JP2016093494A - Magnetic resonance imaging apparatus, image processing apparatus and image processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a magnetic resonance imaging apparatus which can properly removes noises from the whole of pixels even when difference in S/N among a plurality of pixels included in an image: to provide an image processing apparatus: and to provide an image processing method.SOLUTION: A magnetic resonance imaging apparatus 100 includes an image generation part 25, a feature amount calculation part, a correction part and a de-noise processing part. The image generation part generates an image on the basis of magnetic resonance signals generated from a subject. The feature amount calculation part calculates a feature amount relating to a pixel signal value on each of the plurality of pixels included in the image. The correction part corrects the feature amount of each of the plurality of pixels on the basis of sensitivity characteristic of a reception coil receiving magnetic resonance signals. The de-noise processing part reduces noise in the image on the basis of the distribution of the feature amount corrected by the correction part.SELECTED DRAWING: Figure 1

Description

  Embodiments described herein relate generally to a magnetic resonance imaging apparatus, an image processing apparatus, and an image processing method.

  Conventionally, as a technique for removing image noise, a method of calculating a standard deviation of a temporal change in signal intensity for each pixel of a frame, and using an average of all pixels of the calculated standard deviation as a noise magnitude is known. Yes. In such a method, since the average of all pixels is used, for example, when there is a difference in S / N (S / N ratio) between a plurality of pixels included in an image, noise of all pixels can be appropriately removed. It was sometimes difficult.

Japanese Patent Laid-Open No. 7-23926 JP 7-123373 A

Hideaki Hisamitsu, Satoshi Arakawa, "Improvement of compressed video image quality by spatio-temporal ε-filter considering local properties of images" IEICE Tech. Bulletin SIS 2004-15, 2004 Klaas P. Pruessmann et. al, “SENSE, Sensitivity Encoding for Fast MRI,” Magnetic Resonance in Medicine 42: 952-962 (1999)

  The problem to be solved by the present invention is to provide a magnetic resonance imaging apparatus, an image processing apparatus, and an image processing apparatus capable of appropriately removing noise of all pixels even when an S / N difference occurs between a plurality of pixels included in an image. An image processing method is provided.

  A magnetic resonance imaging (MRI) apparatus according to an embodiment includes an image generation unit, a feature amount calculation unit, a correction unit, and a denoising processing unit. The image generation unit generates an image based on a magnetic resonance signal generated from the subject. The feature amount calculation unit calculates a feature amount related to the signal value of each pixel included in the image. The correction unit corrects the feature amount of each of the plurality of pixels based on sensitivity characteristics of a receiving coil that receives the magnetic resonance signal. The denoising processing unit reduces the noise of the image based on the distribution of the feature amount corrected by the correcting unit.

FIG. 1 is a block diagram showing a configuration example of an MRI apparatus according to the first embodiment. FIG. 2 is a block diagram illustrating a configuration example of functions included in the MRI apparatus according to the first embodiment. FIG. 3 is a flowchart showing a processing procedure of an image processing method by the MRI apparatus according to the first embodiment. FIG. 4 is a diagram illustrating an example of the neighborhood R used by the feature vector calculation unit according to the first embodiment. FIG. 5 is a diagram illustrating an example of noise model selection performed by the model selection unit according to the first embodiment. FIG. 6 is a diagram illustrating a setting example of the neighborhood R according to a second modification. FIG. 7 is a diagram illustrating an example of noise removal according to an eleventh modification. FIG. 8 is a flowchart showing a noise removal processing procedure according to an eleventh modification. FIG. 9 is a block diagram illustrating a configuration of an image processing apparatus according to the second embodiment.

  Hereinafter, embodiments of an MRI apparatus, an image processing apparatus, and an image processing method will be described in detail based on the drawings.

(First embodiment)
FIG. 1 is a diagram illustrating a configuration example of an MRI apparatus according to the first embodiment. For example, as shown in FIG. 1, the MRI apparatus 100 includes a static magnetic field magnet 1, a gradient magnetic field coil 2, a gradient magnetic field power source 3, a bed 4, a bed control unit 5, a transmission coil 6, a transmission unit 7, a reception coil 8, and a reception. A unit 9 and a computer system 20 are provided.

  The static magnetic field magnet 1 is formed in a hollow, substantially cylindrical shape (including one having a cross section perpendicular to the central axis of the cylinder is elliptical), and generates a uniform static magnetic field in an imaging space formed inside thereof. . The static magnetic field magnet 1 is, for example, a permanent magnet or a superconducting magnet.

  The gradient magnetic field coil 2 is formed in a hollow, substantially cylindrical shape (including one having a cross section perpendicular to the central axis of the cylinder is elliptical), and is disposed inside the static magnetic field magnet 1. Specifically, the gradient coil 2 is configured by combining three coils corresponding to the x, y, and z axes orthogonal to each other. These three coils generate gradient magnetic fields whose magnetic field strengths change along the x-axis, y-axis, and z-axis orthogonal to each other in the imaging space by currents individually supplied from the gradient magnetic field power supply 3. The direction of the z axis is set to be the same as the direction of the magnetic flux of the static magnetic field.

  The gradient magnetic field power supply 3 generates a gradient magnetic field in the gradient magnetic field coil 2 by supplying electric power to the gradient magnetic field coil 2. Specifically, the gradient magnetic field power source 3 supplies current to each of the three coils included in the gradient magnetic field coil 2 to generate a gradient magnetic field along each of the x-axis, y-axis, and z-axis appropriately. Gradient magnetic fields along the readout direction, the phase encoding direction, and the slice direction that are orthogonal to each other are generated in the imaging space. Hereinafter, the gradient magnetic field along the readout direction is referred to as a readout gradient magnetic field, the gradient magnetic field along the phase encoding direction is referred to as a phase encoding gradient magnetic field, and the gradient magnetic field along the slice direction is referred to as a slice gradient magnetic field. .

  These three directions are used to give spatial position information to the MR signal. Specifically, the readout gradient magnetic field gives position information in the readout direction to the MR signal by changing the frequency of the magnetic resonance (MR) signal in accordance with the position in the readout direction. Further, the phase encoding gradient magnetic field changes the phase of the MR signal along the phase encoding direction, thereby giving position information in the phase encoding direction to the MR signal. The slice gradient magnetic field is used to determine the direction, thickness, and number of slice regions when the imaging region is a slice region. When the imaging region is a volume region, the slice gradient magnetic field is used to determine the slice region. By changing the phase of the MR signal according to the position in the direction, position information along the slice direction is given to the MR signal.

  The bed 4 includes a top plate 4 a on which the subject S is placed, and inserts the top plate 4 a into an imaging space formed inside the static magnetic field magnet 1 and the gradient magnetic field coil 2. For example, the bed 4 is installed such that the longitudinal direction is parallel to the central axis of the static magnetic field magnet 1.

  The bed control unit 5 controls the operation of the bed 4. For example, the bed control unit 5 controls the drive mechanism of the bed 4 to move the top plate 4a in the longitudinal direction, the up-down direction, or the left-right direction.

  The transmission coil 6 is formed in a hollow, substantially cylindrical shape (including one having a cross section perpendicular to the central axis of the cylinder that is elliptical), and is disposed inside the gradient magnetic field coil 2. The transmission coil 6 applies an RF magnetic field to the imaging space by an RF (Radio Frequency) pulse current supplied from the transmission unit 7.

  The transmitter 7 supplies an RF pulse current corresponding to the Larmor frequency to the transmitter coil 6.

  The receiving coil 8 is attached to the subject S placed in the imaging space, and receives an MR signal radiated from the subject S under the influence of the RF magnetic field applied by the transmitting coil 6. The receiving coil 8 outputs the received MR signal to the receiving unit 9. For example, a dedicated coil is used for the receiving coil 8 for each part to be imaged. The dedicated coil here is, for example, a receiving coil for the abdomen, a receiving coil for the head, a receiving coil for the spine, or the like.

  The receiving unit 9 generates MR signal data based on the MR signal received by the receiving coil 8. Specifically, the receiving unit 9 generates MR signal data by digitally converting the MR signal, and transmits the generated MR signal data to the collecting unit 24.

  Here, an example in which the transmission coil 6 applies an RF magnetic field and the reception coil 8 receives an MR signal will be described, but the embodiment is not limited thereto. For example, the transmission coil 6 may further have a reception function for receiving MR signals, and the reception coil 8 may further have a transmission function for applying an RF magnetic field. When the transmission coil 6 has a reception function, the reception unit 9 also generates MR signal data from the MR signal received by the transmission coil 6. When the reception coil 8 has a transmission function, the transmission unit 7 also supplies an RF pulse current to the reception coil 8.

  The computer system 20 performs overall control of the MRI apparatus 100. For example, the computer system 20 includes an input unit 21, a display unit 22, a sequence control unit 23, a collection unit 24, an image generation unit 25, a storage unit 26, and a system control unit 27.

  The input unit 21 receives input of various instructions and various information from the operator. For example, the input unit 21 includes an input device such as a keyboard, a mouse, a trackball, a touch panel, a button, and a switch.

  The display unit 22 displays various information and various images. For example, the display unit 22 displays a GUI (Graphical User Interface) for receiving various instructions and various information input from the operator. For example, the image generated by the image generation unit 25 is displayed. For example, the display unit 22 includes a display device such as a liquid crystal monitor, a CRT (Cathode-Ray Tube) monitor, and a touch panel.

  The sequence control unit 23 executes various scans. Specifically, the sequence control unit 23 executes various scans by driving the gradient magnetic field power supply 3, the transmission unit 7, and the reception unit 9 based on the sequence execution data transmitted from the system control unit 27. Here, the sequence execution data is information defining a pulse sequence indicating a procedure for collecting MR signal data. Specifically, the sequence execution data is transmitted at the timing when the gradient magnetic field power supply 3 supplies a current to the gradient magnetic field coil 2 and the intensity of the supplied current, the timing at which the transmission unit 7 performs RF transmission to the transmission coil 6 and the transmission. The information defines the strength of the RF pulse current, the timing at which the receiving unit 9 detects the MR signal, and the like.

  The collection unit 24 collects MR signal data generated by the reception unit 9 as a result of performing various scans. Specifically, when receiving the MR signal data from the receiving unit 9, the collecting unit 24 performs correction processing such as averaging processing and phase correction processing on the received MR signal data, and outputs the corrected MR signal data. It transmits to the image generation unit 25. In addition, the collection unit 24 transmits the collected image data to the computer system 20. The set of MR signal data collected by the collection unit 24 includes two-dimensional or three-dimensional MR signal data depending on the position information given by the readout gradient magnetic field, phase encoding gradient magnetic field, and slice gradient magnetic field. By arranging in dimension, it is stored in the storage unit 26 of the computer system 20 as data constituting the k-space.

  The image generation unit 25 generates an image based on the MR signal data collected by the collection unit 24. Specifically, when receiving the MR signal data from the collection unit 24, the image generation unit 25 generates an image of the subject S by performing post-processing, that is, reconstruction processing such as Fourier transform, on the received MR signal data. . The image generation unit 25 transmits the generated image data to the computer system 20.

  The storage unit 26 stores various data. For example, the storage unit 26 stores MR signal data collected by the sequence control unit 23 and image data generated by the image generation unit 25 for each subject S. The storage unit 26 stores various programs and various data used when the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27 execute various processes. For example, the storage unit 26 includes a storage device such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, a hard disk, and an optical disk.

  The system control unit 27 performs overall control of the MRI apparatus 100 by controlling each unit included in the MRI apparatus 100. For example, the system control unit 27 receives input of various imaging parameters from the operator via the input unit 21. Then, the system control unit 27 generates sequence execution data based on the received imaging parameter, and transmits the generated sequence execution data to the generated sequence control unit 23, thereby executing various scans.

  Further, as a result of executing various scans, the system control unit 27 stores the MR signal data transmitted from the collection unit 24 and the image data transmitted from the image generation unit 25 in the storage unit 26. In addition, the system control unit 27 reads an image requested by the operator from the storage unit 26 and outputs the read image to the display unit 22. For example, the system control unit 27 operates the bed 4 by controlling the bed control unit 5 based on an instruction received from the operator via the input unit 21.

  Among the above-described units, the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27 are, for example, a processor such as a CPU (Central Processing Unit) and an MPU (Micro Processing Unit), a memory And ASIC (application specific integrated circuits) and FPGA (Field Programmable Gate Array). In this case, for example, the processor included in each of the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27 reads from the storage unit 26 a program that defines the processing procedure of processing performed by each unit. By executing, the processing is executed according to the processing procedure.

  Here, an example in which each of the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27 has a processor has been described, but the embodiment is not limited thereto. The configurations of the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27 may be appropriately distributed or integrated. For example, the MRI apparatus 100 may include one processor, and the processor may perform the processes of the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27. Further, for example, the MRI apparatus 100 may include a plurality of processors, and each processor may perform processing of the sequence control unit 23, the collection unit 24, the image generation unit 25, and the system control unit 27 in an appropriately distributed manner. Good.

  The configuration example of the MRI apparatus 100 according to the first embodiment has been described above. With such a configuration, the MRI apparatus 100 calculates a feature amount related to the signal value of each pixel included in the image, and based on the sensitivity characteristics of the receiving coil that receives the MR signal, The feature vectors of the respective pixels are corrected, and image noise is reduced based on the distribution of the corrected feature vectors.

  In the first embodiment, the MRI apparatus 100 calculates, for each of a plurality of pixels included in an image, a feature vector related to the signal value of the pixel, and based on information indicating a distribution of noise generated in the pixel according to an imaging condition. To calculate a correction map. Then, the MRI apparatus 100 corrects the feature vector of each of the plurality of pixels using the calculated correction map, and reduces image noise based on the distribution of the corrected feature vector.

  In the present embodiment, the feature amount is represented by a feature vector including a plurality of elements. However, the embodiment is not limited to this. For example, the feature amount may be represented by a scalar.

  Here, for example, as a technique for removing image noise, there is a method of using a noise removal device having an adjustment parameter corresponding to the amount of noise in the input image. In this method, it is necessary to give a noise amount to the noise removing device from the outside. Further, for example, there is a method of calculating the standard deviation of the signal intensity with time for each pixel of the frame and setting the average of all the calculated standard deviations as the magnitude of the noise. In this method, it is not necessary to give a noise amount from the outside. However, since this technique uses the average of all pixels, there is no problem when the noise amount of the input image is uniform within the image. For example, there is a difference in S / N between a plurality of pixels included in the image. When this occurs, it may be difficult to properly remove noise from all pixels.

  Generally, in an MRI apparatus, when imaging is performed with a plurality of small receiving coils mounted on the surface of a subject, the S / N is high on the body surface close to the receiving coil, and near the center of the body far from the receiving coil Then, it is known that the S / N between pixels does not become constant, for example, the S / N becomes low. In such a case, the above-described technique cannot cope with the difference in S / N between pixels, and cannot completely remove noise of a pixel having a noise amount larger than the estimated noise amount. In some cases, the characteristics of the signal of a pixel having a noise amount smaller than that amount may be lost.

  On the other hand, in the MRI apparatus 100 according to the first embodiment, the S / N difference between pixels is corrected using information indicating the distribution of noise generated in the pixels according to the imaging conditions, and thus included in the image. Even when an S / N difference occurs between a plurality of pixels, noise of all the pixels can be appropriately removed. Thereby, it is possible to suppress the loss of the signal characteristics in the image by removing the noise. That is, according to the MRI apparatus 100 according to the first embodiment, the noise amount is estimated using the feature vector reflecting the imaging condition, thereby improving the noise amount estimation accuracy and simultaneously improving the performance of the denoising processor. By increasing the noise, it is possible to remove the noise appropriately in all the pixels and remove the noise with which the signal characteristics are not easily lost.

  FIG. 2 is a block diagram illustrating a configuration example of functions included in the MRI apparatus 100 according to the first embodiment. 2 shows an input unit 21, a display unit 22, a sequence control unit 23, a storage unit 26, and a system control unit 27 among the constituent elements of the computer system 20 shown in FIG. For example, as shown in FIG. 2, the system control unit 27 includes a scan control unit 27a, a feature vector calculation unit 27b, a correction map calculation unit 27c, a correction unit 27d, a model selection unit 27e, and a denoising processing unit 27f.

  The scan control unit 27 a receives input of various imaging parameters from the operator via the input unit 21. For example, the system control unit 27 displays a GUI for accepting input of various imaging parameters on the display unit 22, and sets to accept input of various imaging parameters via the displayed GUI. Then, the scan control unit 27a generates sequence execution data based on the received imaging parameter, and transmits the generated sequence execution data to the sequence control unit 23, thereby executing various scans.

  The feature vector calculation unit 27b calculates, for each of a plurality of pixels included in an image generated based on the MR signal generated from the subject S, a feature vector related to the signal value of the pixel. The correction map calculation unit 27c calculates a correction map based on information indicating the distribution of noise generated in the pixels according to the imaging conditions. The correction unit 27d corrects the feature vector of each of the plurality of pixels included in the image using the correction map generated by the correction map calculation unit 27c.

  The denoise processing unit 27f reduces image noise based on the distribution of the feature vectors corrected by the correction unit 27d. In the present embodiment, as an example, the model selection unit 27e selects a noise model from a plurality of noise models based on the distribution of feature vectors corrected by the correction unit 27d. In the present embodiment, the denoising processor 27f uses the noise model selected by the model selector 27e to reduce image noise.

  FIG. 3 is a flowchart showing a processing procedure of the image processing method by the MRI apparatus 100 according to the first embodiment. In the present embodiment, an example will be described in which the MRI apparatus 100 calculates a correction map based on information indicating the distribution of noise generated in a pixel according to the sensitivity characteristic of a receiving coil that receives a magnetic resonance signal. . That is, in the present embodiment, an example will be described in which the above-described imaging condition is a sensitivity characteristic of a receiving coil that receives a magnetic resonance signal.

  Specifically, in the present embodiment, the MRI apparatus 100 uses a plurality of receiving coils as in the SENSE method described in Non-Patent Document 2, and performs imaging using the sensitivity difference between the receiving coils. An image is captured by the high-speed imaging method to be performed. In this imaging method, a sensitivity map representing the sensitivity distribution of the receiving coil is collected in a preparatory scan that is mainly performed before the main scan for collecting diagnostic images.

  For example, as illustrated in FIG. 3, first, the scan control unit 27 a receives an imaging plan setting from the operator (step S <b> 101).

  For example, the scan control unit 27a holds in advance information of a pulse sequence in which initial values of imaging parameters such as TR (Repetition Time) and TE (Echo Time) are set. For example, the scan control unit 27a manages a pulse sequence group including a preparation scan pulse sequence and a main scan pulse sequence for each imaging region and imaging purpose. Then, the scan control unit 27a presents the pulse sequence group to the operator for each imaging region and imaging purpose via the GUI, and receives the selection or change of the pulse sequence group from the operator, thereby executing the target examination. The setting of the inspection pulse sequence group and imaging parameters is accepted.

  Thereafter, when the scan control unit 27a receives an instruction to start the preparation scan from the operator (Yes in step S102), the scan control unit 27a starts executing the preparation scan based on the imaging plan set by the operator.

  In the present embodiment, the scan control unit 27a executes a scan for collecting a sensitivity map as a preparation scan (step S103). Thereafter, the collection unit 24 collects MR signal data obtained by the scan, and the image generation unit 25 generates a sensitivity map based on the collected MR signal data. Note that the preparation scan may include, for example, a scan for shimming, a scan for collecting a positioning image for positioning a diagnostic image, and the like.

  Subsequently, when the scan control unit 27a receives an instruction to start the main scan from the operator (step S104, Yes), the scan control unit 27a starts the main scan based on the imaging plan set by the operator.

  Specifically, the scan control unit 27a executes a scan for collecting diagnostic images as the main scan (step S105). Thereafter, the collection unit 24 collects MR signal data obtained by the scan, and the image generation unit 25 generates a diagnostic image based on the collected MR signal data and the sensitivity map.

  For example, in the main scan, the scan control unit 27a executes a plurality of scans for collecting a plurality of types of diagnostic images. For example, when the examination target site is the heart, the scan control unit 27a performs a scan for collecting a left ventricular short-axis image, a left ventricular two-chamber length in the examination of the cardiac function of the left ventricular system of the heart. A scan for collecting axial images, a scan for collecting left ventricular three-chamber long-axis images, a scan for collecting left ventricular four-chamber long-axis images, and the like are executed. In the examination of the cardiac function of the right ventricular system of the heart, the scan control unit 27a scans to collect a right ventricular short-axis image, scans to collect a right ventricular dual-chamber long-axis image, and right ventricular three A scan for collecting a long-axis image of the cavity, a scan for collecting a long-axis image of the right ventricular four-chamber, etc. are executed.

  Subsequently, the feature vector calculation unit 27b calculates a feature vector for each of a plurality of pixels included in the image generated by the image generation unit 25 (step S106). Here, an example will be described in which the feature vector calculation unit 27b uses a single two-dimensional still image as an input image.

  Specifically, the feature vector calculation unit 27b, for each of a plurality of pixels included in the image, uses the signal value of the pixel and other pixels that are spatially close to the pixel as elements of the feature vector. Based on the signal value, the fluctuation amount of the signal value of the pixel is calculated.

For example, the feature vector v (x, y) includes a representative signal value v ₀ (x, y) that is a signal value representing the coordinate (x, y) and a signal value fluctuation amount v at the coordinate (x, y). ₁ (x, y). For example, v ₀ (x, y) is an average of signal values, and v ₁ (x, y) is a standard deviation of signal values.

In this case, for example, the coordinates indicating each pixel included in the image is (x, y), the neighborhood of (x, y) preset by the operator is R, the number of pixels included in R is N, When the signal value is s (x, y), the feature vector v (x, y) = (v ₀ (x, y), v ₁ (x, y)) is expressed by the following equation (1). .

  Note that the neighborhood R is desirably set so that pixels corresponding to the same body tissue are included therein as much as possible.

  FIG. 4 is a diagram illustrating an example of the neighborhood R used by the feature vector calculation unit 27b according to the first embodiment. For example, as illustrated in FIG. 4, the feature vector calculation unit 27b centers on (x, y) like the pixels 301 to 309 around the pixel 305 when noise removal of the pixel 305 is performed. Let 3 × 3 block be neighborhood R. In this case, N = 9.

Here, if the neighborhood R block is too wide, for example, a feature vector may be calculated from a signal sequence collected over two types of organs. There is a risk of decline. On the other hand, if this block is too narrow, there is a risk that the accuracy of v ₀ (x, y) and v ₁ (x, y) will be reduced, and the accuracy of noise amount estimation performed later will be reduced. For example, the operator of the MRI apparatus 100 selects the vicinity R using the knowledge of the subject so that such a problem does not occur.

  Returning to FIG. 3, subsequently, the correction map calculation unit 27c calculates a correction map for correcting the feature vector v (x, y) based on the sensitivity map generated by the image generation unit 25 (step S107). .

  For example, the value of each coordinate in the correction map is M (x, y). For example, M (x, y) is a positive real number and is set to a large value when the S / N at the coordinates (x, y) is expected to be high. Here, the value of M (x, y) is set using a value representing the sensitivity of the receiving coil at each point of the sensitivity map. The value of each point in the sensitivity map has higher sensitivity as the point is closer to the receiving coil. In other words, the correction map has a larger value as the sensitivity of the receiving coil is higher.

  Subsequently, the correction unit 27d corrects the feature vector v (x, y) calculated by the feature vector calculation unit 27b using the correction map M (x, y) calculated by the correction map calculation unit 27c ( Step S108). Hereinafter, the corrected feature vector is referred to as a corrected feature vector v ′ (x, y).

  For example, with respect to at least one element of the feature vector of each of the plurality of pixels included in the image, the correction unit 27d has a smaller value corresponding to a position having a smaller value in the correction map M (x, y). A correction is made so that a value corresponding to a position having a large value in M (x, y) becomes a larger value. Thereby, the observation conditions for the fluctuation amount of the signal value obtained from the pixels having different S / N can be made uniform.

For example, in the corrected feature vector v ′ (x, y), v ₀ (x, y) after correction is v ₀ ′ (x, y), and v ₁ (x, y) after correction is v ₁ ′ (x , Y), it is expressed by the following equation (2).

  Subsequently, the model selection unit 27e selects one noise model from a plurality of noise models stored in advance in the storage unit 26 using the corrected feature vector v ′ (x, y) (step S109). .

  For example, the model selection unit 27e selects a noise model that most closely approximates the data point group represented by the correction feature vector v ′ (x, y) from among a plurality of noise models. For example, the noise model here outputs a noise amount with respect to an input signal value. The smaller the input signal value is, the smaller the noise amount is output, and the larger the input signal value is, the larger the noise model is. As the amount of noise is output and the input signal value increases, the output amount of noise converges to a constant value.

FIG. 5 is a diagram illustrating an example of noise model selection performed by the model selection unit 27e according to the first embodiment. FIG. 5 shows the distribution of the correction feature vector v ′ (x, y) obtained by the correction unit 27d, the horizontal axis indicates v ₀ ′ (x, y), and the vertical axis indicates v ₁ ′. (X, y) is shown.

  For example, as shown in FIG. 5, it is assumed that a data point 401 represented by a corrected feature vector v ′ (x, y) has been obtained. Here, one data point 401 corresponds to one correction feature vector v ′ (x, y).

  For example, when the value s corresponding to the signal level is input to the storage unit 26, a plurality of noise models that output the value σ corresponding to the noise level are stored. For example, the noise model is expressed by the following equation (3).

  This noise model is represented by a straight line such as a broken line 402 and a broken line 403 shown in FIG. For example, the model selection unit 27e selects, from among a plurality of linear noise models stored in the storage unit 26, a model that passes through the nearest neighborhood of a data point group formed by a plurality of correction feature vectors. For example, the model selection unit 27e is represented by the broken line 402 when the noise model represented by the broken line 402 and the noise model represented by the broken line 403 illustrated in FIG. Select a noise model. As a result, the noise amount based on the S / N as a reference can be predicted from the signal value.

As a specific method, for example, the model selection unit 27e may perform Hough transform using the data point group shown in FIG. 5 as an input. Also, the difference between the estimated value σ of the noise amount estimated using the representative signal value v ₀ ′ (x, y) at each coordinate and the actually observed signal value variation v ₁ ′ (x, y). A noise model that minimizes the sum of absolute values of may be selected. That is, the model selection unit 27e selects the noise model F (s) represented by the following equation (4) when the plurality of noise models stored in the storage unit 26 is Fi (s).

  Returning to FIG. 3, subsequently, the denoising processor 27f estimates the noise amount of each pixel included in the image using the noise model F (s) selected by the model selector 27e, and the obtained noise Noise removal is performed based on the amount (step S110).

  At this time, for example, the denoise processing unit 27f increases the noise removal strength as the amount of noise output from the noise model selected by the model selection unit 27e increases.

Specifically, the denoising processor 27f first obtains F (s (x, y)) using the signal value s (x, y) before noise removal. Next, using the correction map M (x, y), the correction unit 27d performs a correction opposite to the correction performed on v ₁ (x, y), and estimates the noise amount at the coordinates (x, y). The value σ (x, y) is obtained.

  After that, the denoising processor 27f increases the intensity of the denoising process for determining the output at the coordinates (x, y) as the estimated value σ (x, y) of the noise amount increases, and the signal value s ′ (x, y y) is obtained.

  Here, the method of increasing the strength of the denoising process is based on the noise removal method adopted by the denoising processor 27f. For example, as described in Non-Patent Document 1, in the case of processing using a weighted average of only signals whose difference from s (x, y) is equal to or less than the threshold, the threshold may be increased. Further, if filtering processing is used, the filter tap width may be widened, or a coefficient designed to make it difficult for high frequencies to pass through may be employed.

  Then, the denoising processor 27f displays the image from which noise has been removed on the display unit 22 (step S111). Note that the denoising processor 27 f may store the image from which noise has been removed in the storage unit 26.

  In the above-described processing procedure, the example in which the scan for collecting the sensitivity map is executed as the preparation scan has been described, but the embodiment is not limited thereto. For example, a scan that collects a sensitivity map may be performed during the main scan. In addition, a sensitivity map collected in another examination for the same subject may be used.

  As described above, the MRI apparatus 100 according to the first embodiment corrects the signal value of each pixel using information indicating the distribution of noise generated in the pixel according to the sensitivity distribution of the receiving coil 8. Therefore, according to the MRI apparatus 100 according to the first embodiment, even when the S / N difference occurs between a plurality of pixels included in an image due to the non-uniform sensitivity of the receiving coil 8, noise of all pixels is reduced. Can be removed appropriately. Thereby, it is possible to suppress the loss of the signal characteristics in the image by removing the noise.

  In the present embodiment, an example in which a correction map is calculated based on information indicating the distribution of noise generated in pixels according to the sensitivity characteristic of the receiving coil has been described. However, the imaging condition is the sensitivity characteristic of the receiving coil. Not limited to. For example, when an S / N difference occurs between pixels due to other imaging conditions, a correction map may be calculated based on information indicating a noise distribution according to such imaging conditions.

  Further, the above-described embodiment may be modified and implemented as in a modification described below.

(First modification)
For example, in the first embodiment described above, an example in which the feature vector calculation unit 27b uses a single two-dimensional still image as an input image has been described. However, the dimension of the input image is not necessarily limited to two dimensions. Absent. For example, the input image may be three-dimensional volume data. That is, the dimension of the input image may be increased such that the signal value s (x, y, z) of the input image and the correction map M (x, y, z) are used. In addition to this, when inputting a two-dimensional moving image, it is sufficient to handle s (x, y, t) and the correction map M (x, y). In this case, the signal value s (x, y, z, t) of the input image and the correction map M (x, y, z) may be used.

(Second modification)
Further, for example, the shape of the neighborhood R used by the feature vector calculation unit 27b may be freely changed depending on knowledge at the time of imaging. For example, the scan control unit 27a further accepts the setting of the neighborhood R when accepting the setting of the imaging plan from the operator. Then, the feature vector calculating unit 27b calculates a feature vector using the neighborhood R received by the scan control unit 27a.

  For example, when a moving image of the heart is captured by the MRI apparatus, the doctor instructs the subject (patient) not to move, so that the subject can be regarded as hardly moving. As described above, when it can be assumed that the subject does not substantially move and the input image is a two-dimensional moving image, for example, the feature vector calculation unit 27b performs the processing for each of a plurality of pixels included in the image. The amount of change in the signal value of the pixel may be calculated based on the signal value of the pixel and the signal values of other pixels that are close in time to the pixel as elements of the feature vector.

  FIG. 6 is a diagram illustrating a setting example of the neighborhood R according to the second modification. For example, as illustrated in FIG. 6, the feature vector calculation unit 27b has the same position (x, y) in each frame when a plurality of two-dimensional frames captured at different times t are input as input images. The range of the pixels 501 to 503 may be set as the neighborhood R. That is, the feature vector calculation unit 27b may set the neighborhood R along the time direction.

  In addition, for example, when 3D volume data obtained by imaging the lumbar spine can be considered that the same body tissue continues in the direction of the spine, the range of a plurality of pixels arranged along the direction of the spine It may be set as R. That is, the feature vector calculation unit 27b may set the neighborhood R along the shape of the part to be imaged.

  For example, as described above, a pulse sequence group including a pulse sequence for a preparation scan and a pulse sequence for a main scan is managed for each imaging region and imaging purpose in order to assist the operator in setting an imaging plan. In the case of being present, the neighborhood R may be set and managed according to the imaging region and the imaging purpose. In that case, the neighborhood R may be changeable via the GUI, or may not be changed without being displayed on the GUI.

  Note that, for example, when a part whose shape changes greatly over time, such as the heart, is a target part to be inspected, the standard deviation of the signal value may be prominently increased for a pixel with a large movement. obtain. However, in general, the region where MR signals are collected is often set larger than the target region in consideration of aliasing artifacts and the like. Become dominant. In this regard, in the noise removal method described above, a noise model that most closely approximates the feature vector data group is selected, so even if there is a large signal value with a large standard deviation, the result is a portion with little motion. The noise model is selected based on the feature vector of the pixels. For this reason, the amount of noise is also estimated for pixels with a large amount of motion based on other portions with a small amount of motion. Therefore, according to the above-described method, for example, even when a part whose shape changes greatly over time, such as the heart, is a target part to be examined, without damaging the characteristics of the anatomical movement of the target part, Image noise removal can be performed.

(Third Modification)
Further, for example, the feature vector calculation unit 27b does not necessarily have to calculate the feature vector for all pixels included in the image. For example, the calculation may be made only for pixels in which both x and y are even numbers, and when a pixel in which a body tissue is not drawn in advance is known, the calculation from that pixel is not performed. It may be. That is, the feature vector calculation unit 27b may calculate a feature vector by thinning out pixels. Thereby, calculation cost can be reduced.

When the calculated signal value fluctuation amount v ₁ (x, y) is extremely large, there is a possibility that the boundary of the body tissue is included in the neighborhood R used for calculating the variance. is there. Therefore, for example, the correction unit 26d may not use the feature vector calculated by the feature vector calculation unit 26b for such a pixel by using a threshold value set in advance by the operator.

(Fourth modification)
Further, for example, for v ₀ (x, y) among the elements of the feature vector, the median or mode (mode) of the signal value may be used, and for v ₁ (x, y). The variance, the difference between the signal maximum value and the minimum value, or the like may be used. For example, the Rice distribution known as a noise model of an image captured by an MRI apparatus is known to approach a Gaussian distribution when the signal level is high, and is theoretically compatible with the average and standard deviation. It is expected.

(Fifth modification)
Further, for example, the elements of the feature vector are not necessarily limited to two, that is, a signal value representing coordinates and a variation amount in coordinates. For example, the feature vector calculation unit 27b may include the value of the sensitivity map used for calculating the correction map as an element of the feature vector.

  In addition to this, the feature vector calculation unit 27b uses, for example, a sensitivity difference between reception coils using a plurality of reception coils as in the SENSE method described in Non-Patent Document 2. If the image is obtained by a high-speed imaging method that performs imaging, a feature factor indicating the accuracy of signal value calculation in image reconstruction may be included in the element of the feature vector.

  In the case of performing such extension, as shown in the following formula (5), a term of l representing the value of the sensitivity map or g representing the geometry factor is added to the noise model F (s). In equation (5), a, b, c, and d are real numbers.

(Sixth Modification)
Further, for example, the correction map M (x, y) is obtained by a high-speed imaging method in which an input image is captured using a plurality of receiving coils and using a sensitivity difference between the receiving coils. In some cases, it may be calculated so that the position where the calculation accuracy of the signal value in image reconstruction is lower has a larger value. In fact, Non-Patent Document 2 specifies that S / N is the inverse of Geometry factor. Note that the Geometry factor does not affect the signal level of the image, unlike the sensitivity of the receiving coil. Therefore, for example, as shown in the following equation (6), in the equation (2) used for the feature vector, the representative signal value v ₀ (x, y) that is a signal value representing the coordinates is not corrected. It may be changed.

Further, the correction unit 27d may correct the feature amount of each of the plurality of pixels further based on a geometry factor used in the high-speed imaging method. In this case, for example, the correction map calculation unit 27c calculates the value of the correction map M (x, y) by dividing the value of the sensitivity map of the receiving coil by the Geometry factor or the value of the sensitivity map of the receiving coil. It is a composite of two values, like the one obtained by subtracting the Geometry factor. Furthermore, the correction map need not be limited to one, and two correction maps may be used in combination. For example, a correction map M ₀ (x, y) caused by data that affects the signal level and a correction map M ₁ (x, y) caused by data that does not affect the signal level are prepared. As shown in Expression (7), the feature vector may be corrected.

(Seventh Modification)
Further, for example, the noise model F (s) is not necessarily limited to the straight line shape shown in Expression (3). For example, the noise model F (s) may be expressed by a special function such as log (s) or s ^1/2 , a piecewise broken line, or the like.

(Eighth modification)
Further, for example, the noise model F (s) does not necessarily need to include a term corresponding to the signal value as shown in Expression (3). Such a noise model is represented as a horizontal line like a broken line 403 in the example shown in FIG. For example, in the Rice distribution known as a noise model of an image captured by an MRI apparatus, it is known that the Gaussian distribution approaches when the signal level is high, but the standard deviation of the Gaussian distribution does not depend on the signal level. When such a model is used, the input image may be binarized with a signal value, and only the bright part may be used for feature vector calculation. In this method, the calculation amount of the model selection unit 27e can be reduced, so that high-speed processing can be expected.

(Ninth Modification)
For example, the denoising processor 27f may change the noise removal method according to the characteristics of the subject. For example, as described above, when a moving image of the heart is captured by the MRI apparatus, since the doctor instructs the subject (patient) not to move, the subject can be regarded as not moving substantially. In such a case, the denoising processor 27f may perform noise removal using filtering in the time direction as described in Non-Patent Document 1.

  For example, the denoising processor 27f may change the noise removal method according to the imaging conditions. The imaging conditions here include, for example, an imaging part and an imaging method. For example, the denoise processing unit 27f performs noise removal using filtering in the time direction when the imaging region is the heart. In addition, for example, the denoising processor 27f performs noise removal using filtering in the spatial direction when an imaging method using a contrast agent is performed.

(10th modification)
In the sixth modification, the example in which the feature amount is corrected based on the geometry factor used in the high-speed imaging method has been described, but the embodiment is not limited thereto. For example, depending on the type of high-speed imaging method, the sensitivity of the image may be corrected in the process of reconstructing the image without performing sensitivity correction using the sensitivity map. When such a high-speed imaging method is used, the correction unit 27d may correct the feature amount based on the Geometry factor without using the sensitivity map. In this case, for example, the correction map calculation unit 27c calculates the correction map based on the Geometry factor without using the sensitivity map.

  Further, in the examination of the heart, in one examination, in addition to the scan that collects cross-sectional images such as the short axis image and the long axis image, the scan that collects the cine image (moving image) of the heart, A scan may be performed that collects heart volume data. In such a case, the MRI apparatus 100 may switch the noise removal method according to the type of image collected in each scan. Specifically, the MRI apparatus 100 performs noise removal on the cross-sectional image by the method using the two-dimensional still image described in the first embodiment as an input image, and the cine image is the first modified example. The noise removal is performed by the method using the moving image described in the above as an input image, and the volume data is removed by the method using the volume data described in the first modification as an input.

(Eleventh modification)
Further, for example, when the input image is a moving image, the noise removal method may be changed between a region where the subject moves greatly (motion region) and a region where the movement is small (still region).

  In this case, the image generation unit 25 generates a moving image based on the MR signal data. In addition, the denoising processor 27f detects a moving area and a still area from the moving image generated by the image generating section 25, and reduces noise by using different methods for each area. Note that the still area here is an area where the movement is smaller than the movement area, and is not necessarily limited to a completely stationary area.

For example, the denoising processor 27f selects pixels whose temporal fluctuation amount of the signal value of the pixel at the same position in each of a plurality of frames included in the moving image is larger than a threshold based on the noise amount obtained from the noise model. It detects as a pixel. In addition, the denoising processor 27f detects pixels whose temporal fluctuation amount of the signal value of the pixel at the same position in each of the plurality of frames included in the moving image is equal to or less than a threshold based on the noise amount obtained from the noise model. It detects as a pixel.
FIG. 7 is a diagram illustrating an example of noise removal according to the eleventh modification. For example, the diagram shown on the left side of FIG. 7 shows a moving image 602 including a heart 601.

In the center diagram of FIG. 7, the horizontal axis indicates the noise amount σ (x, y) of each pixel calculated by this method, and the vertical axis indicates the temporal fluctuation amount v ₁ (x, y) of the signal value. y). Here, for example, v ₁ (x, y) is the standard deviation of the signal value. When v ₁ (x, y) and σ (x, y) are calculated for each pixel on the image, points such as the point 606 can be arranged by the number of pixels. At this time, since the fluctuation amount v ₁ (x, y) of the signal of the pixel belonging to the non-moving subject is expected to coincide with the noise amount σ (x, y), the point corresponding to the still region has an inclination of 1 It is also expected to ride on the straight line (straight line 605). Since the variation amount v ₁ (x, y) is affected by the motion of the heart 601 in the pixel belonging to the motion region, the pixel corresponding to the point existing above the straight line 605 like the point 607 is in the motion region. The possibility of belonging increases.

Using this property, the denoising processor 27f compares the noise amount σ (x, y) obtained from the noise model with v ₁ (x, y), for example, as shown in the center diagram of FIG. Then, it is determined whether the pixel (x, y) belongs to the motion region or the still region. Specifically, when there is a positive real number A specified in advance by the user and a noise amount σ (x, y) obtained from the noise model, v ₁ (x, y)> Aσ (x, y) If there is, it is determined as a motion region, otherwise it is determined as a still region. In the middle diagram of FIG. 7, the threshold value is represented by a straight line such as a straight line 603.

Then, the denoising processing unit 27f is, v ₁ (x, y) is the threshold value (linear 603) is detected as a pixel region motion larger pixels, v ₁ (x, y) is equal to or less than the threshold value (linear 603) pixels Are detected as pixels in a still region.

  As a result, for example, as shown in the diagram on the right side of FIG. 7, in the moving image 601, a region 604 including the heart 601 and its peripheral portion is detected as a motion region, and a region other than the region 604 is detected as a still region. Is done.

  Thereafter, the denoising processor 27f performs noise removal using different methods for each pixel included in the motion region and each pixel included in the still region. For example, the denoising processor 27f makes the noise removal strength for the pixels included in the motion region weaker than the noise removal strength for the pixels included in the still region.

  FIG. 8 is a flowchart showing a noise removal processing procedure according to the eleventh modification. The processing in steps S201 to S207 shown in FIG. 8 corresponds to the processing in step S110 shown in FIG. That is, in the present modification, after the processing similar to steps S101 to S109 shown in FIG. 3 is executed, the processing of steps S201 to S207 shown in FIG. 8 is executed, and thereafter, the processing in steps S110 and S110 shown in FIG. Similar processing is executed.

  For example, as illustrated in FIG. 8, the denoising processor 27f estimates the noise amount of each pixel included in each frame of the moving image using the noise model F (s) selected by the model selecting unit 27e ( Step S201). Specifically, the denoising processor 27f obtains an estimated value σ (x, y) of the noise amount at the coordinates (x, y) using the noise model F (s) as described above for each frame. .

Thereafter, the denoising processor 27f compares, for each pixel at the same position included in each frame, a numerical value representing the magnitude of the temporal fluctuation amount of the signal value with a threshold value based on the estimated noise amount (step S202). Here, for example, as the numerical value indicating the magnitude of the temporal fluctuation amount of the signal value, the fluctuation amount v ₁ (x, y) of the signal value shown in Expression (1) is used. For example, as the threshold value, a value obtained by multiplying σ (x, y) by a predetermined constant magnification is used. For example, a value that is three times σ (x, y) is used as the threshold value.

  Then, the denoising processor 27f determines that a pixel whose numerical value indicating the magnitude of the fluctuation amount of the signal value is larger than the threshold value is a pixel in the motion region (Yes in step S202). On the other hand, the denoising processor 27f determines that the pixel whose numerical value indicating the magnitude of the fluctuation amount of the signal value is equal to or smaller than the threshold is a still region pixel (No in step S202).

  Thereafter, the denoising processor 27f performs noise removal for each of the pixels in the motion area and the pixels in the still area. At this time, the denoising processor 27f changes the noise removal method between the motion region and the still region so that the noise removal strength for the pixels in the motion region is weaker than the noise removal strength for the pixels in the still region.

  For example, as described in Non-Patent Document 1, the denoising processor 27f performs noise removal by performing processing using a weighted average of only signals whose difference from s (x, y) is equal to or less than a threshold. Do. In this case, for example, the denoising processor 27f sets the threshold value Tm used for noise removal for the pixels in the motion area relatively lower than the threshold value Ts used for noise removal for the pixels in the still area.

  For example, the denoising processor 27f sets the threshold value to Tm = Cm * σ (x, y) for the motion region and sets the threshold value to Ts = Cs * σ (x, y) for the still region. Here, both Cm and Cs are positive real numbers, and Cm <Cs.

  As a result, when the noise amount is equal to or less than the threshold value Tm for the pixels in the motion region (No in step S203), the denoising processor 27f performs a weighted average of the signal values, thereby reducing the relatively weak noise. Removal is performed (step S204). In addition, when the amount of noise is larger than the threshold value Tm (Yes in step S203), the denoising processor 27f performs noise removal with a relatively medium intensity by performing normal averaging. (Step S205).

  In addition, when the noise amount is equal to or less than the threshold value Ts (No in step S206), the denoising processor 27f performs a weighted average of the signal values so that the intensity is relatively moderate. Noise removal is performed (step S205). In addition, when the noise amount is larger than the threshold value Ts (Yes in step S206), the denoising processor 27f performs noise removal with relatively strong intensity by performing normal averaging. (Step S207).

  As described above, by setting the threshold value Tm used for noise removal for the pixels in the motion region relatively lower than the threshold value Ts used for noise removal for the pixels in the still region, the signal in the motion region is changed by noise removal. It becomes difficult to do. As a result, it is possible to suppress the signal in the motion region from being weakened by noise removal.

  Further, in this way, the denoising processor 27f includes the pixel included in the motion region and the pixel included in the still region if the amount of noise obtained from the noise model is the same. The intensity of noise removal for a pixel to be detected is made lower than the intensity of noise removal for a pixel included in the still region.

  Note that the noise removal strength in the motion region is not always smaller than the noise removal strength in the still region. For example, if σ (x1, y1) = σ (x2, y2) at one point (x1, y1) on the motion region and one point (x2, y2) on the still region, Cm <Cs. Therefore, Tm <Ts is always satisfied. However, when σ (x1, y1) is larger than Cs / Cm times σ (x2, y2), Tm> Ts can be satisfied.

  Further, the method for changing the noise removal strength between the motion region and the stationary region is not limited to the above-described method. For example, as described in Non-Patent Document 3, in the time direction filtering, a method of increasing the high frequency gain of the motion region filter compared to the high frequency gain of the stationary region filter may be used. .

(Second Embodiment)
In addition, in 1st Embodiment mentioned above, although embodiment of the MRI apparatus was described, embodiment is not restricted to this. For example, the same embodiment can be applied to an image processing apparatus connected to an MRI apparatus so as to be communicable via a network. Hereinafter, an embodiment of an image processing apparatus will be described as a second embodiment.

  FIG. 9 is a block diagram illustrating a configuration of the image processing apparatus 30 according to the second embodiment. For example, as illustrated in FIG. 9, the image processing apparatus 30 is connected to the MRI apparatus 100 via a network 40 so as to be communicable. The image processing apparatus 30 includes an input unit 31, a display unit 32, a storage unit 33, and a control unit 34. Here, for example, the control unit 34 includes an image acquisition unit 34a, a feature vector calculation unit 34b, a correction map calculation unit 34c, a correction unit 34d, a model selection unit 34e, and a denoising processing unit 34f.

  The image acquisition unit 34a acquires an image captured by the MRI apparatus 100 by communicating with the MRI apparatus 100 via the network 40. For example, the image acquisition unit 34a receives an operation for designating an image to be processed from the operator via the input unit 31, and designates the designated image and the sensitivity map of the receiving coil when the image is captured. Obtained from the MRI apparatus 100. Then, the image acquisition unit 34 a stores the acquired image and sensitivity map in the storage unit 33.

  The feature vector calculation unit 34b, the correction map calculation unit 34c, the correction unit 34d, the model selection unit 34e, and the denoising processing unit 34f are the feature vector calculation unit 27b and the correction map calculation unit 27c described in the first embodiment, respectively. The correction unit 27d, the model selection unit 27e, and the denoising processing unit 27f have the same configuration and functions.

  Here, in the second embodiment, the feature vector calculation unit 34b acquires the image generated by the image acquisition unit 34a from the storage unit 33, and calculates a feature vector for each of a plurality of pixels included in the acquired image. To do. In the second embodiment, the denoising processor 34 f displays an image from which noise has been removed on the display unit 32. Note that the denoising processor 34 f may store the image from which noise has been removed in the storage unit 33.

  The image acquisition unit 34a may acquire MR signal data for sensitivity map and MR signal data for diagnosis collected in the same examination, instead of acquiring an image and a sensitivity map from the MRI apparatus 100. In this case, for example, the control unit 34 generates a sensitivity map from the MR signal data for sensitivity map and the image from the MR signal data for diagnosis, similarly to the image generation unit 25 described in the first embodiment. And the generated sensitivity map and image are stored in the storage unit 33.

  Among the above-described units, the control unit 34 is, for example, a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit), a memory, an ASIC (Application Specific Integrated Circuits), an FPGA (Field Programmable Gate Array), or the like. It is comprised by the electronic circuit. In this case, for example, the processor included in the control unit 34 executes the process according to the processing procedure by reading the program that defines the processing procedure of the processing performed by the control unit 34 from the storage unit 33 and executing the program.

  According to the image processing apparatus 30 according to the second embodiment described above, even when an S / N difference occurs between a plurality of pixels in an image captured by the MRI apparatus 100, noise of all pixels is appropriately removed. can do. Thereby, in the image imaged by the MRI apparatus 100, it can suppress that the characteristic of a signal is lost by removing noise.

  In the MRI apparatus or the image processing apparatus according to the above-described embodiment, the program stored in the storage unit may be installed in each apparatus in advance, or a CD-ROM (Compact Disc Read Only Memory). It may be distributed via a recording medium such as the above or a network and may be appropriately installed in each device. Alternatively, the program is, for example, a hard disk, a memory, a CD-R (Compact Disk Recordable), a CD-RW (Compact Disk Rewritable), or a DVD-RAM (Digital Versatile Disc Random Access Memory) incorporated in or attached to each device. Further, the program may be recorded on a storage medium such as a DVD-R (Digital Versatile Disc Recordable), and appropriately read and executed by a processor included in each device.

  According to at least one embodiment described above, noise of all pixels can be appropriately removed even when an S / N difference occurs between a plurality of pixels included in an image.

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

DESCRIPTION OF SYMBOLS 100 Magnetic Resonance Imaging (MRI) apparatus 25 Image generation part 27 System control part 27b Feature vector calculation part 27c Correction map calculation part 27d Correction part 27f Denoising process part

Claims (20)

An image generator for generating an image based on a magnetic resonance signal generated from the subject;
For each of a plurality of pixels included in the image, a feature amount calculation unit that calculates a feature amount related to the signal value of the pixel;
A correction unit that corrects the feature amount of each of the plurality of pixels based on sensitivity characteristics of a receiving coil that receives the magnetic resonance signal;
A magnetic resonance imaging apparatus comprising: a denoising processor that reduces noise of the image based on a distribution of feature values corrected by the correcting unit. The correction unit corrects the feature amount of each of the plurality of pixels further based on a Geometry factor used in high-speed imaging.
The magnetic resonance imaging apparatus according to claim 1. For each of the plurality of pixels, the feature amount calculation unit is based on the signal value of the pixel and the signal values of other pixels located spatially or temporally close to the pixel as elements of the feature amount. To calculate the fluctuation amount of the signal value of the pixel,
The magnetic resonance imaging apparatus according to claim 1. A correction map calculating unit that calculates a correction map based on sensitivity characteristics of a receiving coil that receives the magnetic resonance signal;
The correction unit corrects the feature amount of each of the plurality of pixels using the correction map;
The magnetic resonance imaging apparatus according to claim 1. The value of the correction map is calculated by dividing the sensitivity map value representing the sensitivity distribution of the receiving coil by the Geometry factor used in the high-speed imaging method.
The magnetic resonance imaging apparatus according to claim 4. The correction map has a larger value as the sensitivity of the receiving coil is higher.
The magnetic resonance imaging apparatus according to claim 4. The correction map is a signal value in image reconstruction when the image is obtained by a high-speed imaging method in which a plurality of receiving coils are used and imaging is performed using a sensitivity difference between the receiving coils. The lower the calculation accuracy, the larger the value.
The magnetic resonance imaging apparatus according to claim 4. For the at least one element of the feature vector of each of the plurality of pixels, the correction unit has a smaller value corresponding to a position having a smaller value in the correction map, and corresponds to a position having a larger value in the correction map. Correct so that the larger the value,
The magnetic resonance imaging apparatus according to claim 4. The denoising processing unit decreases the noise removal strength as the value of the correction map is larger.
The magnetic resonance imaging apparatus according to claim 4. The feature amount is represented by a feature vector including a plurality of elements.
The magnetic resonance imaging apparatus according to claim 1. The feature vector includes a standard deviation of the signal value as an element.
The magnetic resonance imaging apparatus according to claim 10. A model selection unit that selects a noise model from a plurality of noise models based on the distribution of the feature amount corrected by the correction unit;
The denoising processor uses the noise model selected by the model selector to reduce the noise of the image.
The magnetic resonance imaging apparatus according to claim 1. The noise model outputs a noise amount with respect to an input signal value. The smaller the input signal value, the smaller the noise amount is output, and the larger the input signal value, the larger the noise amount is output. In addition, as the input signal value increases, the output noise amount converges to a constant value.
The magnetic resonance imaging apparatus according to claim 12. The model selection unit selects a noise model that most closely approximates a data point group represented by the feature amount corrected by the correction unit from the plurality of noise models.
The magnetic resonance imaging apparatus according to claim 12. The denoising processor increases the noise removal strength as the amount of noise output from the noise model selected by the model selector increases.
The magnetic resonance imaging apparatus according to claim 12. The image generation unit generates a moving image as the image,
The denoising processing unit detects a motion region and a still region having a smaller motion than the motion region from the moving image, and reduces noise by a different method for each of the motion region and the still region.
The magnetic resonance imaging apparatus according to claim 12. The denoising processing unit determines a pixel in which a temporal variation amount of a signal value of a pixel at the same position in each of a plurality of frames included in the moving image is larger than a threshold value based on a noise amount obtained from the noise model. Detect as a pixel of
The magnetic resonance imaging apparatus according to claim 16. The denoising processing unit applies a pixel to a pixel included in the motion region when the amount of noise obtained from the noise model is the same between the pixel included in the motion region and the pixel included in the still region. The noise removal strength is weaker than the noise removal strength for the pixels included in the still region.
The magnetic resonance imaging apparatus according to claim 16 or 17. An image acquisition unit for acquiring an image generated based on a magnetic resonance signal generated from a subject;
For each of a plurality of pixels included in the image, a feature amount calculation unit that calculates a feature amount related to the signal value of the pixel;
A correction unit that corrects the feature amount of each of the plurality of pixels based on sensitivity characteristics of a receiving coil that receives the magnetic resonance signal;
An image processing apparatus comprising: a denoising processor that reduces noise of the image based on the distribution of the corrected feature value. Acquire an image generated based on the magnetic resonance signal generated from the subject,
For each of a plurality of pixels included in the image, a feature amount related to the signal value of the pixel is calculated,
Based on the sensitivity characteristics of the receiving coil that receives the magnetic resonance signal, the feature amount of each of the plurality of pixels is corrected,
Reducing noise of the image based on the distribution of the corrected feature value;
An image processing method.

Images