JP2020119429A - Medical image processing device, medical image processing method and program - Google Patents

Abstract

To improve accuracy of denoising processing.SOLUTION: A medical image processing device of an embodiment comprises: a setting unit; a denoising image generation unit; a differential image generation unit; and a determination unit. The setting unit is configured to set a plurality of denoising strength indicative of a degree of removing a noise component from a medical image, and the denoising image generation unit is configured to remove the noise component from the medical image on the basis of each of the plurality of denoising strength set by the setting unit, and generate a plurality of denoising images. The differential image generation unit is configured to generate a plurality of differential images serving a differential between the medical image and each of the plurality of denoising images generated by the denoising image generation unit. The determination unit is configured to determine an optimal value of the denoising strength on the basis of the plurality of differential images generated by the differential image generation unit.SELECTED DRAWING: Figure 4

Description

Embodiments of the present invention relate to a medical image processing apparatus, a medical image processing method, and a program.

2. Description of the Related Art Conventionally, in medical devices used for image diagnosis and the like, various studies have been conducted for the purpose of improving image quality. For example, there is known a technique of obtaining an image in which noise components are removed or suppressed by performing denoising processing on an image including noise components.

If the noise component locally remains in the image, the noise component may appear as a structure. Therefore, it is necessary to set the degree of noise removal so that the noise component does not remain. However, if the degree of noise removal is excessively increased, the image may be blurred.

U.S. Patent Application Publication No. 2004/0258325

The problem to be solved by the present invention is to improve the accuracy of denoising processing.

The medical image processing apparatus according to the embodiment includes a setting unit, a denoising image generation unit, a difference image generation unit, and a determination unit. The setting unit sets a plurality of denoising strengths indicating the degree of removing noise components from the medical image. The denoising image generation unit removes a noise component from the medical image based on each of the plurality of denoising intensities set by the setting unit to generate a plurality of denoising images. The difference image generation unit generates a plurality of difference images that are differences between the medical image and each of the plurality of denoised images generated by the denoised image generation unit. The determining unit determines an optimum value of the denoising intensity based on the plurality of difference images generated by the difference image generating unit.

The figure which shows an example of a structure of the medical image processing system 1 containing the medical image processing apparatus 200 which concerns on 1st Embodiment. The figure which shows an example of the medical image generation apparatus 100 which concerns on 1st Embodiment. FIG. 3 is a diagram showing an example of an arrangement of receiving coils 108 included in the medical image generating apparatus 100 according to the first embodiment. The figure which shows an example of the medical image processing apparatus 200 which concerns on 1st Embodiment. The figure which shows an example of a structure of the denoising model MDL1 which concerns on 1st Embodiment. The figure which shows an example of the activation function of the activation layer 330 which concerns on 1st Embodiment. FIG. 3 is a diagram showing an example of a configuration for generating a denoised image from an original image according to the first embodiment. FIG. 6 is a diagram showing an example of a configuration for generating a difference image according to the first embodiment. The figure which shows an example of the relationship of the difference image and control function which concern on 1st Embodiment. The figure which shows an example of the relationship between the control function which concerns on 1st Embodiment, and the skewness of a difference image. 3 is a flowchart showing a series of flows of learning processing of the processing circuit 210 according to the first embodiment. 3 is a flowchart showing a series of flow of image processing of the processing circuit 210 according to the first embodiment. The flowchart which shows the series of flows of the image processing of the processing circuit 210 which concerns on 2nd Embodiment. The figure which shows an example of the histogram of the signal value of the difference image which concerns on 2nd Embodiment. The figure which shows an example of the histogram of the signal value of the difference image which concerns on 2nd Embodiment. The figure which shows an example of the histogram of the signal value of the difference image which concerns on 2nd Embodiment. The figure which shows an example of the histogram of the signal value of the difference image which concerns on 2nd Embodiment. The figure which shows the medical image generation apparatus 100 which concerns on the modification of 1st and 2nd embodiment.

Hereinafter, embodiments of a medical image processing apparatus, a medical image processing method, and a program will be described in detail with reference to the drawings.

(First embodiment)
FIG. 1 is a diagram showing an example of the configuration of a medical image processing system 1 including a medical image processing apparatus 200 according to the first embodiment. As shown in FIG. 1, the medical image processing system 1 includes, for example, a medical image generating apparatus 100 and a medical image processing apparatus 200. The medical image generation apparatus 100 and the medical image processing apparatus 200 are connected to each other via a network NW. The network NW includes, for example, a WAN (Wide Area Network), a LAN (Local Area Network), the Internet, a dedicated line, a wireless base station, a provider, and the like.

The medical image generating apparatus 100 includes, for example, an MRI (Magnetic Resonance Imaging) apparatus and a CT (Computed Tomography) apparatus. The MRI apparatus, for example, applies a magnetic field to a subject (for example, a human body), receives an electromagnetic wave generated from hydrogen atomic nuclei in the subject by a nuclear magnetic resonance phenomenon using a coil, and receives a signal based on the received electromagnetic wave. To generate a medical image (MR image). The CT device irradiates the subject with X-rays from an X-ray tube that rotates around the subject, detects the X-rays that have passed through the subject, and reconstructs a signal based on the detected X-rays. As a result, a medical image (CT image) is generated. In the following description, the medical image generating apparatus 100 will be described as an MRI apparatus as an example.

The medical image processing apparatus 200 is realized by one or a plurality of processors. For example, the medical image processing apparatus 200 may be a computer included in a cloud computing system, or a computer that operates independently without depending on other devices (stand-alone computer).

[Example of Configuration of Medical Image Generating Device (MRI Device)]
FIG. 2 is a diagram illustrating an example of the medical image generation apparatus 100 according to the first embodiment. As shown in FIG. 2, the medical image generating apparatus 100 includes, for example, a static magnetic field magnet 101, a gradient magnetic field coil 102, a gradient magnetic field power source 103, a bed 104, a bed control circuit 105, a transmission coil 106, and a transmission coil 106. The circuit 107, the receiving coil 108, the receiving circuit 109, the sequence control circuit 110, and the console device 120 are provided.

The static magnetic field magnet 101 is a magnet formed in a hollow and substantially cylindrical shape. The static magnetic field magnet 101 generates a uniform static magnetic field in the internal space. The static magnetic field magnet 101 is, for example, a permanent magnet or a superconducting magnet. The gradient magnetic field coil 102 is a coil formed in a hollow and substantially cylindrical shape, and is arranged inside the static magnetic field magnet 101. The gradient magnetic field coil 102 is formed by combining three coils corresponding to x, y, and z axes that are orthogonal to each other. The z-axis direction represents the longitudinal direction of the top plate 104a of the bed 104, and the x-axis direction is orthogonal to the z-axis direction and parallel to the floor of the room where the medical image generating apparatus 100 is installed. And the y-axis direction represents the axial direction that is the direction perpendicular to the floor surface. The three coils corresponding to the respective axial directions individually receive currents from the gradient magnetic field power source 103 to generate gradient magnetic fields whose magnetic field strength changes along the x, y, and z axes. The z-axis direction is the same as the static magnetic field.

The gradient magnetic field power supply 103 supplies a current to the gradient magnetic field coil 102. Here, the gradient magnetic fields of the x, y, and z axes generated by the gradient magnetic field coil 102 respectively correspond to the slice selection gradient magnetic field Gs, the phase encoding gradient magnetic field Ge, and the readout gradient magnetic field Gr, respectively. .. The slice selection gradient magnetic field Gs is used to arbitrarily determine an imaging cross section. The phase encoding gradient magnetic field Ge is used to change the phase of the magnetic resonance signal according to the spatial position. The readout gradient magnetic field Gr is used to change the frequency of the magnetic resonance signal according to the spatial position.

The bed 104 includes a top plate 104a on which the subject OB is placed. Under the control of the bed control circuit 105, the bed 104 inserts the top plate 104a into the imaging opening with the subject OB placed thereon. Usually, the bed 104 is installed so that the longitudinal direction thereof is parallel to the central axis of the static magnetic field magnet 101. Under the control of the console device 120, the bed control circuit 105 drives the bed 104 to move the table 104a in the longitudinal direction and the vertical direction.

The transmission coil 106 is arranged inside the gradient magnetic field coil 102. The transmission coil 106 receives supply of an RF (Radio Frequency) pulse from the transmission circuit 107 and generates a high frequency magnetic field. The transmission circuit 107 supplies the transmission coil 106 with an RF pulse corresponding to the Larmor frequency determined by the type of the target atomic nucleus and the strength of the magnetic field.

The receiving coil 108 is arranged inside the gradient magnetic field coil 102. The receiving coil 108 receives the magnetic resonance signal emitted from the subject OB under the influence of the high frequency magnetic field. The magnetic resonance signal includes, for example, a signal intensity component and a phase component. Upon receiving the magnetic resonance signal, the receiving coil 108 outputs the received magnetic resonance signal to the receiving circuit 109. In the first embodiment, the receiving coil 108 is a coil array having a plurality of receiving coils. Hereinafter, each coil forming the coil array will be described as a coil element. Further, although the transmission coil 106 and the reception coil 108 have been described separately, for example, one RF (Radio Frequency) coil may have both a transmission function and a reception function. Further, the transmission coil 106 and the reception coil 108 can be in the form of a whole-body coil housed in a stand including the static magnetic field magnet 101 or a local coil arranged near the body surface of the subject OB. The form does not matter in the embodiment.

FIG. 3 is a diagram showing an example of the arrangement of the receiving coils 108 included in the medical image generating apparatus 100 according to the first embodiment. FIG. 3 shows an example in which the receiving coil 108 includes eight coil elements 108a to 108h. These coil elements 108a to 108h are arranged so as to surround the subject OB. Each of the coil elements 108 a to 108 h receives the magnetic resonance signal emitted from the subject OB and outputs it to the receiving circuit 109.

The reception circuit 109 generates magnetic resonance data for each magnetic resonance signal output by the coil elements 108a to 108h. For example, the receiving circuit 109 generates a magnetic resonance data set that is a digital signal by performing analog-digital conversion on each of the magnetic resonance signals that are analog signals output by the coil elements 108a to 108h. The receiving circuit 109 also transmits the generated magnetic resonance data set to the sequence control circuit 110. The receiving circuit 109 may be provided on the gantry device side including the static magnetic field magnet 101, the gradient magnetic field coil 102, and the like.

The sequence control circuit 110 drives the gradient magnetic field power supply 103, the transmission circuit 107, and the reception circuit 109 based on the sequence information output from the console device 120 to image the subject OB. The sequence information is information that defines the procedure for performing the imaging process. The sequence information includes the intensity of the current supplied by the gradient magnetic field power supply 103 to the gradient magnetic field coil 102 and the timing of supplying the current, the intensity of the RF pulse transmitted by the transmission circuit 107 to the transmission coil 106, the timing of applying the RF pulse, and the reception. It includes information defining the timing at which the circuit 109 detects the magnetic resonance signal.

When the sequence control circuit 110 drives the gradient magnetic field power supply 103, the transmission circuit 107, and the reception circuit 109 to image the subject OB and receives the magnetic resonance data from the reception circuit 109, the received magnetic resonance data is sent to the console device. Transfer to 120.

The console device 120 controls the entire medical image generating apparatus 100 and collects magnetic resonance data. For example, the console device 120 includes a communication interface 122, an input interface 124, a display 126, a processing circuit 130, and a memory (storage) 150.

The communication interface 122 includes a communication interface such as a NIC (Network Interface Card). The communication interface 122 communicates with the medical image processing apparatus 200 via the network NW, and transmits/receives information to/from the medical image processing apparatus 200. The communication interface 122 outputs the received information to the processing circuit 130. Further, the communication interface 122 may be controlled by the processing circuit 130 and may transmit information to another device connected via the network NW.

The input interface 124 is an interface that receives various input operations from an operator. When receiving the input operation, the input interface 124 converts the received input operation into an electric signal and outputs the electric signal to the processing circuit 130. For example, the input interface 124 is realized by a mouse, keyboard, trackball, switch, button, joystick, touch panel, or the like. Further, the input interface 124 may be realized by, for example, a user interface such as a microphone that receives a voice input. When the input interface 124 is a touch panel, the display 126 described below may be formed integrally with the input interface 124.

The display 126 displays various information. For example, the display 126 displays an image generated by the processing circuit 130, a GUI (Graphical User Interface) for receiving various input operations from an operator, and the like. For example, the display 126 is an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube) display, an organic EL (Electroluminescence) display, or the like.

The processing circuit 130 includes, for example, an acquisition function 132, a generation function 134, a communication control function 136, and a display control function 138. These functions (components) are realized by, for example, a processor (or processor circuit) such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) executing a program (software) stored in the memory 150. To be done. Further, some or all of the functions of the processing circuit 130 are realized by hardware (circuit section: circuitry) such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), and an FPGA (Field-Programmable Gate Array). It may be realized by the cooperation of software and hardware. Further, the above program may be stored in the memory 150 in advance, or may be stored in a removable storage medium such as a DVD or a CD-ROM, and the storage medium is attached to the drive device of the console device 120. Therefore, it may be installed in the memory 150 from the storage medium.

The memory 150 is realized by, for example, a RAM (Random Access Memory), a semiconductor memory element such as a flash memory, a hard disk, an optical disk, or the like. These non-transitory storage media may be realized by other storage devices such as NAS (Network Attached Storage) and external storage server devices connected via the network NW. Further, these non-transitory storage media may be realized by a storage device such as a ROM (Read Only Memory) or a register.

The acquisition function 132 acquires magnetic resonance data from the sequence control circuit 110. As described above, the magnetic resonance data is data obtained by analog-to-digital conversion of the electromagnetic wave signal (magnetic resonance signal) generated in the object OB by the nuclear magnetic resonance phenomenon. Data obtained by arranging the magnetic resonance data according to the amount of phase encoding and the amount of frequency encoding given by the gradient magnetic field is also referred to as k-space data. The k-space represents a frequency space in which the one-dimensional waveform is collected when the magnetic resonance signal is repeatedly collected by the receiving coil 108 as the one-dimensional waveform.

The generation function 134 performs a reconstruction process including a process such as a Fourier transform (for example, an inverse Fourier transform) on the k-space data acquired by the acquisition function 132 to obtain a medical image reconstructed from the k-space data. The MR image (hereinafter referred to as the original image) is generated. The generation function 134 performs reconstruction processing including processing such as Fourier transform (for example, inverse Fourier transform) on the magnetic resonance data for each magnetic resonance signal output by the coil elements 108a to 108h, thereby generating the original image. To generate. The generation function 134 generates an original image using a synthesis algorithm such as the sum of squares (SOS) method. When the medical image generation apparatus 100 is an apparatus that reconstructs an original image by parallel imaging, the generation function 134 may perform processing such as unfolding an image in the reconstruction processing.

When the original image is generated by the generation function 134 by the reconstruction function, the communication control function 136 causes the communication interface 122 to communicate with the medical image processing apparatus 200, and the medical image processing apparatus 200 of the communication partner reconstructs the original image. Send the image. Further, the communication control function 136 may cause the communication interface 122 to communicate with the medical image processing apparatus 200 and receive various kinds of information from the medical image processing apparatus 200 of the communication partner.

The display control function 138 causes the display 126 to display the medical image received from the medical image processing apparatus 200. Further, the display control function 138 may display the original image generated by the generation function 134 on the display 126.

[Example of configuration of medical image processing apparatus]
FIG. 4 is a diagram illustrating an example of the medical image processing apparatus 200 according to the first embodiment. The medical image processing apparatus 200 generates an image in which noise is removed or reduced by performing denoising processing on the original image received from the medical image generating apparatus 100 (hereinafter, denoising image). The medical image processing apparatus 200 also has a function of determining an optimum value of the degree of noise removal (hereinafter, denoising intensity) in the above-described denoising process. In the following, removing or reducing noise is simply referred to as removing noise. As shown in FIG. 4, the medical image processing apparatus 200 includes, for example, a communication interface 202, an input interface 204, a display 206, a processing circuit 210, and a memory 230.

The communication interface 202 includes, for example, a communication interface such as NIC. For example, the communication interface 202 communicates with the medical image generating apparatus 100 via the network NW, and receives the reconstructed original image from the medical image generating apparatus 100. The communication interface 202 outputs the received original image to the processing circuit 210. Further, the communication interface 202 may be controlled by the processing circuit 210 to transmit information to the medical image generating apparatus 100 and other devices connected via the network NW. The other device may be, for example, a terminal device that can be used by an image interpreter such as a doctor or a nurse.

The input interface 204 receives various input operations from the operator, converts the received input operations into electric signals, and outputs the electric signals to the processing circuit 210. For example, the input interface 204 is realized by a mouse, keyboard, trackball, switch, button, joystick, touch panel, or the like. Further, the input interface 204 may be realized by, for example, a user interface such as a microphone that receives a voice input. When the input interface 204 is a touch panel, the display 206 described later may be formed integrally with the input interface 204.

The display 206 displays various information. For example, the display 206 displays a denoising image generated by the processing circuit 210, a GUI for receiving various input operations from the operator, and the like. For example, the display 206 is an LCD, a CRT display, an organic EL display, or the like.

The processing circuit 210 includes, for example, an acquisition function 212, a derivation function 214, a parameter adjustment function 216, a denoising function 218, a difference image generation function 220, a denoising intensity determination function 222, an output control function 224, and a learning function. 226 and are executed. The derivation function 214 is an example of a “derivation unit”, the parameter adjustment function 216 is an example of a “setting unit”, the denoising function 218 is an example of a “Denoise image generation unit”, and the difference image generation function 220. Is an example of a “difference image generation unit”, and the denoising intensity determination function 222 is an example of a “determination unit”.

These functions (components) are realized, for example, by a processor (or processor circuit) such as a CPU or GPU executing a program (software) stored in the memory 230. Further, some or all of these plurality of functions may be realized by hardware (circuit section) such as LSI, ASIC, FPGA, or may be realized by cooperation of software and hardware. Good. Further, the above program may be stored in the memory 230 in advance, or may be stored in a removable storage medium such as a DVD or a CD-ROM, and the storage medium may be attached to the drive device of the medical image processing apparatus 200. Then, it may be installed from the storage medium to the memory 230.

The memory 230 is realized by, for example, a RAM, a semiconductor memory device such as a flash memory, a hard disk, an optical disk, or the like. These non-transitory storage media may be realized by another storage device connected via the network NW such as a NAS or an external storage server device. Further, these non-transitory storage media may be realized by a storage device such as a ROM or a register. In the memory 230, for example, denoising model information 232, original image information (hereinafter, original image information 234), denoising image information (hereinafter, denoising image information 236), and difference image information (to be described below, difference image). Information 238) and the like are stored.

The denoising model information 232 is information (program or data structure) defining a denoising model MDL1 described later. The denoising model MDL1 is a model learned to output an image from which noise is removed when a certain image is input. The denoising model MDL1 includes, for example, one or more DNNs (Deep Neural Networks(s)).

In the denoising model information 232, for example, neurons (units or nodes) included in each of the input layer, one or more hidden layers (intermediate layers), and the output layer that configure each DNN included in the denoising model MDL1 are mutually included. It includes connection information about how to connect, weight information about how many connection coefficients are given to the data input/output between the connected neurons, and the like. The connection information is, for example, information such as the number of neurons included in each layer, information specifying the type of neuron to which each neuron is connected, an activation function that realizes each neuron, and a gate provided between neurons in the hidden layer. including. The activation function that realizes the neuron may be, for example, a function (ReLU (Rectified Linear Unit) function, ELU (Exponential Linear Units) function, clipping function) that switches operation according to an input code, or a sigmoid function or , A step function, a hyperpolic tangent function, or an identity function. The gate selectively passes or weights the data transmitted between the neurons, depending on, for example, the value returned by the activation function (eg, 1 or 0). The coupling coefficient is a parameter of the activation function. For example, in a hidden layer of a neural network, when data is output from a neuron of a certain layer to a neuron of a deeper layer, a weight given to the output data. including. Further, the coupling coefficient may include a bias component unique to each layer.

For example, the denoising model MDL1 may be realized by a CNN (Convolutional Neural Network). The CNN may be configured in multiple layers such as a convolutional layer and an activation layer.

FIG. 5 is a diagram showing an example of the configuration of the denoising model MDL1. As shown in FIG. 5, the denoising model MDL1 may include, for example, an input layer 310, one or more convolutional layers 320, one or more activation layers 330, and an output layer 340.

For example, when the original image is a matrix having elements corresponding to each pixel, the matrix corresponding to the original image is input to the input layer 310. The input layer 310 adds a bias component to the input matrix at appropriate times, and outputs the matrix to the convolutional layer 320 in the subsequent stage.

The convolutional layer 320 repeats a product-sum operation while sliding a linear transformation matrix called a filter or a kernel by a certain predetermined stride amount with respect to the input matrix, and transforms the input matrix into a linear transformation matrix. A matrix including a plurality of elements in which the sum of products is associated as an element value is generated. At this time, the convolutional layer 320 performs padding (for example, zero padding) that interpolates an element of an arbitrary value around the input matrix, and inputs the matrix input to the convolutional layer 320 to the input layer 310. It may be converted into a matrix having the same number of rows and columns as the matrix of the original image. Then, the convolutional layer 320 outputs the generated matrix to the activation layer 330.

The activation layer 330 performs activation function calculation processing on each element of the matrix input from the convolutional layer 320, and outputs the calculated matrix to the subsequent layer.

FIG. 6 is a diagram showing an example of the activation function of the activation layer 330. As shown in FIG. 6, the activation function of the activation layer 330 may be a Soft-Shrinkage function. The Soft-Shrinkage function is represented by the following mathematical expression (1), for example. The activation function of the activation layer 330 may be a Hard-Shrinkage function instead of the Soft-Shrinkage function.

The Soft-Shrinkage function or Hard-Shrinkage function outputs zero when the element value x that is an input value is within a range of a predetermined positive/negative threshold ±T centered on zero, and outputs the element value that is an input value. When x exceeds the threshold value T or is less than the threshold value T, it is a function that outputs a value proportional to the element value x. By applying the Soft-Shrinkage function or the Hard-Shrinkage function to the activation function of the activation layer 330, an image signal whose amplitude is smaller than the threshold value T, that is, a weak image signal with high probability as noise is activated. It can be zero at the output of the digitization function.

The threshold value T is a parameter that changes according to the level of noise (signal strength or signal power) included in the input image, and is represented by the following mathematical expression (2), for example.

G in the equation (2) represents the level of noise included in the input image, and is a signal (hereinafter, a control signal) that controls the value of the threshold T. The control signal G increases as the noise level included in the input image increases, and the control signal G decreases as the noise level included in the input image decreases. The level of noise included in the input image may be determined by the magnitude of the signal-noise ratio (SNR) of the input image.

Α in the equation (2) is a weighting coefficient by which the noise level is multiplied. The weighting factor α is determined by machine learning. By varying the threshold value T as shown in Expression (2), the activation function of the activation layer 330 can be made an activation function that reacts to a signal having a certain intensity or more. As a result, noise can be accurately removed from the original image even if the signal strength of the noise included in the original image varies.

The output layer 340 outputs the matrix processed by the convolutional layer 320 and the activation layer 330 in the previous stage.

The denoising model MDL1 illustrated in FIG. 5 is merely an example, and a pooling layer or the like may be included, for example. The pooling layer compresses (reduces) the number of dimensions of the matrix by replacing the element values of the input matrix with a representative value such as the average value or the maximum value of all element values included in the matrix. The pooling layer outputs a matrix in which the number of dimensions is compressed to the subsequent layer.

The acquisition function 212 causes the communication interface 202 to communicate with the medical image generation apparatus 100, and acquires the reconstructed original image from the medical image generation apparatus 100 of the communication partner. The acquisition function 212 stores the acquired original image in the memory 230 as the original image information 234.

The derivation function 214 derives an index value regarding noise included in the original image based on the original image acquired by the acquisition function 212. For example, the derivation function 214 derives the SNR as an index value regarding noise. The SNR is an index value obtained by dividing the image signal intensity by the noise signal intensity. Details of the method of deriving the SNR will be described later. The derivation function 214 may derive data related to SNR, such as the signal strength of the original image, the magnitude of noise, and the gain during normalization processing. The derivation function 214 may derive an index value regarding noise included in the original image based on the incidental information of the original image.

The parameter adjustment function 216 adjusts the control function G, which is an internal parameter of the denoising model MDL1, based on the SNR derived by the derivation function 214. That is, the parameter adjustment function 216 adjusts the control function G based on the SNR that is the index value derived by the derivation function 214. The parameter adjustment function 216 sets the control function G higher as the signal-to-noise ratio becomes smaller, and sets the control function G lower as the signal-to-noise ratio becomes larger.

Here, the parameter adjustment function 216 sets a plurality of control functions G based on the SNR derived by the derivation function 214 in order to adjust the parameters more accurately. For example, the parameter adjusting function 216 sets a rough range of the control function G based on the magnitude of the SNR derived by the deriving function 214, and sets a plurality of control functions G included in the set range of the control function G. Set. The parameter adjustment function 216 inputs the plurality of control functions G set for at least one node of the activation layer 330.

The denoising function 218 uses the denoising model MDL1 indicated by the denoising model information 232 to generate a denoising image in which noise is removed from the original image obtained by the obtaining function 212. The denoising function 218 acquires the matrix output from the output layer 340 of the denoising model MDL1 as a denoising image in which noise is removed from the original image. The denoising function 218 stores the generated denoising image in the memory 230 as denoising image information 236.

FIG. 7 is a diagram showing an example of a configuration for generating a denoised image from an original image according to the first embodiment. When generating a denoising image, the denoising function 218 can adjust the value of the internal parameter set in the denoising model MDL1 based on the parameter adjusted by the parameter adjusting function 216. That is, the denoising function 218 can change the denoising intensity of the original image by the plurality of control functions G adjusted based on the SNR of the original image.

In FIG. 7, denoising processing (that is, denoising processing with four denoising intensities) based on the four control functions G1 to G4 set by the parameter adjusting function 216 is performed to generate four denoising images B1 to B4, respectively. It shows an example. Here, the control functions G1 to G4 are G1<G2<G3<G4. That is, the denoising intensity increases from the control function G1 to G4.

As shown in FIG. 7, the denoising function 218 generates a denoising image B1 by performing denoising processing of the control function G1 on the original image A. The denoising function 218 performs denoising processing of the control function G2 on the original image A to generate a denoising image B2. The denoising function 218 performs denoising processing of the control function G3 on the original image A to generate a denoising image B3. The denoising function 218 performs denoising processing of the control function G4 on the original image A to generate a denoising image B4. In this way, the denoising function 218 removes the noise component from the original image A based on each of the plurality of denoising strengths set by the parameter adjusting function 216 to generate a plurality of denoising images.

Note that the denoising model MDL1 may be realized as a part of the denoising function 218 by the processor executing the denoising model MDL1, for example. The denoising function 218 is not limited to the one using a neural network. The denoising function 218 may remove or reduce noise using a model generated by any machine learning, such as a logistic regression analysis, a decision tree analysis, a support vector machine-based technique, or the like. Further, the denoising function 218 may be one that removes or reduces noise by wavelet transform on the original image.

The difference image generation function 220 generates an image (hereinafter, a difference image) that is a difference between the original image acquired by the acquisition function 212 and the denoising image generated by the denoising function 218. For example, the difference image generation function 220 generates a difference image by subtracting the denoising image from the original image for the signal component of each pixel. The difference image generation function 220 stores the generated difference image in the memory 230 as difference image information 238.

FIG. 8 is a diagram showing an example of a configuration for generating a difference image according to the first embodiment. As shown in FIG. 8, when the four denoising images B1 to B4 are generated by the denoising function 218, the difference image generating function 220 subtracts the denoising image B1 from the original image A to generate the difference image C1. To do. The difference image generation function 220 generates a difference image C2 by subtracting the denoising image B2 from the original image A. The difference image generation function 220 generates a difference image C3 by subtracting the denoising image B3 from the original image A. The difference image generation function 220 generates a difference image C4 by subtracting the denoising image B4 from the original image A.

Basically, in the denoising process, as the control function G becomes larger (as the denoising intensity becomes higher), more noise components are removed. However, if the denoising intensity becomes too high, the components of the structure to be imaged, which should not be originally removed (hereinafter, structure components), will be removed. FIG. 9 is a diagram showing the relationship between the control function and the difference image according to the first embodiment. Here, suppose that an optimal control function (optimal denoising intensity) exists between the control function G2 and the control function G3. The optimum control function is a control function that can remove the noise component to the maximum without removing the structure component.

As shown in FIG. 9, the difference image C1 between the original image A and the denoising image B1 obtained by the denoising process with the control function G1 (OP>G1) smaller than the optimum value OP includes only the noise component. Further, the difference image C2 between the original image A and the denoising image B2 obtained by the denoising process with the control function G2 (OP>G2>G1) smaller than the optimum value OP includes only the noise component. The noise component of the difference image C2 is larger than the noise component of the difference image C1.

As shown in FIG. 9, the difference image C3 between the original image A and the denoising image B3 obtained by the denoising process with the control function G3 (G3>OP) larger than the optimum value OP is added to the noise component. It contains structural components that should not be contained originally. Further, the difference image C4 between the original image A and the denoising image B4 obtained by the denoising process with the control function G4 (G4>G3) larger than the optimum value OP should not be originally included in addition to the noise component. Includes structural components. The structure component of the difference image C4 is larger than the structure component of the difference image C3. As described above, the difference image generation function 220 generates a plurality of difference images which are differences between the original image and each of the plurality of denoising images.

The denoising strength determination function 222 determines the optimum value OP of the denoising strength (for example, a control function) based on the difference image generated by the difference image generation function 220. For example, the denoising strength determination function 222 obtains, for each difference image, an index value having a correlation with the noise distribution. For example, the skewness (Skewness) is calculated for the luminance value of each pixel on the entire surface of the image, and the optimum value OP of the control function is determined by comparing the difference images.

FIG. 10 is a diagram showing the relationship between the control function and the skewness of the difference image according to the first embodiment. In the denoising process using the denoising model MDL1, basically, the noise component according to the Gaussian distribution is removed. Therefore, in the vicinity of the optimum value OP, it is assumed that the distortion factor of the difference image approaches 0 as a result of the noise component that follows this Gaussian distribution being appropriately removed over the entire image. That is, the skewness of the difference image tends to decrease near the optimum value OP. Therefore, for example, the denoising strength determination function 222 compares the respective skewnesses of the plurality of difference images with each other, and determines the value of the control function used to generate the difference image with the smallest skewness as the optimum value OP. To do.

As described above, the denoising strength determination function 222 determines the optimum value of the denoising strength by mutually comparing the index values having a correlation with the noise distribution of each of the plurality of difference images.

The denoising strength determination function 222 takes a difference between a plurality of difference images (for example, a difference between the difference image C1 and the difference image C2, a difference between the difference image C1 and the difference image C3, etc.), and calculates the difference image. The optimum value OP of the control function may be determined based on the index value having a correlation with the noise distribution. For example, the denoising strength determination function 222 determines a difference between a plurality of difference images, and determines the average value of the set of control functions that minimizes the standard deviation of the brightness values of the difference images as the optimum value OP of the control function. To do. The reason for determining the optimum value OP in this way is that the small standard deviation indicates that the change is small, and the effect of removing the noise component in the denoising process using the denoising model MDL1 is saturated (that is, This is because it is assumed that a noise component is removed to the maximum). Note that the denoising strength determination function 222 does not need to take the difference between all the denoised images, but may take only the difference between the denoised images whose control functions are close. As the index value having a correlation with the noise distribution, peak count, kurtosis, and average may be adopted in addition to the above-mentioned skewness and standard deviation when it can be determined that there is a correlation with the noise distribution. ..

When the optimum value OP is determined by the denoising intensity determination function 222, the output control function 224 causes the communication interface 202 to communicate with the medical image generating apparatus 100, and causes the medical image generating apparatus 100 of the communication partner to communicate the original image. The denoising image obtained by performing the denoising process with the optimum value OP is transmitted. Further, the output control function 224 may display the denoised image on the display 206.

The learning function 226 inputs an image to be given learning data (hereinafter, a learning image) to the denoising model MDL1, and the denoising image output by the denoising model MDL1 is an image to be teaching data (hereinafter, a teacher data). The denoising model MDL1 is learned so as to approach the image). For example, the teacher image may be an image whose SNR is increased by increasing the sampling frequency. The learning image may be an image obtained by adding known noise to the teacher image. The known noise may be Gaussian noise, for example.

For example, the learning function 226 uses the element values of the linear transformation matrix of the convolutional layer 320 and the nodes of the activation layer 330 so that the difference between the denoising image output by the denoising model MDL1 and the teacher image becomes small. Various parameters such as the weighting coefficient α of the activation function are learned by using a gradient method such as SGD (Stochastic Gradient Descent), Momentum SGD, AdaGrad, RMSprop, AdaDelta, and Adam (Adaptive moment estimation). Details of the learning process will be described later.

[Processing flow (learning process)]
The processing flow of the processing circuit 210 according to the first embodiment will be described below. The processing of the processing circuit 210 includes learning processing for learning the denoising model MDL1 and image processing for performing noise removal processing using the learned denoising model MDL1. In the following, first, the learning process of the processing circuit 210 will be described. FIG. 11 is a flowchart showing a series of flows of learning processing of the processing circuit 210 according to the first embodiment. The process of this flowchart is performed, for example, when the operator of the medical image processing apparatus 200 operates the input interface 204 to instruct to start the learning process.

First, the learning function 226 sets various parameters such as the control signal G for the denoising model MDL1 and the weighting coefficient α of the activation function of each node of the activation layer 330 to initial values (step S100). For example, the control signal G and the weighting factor α are set to 1.

Next, the learning function 226 inputs the learning image to the denoising model MDL1 and obtains the processing result thereof (step S102). The learning image is, for example, an image in which known noise is added to a teacher image that is a noise-free image or an image having a high SNR. The learning image is obtained, for example, by adding the noise and the pixel value of the teacher image.

For example, the learning function 226 inputs the pixel value of each pixel of the learning image to each input end of the input layer of the denoising model MDL1. This pixel value propagates through the nodes in the middle layer of the denoising model MDL1 while changing its value from the input layer to the output layer by the calculation of weighted addition, bias addition, and activation function processing. Go Then, the pixel value of the image processed by the denoising model MDL1 (hereinafter referred to as the processing result image) is output to the output end of the output layer as the processing result.

Next, the learning function 226 calculates a training error which is an error between the processing result image and the teacher image (step S104). The training error is, for example, the mean square error of each pixel of the processing result image and the teacher image, the sum of the square error of each pixel, or the like.

Next, the learning function 226 updates the internal parameter of the denoising model MDL1 so that the calculated training error becomes small by using, for example, the error back propagation method (step S106). Next, the learning function 226 performs verification using, for example, a correct image prepared separately from the teacher image, which is an image containing no noise or an image having a high SNR, and a verification image obtained by adding noise to this correct image. A generalization error which is an error between the processing result image when the image is input to the denoising model MDL1 and the correct image is calculated (step S108). Next, the learning function 226 determines whether the generalization error has reached the minimum value (step S110).

When the learning function 226 determines that the generalization error has not reached the minimum value, the learning function 226 repeats the processing from step S102 again. On the other hand, when the learning function 226 determines that the generalization error has reached the minimum value, the learning function 226 stores the internal parameter of the denoised model MDL1 after learning in the memory 230 (step S112). With the above, the processing of this flowchart is completed.

[Processing flow (image processing)]
Next, the image processing of the processing circuit 210 will be described. FIG. 12 is a flowchart showing a series of flow of image processing of the processing circuit 210 according to the first embodiment. The process of this flowchart is performed, for example, when the original image transmitted by the medical image generation apparatus 100 is acquired by the acquisition function 212.

First, the derivation function 214 derives the SNR of the original image acquired by the acquisition function 212 (step S200). For example, with respect to the original image acquired by the acquisition function 212, the derivation function 214 supplies an RF pulse to the transmission coil 106 with the gradient magnetic field coil 102 generating the gradient magnetic field, and the high frequency magnetic field is generated from the transmission coil 106. The original image obtained when output (hereinafter, referred to as an original image with RF pulse) and the transmission of the RF pulse to the transmission coil 106 without supplying the RF pulse to the transmission coil 106 in a state where the gradient magnetic field coil 102 generates the gradient magnetic field. The SNR is derived based on the pixel value difference from the original image (hereinafter referred to as the original image without RF pulse) obtained when the high-frequency magnetic field is not output from the coil 106.

Further, the derivation function 214, with respect to the original image acquired by the acquisition function 212, the difference between the pixel values of the two or more original images with RF pulses obtained when the high-frequency magnetic field is output toward the same object OB. The SNR of the original image may be derived based on At this time, the derivation function 214 may derive the SNR of the original image based on the difference between the pixel values of the two original images with RF pulses close to the central slice. For example, when the main scan is a sequence in which the scan is repeated 10 times, the central slice is the original image obtained by the 5th or 6th scan.

As described above, under the condition that the object OB is the same, by taking the difference between the two original images obtained by scanning a plurality of times, the magnetic resonance signal components are canceled each other, and the heat of the receiving system is reduced. The SNR can be calculated based on the random noise component caused by noise.

Next, the parameter adjustment function 216 sets a plurality of control functions G for the original image based on the SNR derived by the derivation function 214 (step S202). For example, the parameter adjustment function 216 sets the range of the control function G based on the magnitude of the SNR derived by the derivation function 214, and sets the plurality of control functions G within the set range of the control function G.

Next, the denoising function 218 uses the denoising model MDL1 in which the threshold T for each control function G set by the parameter adjusting function 216 is set as an internal parameter, performs denoising processing on the original image, and performs a plurality of denoising processes. A denoised image is generated (step S204).

Next, the difference image generation function 220 generates a plurality of difference images based on the original image acquired by the acquisition function 212 and the plurality of denoised images generated by the denoising function 218 (step S206). For example, the difference image generation function 220 generates a difference image by subtracting the denoising image from the original image for the signal component of each pixel.

Next, the denoising strength determination function 222 calculates an index value having a correlation with the noise distribution for each difference image (step S208).

Next, the denoising strength determination function 222 determines the optimum value of the control function based on the calculated index value of each difference image (step S210).

Next, the denoising function 218 uses the denoising model MDL1 set as an internal parameter based on the optimum value OP of the control function determined by the denoising strength determining function 222, performs denoising processing on the original image, and optimizes it. A denoising image based on the value is generated (step S212). Note that, without generating the denoising image, the denoising image generated in step S204 by the denoising process using the control function closest to the determined optimum value OP is used as the denoising image based on the optimum value. Good.

Next, the output control function 224 controls the communication interface 202 to transmit the denoised image based on the optimum value to the medical image generation apparatus 100 (step S214). At this time, the output control function 224 may display the denoised image on the display 206. When the communication interface 122 of the medical image generation apparatus 100 receives the denoised image from the medical image processing apparatus 200, the display control function 138 of the medical image generation apparatus 100 may display the denoised image on the display 126. Further, the output control function 224 may control the communication interface 202 to transmit the denoised image to the terminal device that can be used by the image interpreter.

According to the first embodiment described above, the optimum value of the denoising intensity is determined based on the index value having the correlation with the noise distribution obtained from the difference image, and the denoising process with the determined optimum value of the denoising intensity is performed. By doing so, it is possible to remove the noise component to the maximum without removing the structure component. Thereby, even when the noise distribution of the image is non-uniform, the accuracy of the denoising process can be improved.

(Second embodiment)
The second embodiment will be described below. In the above-described first embodiment, a configuration in which the denoising strength determination function 222 of the medical image processing apparatus 200 determines the optimum value OP of the denoising strength based on an index value having a correlation with the noise distribution obtained from the difference image will be described. did. On the other hand, the denoising strength determination function 222 of the present embodiment determines the optimum value OP of the denoising strength based on the histogram of the signal value for each pixel of the difference image. Therefore, for the configuration and the like, the drawings described in the first embodiment and the related description are cited, and the detailed description is omitted.

[Processing flow (image processing)]
Next, the image processing of the processing circuit 210 will be described. FIG. 13 is a flowchart showing a series of flow of image processing of the processing circuit 210 according to the second embodiment. The process of this flowchart is performed, for example, when the original image transmitted by the medical image generation apparatus 100 is acquired by the acquisition function 212.

First, the derivation function 214 derives the SNR of the original image acquired by the acquisition function 212 (step S300). Next, the parameter adjustment function 216 sets a plurality of control functions G based on the SNR derived by the derivation function 214 (step S302). For example, the parameter adjustment function 216 sets the range of the control function G based on the magnitude of the SNR derived by the derivation function 214, and sets the plurality of control functions G included in the set range of the control function G. ..

Next, the denoising function 218 uses the denoising model MDL1 in which the threshold value T is set as an internal parameter for each control function G set by the parameter adjusting function 216, performs denoising processing on the original image, and performs a plurality of denoising processes. A denoised image is generated (step S304).

Next, the difference image generation function 220 generates a plurality of difference images based on the original image acquired by the acquisition function 212 and the plurality of denoised images generated by the denoising function 218 (step S306). For example, the difference image generation function 220 generates a difference image by subtracting the denoising image from the original image for the signal component of each pixel.

Next, the denoising strength determination function 222 generates a histogram of signal values of the entire image for each difference image (step S308). For example, the denoising intensity determination function 222 generates a histogram of brightness values for each pixel for each difference image.

Next, the denoising strength determination function 222 determines the optimum value of the denoising strength based on the histogram of the signal values of the calculated difference images (step S310). For example, the denoising strength determination function 222 obtains, for each difference image, an index value having a correlation with the noise distribution. 14 to 17 are diagrams showing histograms of signal values indicating the brightness value of each pixel of the difference images C1 to C4. For example, the denoising strength determination function 222 calculates the skewness of a signal value within a certain range (for example, a signal value in a predetermined range around the signal value 0) for each histogram of the difference images C1 to C4, The optimum value OP of the denoising intensity (for example, control function) is determined by comparing the difference images. That is, in the first embodiment, the skewness is calculated with respect to the brightness value of each pixel in the difference image (the brightness value of each pixel is individually evaluated), whereas in the second embodiment, the skewness is calculated. A histogram of the brightness value for each pixel is generated, and the skewness is calculated in this histogram (the brightness values within a certain range are collectively evaluated). For example, the denoising strength determination function 222 compares the respective skewnesses of the plurality of difference images with each other, and determines the value of the control function used to generate the difference image with the smallest skewness as the optimum value OP.

Further, as shown in FIGS. 16 and 17, in the difference images C3 and C4, portions P1 and P2 in which an increase in frequency due to the structure component is observed occur. The denoising intensity determination function 222 determines the maximum denoising intensity (for example, the difference image) among the denoising intensities associated with the difference images in which the portions P1 and P2 in which the frequency increase due to such a structural component is observed do not occur. The denoising strength associated with C2) may be determined as the optimal value OP of the denoising strength.

Next, the denoising function 218 uses the denoising model MDL1 in which the internal parameters are set based on the optimum value OP of the denoising strength determined by the denoising strength determining function 222, performs denoising processing on the original image, and optimizes A denoising image based on the value is generated (step S312).

Next, the output control function 224 controls the communication interface 202 to transmit the denoised image based on the optimum value to the medical image generation apparatus 100 (step S314). At this time, the output control function 224 may display the denoised image on the display 206. When the communication interface 122 of the medical image generation apparatus 100 receives the denoised image from the medical image processing apparatus 200, the display control function 138 of the medical image generation apparatus 100 may display the denoised image on the display 126.

According to the second embodiment described above, by determining the optimum value of the denoising intensity based on each histogram of the difference image and performing the denoising process with the determined optimum value of the denoising intensity, the structural component The noise component can be removed to the maximum without being removed. Thereby, the accuracy of the denoising process can be improved.

(Modifications of the first and second embodiments)
Hereinafter, modified examples of the first and second embodiments will be described. In the above-described first and second embodiments, the medical image generating apparatus 100 and the medical image processing apparatus 200 are described as being different apparatuses, but the present invention is not limited to this. For example, the medical image processing apparatus 200 may be realized by a function of the console device 120 of the medical image generating apparatus 100. That is, the medical image processing apparatus 200 may be a virtual machine virtually realized by the console device 120 of the medical image generating apparatus 100. In this case, the medical image generation device 100 is an example of a “medical image processing device”.

FIG. 18 is a diagram showing a medical image generation apparatus 100 according to a modified example of the first and second embodiments. As shown in FIG. 18, the processing circuit 130 of the console device 120 has a derivation function 214 and a parameter in addition to the acquisition function 132, the generation function 134, the communication control function 136, and the display control function 138 described above. The adjusting function 216, the denoising function 218, the difference image generating function 220, the denoising intensity determining function 222, and the learning function 226 may be executed. Further, the memory 150 of the console device 120 may store denoising model information 232, original image information 234, denoising image information 236, and difference image information 238.

According to the modified examples of the first and second embodiments described above, the accuracy of the denoising process can be improved by the medical image generating apparatus 100 alone.

Any of the embodiments described above can be expressed as follows.
Storage for storing programs,
And a processor,
The processor, by executing the program,
Set multiple denoising strengths that indicate the degree to which noise components are removed from medical images,
A noise component is removed from the medical image based on each of the plurality of set denoising strengths to generate a plurality of denoising images,
Generating a plurality of difference images that are differences between the medical image and each of the generated plurality of denoising images,
Determining an optimum value of the denoising intensity based on the plurality of generated difference images,
Medical image processing device.

According to at least one embodiment described above, the processing circuit 210 sets a plurality of parameter adjustment functions 216 for setting a plurality of denoising intensities indicating a degree of removing a noise component from a medical image, and a plurality of parameter adjustment functions 216. A denoising function 218 that removes a noise component from a medical image to generate a plurality of denoising images based on each denoising intensity, and a difference between the medical image and each of the plurality of denoising images generated by the denoising function 218. A difference image generation function 220 that generates a plurality of difference images, and a denoising strength determination function 222 that determines an optimum value of the denoising strength based on the plurality of difference images generated by the difference image generation function 220. Therefore, the accuracy of the denoising process can be improved.

Although some embodiments of the present invention 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, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and modifications thereof are included in the invention described in the claims and equivalents thereof as well as included in the scope and the gist of the invention.

DESCRIPTION OF SYMBOLS 1... Medical image processing system, 100... Medical image generating apparatus, 101... Static magnetic field magnet, 102... Gradient magnetic field coil, 104... Bed, 105... Bed control circuit, 106... Transmission coil, 107... Transmission circuit, 108... Reception coil , 108a to 108h... Coil element, 109... Receiving circuit, 110... Sequence control circuit, 120... Console device, 200... Medical image processing device, 202... Communication interface, 204... Input interface, 206... Display, 210... Processing circuit, 212... Acquisition function, 214... Derivation function, 216... Parameter adjustment function, 218... Denoise function, 220... Difference image generation function, 222... Denoise intensity determination function, 224... Output control function, 226... Learning function, 230... Memory

Claims (9)

A setting unit that sets a plurality of denoising strengths indicating the degree of removing noise components from a medical image,
A denoising image generation unit that removes a noise component from the medical image based on each of the plurality of denoising strengths set by the setting unit to generate a plurality of denoising images,
A difference image generating unit that generates a plurality of difference images that are differences between the medical image and each of the plurality of denoising images generated by the denoising image generating unit;
A determining unit that determines an optimum value of the denoising intensity based on the plurality of difference images generated by the difference image generating unit;
A medical image processing apparatus comprising: The determination unit determines the optimum value of the denoising intensity by mutually comparing index values having a correlation with the noise distribution of each of the plurality of difference images,
The medical image processing apparatus according to claim 1. The denoising image generation unit removes a noise component from the medical image based on a model learned to output an image in which a noise component is removed from the image when the image is input,
The medical image processing apparatus according to claim 1. The setting unit sets an internal parameter of the model as the denoising strength,
The medical image processing apparatus according to claim 3. The internal parameter is a parameter of an activation function of at least one node included in the model,
The medical image processing apparatus according to claim 4. Further comprising a derivation unit that derives an index value related to a noise component included in the medical image,
The setting unit sets a plurality of the denoising strengths based on the index value derived by the deriving unit,
The medical image processing apparatus according to claim 1. The deriving unit derives a signal-to-noise ratio of the medical image as the index value,
The medical image processing apparatus according to claim 6. Computer
Set multiple denoising strengths that indicate the degree to which noise components are removed from medical images,
Based on each of the plurality of denoising intensity set, remove the noise component from the medical image, to generate a plurality of denoising images,
Generating a plurality of difference images that are differences between the medical image and each of the generated plurality of denoising images,
Determining an optimum value of the denoising intensity based on the plurality of generated difference images,
Medical image processing method. On the computer,
Lets you set multiple denoising strengths that indicate the degree to which noise components are removed from medical images,
A noise component is removed from the medical image based on each of the plurality of set denoising strengths to generate a plurality of denoising images,
Generating a plurality of difference images that are differences between the medical image and each of the generated plurality of denoising images,
Based on the plurality of generated difference images, determine the optimum value of the denoising intensity,
program.

Images