n_low.

[0107] Next, the machine learning function 220 generates a training data set for each learned model MDL (step S210).

[0108] For example, the machine learning function 220 generates a dataset as a first training dataset in which a signal image s_low separated from a reference material image a_low derived from a low dose is matched as a target with a signal image s_high separated from a reference material image a_high derived from a high dose, or generates a dataset in which a signal image s_low separated from a reference material image b_low derived from a low dose is matched as a target with a signal image s_high separated from a reference material image b_high derived from a high dose.

[0109] In addition, the machine learning function 220 generates, as a second training dataset, a dataset in which a noise image n_low separated from a reference material image a_low derived from a low dose is matched as a target with a noise image n_high separated from a reference material image a_high derived from a high dose, or generates a dataset in which a noise image n_low separated from a reference material image b_low derived from a low dose is matched as a target with a noise image n_high separated from a reference material image b_high derived from a high dose.

[0110] Next, the machine learning function 220 trains each trained model MDL using the training data set (step S212).

[0111] For example, the machine learning function 220 uses a first training data set to learn a first trained model MDLα. Specifically, the machine learning function 220 inputs a signal image s_low to the first trained model MDLα. The machine learning function 220 calculates the difference between an image output by the first trained model MDLα in response to the input of the signal image s_low and a target signal image s_high. Then, the machine learning function 220 adjusts the parameters (weight coefficients and bias components) of the first trained model MDLα by performing error backpropagation using a stochastic gradient descent method or the like based on the difference.

[0112] Furthermore, the machine learning function 220 uses the second training dataset to train a second trained model MDLβ. Specifically, the machine learning function 220 inputs the noise image n_low to the second trained model MDLβ. The machine learning function 220 calculates the difference between the image output by the second trained model MDLβ in response to the input of the noise image n_low and the target noise image n_high. Then, the machine learning function 220 adjusts the parameters (weight coefficients and bias components) of the second trained model MDLβ by performing error backpropagation using a stochastic gradient descent method or the like based on the difference.

[0113] In this way, the machine learning function 220 repeats the above process until the number of training iterations reaches a specified number, and when the number of iterations reaches the specified number, it stores model definition data in which the first trained model MDLα and the second trained model MDLβ are defined in the memory 208.

[0114] Next, the output control function 222 transmits model definition data in which the first trained model MDLα and the second trained model MDLβ trained by the machine learning function 220 are defined to each X-ray CT device 100 via the communication interface 202 (step S214). This ends the processing of this flowchart.

[0115] According to the first embodiment described above, the X-ray CT device 100 acquires projection data at a high tube voltage (high kVp) and projection data at a low tube voltage (low kVp) of X-rays irradiated onto the subject P, discriminates between reference materials a and b based on the projection data at a high tube voltage (high kVp) and the projection data at a low tube voltage (low kVp), and generates reference material images a and b by reconstruction processing.

[0116] The X-ray CT device 100 separates each of the reference material image a and the reference material image b into a signal image s and a noise image n by whitening transformation. The X-ray CT device 100 inputs the signal image s to the first trained model MDLα and inputs the noise image n to the second trained model MDLβ. The X-ray CT device 100 acquires the denoised signal image s from the first trained model MDLα and acquires the denoised noise image n from the second trained model MDLβ.

[0117] The first trained model MDLα is a machine learning model trained based on a first training dataset that includes a dataset in which a signal image s_high separated from a high-dose reference material image a_high is matched as a target to a signal image s_low separated from a low-dose reference material image a_low, and a dataset in which a signal image s_high separated from a high-dose reference material image b_high is matched as a target to a signal image s_low separated from a low-dose reference material image b_low.

[0118] The second trained model MDLβ is a machine learning model trained based on a second training dataset that includes a dataset in which a noise image n_low separated from a low-dose reference material image a_low is matched as a target with a noise image n_high separated from a high-dose reference material image a_high, and a dataset in which a noise image n_low separated from a low-dose reference material image b_low is matched as a target with a noise image n_high separated from a high-dose reference material image b_high.

[0119] The X-ray CT apparatus 100 generates denoised reference material images a' and b' by combining the denoised signal image s and the denoised noise image n for each reference material image using an inverse whitening transform. Then, the X-ray CT apparatus 100 displays the denoised reference material images a' and b' on the display 142 or transmits them to an external device via the communication interface 144, for example.

[0120] In other words, the X-ray CT device 100 (i) separates the reference material image into a signal image s and a noise image n using a whitening transform, (ii) denoises each of the signal image s and the noise image n using the trained model MDL, and (iii) combines the denoised signal image s and the denoised noise image n using an inverse whitening transform to generate a denoised reference material image.

[0121] As mentioned above, the whitening transform cannot completely separate signal and noise; signal image s contains noise components, and noise image n contains signal components. Therefore, the MDL trained model is used to remove the noise components from signal image s and the signal components from noise image n, and then combine them to generate a denoised reference material image. This process can improve the accuracy of noise reduction in spectral imaging.

[0122] (Modification of the first embodiment) A modified example of the first embodiment will be described below. In the above-described first embodiment, the medical images generated by the reconstruction process are described as reference material images, but this is not limited to this. For example, the reconstruction process may generate a virtual monochromatic X-ray image, a virtual plain image, an iodine map, an effective atomic number image, or an electron density image. In this case, the medical images included in each training data set are also assumed to be these virtual monochromatic X-ray images, virtual plain images, iodine maps, effective atomic number images, or electron density images.

[0123] (Second embodiment) The second embodiment will be described below. In the above-described first embodiment, the X-ray CT device 100 has been described as a CT device using DE (Dual Energy). In contrast, the second embodiment differs from the first embodiment in that the X-ray CT device 100 is a PCCT (Photon Counting CT) device. The following description will focus on the differences from the first embodiment, and will omit a description of the points in common with the first embodiment. In the description of the second embodiment, the same parts as in the first embodiment will be described with the same reference numerals.

[0124] Fig. 8 is a diagram showing an example of the configuration of the DAS 116 of the X-ray CT apparatus 100 according to the second embodiment. The DAS 116 of the X-ray CT apparatus 100, which is a photon-counting CT apparatus, has readout channels the number of which corresponds to the number of X-ray detection elements. These multiple readout channels are implemented in parallel in an integrated circuit such as an application specific integrated circuit (ASIC). Fig. 8 shows the configuration of only DAS 116-1 for one readout channel.

[0125] The DAS116-1 has a preamplifier circuit 61, a waveform shaping circuit 63, a plurality of pulse height discriminator circuits 65, a plurality of counter circuits 67, and an output circuit 69. The preamplifier circuit 61 amplifies the detected electrical signal DS (current signal) from the connected X-ray detection element. For example, the preamplifier circuit 61 converts the current signal from the connected X-ray detection element into a voltage signal having a voltage value (peak value) proportional to the amount of charge of the current signal. A waveform shaping circuit 63 is connected to the preamplifier circuit 61. The waveform shaping circuit 63 shapes the waveform of the voltage signal from the preamplifier circuit 61. For example, the waveform shaping circuit 63 reduces the pulse width of the voltage signal from the preamplifier circuit 61.

[0126] A number of counting channels corresponding to the number of energy bands (energy bins) are connected to the waveform shaping circuit 63. When n energy bins are set, n counting channels are provided in the waveform shaping circuit 63. Each counting channel has a pulse height discrimination circuit 65-n and a counting circuit 67-n.

[0127] Each of the pulse height discrimination circuits 65-n discriminates the energy of the X-ray photons detected by the X-ray detection elements, which is the pulse height value of the voltage signal from the waveform shaping circuit 63. For example, the pulse height discrimination circuit 65-n includes a comparison circuit 653-n. The voltage signal from the waveform shaping circuit 63 is input to one input terminal of each of the comparison circuits 653-n. Reference signals TH (reference voltage values) corresponding to different thresholds are supplied from the control device 118 to the other input terminal of each of the comparison circuits 653-n. For example, the comparison circuit 653-1 for energy bin bin1 is supplied with the reference signal TH-1, the comparison circuit 653-2 for energy bin bin2 is supplied with the reference signal TH-2, and the comparison circuit 653-n for energy bin n is supplied with the reference signal TH-n. Each of the reference signals TH has an upper reference value and a lower reference value. Each of the comparison circuits 653-n outputs an electrical pulse signal when the voltage signal from the waveform shaping circuit 63 has a peak value corresponding to an energy bin corresponding to each of the reference signals TH. For example, the comparison circuit 653-1 outputs an electrical pulse signal when the peak value of the voltage signal from the waveform shaping circuit 63 is the peak value corresponding to energy bin bin1 (when it is between reference signals TH-1 and TH-2). On the other hand, the comparison circuit 653-1 for energy bin bin1 does not output an electrical pulse signal when the peak value of the voltage signal from the waveform shaping circuit 63 is not the peak value corresponding to energy bin bin1. Furthermore, for example, the comparison circuit 653-2 outputs an electrical pulse signal when the peak value of the voltage signal from the waveform shaping circuit 63 is the peak value corresponding to energy bin bin2 (when it is between reference signals TH-2 and TH-3).

[0128] The counter circuit 67-n counts the electrical pulse signals from the pulse-height discriminator circuit 65-n at a readout period that coincides with the view switching period. For example, the counter circuit 67-n receives a trigger signal TS from the control device 118 at the timing of switching each view. The counter circuit 67-n, triggered by the supply of the trigger signal TS, increments the count number stored in its internal memory each time an electrical pulse signal is input from the pulse-height discriminator circuit 65-n. The counter circuit 67-n reads out the count number data (i.e., count data) stored in its internal memory and supplies it to the output circuit 69 when the next trigger signal is supplied. The counter circuit 67-n also resets the count number stored in its internal memory to its initial value each time the trigger signal TS is supplied. In this manner, the counter circuit 67-n counts the count number for each view.

[0129] The output circuit 69 is connected to counting circuits 67-n for multiple readout channels mounted on the X-ray detector 115. The output circuit 69 integrates count data from the counting circuits 67-n for multiple readout channels for each of multiple energy bins to generate count data for multiple readout channels for each view. The count data for each energy bin is a collection of data with a count number defined by the channel, segment (column), and energy bin. The count data for each energy bin is transmitted to the console device 140 on a view-by-view basis. The count data for each view is called a count data set CS.

[0130] The DAS 116 having such a configuration collects count data indicating the number of counts (count values) of X-ray photons detected by the X-ray detector 115 for each set energy bin, and obtains the count data for each energy bin as detection data.

[0131] The system control function 151 according to the second embodiment may set the energy bins to be referenced by the reconstruction function 153, for example, based on an input operation received by the input interface 143. The system control function 151 may also set the energy bins automatically, regardless of the input operation received by the input interface 143.

[0132] The pre-processing function 152 according to the second embodiment generates projection data from the count data for each energy bin by performing predetermined pre-processing on the detection data (count data) output by the DAS 116. In other words, the pre-processing function 152 generates projection data for each energy bin. The predetermined pre-processing may include, for example, logarithmic conversion, offset correction, inter-channel sensitivity correction, beam hardening correction, scattered radiation correction, dark count correction, and the like.

[0133] The reconstruction function 153 performs a predetermined reconstruction process on the projection data generated by the preprocessing function 152, and generates a CT image from the projection data.

[0134] For example, the reconstruction function 153 may generate a CT image for each energy bin (hereinafter referred to as a bin image) by reconstructing projection data for each energy bin.

[0135] In addition, the reconstruction function 153 may accumulate the projection data for each energy bin, calculate the amount of X-ray absorption based on the sum of the projection data and the response function stored in the memory 141, and generate a counting image (also called an integral image) based on the amount of X-ray absorption.

[0136] Furthermore, as in the first embodiment, the reconstruction function 153 may extract only the components of the reference material (e.g., iodine) from the projection data of each energy bin, and reconstruct the projection data of each energy bin from which the components of the reference material have been extracted, thereby generating a CT image (reference material image) in which the reference material is emphasized.

[0137] Furthermore, in addition to the reference material image, the reconstruction function 153 may generate a virtual monochromatic X-ray image, a virtual plain image, an iodine map, an effective atomic number image, an electron density image, a CT image reconstructed for each energy bin (bin image), and the like.

[0138] Furthermore, when an NR (Normal Resolution) mode and an SHR (Super High Resolution) mode are set in the X-ray CT device 100, which is a PCCT, the reconstruction function 153 may generate CT images with resolutions corresponding to each mode. The NR mode is a mode in which the resolution is equal to or less than a threshold, and the SHR mode is a mode in which the resolution exceeds the threshold. A CT image generated in the NR mode is also called an NR image, and a CT image generated in the SHR mode is also called an SHR image.

[0139] The predetermined reconstruction processing may include, for example, a filtered back projection method, an iterative reconstruction method, etc. The reconstruction function 153 stores the reconstructed CT image in the memory 141. When preprocessing is not performed by the preprocessing function 152, the reconstruction function 153 may perform reconstruction processing using detection data (count data).

[0140] 9 is a diagram showing an example of the functional configuration of the reconstruction function 153 according to the second embodiment. The reconstruction function 153 has, for example, a response function generation function 1531, an X-ray absorption amount calculation function 1532, and a reconstruction processing function 1533.

[0141] The response function generation function 1531 generates response function data representing detector response characteristics. For example, the response function generation function 1531 measures the response (i.e., detection energy and detection intensity) of a standard detection system to multiple monochromatic X-rays having multiple incident X-ray energies through predictive calculations, experiments, and a combination of predictive calculations and experiments, and generates a response function based on the measured values ​​of the detection energy and detection intensity. The response function generation function 1531 may also generate response function data based on actual measurement values ​​collected during calibration or the like. The response function defines the relationship between the detection energy of each incident X-ray and the output response of the system. For example, the response function defines the relationship between the detection energy of each incident X-ray and the detection intensity. The generated response function data is stored in the memory 141.

[0142] The X-ray absorption amount calculation function 1532 calculates the X-ray absorption amount for each of the multiple reference materials based on count data for multiple energy bins included in the projection data, the energy spectrum of X-rays incident on the subject P, and the response function stored in the memory 141. The X-ray absorption amount calculation function 1532 calculates the X-ray absorption amount based on the count data and the energy spectrum of X-rays incident on the subject P using the response function, thereby enabling calculation of the X-ray absorption amount without being affected by the response characteristics of the X-ray detector 115 and the DAS 116. As described in the first embodiment, this process of obtaining the X-ray absorption amount for each reference material is also called material decomposition. The reference material can be any material, such as calcium, calcification, bone, fat, muscle, air, organs, lesions, hard tissue, soft tissue, or contrast material. The type of reference material to be calculated may be determined in advance by an operator or the like via the input interface 143. The X-ray absorption amount indicates the amount of X-rays absorbed by the reference material. For example, the amount of X-ray absorption is determined by a combination of the X-ray attenuation coefficient and the X-ray transmission path length.

[0143] The reconstruction processing function 1533 reconstructs a photon-counting CT image representing the spatial distribution of the reference material to be imaged among the plurality of reference materials based on the X-ray absorption amount for each of the plurality of reference materials calculated by the X-ray absorption amount calculation function 1532, and stores the generated CT image in the memory 141. The reference material to be imaged may be of one type or of multiple types. The type of reference material to be imaged may be determined by an operator or the like via the input interface 143.

[0144] Projection data obtained by the photon-counting CT device contains information about the energy of X-rays attenuated by passing through the subject P. Therefore, the reconstruction processing function 1533 can, for example, reconstruct a CT image of a specific energy component. The reconstruction processing function 1533 can also, for example, reconstruct CT images of each of a plurality of energy components. Furthermore, the reconstruction processing function 1533 can, for example, assign a color tone corresponding to the energy component to each pixel of the CT image of each energy component, and superimpose a plurality of CT images color-coded according to the energy component.

[0145] The X-ray CT device 100, which is a photon-counting CT device, may generate a reference material image, a virtual monochromatic X-ray image, a virtual plain image, an iodine map, an effective atomic number image, an electron density image, or an NR image or an SHR image through reconstruction processing, in addition to or instead of generating a bin image or a counting image through reconstruction processing, as in the first embodiment described above.

[0146] When the X-ray CT device 100 generates a bin image for each energy bin, the learning device 200 according to the second embodiment prepares machine learning models MDL in the same number as the number of the energy bins and trains each of them. For example, when there are four energy bins, the number of machine learning models MDL is four. Furthermore, when the X-ray CT device 100 generates a counting image, the learning device 200 prepares and trains one machine learning model MDL regardless of the number of energy bins.

[0147] According to the second embodiment described above, when the X-ray CT device 100 is a photon-counting CT device, the machine learning model MDL can be trained using a training data set that employs a reference material image, a virtual monochromatic X-ray image, a virtual plain image, an iodine map, an effective atomic number image, an electron density image, a bin image, a counting image, an NR image, and an SHR image.

[0148] (Third embodiment) The third embodiment will be described below. In the first and second embodiments described above, of the signal image s and the noise image n separated from the reference material image by the whitening transform, the signal image s is denoised using the first trained model MDLα, and the noise image n is denoised using the second trained model MDLβ.

[0149] In contrast, the third embodiment differs from the first and second embodiments in that both the signal image s and the noise image n are denoised using either the first trained model MDLα or the second trained model MDLβ. The following description will focus on the differences from the first and second embodiments, and will omit a description of the points in common with the first and second embodiments. In the description of the third embodiment, the same parts as those in the first and second embodiments will be denoted by the same reference numerals.

[0150] FIG. 10 is a diagram schematically showing the flow of a series of processes in the X-ray CT apparatus 100 in the third embodiment.

[0151] The denoising processing function 156 according to the third embodiment denoises each of the signal image s and the noise image n using either the first trained model MDLα or the second trained model MDLβ, for example, as the processing of step S108. In the third embodiment, as an example, the first trained model MDLα is used, but the other second trained model MDLβ may also be used.

[0152] Specifically, the denoising processing function 156 inputs a signal image s separated from the reference material image a into the first trained model MDLα, thereby obtaining a denoising signal image s of the reference material image a from the first trained model MDLα.

[0153] Similarly, the denoising processing function 156 inputs a signal image s separated from the reference material image b to the first trained model MDLα, thereby obtaining a denoising signal image s of the reference material image b from the first trained model MDLα.

[0154] Furthermore, the denoising processing function 156 inputs the noise image n separated from the reference material image a to the first trained model MDLα, thereby obtaining a denoised noise image n of the reference material image a from the first trained model MDLα.

[0155] Similarly, the denoising processing function 156 inputs a noise image n separated from the reference material image b to the first trained model MDLα, thereby obtaining a denoised noise image n of the reference material image b from the first trained model MDLα.

[0156] In this way, in the third embodiment, both the signal image s and the noise image n are input to the first trained model MDLα, thereby denoising both the signal image s and the noise image n.

[0157] FIG. 11 is a diagram schematically illustrating the flow of a series of processes performed by the learning device 200 according to the third embodiment.

[0158] The machine learning function 220 according to the third embodiment generates a first training data set for training the first trained model MDLα as the process of step S210. For example, the machine learning function 220 generates the data sets (i) to (iv) as the first training data set.

[0159] (i) A data set in which a signal image s_high separated from a reference material image a_high derived from a high dose is matched as a target to a signal image s_low separated from a reference material image a_low derived from a low dose. (ii) A data set in which a signal image s_high separated from a reference material image b_high derived from a high dose is matched as a target to a signal image s_low separated from a reference material image b_low derived from a low dose. (iii) A dataset in which the noise image n_low separated from the reference material image a_low derived from a low dose is matched with the noise image n_high separated from the reference material image a_high derived from a high dose as the target. (iv) A data set in which a noise image n_low separated from a reference material image b_low derived from a low dose is matched with a noise image n_high separated from a reference material image b_high derived from a high dose as a target.

[0160] In step S212, the machine learning function 220 trains a first trained model MDLα using a first training data set including (i) to (iv).

[0161] According to the third embodiment described above, the X-ray CT apparatus 100 separates each of the reference material image a and the reference material image b into a signal image s and a noise image n by whitening transformation. The X-ray CT apparatus 100 inputs the signal image s and the noise image n to a first trained model MDLα trained based on a first training data set, and acquires a denoised signal image s and a denoised noise image n from the first trained model MDLα.

[0162] In the first or second embodiment described above, the first training data set includes only the data sets (i) to (ii) derived from the signal image s. In contrast, in the third embodiment, the first training data set includes not only the data sets (i) to (ii) derived from the signal image s, but also the data sets (iii) to (iv) derived from the noise image n. By solely using the first trained model MDLα trained using such a first training data set, it is possible to denoise not only the signal image s but also the noise image n.

[0163] In the third embodiment described above, the first trained model MDLα is used alone to denoise both the signal image s and the noise image n, but this is not limiting. For example, the second trained model MDLβ may be used alone to denoise both the signal image s and the noise image n. In this case, a second training data set including (i) to (iv) is used to train the second trained model MDLβ.

[0164] (Fourth embodiment) The fourth embodiment will be described below. The fourth embodiment differs from the first to third embodiments in that either the signal image s or the noise image n is denoised using either the first trained model MDLα or the second trained model MDLβ. The following description will focus on the differences from the first to third embodiments, and will omit a description of the points in common with the first to third embodiments. In the description of the fourth embodiment, the same parts as those in the first to third embodiments will be denoted by the same reference numerals.

[0165] FIG. 12 is a diagram schematically showing the flow of a series of processes in the X-ray CT apparatus 100 in the fourth embodiment.

[0166] The denoising processing function 156 according to the fourth embodiment denoises only one of the signal image s and the noise image n, for example, using either the first trained model MDLα or the second trained model MDLβ as the processing of step S108. In the fourth embodiment, as an example, a description will be given assuming that only the first trained model MDLα is used to denoise only the signal image s. Note that only the first trained model MDLα may be used to denoise only the noise image n. Furthermore, only the second trained model MDLβ may be used to denoise only the signal image s or only the noise image n.

[0167] For example, the denoising processing function 156 inputs a signal image s separated from the reference material image a into the first trained model MDLα, thereby obtaining a denoising signal image s of the reference material image a from the first trained model MDLα.

[0168] Furthermore, the denoising processing function 156 inputs the noise image n separated from the reference material image a to the first trained model MDLα, thereby obtaining a denoised noise image n of the reference material image a from the first trained model MDLα.

[0169] Next, in the process of step S110, the conversion function 155 according to the fourth embodiment generates a denoised reference material image by combining the denoised signal image s and the noise image n for each reference material image using inverse whitening conversion.

[0170] For example, the transform function 155 generates the denoised reference material image a' by combining the denoised signal image s and the noise image n derived from the reference material image a using an inverse whitening transform. Similarly, the transform function 155 generates the denoised reference material image b' by combining the denoised signal image s and the noise image n derived from the reference material image b using an inverse whitening transform.

[0171] In this way, in the fourth embodiment, only the signal image s is input to the first learned model MDLα, thereby denoising only the signal image s.

[0172] FIG. 13 is a diagram schematically illustrating the flow of a series of processes performed by the learning device 200 according to the fourth embodiment.

[0173] In the machine learning function 220 according to the fourth embodiment, when the conversion function 218 separates the four types of reference material images into signal image s and noise image n by whitening conversion as processing in S208, it determines whether the image for which denoising is requested (hereinafter referred to as the denoising request image) is signal image s or noise image n (step S216).

[0174] For example, the denoised request image may be determined in response to a user's operation on the input interface 204. Alternatively, the communication interface 202 may receive the denoised request image from the X-ray CT apparatus 100. Alternatively, the denoised request image may be pre-programmed. In the fourth embodiment, as described above, it is assumed that the request image is a signal image s.

[0175] When the denoising request image is a signal image s, the machine learning function 220 extracts only the signal image s from the plurality of signal images s and noise images n in step S210, and generates a first training data set that includes only the signal image s. For example, the machine learning function 220 generates a first training data set that includes the above-mentioned (i) to (ii).

[0176] In step S212, the machine learning function 220 trains a first trained model MDLα using a first training data set including (i) to (ii).

[0177] According to the fourth embodiment described above, the X-ray CT device 100 separates a reference material image into a signal image s and a noise image n by whitening transformation. The X-ray CT device 100 inputs the signal image s to a first trained model MDLα trained based on a first training data set, and acquires a denoised signal image s from the first trained model MDLα. The X-ray CT device 100 then combines the denoised signal image s and the noise image n by inverse whitening transformation to generate a denoised reference material image. In this way, a denoised reference material image can be generated by denoising only one of the signal image s and the noise image n, rather than denoising both of them.

[0178] In the fourth embodiment described above, the first trained model MDLα is used alone to denoise the signal image s, but this is not limiting. For example, the second trained model MDLβ may be used alone to denoise the signal image s.

[0179] (Fifth embodiment) The fifth embodiment will be described below. In the first to fourth embodiments described above, the signal image s and / or the noise image n are denoised. In contrast, the fifth embodiment differs from the first to fourth embodiments in that not only are the signal image s and / or the noise image n denoised, but also the reference material image is denoised.

[0180] The following description will focus on the differences from the first to fourth embodiments, and will omit a description of the commonalities with the first to fourth embodiments. In the description of the fifth embodiment, the same parts as those in the first to fourth embodiments will be denoted by the same reference numerals.

[0181] Fig. 14 is a flowchart showing an example of the flow of a series of processes in the X-ray CT apparatus 100 in the fifth embodiment. Fig. 15 is a diagram schematically showing the flow of a series of processes in the X-ray CT apparatus 100 in the fifth embodiment. The processing in this flowchart is executed after learning of the machine learning model MDL is completed.

[0182] First, the preprocessing function 152 acquires projection data (step S300) by performing preprocessing on the detection data output by the DAS 116. The acquired projection data includes projection data at a high tube voltage (kVp equal to or greater than a threshold) and projection data at a low tube voltage (kVp less than a threshold).

[0183] Next, the reconstruction function 153 performs material decomposition based on the acquired projection data at high tube voltage (high kVp) and projection data at low tube voltage (low kVp) (step S302).

[0184] Next, the reconstruction function 153 performs reconstruction processing on the material-decomposed projection data to generate a reference material image (step S304).

[0185] Next, the conversion function 155 separates the reference material image into a high-tube voltage image x and a low-tube voltage image y by image conversion such as grayscale conversion or color channel decomposition (step S306).

[0186] The high-tube voltage image x is a reference material image reconstructed from projection data in which the tube voltage of the X-ray tube 111 is equal to or greater than a threshold (e.g., 80 keV), and the low-tube voltage image y is a reference material image reconstructed from projection data in which the tube voltage of the X-ray tube 111 is less than a threshold (e.g., 50 keV). The threshold may be, for example, approximately 60 keV to 70 keV. The high-tube voltage image x is an example of a "high-tube voltage medical image," and the low-tube voltage image y is an example of a "low-tube voltage medical image."

[0187] For example, when reference material images a and b are generated, the conversion function 155 separates the reference material image a into a high tube voltage image x and a low tube voltage image y, and also separates the reference material image b into a high tube voltage image x and a low tube voltage image y.

[0188] Next, the denoising function 156 denoises each of the high-tube voltage image x and the low-tube voltage image y using the learned model MDL (step S308).

[0189] For example, the denoising processing function 156 reads out the third trained model MDLκ and the fourth trained model MDLλ defined by the model definition data stored in the memory 141, inputs the high tube voltage image x to the third trained model MDLκ, and inputs the low tube voltage image y to the fourth trained model MDLλ.

[0190] The third trained model MDLκ and the fourth trained model MDLλ may be implemented by a deep neural network such as a convolutional neural network. Instead of a neural network, the third trained model MDLκ and the fourth trained model MDLλ may be implemented by other machine learning models such as a support vector machine, a decision tree, a random forest, or a logistic regression.

[0191] The third trained model MDLκ and the fourth trained model MDLλ are trained to output an image with reduced noise (i.e., a denoised image) from an input image.

[0192] Therefore, the denoising processing function 156 inputs the high tube voltage image x separated from the reference material image a into the third trained model MDLκ, thereby obtaining a denoised image of the high tube voltage image x separated from the reference material image a (hereinafter referred to as the denoised high tube voltage image x) from the third trained model MDLκ.

[0193] Similarly, the denoising processing function 156 inputs the high tube voltage image x separated from the reference material image b into the third trained model MDLκ, thereby obtaining a denoised high tube voltage image x, which is a denoised image of the high tube voltage image x separated from the reference material image b, from the third trained model MDLκ.

[0194] Furthermore, the denoising processing function 156 inputs the low tube voltage image y separated from the reference material image a to the fourth trained model MDLλ, thereby obtaining a denoised image of the low tube voltage image y separated from the reference material image a (hereinafter referred to as the denoised low tube voltage image y) from the fourth trained model MDLλ.

[0195] Similarly, the denoising processing function 156 inputs the low tube voltage image y separated from the reference material image b to the fourth trained model MDLλ, thereby obtaining a denoised low tube voltage image y, which is a denoised image of the low tube voltage image y separated from the reference material image b, from the fourth trained model MDLλ.

[0196] Next, the conversion function 155 generates a denoised reference material image by combining the denoised high-tube voltage image x and the denoised low-tube voltage image y for each reference material image using inverse image conversion (inverse conversion such as grayscale conversion or color channel decomposition) (step S310).

[0197] For example, the transformation function 155 generates a denoised reference material image a' by combining a denoised high-tube voltage image x and a denoised low-tube voltage image y derived from the reference material image a through inverse image transformation. Similarly, the transformation function 155 generates a denoised reference material image b' by combining a denoised high-tube voltage image x and a denoised low-tube voltage image y derived from the reference material image b through inverse image transformation.

[0198] Next, the transformation function 155 separates each of the denoised reference material image a' and the denoised reference material image b' into a signal image s and a noise image n by whitening transformation (step S312).

[0199] For example, the transformation function 155 separates the denoised reference material image a' into a signal image s and a noise image n by a whitening transformation, and also separates the denoised reference material image b' into a signal image s and a noise image n.

[0200] Next, the denoising function 156 denoises each of the signal image s and the noise image n using the first trained model MDLα and the second trained model MDLβ (step S314).

[0201] The denoising processing function 156 inputs the signal image s separated from the denoised reference material image a' into the first trained model MDLα, thereby obtaining a denoised signal image s of the signal image s separated from the denoised reference material image a' from the first trained model MDLα.

[0202] Similarly, the denoising processing function 156 inputs the signal image s separated from the denoised reference material image b' into the first trained model MDLα, thereby obtaining a denoised signal image s of the signal image s separated from the denoised reference material image b' from the first trained model MDLα.

[0203] Furthermore, the denoising processing function 156 inputs the noise image n separated from the denoised reference material image a' into the second trained model MDLβ, thereby obtaining a denoised noise image n of the noise image n separated from the denoised reference material image a' from the second trained model MDLβ.

[0204] Similarly, the denoising processing function 156 inputs the noise image n separated from the denoised reference material image b' to the second trained model MDLβ, thereby obtaining a denoised noise image n of the noise image n separated from the denoised reference material image b' from the second trained model MDLβ.

[0205] Next, the transform function 155 generates a denoised reference material image (hereinafter referred to as a double denoised reference material image) by combining the denoised signal image s and the denoised noise image n for each denoised reference material image using an inverse whitening transform (step S316). The double denoised reference material image is an example of a "double denoised medical image".

[0206] For example, the transformation function 155 generates a double denoised reference material image a'' by combining the denoised signal image s and the denoised noise image n derived from the denoised reference material image a' using an inverse whitening transformation. Similarly, the transformation function 155 generates a double denoised reference material image b'' by combining the denoised signal image s and the denoised noise image n derived from the denoised reference material image b' using an inverse whitening transformation.

[0207] Next, the output control function 157 outputs the double denoised image (step S318). For example, the output control function 157 may cause the display 142 to display the double denoised reference material image a" and the double denoised reference material image b". The output control function 157 may also transmit the double denoised reference material image a" and the double denoised reference material image b" to an external device (e.g., a computer used by a doctor, technician, etc.) via the communication interface 144. This completes the processing of this flowchart.

[0208] Fig. 16 is a flowchart showing the flow of a series of processes in the learning device 200 according to the fifth embodiment. Fig. 17 is a diagram schematically showing the flow of a series of processes in the learning device 200 according to the fifth embodiment. The processes in this flowchart are executed when learning the machine learning model MDL.

[0209] First, the acquisition function 212 acquires projection data in which the X-ray dose is equal to or greater than a threshold, that is, high-dose projection data (step S400).

[0210] The acquisition function 212 acquires projection data at a high tube voltage (high kVp) and projection data at a low tube voltage (low kVp) as high-dose projection data.

[0211] Next, the generating function 214 generates projection data in which the X-ray dose is less than a threshold, that is, low-dose projection data, from the high-dose projection data acquired by the acquiring function 212 (step S402).

[0212] For example, the generation function 214 may generate low-dose, high-tube voltage (high kVp) projection data from high-dose, high-tube voltage (high kVp) projection data by noise simulation, and may generate low-dose, low-tube voltage (low kVp) projection data from high-dose, low-tube voltage (low kVp) projection data.

[0213] Next, the reconstruction function 216 performs material decomposition based on a combination of high-dose, high-tube voltage (high kVp) projection data and high-dose, low-tube voltage (low kVp) projection data, and also performs material decomposition based on a combination of low-dose, high-tube voltage (high kVp) projection data and low-dose, low-tube voltage (low kVp) projection data (step S404).

[0214] For example, the reconstruction function 216 uses the attenuation coefficients of two different types of reference materials a and b to distinguish between reference materials a and b from a combination of high-dose, high-tube voltage (high kVp) projection data and high-dose, low-tube voltage (low kVp) projection data, and to distinguish between reference materials a and b from a combination of low-dose, high-tube voltage (high kVp) projection data and low-dose, low-tube voltage (low kVp) projection data.

[0215] Next, the reconstruction function 216 performs reconstruction processing on the material-decomposed projection data to generate reference material images, which are CT images in which each reference material is enhanced or suppressed (step S406).

[0216] For example, the reconstruction function 216 performs reconstruction processing on high-dose projection data in which the reference materials a and b have been differentiated to generate a reference material image a_high and a reference material image b_high. Similarly, the reconstruction function 216 performs reconstruction processing on low-dose projection data in which the reference materials a and b have been differentiated to generate a reference material image a_low and a reference material image b_low.

[0217] Next, the transformation function 218 separates the four types of reference material images into a signal image s and a noise image n by whitening transformation (step S408).

[0218] For example, the transformation function 218 separates the reference material image a_high into a signal image s_high and a noise image n_high by whitening transformation, while separating the reference material image b_high into a signal image s_high and a noise image n_high. Similarly, the transformation function 155 separates the reference material image a_low into a signal image s_low and a noise image n_low by whitening transformation, while separating the reference material image b_low into a signal image s_low and a noise image n_low.

[0219] Next, the machine learning function 220 generates a first training data set and a second training data set (step S410).

[0220] For example, the machine learning function 220 generates a first training dataset that includes only datasets (i) to (ii) derived from the signal image s, and generates a second training dataset that includes only datasets (iii) to (iv) derived from the noise image n.

[0221] Next, the machine learning function 220 trains a first trained model MDLα using the first training data set, and trains a second trained model MDLβ using the second training data set (step S412).

[0222] Meanwhile, the conversion function 218 separates the four types of reference material images into high-tube voltage images x and low-tube voltage images y through image conversion (step S414).

[0223] For example, the conversion function 218 separates the reference material image a_high into a high-tube voltage image x_high and a low-tube voltage image y_high, while separating the reference material image b_high into a high-tube voltage image x_high and a low-tube voltage image y_high, through image conversion. Similarly, the conversion function 155 separates the reference material image a_low into a high-tube voltage image x_low and a low-tube voltage image y_low, while separating the reference material image b_low into a high-tube voltage image x_low and a low-tube voltage image y_low, through image conversion.

[0224] Next, the machine learning function 220 generates a third training data set and a fourth training data set (step S416).

[0225] For example, the machine learning function 220 generates a third training dataset that includes only (v) to (vi), and a fourth training dataset that includes only (vii) to (viii).

[0226] (v) A dataset in which the high tube voltage image x_high separated from the reference material image a_high derived from a high dose is matched as the target to the high tube voltage image x_low separated from the reference material image a_low derived from a low dose. (vi) A dataset in which the high tube voltage image x_high separated from the reference material image b_high derived from a high dose is matched as the target to the high tube voltage image x_low separated from the reference material image b_low derived from a low dose. (vii) A dataset in which the low tube voltage image y_low separated from the reference material image a_low derived from a low dose is matched as the target with the low tube voltage image y_high separated from the reference material image a_high derived from a high dose. (viii) A dataset in which the low tube voltage image y_low separated from the reference material image b_low derived from a low dose is matched as the target with the low tube voltage image y_high separated from the reference material image b_high derived from a high dose.

[0227] Next, the machine learning function 220 trains a third trained model MDLκ using the third training data set, and trains a fourth trained model MDLλ using the fourth training data set (step S418).

[0228] For example, the machine learning function 220 inputs the high-tube voltage image x_low of the third training data set into the third trained model MDLκ. In response to the input of the high-tube voltage image x_low, the machine learning function 220 calculates the difference between the image output by the third trained model MDLκ and the target high-tube voltage image x_high. Then, the machine learning function 220 adjusts the parameters (weighting coefficients and bias components) of the third trained model MDLκ by performing error backpropagation using a stochastic gradient descent method or the like based on the difference.

[0229] Furthermore, the machine learning function 220 inputs the low tube voltage image y_low of the fourth training data set into the fourth trained model MDLλ. In response to the input of the low tube voltage image y_low, the machine learning function 220 calculates the difference between the image output by the fourth trained model MDLλ and the target low tube voltage image y_high. Then, the machine learning function 220 adjusts the parameters (weighting coefficients and bias components) of the fourth trained model MDLλ by performing error backpropagation using a stochastic gradient descent method or the like based on the difference.

[0230] In this way, the machine learning function 220 repeats the above process until the number of training iterations reaches a specified number, and when the number of iterations reaches the specified number, it stores model definition data defining the first trained model MDLα, the second trained model MDLβ, the third trained model MDLκ, and the fourth trained model MDLλ in the memory 208.

[0231] Next, the output control function 222 transmits model definition data that defines the first trained model MDLα, the second trained model MDLβ, the third trained model MDLκ, and the fourth trained model MDLλ trained by the machine learning function 220 to each X-ray CT device 100 via the communication interface 202 (step S420). This ends the processing of this flowchart.

[0232] According to the fifth embodiment described above, not only the signal image s and / or the noise image n are denoised, but also the reference material image is denoised. By performing denoising twice in this way, the accuracy of noise reduction in spectral imaging can be further improved.

[0233] (Sixth embodiment) The sixth embodiment will be described below. The sixth embodiment differs from the first to fifth embodiments in that both the high tube voltage image x and the low tube voltage image y are denoised using either the third trained model MDLκ or the fourth trained model MDLλ. The following description will focus on the differences from the first to fifth embodiments, and will omit a description of the points in common with the first to fifth embodiments. Note that in the description of the sixth embodiment, the same parts as those in the first to fifth embodiments will be denoted by the same reference numerals.

[0234] FIG. 18 is a diagram schematically illustrating the flow of a series of processes performed by the learning device 200 according to the sixth embodiment.

[0235] The machine learning function 220 according to the sixth embodiment generates a third training data set for training the third trained model MDLκ in step S416. For example, the machine learning function 220 generates a data set including the above items (v) to (viii) as the third training data set.

[0236] In step S418, the machine learning function 220 trains a third trained model MDLκ using a third training data set including (v) to (viii).

[0237] According to the sixth embodiment described above, the X-ray CT device 100 separates each of the reference material image a and the reference material image b into a high-tube voltage image x and a low-tube voltage image y by image conversion. The X-ray CT device 100 inputs the high-tube voltage image x and the low-tube voltage image y to a third trained model MDLκ trained based on a third training data set, and acquires a denoised high-tube voltage image x and a denoised low-tube voltage image y from the third trained model MDLκ.

[0238] In the fifth embodiment described above, the third training data set includes only the data sets (v) to (vi) derived from the high tube voltage image x. In contrast, in the sixth embodiment, the third training data set includes not only the data sets (v) to (vi) derived from the high tube voltage image x, but also the data sets (vii) to (viii) derived from the low tube voltage image y. By using the third trained model MDLκ trained using such a third training data set alone, it is possible to denoise not only the high tube voltage image x but also the low tube voltage image y.

[0239] In the third embodiment described above, the third trained model MDLκ is used alone to denoise both the high-tube voltage image x and the low-tube voltage image y, but this is not limited to this. For example, the fourth trained model MDLλ may be used alone to denoise both the high-tube voltage image x and the low-tube voltage image y. In this case, a fourth training data set including (v) to (viii) is used to train the fourth trained model MDLλ.

[0240] (Seventh embodiment) The seventh embodiment will be described below. The seventh embodiment differs from the first to sixth embodiments in that either the high tube voltage image x or the low tube voltage image y is denoised using either the third trained model MDLκ or the fourth trained model MDLλ. The following description will focus on the differences from the first to sixth embodiments, and will omit a description of the points in common with the first to sixth embodiments. Note that in the description of the seventh embodiment, the same parts as those in the first to sixth embodiments will be denoted by the same reference numerals.

[0241] FIG. 19 is a diagram schematically illustrating the flow of a series of processes performed by the learning device 200 according to the seventh embodiment.

[0242] In the seventh embodiment, when the conversion function 218 separates the four types of reference material images into high-tube voltage images x and low-tube voltage images y by image conversion as processing in S414, the machine learning function 220 determines whether the denoising request image is the high-tube voltage image x or the low-tube voltage image y (step S422).

[0243] In the seventh embodiment, it is assumed that the denoising request image is a high tube voltage image x.

[0244] If the denoising request image is high tube voltage image x, in the process of S416, the machine learning function 220 extracts only high tube voltage image x from the multiple high tube voltage images x and low tube voltage images y, and generates a third training data set that includes only high tube voltage image x. For example, the machine learning function 220 generates a third training data set that includes the above-mentioned (v) to (vi).

[0245] In the process of S418, the machine learning function 220 trains a third trained model MDLκ using a third training dataset including (v) to (vi).

[0246] According to the seventh embodiment described above, the X-ray CT device 100 separates each of the reference material image a and the reference material image b into a high-tube voltage image x and a low-tube voltage image y by image transformation. The X-ray CT device 100 inputs the high-tube voltage image x to a third trained model MDLκ trained based on a third training data set, and acquires a denoised high-tube voltage image x from the third trained model MDLκ. The X-ray CT device 100 then combines the denoised high-tube voltage image x and the low-tube voltage image y by inverse image transformation to generate a denoised reference material image. In this way, a denoised reference material image can be generated by denoising only one of the high-tube voltage image x and the low-tube voltage image y, rather than denoising both of them.

[0247] In the seventh embodiment described above, the third trained model MDLκ is used alone to denoise the high-tube voltage image x, but this is not limiting. For example, the fourth trained model MDLλ may be used alone to denoise the high-tube voltage image x.

[0248] (Other embodiments) Other embodiments will be described below. In the above-described embodiment, the processing circuitry 150 of the console device 140 of the X-ray CT apparatus 100 has been described as having a system control function 151, a preprocessing function 152, a reconstruction function 153, an image processing function 154, a conversion function 155, a denoising function 156, and an output control function 157, but this is not limited to this. For example, among these multiple functions, in particular, some or all of the preprocessing function 152, the reconstruction function 153, the image processing function 154, the conversion function 155, and the denoising function 156 may be provided by the learning device 200 or another external device.

[0249] Furthermore, the processing circuitry 210 of the learning device 200 has been described as including the acquisition function 212, the generation function 214, the reconstruction function 216, the conversion function 218, the machine learning function 220, and the output control function 222, but is not limited to this. For example, among these multiple functions, particularly some or all of the generation function 214, the reconstruction function 216, the conversion function 218, and the machine learning function 220 may be provided by the processing circuitry 150 of the X-ray CT apparatus 100.

[0250] It is also possible to use a machine learning model MDL trained on projection data acquired by a different acquisition method. For example, a single machine learning model MDL trained on monochromatic 70 keV images created from projection data acquired by a DECT device can be applied to three reference material images of iodine, water, and bone generated by a PCCT device.

[0251] It is also possible to use a machine learning model MDL trained on projection data collected by a different detector. For example, a machine learning model MDL trained on counting images generated from projection data collected by an EID (Energy Integral Detector) may be applied to counting images created from projection data collected by a PCCT device.

[0252] Although several embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0253] 100...X-ray CT device, 110...gantry device, 130...bed device, 140...console device, 141...memory, 142...display, 143...input interface, 144...communication interface, 145...speaker, 150...processing circuit, 151...system control function, 152...preprocessing function, 153...reconstruction function, 154...image processing function, 155...conversion function, 156...denoising function, 157...output control function, 200...learning device, 202...communication interface, 204...input interface, 206...output interface, 208...memory, 210...processing circuit, 212...acquisition function, 214...generation function, 216...reconstruction function, 218...conversion function, 220...machine learning function, 222...output control function

Claims

1. acquiring projection data of X-rays irradiated onto a subject; generating a medical image from the projection data by a reconstruction process; separating the medical image into a signal image having more signal components than noise components and a noise image having more noise components than signal components by a whitening transformation; denoising at least one of the signal image and the noise image using a first trained model, which is a machine learning model trained based on a training dataset in which a first output image having less noise than the first input image is associated as a target with respect to the first input image; When both the signal image and the noise image are denoised, combining the denoised signal image and the denoised noise image by an inverse whitening transform to generate a denoised medical image, which is the denoised medical image; When one of the signal image and the noise image is denoised, combining the denoised one of the signal image and the noise image with the other undenoised one of the signal image and the noise image by the inverse whitening transformation to generate the denoised medical image; A medical information processing method including:

2. generating a plurality of different types of medical images based on attenuation coefficients of a plurality of predetermined reference materials; further comprising separating each of the plurality of medical images into the signal image and the noise image by the whitening transformation. The medical information processing method according to claim 1 .

3. The plurality of medical images include at least any of a reference material image, a virtual monochromatic X-ray image, a virtual plain image, an iodine map, an effective atomic number image, an electron density image, an image reconstructed for each energy bin in spectral imaging, an image reconstructed under a normal resolution mode in which the resolution is equal to or less than a threshold, or an image reconstructed under a super high resolution mode in which the resolution exceeds the threshold. The medical information processing method according to claim 1 or 2.

4. separating the medical image by image conversion into a high-tube-voltage medical image, which is the medical image reconstructed from the projection data in which the tube voltage, which is the voltage of an X-ray tube that generates the X-rays, is equal to or greater than a threshold, and a low-tube-voltage medical image, which is the medical image reconstructed from the projection data in which the tube voltage is less than the threshold; denoising at least one of the high-tube voltage medical image and the low-tube voltage medical image using a second trained model, which is a machine learning model trained based on a training data set in which a second output image having less noise than a second input image is associated as a target with respect to the second input image; When both the high-tube voltage medical image and the low-tube voltage medical image are denoised, combining the denoised high-tube voltage medical image and the denoised low-tube voltage medical image by inverse image transformation to generate the denoised medical image; When one of the high tube voltage medical image and the low tube voltage medical image is denoised, the denoised one of the high tube voltage medical image and the low tube voltage medical image is combined with the other undenoised image by the inverse image conversion to generate the denoised medical image. The medical information processing method according to claim 1 or 2.

5. Separating the denoised medical image into the signal image and the noise image by the whitening transform; denoising at least one of the signal image and the noise image using the first trained model; When both the signal image and the noise image are denoised, combining the denoised signal image and the denoised noise image by the inverse whitening transform to generate a double denoised medical image, which is the denoised medical image; When one of the signal image and the noise image is denoised, combining the denoised one of the signal image and the noise image with the other undenoised one of the signal image and the noise image by the inverse whitening transform to generate the double denoised medical image. The medical information processing method according to claim 4.

6. an acquisition unit that acquires projection data of X-rays irradiated onto a subject; a reconstruction unit that generates a medical image from the projection data by reconstruction processing; a conversion unit that separates the medical image into a signal image having more signal components than noise components and a noise image having more noise components than signal components by whitening conversion; a denoising processing unit that denoises at least one of the signal image and the noise image using a first trained model, which is a machine learning model trained based on a training data set in which a first output image having less noise than the first input image is associated as a target with respect to a first input image; The conversion unit When both the signal image and the noise image are denoised, combining the denoised signal image and the denoised noise image by an inverse whitening transform to generate a denoised medical image, which is the medical image that has been denoised; When one of the signal image and the noise image is denoised, the denoised one of the signal image and the noise image is combined with the other undenoised one of the signal image and the noise image by the inverse whitening transformation to generate the denoised medical image. Medical information processing equipment.

7. A program to be executed by a computer, acquiring projection data of X-rays irradiated onto a subject; generating a medical image from the projection data by a reconstruction process; separating the medical image into a signal image having more signal components than noise components and a noise image having more noise components than signal components by a whitening transformation; denoising at least one of the signal image and the noise image using a first trained model, which is a machine learning model trained based on a training dataset in which a first output image having less noise than the first input image is associated as a target with respect to the first input image; When both the signal image and the noise image are denoised, combining the denoised signal image and the denoised noise image by an inverse whitening transform to generate a denoised medical image, which is the denoised medical image; When one of the signal image and the noise image is denoised, combining the denoised one of the signal image and the noise image with the other undenoised one of the signal image and the noise image by the inverse whitening transformation to generate the denoised medical image; Programs including.

Images