JP7346270B2 - Medical information processing device, medical information processing method, and program - Google Patents

Description

Embodiments of the present invention relate to a medical information processing device, a medical information processing method, and a program.

Conventionally, an image is generated by taking the sum of squares (SoS) of signals obtained by multiple receiving coils installed in an MRI (Magnetic Resonance Imaging) device, and then denoising is performed on the generated image. 2. Description of the Related Art There is a known technique for obtaining an image in which noise components are removed or suppressed through processing. For example, a deep neural network can be used for the denoising process.

US Patent Application Publication No. 2018/0018757

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

The medical information processing device of the embodiment includes a deriving section, an adjusting section, a removing section, and a combining section. The derivation unit derives an index value regarding noise included in data corresponding to the magnetic resonance signals collected by each of the plurality of receiving coils. The adjustment section adjusts the degree to which noise is removed from data corresponding to the magnetic resonance signal based on the index value derived by the derivation section. The removal section removes noise from the data corresponding to the magnetic resonance signal based on the degree adjusted by the adjustment section. The synthesizing section synthesizes data corresponding to the magnetic resonance signals collected by each of the plurality of receiving coils, from which noise has been removed by the removing section.

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 a first embodiment. FIG. 1 is a diagram showing an example of a medical image generation device 100 according to a first embodiment. FIG. 3 is a diagram showing an example of the arrangement of RF coils 108 provided in the medical image generation apparatus 100 according to the first embodiment. FIG. 1 is a diagram showing an example of a medical image processing apparatus 200 according to a first embodiment. The figure which shows an example of the structure of denoising model MDL1 based on 1st Embodiment. FIG. 3 is a diagram showing an example of an activation function of the activation layer 330 according to the first embodiment. FIG. 3 is a diagram illustrating an example of a configuration for generating a denoised image and a composite image from an original image according to the first embodiment. 7 is a flowchart showing a series of learning processes of the processing circuit 210 in the first embodiment. 7 is a flowchart showing a series of image processing steps of the processing circuit 210 in the first embodiment. The figure which shows an example of the medical image generation apparatus 100 based on the modification of 1st Embodiment. The figure which shows an example of the medical image processing apparatus 200 based on 2nd Embodiment. 7 is a flowchart showing a series of image processing steps of the processing circuit 210 in the second embodiment. The figure which shows the medical image generation apparatus 100 based on the modification of 2nd Embodiment.

Hereinafter, embodiments of a medical information processing device, a medical information 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 generation device 100 and a medical image processing device 200. The medical image generation device 100 and the medical image processing device 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 processing device 200 is an example of a “medical information processing device”.

The medical image generation device 100 includes, for example, an MRI device. An MRI apparatus, for example, applies a magnetic field to a subject (for example, a human body), uses a coil to receive electromagnetic waves generated from hydrogen nuclei within the subject due to the nuclear magnetic resonance phenomenon, and generates a signal based on the received electromagnetic waves. A medical image (MR image) is generated by reconstructing the image. In the following description, the medical image generation apparatus 100 will be described as an MRI apparatus as an example.

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

[Configuration example of medical image generation device (MRI device)]
FIG. 2 is a diagram showing an example of the medical image generation device 100 according to the first embodiment. As shown in FIG. 2, the medical image generation apparatus 100 includes, for example, a static magnetic field magnet 101, a gradient magnetic field coil 102, a gradient magnetic field power supply 103, a bed 104, a bed control circuit 105, a transmission circuit 107, and an RF It includes a coil 108, a receiving circuit 109, a sequence control circuit 110, and a console device 120.

The static field magnet 101 is a hollow, substantially cylindrical magnet. The static magnetic field magnet 101 generates a uniform static magnetic field in the internal space. The static field magnet 101 is, for example, a permanent magnet or a superconducting magnet. The gradient magnetic field coil 102 is a hollow, substantially cylindrical coil, and is arranged inside the static magnetic field magnet 101. The gradient magnetic field coil 102 is formed by combining three coils corresponding to the 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 an axial direction that is orthogonal to the z-axis direction and parallel to the floor surface of the room in which the medical image generation device 100 is installed. , and the y-axis direction represents the axial direction that is perpendicular to the floor surface. The three coils corresponding to each axial direction individually receive current from the gradient magnetic field power supply 103 to generate gradient magnetic fields whose magnetic field strengths vary along each of the x, y, and z axes. Note that the z-axis direction is the same direction as the static magnetic field.

The gradient magnetic field power supply 103 supplies 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 correspond to, for example, a slice selection gradient magnetic field Gs, a phase encoding gradient magnetic field Ge, and a readout gradient magnetic field Gr, respectively. . The slice selection gradient magnetic field Gs is used to arbitrarily determine an imaging section. The phase encoding gradient magnetic field Ge is used to change the phase of the magnetic resonance signal depending on the spatial position. The readout gradient magnetic field Gr is used to change the frequency of the magnetic resonance signal depending on 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 cavity (imaging port) of the gradient magnetic field coil 102 with the subject OB placed thereon. Usually, the bed 104 is installed so that its longitudinal direction is parallel to the central axis of the static magnetic field magnet 101. The bed control circuit 105 drives the bed 104 under the control of the console device 120 to move the top plate 104a in the longitudinal direction and the vertical direction.

The RF coil 108 receives an RF (Radio Frequency) pulse from the transmission circuit 107 and generates a high frequency magnetic field. The transmitting circuit 107 supplies the RF coil 108 with an RF pulse corresponding to the Larmor frequency determined by the type of target atomic nucleus and the strength of the magnetic field. Further, the RF coil 108 receives magnetic resonance signals emitted from the subject OB due to the influence of the high frequency magnetic field. A magnetic resonance signal includes, for example, a signal intensity component and a phase component. Upon receiving the magnetic resonance signal, the RF coil 108 outputs the received magnetic resonance signal to the receiving circuit 109 .

The RF coil 108 is, for example, a whole-body coil that is housed in the frame of the MRI apparatus and configured to surround the subject OB, or a local coil that is provided for each imaging region of the subject OB. Although a local coil will be described below as an example of the RF coil 108, it is not intended to limit the type of the RF coil 108. Further, transmission and reception may be performed using different RF coils, or the RF coil 108 may be configured to serve both as a transmitter and receiver. Note that in the first embodiment, the RF coil 108 is a coil array having a plurality of coil elements.

FIG. 3 is a diagram showing an example of the arrangement of the RF coil 108 provided in the medical image generation apparatus 100 according to the first embodiment. FIG. 3 shows an example in which the RF coil 108 includes eight coil elements 108a to 108h. These coil elements 108a to 108h are arranged, for example, so as to surround the subject OB along the z-axis. Each coil element 108a to 108h receives a magnetic resonance signal emitted from the subject OB and outputs it to the receiving circuit 109.

The receiving 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 set of magnetic resonance data that are digital signals by digitally converting each of the magnetic resonance signals that are analog signals output by the coil elements 108a to 108h. Further, the receiving circuit 109 transmits the generated set of magnetic resonance data to the sequence control circuit 110. Note that the receiving circuit 109 may be provided on the side of the gantry device that includes the static magnetic field magnet 101, the gradient magnetic field coil 102, and the like.

The sequence control circuit 110 images the object OB by driving the gradient magnetic field power supply 103, the transmitting circuit 107, and the receiving circuit 109 based on sequence information output by the console device 120. Sequence information is information that defines a procedure for performing imaging processing. The sequence information includes the magnitude of the current that the gradient magnetic field power supply 103 supplies to the gradient magnetic field coil 102, the timing of supplying the current, the strength of the RF pulse that the transmitting circuit 107 transmits to the RF coil 108, the timing of applying the RF pulse, and the reception. It includes information defining the timing at which the circuit 109 detects a magnetic resonance signal.

Note that the sequence control circuit 110 drives the gradient magnetic field power supply 103, the transmitting circuit 107, and the receiving circuit 109, and upon receiving magnetic resonance data from the receiving circuit 109, transfers the received magnetic resonance data to the console device 120.

The console device 120 controls the entire medical image generation device 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, for example, 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 and receives information to and from the medical image processing apparatus 200. Communication interface 122 outputs the received information to processing circuit 130. Furthermore, the communication interface 122 may transmit information to other devices connected via the network NW under the control of the processing circuit 130.

The input interface 124 is an interface that accepts various input operations from an operator. Upon receiving an input operation, the input interface 124 converts the received input operation into an electrical signal and outputs it 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. Furthermore, the input interface 124 may be realized by a user interface that accepts voice input, such as a microphone. When the input interface 124 is a touch panel, a display 126, which will be described later, may be formed integrally with the input interface 124.

Display 126 displays various information. For example, the display 126 displays an image generated by the processing circuit 130, a GUI (Graphical User Interface), etc. for accepting various input operations from an operator. 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. be done. Further, some or all of the functions of the processing circuit 130 are realized by hardware (circuitry) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), and FPGA (Field-Programmable Gate Array). Alternatively, it may be realized by cooperation of software and hardware. Further, the above program may be stored in advance in the memory 150 or in a removable storage medium such as a DVD or CD-ROM, and the storage medium is installed in the drive device of the console device 120. It may also be installed into memory 150 from a 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 connected via the network NW, such as NAS (Network Attached Storage) or external storage server devices. 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 magnetic resonance data obtained by digitizing electromagnetic wave signals (magnetic resonance signals) generated within the subject OB due to the nuclear magnetic resonance phenomenon. In addition, magnetic resonance data arranged in two dimensions or three dimensions according to the information given by the aforementioned gradient magnetic field Gs for slice selection, gradient magnetic field Ge for phase encoding, and gradient magnetic field Gr for readout, for example, is arranged in k space. Also called data.

The generation function 134 performs reconstruction processing including processing such as Fourier transformation (for example, inverse Fourier transformation) on the k-space data acquired by the acquisition function 132, thereby generating an MR image reconstructed from the k-space data. generate. 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 coil element 108a. A set of MR images (hereinafter referred to as original images) is generated for each magnetic resonance signal output by 108h. As shown in FIG. 3, the generation function 134 generates original images a to h as a set of original images of the magnetic resonance signals output by the coil elements 108a to 108h, respectively.

When the original images a to h are generated by reconstruction by the generation function 134, the communication control function 136 causes the communication interface 122 to communicate with the medical image processing apparatus 200, and causes the medical image processing apparatus 200 of the communication partner to perform reconstruction. The resulting original images a to h are then sent. Further, the communication control function 136 may cause the communication interface 122 to communicate with the medical image processing apparatus 200 and receive various information from the medical image processing apparatus 200 with which the communication interface 122 communicates.

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 images a to h generated by the generation function 134 on the display 126.

[Example of configuration of medical image processing device]
FIG. 4 is a diagram showing an example of the medical image processing apparatus 200 according to the first embodiment. The medical image processing device 200 performs denoising processing to remove or reduce noise on the original images a to h received from the medical image generation device 100, and synthesizes the images after the denoising processing to create a final image ( Hereinafter, a composite image) is generated. In the following, removing or reducing noise will be 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 a NIC. For example, the communication interface 202 communicates with the medical image generation device 100 via the network NW, and receives the reconstructed medical image from the medical image generation device 100. Communication interface 202 outputs the received medical image to processing circuit 210. Further, the communication interface 202 may transmit information to the medical image generation device 100 and other devices connected via the network NW under the control of the processing circuit 210. The other device may be, for example, a terminal device that can be used by an image interpreter such as a doctor or nurse.

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 circuit 210. For example, the input interface 204 is implemented using 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 that accepts voice input such as a microphone. When the input interface 204 is a touch panel, a display 206, which will be described later, may be formed integrally with the input interface 204.

Display 206 displays various information. For example, the display 206 displays an image generated by the processing circuit 210 (a denoised image or a composite image, which will be described later), or displays a GUI for accepting various input operations from an operator. For example, the display 206 is an LCD, a CRT display, an organic EL display, or the like.

The processing circuit 210 executes, for example, an acquisition function 212, a derivation function 214, a parameter adjustment function 216, a denoising function 218, an image synthesis function 220, an output control function 222, and a learning function 224. The derivation function 214 is an example of a "derivation section," the parameter adjustment function 216 is an example of an "adjustment section," the denoising function 218 is an example of a "removal section," and the image synthesis function 220 is an example of a "removal section." This is an example of the "synthesizing section".

These functions (components) are realized by, for example, a processor (or processor circuit) such as a CPU or GPU executing a program (software) stored in the memory 230. Furthermore, some or all of these multiple functions may be realized by hardware (circuitry) such as LSI, ASIC, or FPGA, or may be realized by collaboration between software and hardware. good. Further, the above program may be stored in advance in the memory 230 or in a removable storage medium such as a DVD or CD-ROM, and the storage medium is installed in the drive device of the medical image processing apparatus 200. The program may be installed into the memory 230 from the storage medium by being installed.

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 other storage devices 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. The memory 230 stores, for example, denoising model information 232, original image information (hereinafter referred to as original image information 234), denoised image information (hereinafter referred to as denoised image information 236), and composite image information (hereinafter referred to as composite image information 238). ) etc. are stored.

The denoising model information 232 is information (program or data structure) that defines a denoising model MDL1, which will be described later. The denoising model MDL1 is a model that is trained to output an image from which noise has been removed when a certain image is input. The denoising model MDL1 includes, for example, one or more DNNs (Deep Neural Network(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 layer), and output layer constituting each DNN included in the denoising model MDL1 are This includes connection information on how the neurons are connected, and weight information on how many connection coefficients are given to data input and output between connected neurons. Connection information includes, 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, activation functions that realize each neuron, and gates installed between neurons in hidden layers. including. The activation function that realizes the neuron may be, for example, a function that switches operations depending on the input code (ReLU (Rectified Linear Unit) function, ELU (Exponential Linear Units) function, clipping function), a sigmoid function, , a step function, a hyperporic tangent function, or an identity function. The gate selectively passes or weights data communicated between neurons, for example, depending on the value returned by the activation function (eg, 1 or 0). The coupling coefficient includes, for example, a weight given to output data in a hidden layer of a neural network when data is output from neurons in a certain layer to neurons in a deeper layer. Further, the coupling coefficient may include a bias component specific to each layer.

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

FIG. 5 is a diagram showing an example of the configuration of 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, if 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 bias components to the input matrix at appropriate times, and outputs the result to the convolution layer 320 at the subsequent stage.

The convolution layer 320 repeats a product-sum operation on the input matrix while sliding a linear transformation matrix called a filter or kernel by a predetermined stride amount, and calculates the relationship between the input matrix and the linear transformation matrix. Generates a matrix containing multiple elements that are associated with sum of products as element values. At this time, the convolution layer 320 performs padding (for example, zero padding) to interpolate elements of arbitrary values around the input matrix, so that the matrix input to the convolution layer 320 is input 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. The convolution layer 320 then outputs the generated matrix to the activation layer 330.

The activation layer 330 calculates an activation function for each element of the matrix input from the convolution layer 320, and outputs the calculated matrix to a subsequent layer.

FIG. 6 is a diagram illustrating an example of an 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 expressed, for example, by the following equation (1). Note that the activation function of the activation layer 330 may be a hard-shrinkage function instead of a soft-shrinkage function.

The Soft-Shrinkage function or the Hard-Shrinkage function outputs zero when the element value x, which is the input value, is within the range of a predetermined positive and negative threshold value ±T centered on zero, and returns the element value, which is the input value. This is a function that outputs a value proportional to the element value x when x exceeds a positive threshold value +T or is less than a negative threshold value -T. By applying a Soft-Shrinkage function or a Hard-Shrinkage function to the activation function of the activation layer 330, an image signal whose amplitude is smaller than the threshold T, that is, a weak image signal that is highly likely to be noise, is activated. can be zero at the output of the function.

The threshold T is a parameter that varies depending on the level of noise (signal strength or signal power) included in the input image, and is expressed, for example, by the following equation (2).

G in Equation (2) represents the level of noise contained in the input image, and is a signal that controls the value of the threshold T (hereinafter referred to as a control signal). The higher the level of noise included in the input image, the higher the control signal G becomes, and the lower the level of noise included in the input image, the smaller the control signal G becomes. The level of noise included in an input image may be determined by the signal-to-noise ratio (SNR) of the input image.

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

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

Note that the denoising model MDL1 illustrated in FIG. 5 is just an example, and may include, for example, a pooling layer. The pooling layer compresses (reduces) the number of dimensions of the matrix by replacing the element values of the input matrix with representative values such as the average value or maximum value of all element values included in the matrix. The pooling layer outputs a matrix with a reduced number of dimensions to a 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 images a to h from the medical image generation apparatus 100 with which the communication interface 202 communicates. The acquisition function 212 stores the acquired original images a to h in the memory 230 as original image information 234.

Based on the original images a to h acquired by the acquisition function 212, the derivation function 214 derives an index value regarding noise contained in each of the original images a to h. That is, the derivation function 214 derives, for each image, an index value related to noise included in the plurality of images based on the magnetic resonance signals collected from the subject OB by each of the plurality of receiving coils. For example, the derivation function 214 derives SNR as an index value regarding noise. SNR is an index value obtained by dividing the signal strength of an image by the signal strength of noise. The details of the SNR derivation method will be described later. The derivation function 214 may derive data related to SNR, such as the signal strength of the original images a to h, the magnitude of noise, and the gain during normalization processing. The derivation function 214 may derive index values regarding noise included in each of the original images a to h based on the supplementary information of the original images a to h.

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 degree of noise removal from each of the plurality of original images a to h based on the SNR, which is the index value derived by the derivation function 214. The parameter adjustment function 216 increases the degree as the signal-to-noise ratio decreases, and decreases the degree as the signal-to-noise ratio increases. Parameter adjustment function 216 inputs the adjusted control function G for at least one node of activation layer 330 .

Note that the parameter adjustment function 216 may collectively determine the degree of noise removal (denoising strength) for the plurality of original images a to h. That is, one determined degree can be applied to all of the plurality of original images a to h. In this case, the derivation function 214 determines the noise contained in the data obtained by performing imaging under conditions where the subject OB is placed on the top plate 104a and no RF pulse is applied. For example, the derivation function 214 quantifies the distribution of Gaussian noise in an image space that is reconstructed from k-space data collected under conditions in which no RF pulses are applied. The parameter adjustment function 216 may determine the degree of noise removal based on the noise quantified in this way. Note that since the intensity and distribution of noise may change depending on the imaging conditions, the degree of noise removal may be adjusted each time the imaging conditions are changed.

The denoising function 218 uses the denoising model MDL1 indicated by the denoising model information 232 to generate a plurality of images (hereinafter referred to as denoising images) from which noise has been removed from each of the original images a to h acquired by the acquisition function 212. generate. FIG. 7 is a diagram illustrating an example of a configuration for generating a denoised image and a composite image from an original image according to the first embodiment. As shown in FIG. 7, the denoising function 218 removes noise from the original images a to h acquired by the acquisition function 212 to generate denoised images a to h, respectively. That is, the denoising function 218 obtains the matrix output by the output layer 340 of the denoising model MDL1 as denoised images a to h obtained by removing noise from the original images a to h. The denoising function 218 stores the generated denoised images a to h in the memory 230 as denoised image information 236.

When generating denoised images a to h, the denoising function 218 can adjust the values of the internal parameters set in the denoising model MDL1 based on the parameters adjusted by the parameter adjustment function 216. That is, the denoising function 218 can vary the degree of denoising of each original image using the control function G adjusted based on the SNR of each of the original images a to h. The denoising model MDL1 may be implemented as part of the denoising function 218, for example, by a processor executing the denoising model MDL1. The denoising function 218 removes noise from the data corresponding to the magnetic resonance signal based on a model learned to output data from which noise has been removed when the data corresponding to the magnetic resonance signal is input. Remove.

Note that the denoising function 218 is not limited to one using a neural network. Denoising functionality 218 may remove or reduce noise using any machine learning generated model, such as, for example, logistic regression analysis, decision tree analysis, or techniques based on support vector machines. Further, the denoising function 218 may remove or reduce noise by performing wavelet transform on the original image.

Further, the denoising function 218 is not limited to one using machine learning technology. The denoising function 218, for example, statistically analyzes the image signal and restores a signal from which noise has been removed from an image signal containing noise based on the sparsity of the signal when the image signal is projected onto the noise removal space. Then, denoising processing may be performed. Further, the denoising function 218 may perform denoising processing using an arbitrary filtering technique. Further, the denoising function 218 may perform denoising processing by averaging the image signal or extracting feature amounts from the image signal.

The image composition function 220 generates a composite image by combining the denoised images a to h generated by the denoising function 218. The image composition function 220 generates a composite image using, for example, a composition algorithm using a sum of squares. As shown in FIG. 7, the image synthesis function 220 synthesizes a plurality of denoised images a to h to generate one composite image. This composite image becomes the final captured image of the medical image generation device 100. Note that when the medical image generation apparatus 100 is an apparatus that reconstructs MR images by parallel imaging, the image synthesis function 220 performs processing such as unfolding folded images in reconstruction processing based on denoised images a to h. It's fine. The image composition function 220 stores the generated composite image in the memory 230 as composite image information 238.

When a composite image is generated by the image composition function 220, the output control function 222 causes the communication interface 202 to communicate with the medical image generation apparatus 100, and transmits the composite image to the medical image generation apparatus 100 that is the communication partner. Further, the output control function 222 may display the composite image on the display 206.

The learning function 224 inputs an image to be used as training data (hereinafter referred to as a learning image) to the denoising model MDL1, and the denoised image outputted by the denoising model MDL1 is inputted as an image to be used as training data (hereinafter referred to as a training image). The denoising model MDL1 is learned so as to approach the image). For example, the teacher image may be an image with a high SNR, such as 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, for example, Gaussian noise.

For example, the learning function 224 adjusts the element values of the linear transformation matrix of the convolution layer 320 and the values of each node of the activation layer 330 so that the difference between the denoised 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 using gradient methods such as SGD (Stochastic Gradient Descent), Momentum SGD, AdaGrad, RMSprop, AdaDelta, and Adam (Adaptive moment estimation).

Note that in the above, each function of the processing circuit 210 performs denoising processing on the original images a to h received from the medical image generation device 100, and generates a composite image by combining the images after the denoising processing. Although the case has been described as an example, the present invention is not limited to this. For example, each function of the processing circuit 210 denoises data before reconstruction (data corresponding to magnetic resonance signals output by the coil elements 108a to 108h) such as k-space data received from the medical image generation apparatus 100. A composite image may be generated by performing processing and combining the data after denoising processing.

That is, the derivation function 214 derives an index value regarding noise included in data corresponding to magnetic resonance signals collected by each of the plurality of receiving coils. The parameter adjustment function 216 adjusts the degree to which noise is removed from the data corresponding to the magnetic resonance signal, based on the index value derived by the derivation function 214. The denoising function 218 removes noise from the data corresponding to the magnetic resonance signal based on the degree adjusted by the parameter adjustment function 216. The image synthesis function 220 synthesizes data corresponding to the magnetic resonance signals collected at each of the plurality of receiving coils, from which noise has been removed by the denoising function 218. The data corresponding to the magnetic resonance signals includes data of a plurality of images based on the magnetic resonance signals.

[Processing flow (learning process)]
The processing flow of the processing circuit 210 in 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 processing for removing noise using the learned denoising model MDL1. In the following, first, the learning process of the processing circuit 210 will be explained. FIG. 8 is a flowchart showing a series of learning processes performed by the processing circuit 210 in the first embodiment. The processing in this flowchart is performed, for example, when the operator of the medical image processing apparatus 200 operates the input interface 204 to instruct the start of the learning process.

First, the learning function 224 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 coefficient α are set to 1.

Next, the learning function 224 inputs the learning image to the denoising model MDL1 and obtains the processing result (step S102). The learning image is, for example, an image in which known noise is added to a teacher image that does not include noise or has a high SNR. The learning image is obtained, for example, by adding this noise and the pixel values of the teacher image.

For example, the learning function 224 inputs the pixel value of each pixel of the learning image to each input terminal of the input layer of the denoising model MDL1. This pixel value is propagated through each node of the intermediate layer of the denoising model MDL1 from the input layer to the output layer while changing the value by weighted addition, bias addition, and activation function processing operations. go. Then, the pixel values of the image processed by the denoising model MDL1 (hereinafter referred to as a processing result image) are output as a processing result to the output end of the output layer.

Next, the learning function 224 calculates a training error that is an error between the processed 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 errors of each pixel, and the like.

Next, the learning function 224 uses, for example, an error backpropagation method to update the internal parameters of the denoising model MDL1 so that the calculated training error becomes smaller (step S106). Next, the learning function 224 performs verification using, for example, a correct image that is a noise-free image or a high SNR image prepared separately from the teacher image, and a verification image in which noise is added to the correct image. A generalization error, which is an error between the processed image when the image is input to the denoising model MDL1 and the correct image, is calculated (step S108). Next, the learning function 224 determines whether the generalization error has reached the minimum value (step S110).

If the learning function 224 determines that the generalization error has not reached the local minimum value, it repeats the processing from step S102 onwards. On the other hand, if the learning function 224 determines that the generalization error has reached the minimum value, it stores the internal parameters of the learned denoising model MDL1 in the memory 230 (step S112). With the above steps, the processing of this flowchart is completed.

[Processing flow (image processing)]
Next, image processing by the processing circuit 210 will be explained. FIG. 9 is a flowchart showing a series of image processing steps performed by the processing circuit 210 in the first embodiment. The processing in this flowchart is performed, for example, when the original images a to h transmitted by the medical image generation apparatus 100 are acquired by the acquisition function 212.

First, the derivation function 214 derives the SNR of each of the original images a to h acquired by the acquisition function 212 (step S200). For example, for each of the original images a to h acquired by the acquisition function 212, the derivation function 214 supplies an RF pulse to the RF coil 108 in a state where the gradient magnetic field coil 102 generates a gradient magnetic field, and generates a high-frequency magnetic field. The original image obtained as a result of outputting (hereinafter referred to as the original image with RF pulse) and the original image obtained without outputting a high-frequency magnetic field (hereinafter referred to as RF pulse) with the gradient magnetic field coil 102 generating a gradient magnetic field. The SNR is derived based on the difference in pixel values from the original image without pulses (referred to as the original image without pulses).

Further, the derivation function 214 calculates, for each of the original images a to h acquired by the acquisition function 212, pixels of the original image with two or more RF pulses obtained as a result of outputting a high-frequency magnetic field toward the same object OB. The SNR of the original image may be derived based on the difference in values. At this time, the derivation function 214 may derive the SNR of the original image based on the difference in pixel values of two original images with RF pulses near the center slice. For example, if the main scan is a sequence in which scans are repeated 10 times, the center slice is the original image obtained by the fifth or sixth scan.

In this way, by taking the difference between two original images obtained by scanning multiple times under the condition that the object OB is the same, the magnetic resonance signal components cancel each other out, and the heat in the receiving system is reduced. SNR can be determined based on random noise components caused by noise.

Next, the parameter adjustment function 216 determines a control function G based on the SNR derived by the derivation function 214 for each of the original images a to h, and determines a control function G based on the SNR derived by the denoising function 214, By inputting the adjusted control function G, the value of the threshold T, which is an internal parameter of the denoising model MDL1, is adjusted (step S202).

Next, the denoising function 218 removes noise from each of the original images a to h using the denoising model MDL1 in which the threshold T adjusted by the parameter adjustment function 216 is set as an internal parameter, and ~h is generated (step S204).

Next, the image synthesis function 220 synthesizes the denoised images a to h generated by the denoising function 218 to generate a composite image that is the final captured image (step S206).

Next, the output control function 222 controls the communication interface 202 to transmit the composite image generated by the image composition function 220 to the medical image generation apparatus 100 (step S208). At this time, the output control function 222 may display the composite image on the display 206. When the communication interface 122 of the medical image generation device 100 receives the composite image from the medical image processing device 200, the display control function 138 of the medical image generation device 100 may cause the display 126 to display the composite image. Further, the output control function 222 may control the communication interface 202 to transmit the composite image to a terminal device that can be used by a person who interprets the image.

According to the first embodiment described above, denoising processing is performed on each of the original images a to h based on the magnetic resonance signals output by the coil elements 108a to 108h, and denoised images from which noise has been removed are synthesized. By generating a composite image by doing this, it is possible to prevent a decrease in the accuracy of the denoising process due to the difference in the noise distribution of the original images a to h. Further, in the denoising process of this embodiment, the degree of noise removal can be made variable depending on the amount of noise included in each of the original images a to h. This denoising process has no effect on signal components other than noise. Therefore, it is possible to improve the accuracy of the denoising process and maintain the quality of the final image.

(Modified example of the first embodiment)
Modifications of the first embodiment will be described below. In the first embodiment described above, the medical image generation device 100 and the medical image processing device 200 are described as being different devices from each other, but the present invention is not limited to this. For example, the medical image processing device 200 may be realized by one function of the console device 120 of the medical image generation device 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 generation apparatus 100. In this case, the medical image generation device 100 is an example of a "medical information processing device."

FIG. 10 is a diagram showing a medical image generation apparatus 100 according to a modification of the first embodiment. As shown in FIG. 10, the processing circuit 130 of the console device 120 includes the above-described acquisition function 132, generation function 134, communication control function 136, and display control function 138, as well as the derivation function 214 and An adjustment function 216, a denoising function 218, an image composition function 220, and a learning function 224 may be performed. Further, the memory 150 of the console device 120 may store denoised model information 232, original image information 234, denoised image information 236, and composite image information 238.

According to the modification of the first embodiment described above, the accuracy of denoising processing can be improved by the medical image generation apparatus 100 alone.

(Second embodiment)
The second embodiment will be described below. In the first embodiment described above, denoising processing is performed on each of the original images a to h based on the magnetic resonance signals output by the plurality of coil elements 108a to 108h, and denoised images from which noise has been removed are synthesized. I explained the configuration. In this embodiment, the processing circuit 210 of the medical image processing apparatus 200 adds noise components to a composite image generated by combining original images without performing denoising processing to equalize the noise intensity distribution. Addition processing is performed, and denoising processing is performed on the processed composite image. Therefore, regarding the configuration and the like, the drawings and related descriptions described in the first embodiment will be referred to, and detailed description will be omitted.

[Example of configuration of medical image processing device]
FIG. 11 is a diagram showing an example of a medical image processing apparatus 200 according to the second embodiment. The processing circuit 210 of the medical image processing apparatus 200 includes an acquisition function 212, a derivation function 214, a parameter adjustment function 216, a denoising function 218, an image synthesis function 220, an output control function 222, and a learning function 224. Then, the noise equalization function 226 is executed. The noise equalization function 226 is an example of an "equalization processing section."

The noise equalization function 226 estimates the noise intensity distribution from the composite image generated by combining the original images by the image synthesis function 220, and equalizes the noise intensity distribution of the composite image from the estimated noise intensity distribution. By generating a noise distribution (hereinafter referred to as a uniform distribution) for the composite image and adding a signal indicated by the uniform distribution to the composite image, the intensity distribution of noise included in the composite image is made uniform. The denoising function 218 then performs denoising processing on the composite image whose noise intensity distribution has been made uniform by the noise equalization function 226.

[Processing flow (image processing)]
Next, image processing by the processing circuit 210 will be explained. FIG. 12 is a flowchart showing a series of image processing steps performed by the processing circuit 210 in the second embodiment. The processing in this flowchart is performed, for example, when the original images a to h transmitted by the medical image generation apparatus 100 are acquired by the acquisition function 212.

First, the image synthesis function 220 synthesizes the original images a to h acquired by the acquisition function 212 to generate a composite image (step S300). That is, the image synthesis function 220 generates a composite image by synthesizing a plurality of original images a to h based on magnetic resonance signals collected from the subject OB by each of a plurality of receiving coils (a plurality of coil elements). Note that the generation function 134 executed by the processing circuit 130 of the console device 120 of the medical image generation device 100 synthesizes the original images a to h to generate a composite image, and sends the composite image to the image composition function 220. If there is, the process of step S300 does not need to be performed.

Next, the noise equalization function 226 estimates the noise intensity distribution of the composite image generated by the image composition function 220 (step S302). For example, the noise equalization function 226
The noise intensity distribution of the composite image is estimated based on a comparison between each of the original images a to h acquired by the acquisition function 212 and a composite image obtained by combining the original images a to h.

Next, the noise equalization function 226 generates a uniformity distribution for equalizing the noise intensity distribution of the composite image generated by the image composition function 220, based on the estimated noise intensity distribution (step S304). The noise equalization function 226 may store the generated equalization distribution in the memory 230 as noise equalization information. Next, the noise equalization function 226 equalizes the noise intensity distribution of the composite image by adding the generated uniform distribution to the composite image generated by the image composition function 220 (step S306).

Next, the denoising function 218 uses the denoising model MDL1 indicated by the denoising model information 232 to perform denoising processing to remove noise from the composite image whose noise intensity distribution has been made uniform by the noise equalizing function 226 ( Step S308).

Next, the output control function 222 controls the communication interface 202 to transmit the composite image subjected to the denoising process by the denoising function 218 to the medical image generation apparatus 100 (step S310). At this time, the output control function 222 may display the composite image on the display 206. When the communication interface 122 of the medical image generation device 100 receives the composite image from the medical image processing device 200, the display control function 138 of the medical image generation device 100 may cause the display 126 to display the composite image. Further, the output control function 222 may control the communication interface 202 to transmit the composite image to a terminal device that can be used by a person who interprets the image.

According to the second embodiment described above, processing is performed to add a noise component in order to equalize the noise intensity distribution to a composite image generated by combining original images, and the composite image after processing is By performing denoising processing on an image, the accuracy of the denoising processing can be improved.

(Modified example of second embodiment)
Hereinafter, a modification of the second embodiment will be described. In the second embodiment described above, the medical image generation device 100 and the medical image processing device 200 are described as being different devices from each other, but the present invention is not limited to this. For example, the medical image processing device 200 may be realized by one function of the console device 120 of the medical image generation device 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 generation apparatus 100. In this case, the medical image generation device 100 is an example of a "medical information processing device."

FIG. 13 is a diagram showing a medical image generation apparatus 100 according to a modification of the second embodiment. As shown in FIG. 13, the processing circuit 130 of the console device 120 includes, in addition to the above-described acquisition function 132, generation function 134, communication control function 136, and display control function 138, a derivation function 214 and a parameter An adjustment function 216, a denoising function 218, an image composition function 220, a learning function 224, and a noise equalization function 226 may be performed. Further, the memory 150 of the console device 120 may store denoising model information 232, original image information 234, denoising image information 236, composite image information 238, and noise equalization information 240.

According to the modification of the second embodiment described above, the accuracy of denoising processing can be improved by the medical image generation apparatus 100 alone.

Any of the embodiments described above can be expressed as follows.
storage for storing programs;
comprising a processor;
By executing the program, the processor:
Deriving an index value regarding noise contained in data corresponding to magnetic resonance signals collected by each of the plurality of receiving coils,
Adjusting the degree of noise removal from data corresponding to the magnetic resonance signal based on the derived index value,
removing noise from data corresponding to the magnetic resonance signal based on the adjusted degree;
combining data corresponding to the magnetic resonance signals collected by each of the plurality of receiving coils from which the noise has been removed;
Medical information processing equipment.

Any of the embodiments described above can be expressed as follows.
storage for storing programs;
comprising a processor;
By executing the program, the processor:
generating a composite image by combining a plurality of images based on magnetic resonance signals collected from the subject by each of the plurality of receiving coils;
Estimating the intensity distribution of noise included in the generated composite image,
generating a homogenization distribution that homogenizes the estimated noise intensity distribution;
By adding a signal indicated by the generated uniform distribution to the composite image, the intensity distribution of noise included in the composite image is made uniform;
removing noise from the composite image in which the intensity distribution of the noise has been made uniform;
Medical information processing equipment.

According to at least one embodiment described above, the processing circuit 210 calculates, for each image, an index value regarding noise contained in the plurality of images based on the magnetic resonance signals collected from the subject by each of the plurality of RF coils 108. a parameter adjustment function 216 that adjusts the degree to which noise is removed from each of the plurality of images based on the index value derived by the derivation function 214; By including a denoising function 218 that removes noise from each of the plurality of images based on the denoising function 218, denoising accuracy can be improved.

According to at least one embodiment described above, the processing circuit 210 synthesizes a plurality of images based on magnetic resonance signals collected from the subject by each of the plurality of RF coils 108 to generate a composite image. The function 220 estimates the intensity distribution of noise included in the composite image generated by the image synthesis function 220, generates a homogenization distribution that homogenizes the estimated noise intensity distribution, and uses the generated homogenization distribution to The noise equalization function 226 equalizes the intensity distribution of noise included in the composite image by adding the indicated signal to the composite image, and the noise equalization function 226 equalizes the noise intensity distribution from the composite image. By including a denoising function 218 that removes noise, denoising accuracy can be improved.

Although several embodiments of the invention have been described, these embodiments are presented by way of example 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 changes can be made without departing from the gist of the invention. These embodiments and their modifications are included within the scope and gist of the invention as well as within the scope of the invention described in the claims and its equivalents.

DESCRIPTION OF SYMBOLS 1...Medical image processing system, 100...Medical image generation device, 101...Static magnetic field magnet, 102...Gradient magnetic field coil, 104...Bed, 105...Bed control circuit, 107...Transmission circuit, 108...RF coil, 108a-h... 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... Denoising function, 220... Image synthesis function, 222... Output control function, 224... Learning function, 226... Noise equalization function, 230... Memory

Claims (6)

a derivation unit that derives an index value regarding noise contained in data corresponding to magnetic resonance signals collected by each of the plurality of receiving coils;
an adjustment unit that adjusts the degree to which noise is removed from data corresponding to the magnetic resonance signal based on the index value derived by the derivation unit;
a removing unit that removes noise from data corresponding to the magnetic resonance signal based on the degree adjusted by the adjusting unit;
a synthesizing section that synthesizes data corresponding to the magnetic resonance signals collected by each of the plurality of receiving coils from which noise has been removed by the removing section ;
The data corresponding to the magnetic resonance signals includes data of a plurality of images based on each of the magnetic resonance signals collected by each of the plurality of receiving coils,
The derivation unit derives a plurality of signal-to-noise ratios corresponding to each of the data of the plurality of images as the index value,
In each of the plurality of image data, the adjustment unit increases the degree as the signal-to-noise ratio becomes smaller, and decreases the degree as the signal-to-noise ratio becomes larger.
Medical information processing equipment. The removal unit removes noise from the data corresponding to the magnetic resonance signal based on a model learned to output data from which noise has been removed when data corresponding to the magnetic resonance signal is input. remove,
The medical information processing device according to claim 1 . The adjustment unit adjusts the degree by adjusting internal parameters of the model.
The medical information processing device according to claim 2 . the internal parameter is a parameter of an activation function of at least one node included in the model;
The medical information processing device according to claim 3 . The computer is
Deriving an index value regarding noise contained in data corresponding to magnetic resonance signals collected by each of the plurality of receiving coils,
Adjusting the degree of noise removal from data corresponding to the magnetic resonance signal based on the derived index value,
removing noise from data corresponding to the magnetic resonance signal based on the adjusted degree;
combining data corresponding to the magnetic resonance signals collected by each of the plurality of receiving coils from which the noise has been removed;
A medical information processing method ,
The data corresponding to the magnetic resonance signals includes data of a plurality of images based on each of the magnetic resonance signals collected by each of the plurality of receiving coils,
Deriving a plurality of signal-to-noise ratios corresponding to each of the data of the plurality of images as the index value,
In each of the plurality of image data, the smaller the signal-to-noise ratio, the higher the degree, and the larger the signal-to-noise ratio, the lower the degree.
Medical information processing method . to the computer,
deriving an index value regarding noise contained in data corresponding to magnetic resonance signals collected by each of the plurality of receiving coils;
adjusting the degree of noise removal from data corresponding to the magnetic resonance signal based on the derived index value;
removing noise from data corresponding to the magnetic resonance signal based on the adjusted degree;
combining data corresponding to the magnetic resonance signals collected by each of the plurality of receiving coils from which the noise has been removed;
A program ,
The data corresponding to the magnetic resonance signals includes data of a plurality of images based on each of the magnetic resonance signals collected by each of the plurality of receiving coils,
Deriving a plurality of signal-to-noise ratios corresponding to each of the data of the plurality of images as the index value,
In each of the plurality of image data, the smaller the signal-to-noise ratio, the higher the degree, and the larger the signal-to-noise ratio, the lower the degree.
program .

Images