Medical image processing apparatus, medical imaging apparatus, and noise reduction method in medical image

Abstract

The present application provides a medical image processing apparatus, a medical imaging apparatus, and a noise reduction method in a medical image. A technology is provided that can effectively and appropriately perform noise removal even for an image including various noises and artifacts, and an image in which a noise pattern changes depending on an imaging condition. Based on a noise removal technology using AI, noise characteristics including artifacts are analyzed for each image, image classification based on the analysis result is performed, an optimal noise processing neural network is applied for each classification, and noise and artifacts are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to an image diagnostic apparatus such as a magnetic resonance imaging (MRI) apparatus, a CT apparatus, and the like, and particularly relates to a noise reduction technique of an image in the image diagnostic apparatus. BACKGROUND

[0002] As an image diagnostic apparatus, a medical image imaging apparatus such as a magnetic resonance imaging (MRI) apparatus, a CT apparatus, and an ultrasonic imaging apparatus is widely used. In order to reduce a burden on a subject and increase the number of examinations, the importance of high-speed imaging is increasing. In general, in an imaging apparatus, high-speed imaging can be achieved by reducing the amount of data to be acquired. For example, in an MRI apparatus, various high-speed imaging methods such as parallel imaging, compressed sensing, and the like, in which echo signals required for a contrast image matrix are measured (undersampled) and are restored by an operation, are proposed.

[0003] However, in high-speed imaging, there is a problem in that the noise of an obtained image becomes large, that is, the SNR (signal-to-noise ratio) becomes low. This problem is common in a CT apparatus and an ultrasonic apparatus as well.

[0004] In recent years, research and development of quality improvement using AI are actively being conducted, and a high noise reduction effect is exhibited. In Patent Literature 1, an image processing system using a neural network (NN) is disclosed, and it is proposed that the internal parameters of the NN are adjusted on the basis of data associated with an input image, whereby even an image having a different noise amount from a learning image can improve the noise reduction performance.

[0005] PRIOR ART LITERATURE

[0006] PATENT LITERATURE

[0007] Patent Literature 1: Japanese Patent Application Publication No. 2018-206382 SUMMARY

[0008] -PROBLEMS TO BE SOLVED BY THE INVENTION-

[0009] In an image acquired by a medical image imaging apparatus, various quality deteriorations occur due to an imaging method, an imaging condition, and the like. In particular, in a case where an imaging site, a patient's body size, and the like are different, the characteristics of noise change. For example, in an MRI apparatus, if an imaging condition is decided in coordination with a patient to adjust a range of imaging and the like, in conjunction therewith, the characteristics of the spatial distribution of noise change. In addition, depending on the patient's body shape, the spatial distribution of the sensitivity of signal acquisition is different, in conjunction therewith, the characteristics of noise change. For example, as described in Patent Literature 1, even if adjustment corresponding to the size of noise is performed, if processing considering the spatial distribution of noise is not performed, it can be impossible to perform appropriate noise reduction processing, and in this case, a new problem of occurrence of blurring of an image and the like occurs.

[0010] For example, in an MRI apparatus, between a plurality of images taken while making imaging parameters such as a measurement matrix size, a reconstruction matrix size, etc. different, even if the same noise pattern such as Gaussian noise is mixed in the imaging data, if the resulting images are compared, the noise patterns are different, and even if both are processed with the same noise reduction method, sometimes an appropriate noise cannot be obtained.

[0011] The present application has an object to provide a technology capable of effectively and appropriately performing noise reduction even for images in which noise patterns change due to differences in imaging parameters.

[0012] - Means for solving the problem -

[0013] The present application is based on a noise reduction technology using AI, analyzes noise characteristics for each image, performs image classification based on the analysis results, applies the best neural network for noise processing for each classification, and thereby reduces noise.

[0014] That is, the image processing apparatus of the present application is input with an image acquired by a medical imaging device, and outputs an image in which noise is reduced, the medical image processing apparatus having: a plurality of processors for noise reduction; an analysis section that analyzes a signal of the input image and a pattern of noise to classify the input image; and a processor selection section that selects one or a plurality of processors from the plurality of processors based on a classification result of the analysis section, and causes the selected processor to start.

[0015] Further, the noise reduction technology of the present application is a noise reduction method that reduces noise contained in an image acquired by a medical imaging device, including the steps of: analyzing a pattern of noise of the image to classify the input image; selecting a given processor from a plurality of processors for noise reduction prepared for each classification based on a classification result of the step of classifying; and performing noise reduction processing of the input image by the selected processor.

[0016] In addition, the characteristics of noise are mainly determined by the size of noise and the spatial distribution of noise, and the pattern of noise refers to the size of noise and the spatial distribution of noise.

[0017] - Effects of the Invention -

[0018] According to the present application, by selecting and applying a processor most suitable for the noise characteristics of the input image from a plurality of processors prepared in advance according to noise characteristics, inappropriate noise reduction processing can be prevented, and effective noise reduction processing can be achieved. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a diagram showing the overall structure of the medical image processing apparatus of the present application.

[0020] Figure 2 is a diagram showing an outline of the process of the medical image processing apparatus. Figure 1

[0021] Figure 3 is a diagram showing a configuration example of the processor.

[0022] Figure 4 is a diagram explaining a learning step of a plurality of processors.

[0023] Figure 5 is a diagram showing a flow of the process of the medical image processing apparatus using the embodiment.

[0024] Figure 6 is a diagram showing a flow of the noise processing of the embodiment.

[0025] Figure 7 (A), (B) of FIG. 1 are diagrams showing examples of images in which noise patterns are different.

[0026] Figure 8 is a diagram showing that a region in which a range in a histogram is extracted is imaged.

[0027] Figure 9 (A), (B) of FIG. 3 are diagrams showing data of data transformed into a space other than an image space with respect to images in which noise patterns are different.

[0028] Figure 10 (A) to (D) of FIG. 4 are diagrams showing images corresponding to Figure 8 , Figure 9 of FIG. 3 with respect to different images in which noise patterns are different.

[0029] Figure 11 is a diagram showing an example of a GUI in which a user selection of a noise pattern to be analyzed is accepted.

[0030] Figure 12 (A), (B) of FIG. 6 are diagrams showing modification examples of the configuration of the processor, respectively.

[0031] Figure 13 is a diagram showing a medical imaging apparatus as a medical treatment of an MRI apparatus.

[0032] -Explanation of Reference Numerals-

[0033] 1: medical imaging apparatus, 2: image processing apparatus, 21: analysis section, 22: processor selection section, 23: processor, 24: control section, 25: image synthesizing section, 3: display apparatus, 4: input apparatus, 5: storage apparatus, 100: MRI apparatus, 200: image processing section. DETAILED DESCRIPTION

[0034] ​The following describes embodiments of a medical image processing apparatus and a medical imaging apparatus of the present application.

[0035] <Embodiment of Medical Image Processing Apparatus>

[0036] Figure 1 is a diagram showing a configuration example of one embodiment of a medical image processing apparatus. The medical image processing apparatus 2 performs processing of inputting a medical image acquired by a medical imaging apparatus (hereinafter also referred to simply as an imaging apparatus) 1, and performing processing for reducing noise contained in the medical image, and is provided with: a plurality of processors 23 that learn from the noise characteristics (noise pattern) of a medical image to perform optimal noise reduction processing; a noise pattern analysis section (hereinafter referred to as an analysis section) 21 that inputs a medical image as a processing target from the imaging apparatus 1, and analyzes the characteristics of the image (input image) to classify the input image into a given pattern; a processor selection section 22 that selects a processor corresponding to the pattern classified by the analysis section 21 from among the plurality of processors 23; and a control section 24 that controls the actions of the analysis section 21, the processor selection section 22, and the processors 23.

[0037] The medical imaging apparatus 1 is a general medical imaging apparatus such as an MRI apparatus, a CT apparatus, an ultrasonic imaging apparatus, or the like. In a case where the imaging apparatus 1 is provided with an image processing section that processes an image captured by such an imaging apparatus 1, the medical image processing apparatus 2 can function as part of the image processing section of the medical imaging apparatus 1, or can be a device independent of the imaging apparatus 1, and is connected directly or via a network. In addition, it is also possible to connect with a storage device 5 such as a medical image database that stores medical images, and in this case, a medical image is acquired from the database (storage device 5).

[0038] Further, the medical image processing apparatus 2 can be provided with a display device 3 and an input device 4, and can receive a condition or an instruction for processing by a user via the input device 4, or display a processing result on the display device 3.

[0039] The processors 23 are constituted by a plurality of processors 23-1, 23-2,... 23-N whose processing contents differ depending on the noise pattern of an input image to be processed, and each processor is constituted by a neural network (CNN) provided with an input layer, an intermediate layer, and an output layer. In processing using such a CNN, a high-quality (noise-reduced) image as a target is obtained by repeatedly performing convolution processing of an image in the intermediate layer. Optimization of the convolution processing, that is, optimization of the weights of the nodes of each layer, the count of the convolution, is performed by data learning, but the processors 23-1, 23-2,... 23-N learn in advance a plurality of images whose noise patterns differ as learning data, and set the weights of the nodes of each layer, the coefficients of the convolution, and the like so that appropriate noise reduction processing is performed for each noise pattern. In addition, the number of intermediate layers can differ for each processor.

[0040] The analysis section 21 analyzes and classifies the noise pattern of the input image before the processing by the processor 23 described above. In the analysis, a method of using data that transforms an image into a space different from the image space and using a characteristic value in the space, a method of a characteristic of the image itself such as texture analysis, and the like can be used, and these methods can also be appropriately combined.

[0041] The learning step of the processor of the plurality of processors 23-1 to 23-N classifies the noise of a large number of images in advance by the same method as the analysis and classification by the analysis section 21 described above, and learns using each classified learning data set. The analysis and classification of this learning step can be performed by the analysis section 21 of the medical image processing apparatus 2 of the present embodiment, and can also be performed using another image processing apparatus having the same structure as the analysis section 21.

[0042] The processor selection section 22 selects the processor most suitable for processing the input image on the basis of the analysis result of the analysis section 21.

[0043] The control section 24 controls the operation of each section described above, and when the processor selection section 22 selects a given processor 23 for the input image, controls the operation so that the selected processor 23 processes the input image.

[0044] The image processing apparatus 2 is configured of a computer or the like provided with a processor such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a memory, and the CPU reads and executes a program saved in the memory, whereby the functions of the processors 23-1 to 23-N, the analysis section 21, and the processor selection section 22 are realized by software. In addition, a part or all of the processors 23-1 to 23-N, the analysis section 21, and the processor selection section 22 can also be realized by hardware. For example, a circuit design can be performed so that the functions of the processors 23-1 to 23-N, the analysis section 21, and the processor selection section 22 are realized using a custom IC such as an ASIC (Application Specific Integrated Circuit), a programmable IC such as an FPGA (Field-Programmable Gate Array).

[0045] Figure 2An outline of the processing of the medical image processing apparatus 2 having the above-described structure is shown. As shown in the drawing, the basic processing performed by the medical image processing apparatus 2 includes, after an image to be processed (input image) is input, a step S201 of analyzing noise, a step S202 of classifying the image based on the analysis result, a step S203 of selecting the best processor for the classified image, and a step S204 of applying the selected processor to reduce the noise.

[0046] As a premise of these processes, there is a process of learning each processor (learning step), but the learning step does not necessarily need to be performed in the same medical image processing apparatus 2, and can be performed by an image processing apparatus different therefrom. Among them, learning is performed using an image classified in the same method as the analysis method performed by the analysis section 21 of the medical image processing apparatus 2 as a learning set.

[0047] Further, the image after the noise reduction can be displayed on the display apparatus 3 in the case where the medical image processing apparatus 2 is provided with the display apparatus 3, can be transferred to the imaging apparatus 1, or can be displayed on the display apparatus of the imaging apparatus 1.

[0048] Based on the outline of the structure and the processing of the medical image processing apparatus 2 described above, a specific embodiment of the processing is described below.

[0049] <EMBODIMENT>

[0050] In the present embodiment, the processing in the case where the image is an image obtained by an MRI apparatus and the pattern of the noise mainly differs depending on the size of the measurement matrix is described.

[0051] [Learning of Processors]

[0052] The learning of the processors 23 is performed, and each processor 23-1 to 23-N is made as a learning model. Figure 3 An example of a CNN (Convolutional Neural Network) constituting the processor is shown. The example shown is an example of an SRCNN (Super Resolution CNN), and in the drawing, f1 to f3 are the kernel size of the intermediate layer, and N1 to N3 are the depth of the layer. Based on the configuration of this CNN, the kernel size and the coefficients of the kernel (convolution coefficients) are optimized by learning, but the number of layers, the depth, and the like can be further changed.

[0053] As Figure 4As shown, learning is performed using a large number of medical images obtained by full sampling as teaching images, and images (hereinafter referred to as noisy images) obtained by undersampling that reduces the sampling and the teaching images as a set. The analysis section 21 analyzes the noise pattern included in the noisy images, and classifies the noise patterns into a plurality of patterns (1 to N) based on the noise patterns. The analysis, for example, calculates a characteristic value based on the teaching images, and classifies the characteristic values (in addition to numerical values, vectors, matrices are also included) into a plurality of types. The characteristic value can be a value obtained by numerically converting the texture of the image itself using a Gray Level Co-occurrence Matrix (GLCM), or a value obtained by extracting a characteristic value using a result of converting the image into spatial data other than the image space by performing Fourier transform, wavelet transform, or discrete cosine transform. A plurality of characteristic values can also be combined to perform classification. The details of the analysis method will be described later.

[0054] After classifying the images in this way, the teaching images and the noisy images are learned as a learning data set for each noise pattern. A small block image is cut out from the learning data set, and learning is performed on a set of more than several ten thousand images to create a processor for each pattern.

[0055] In addition, at the time of classification, not only the characteristic values extracted from the images, the transformed images, but also information on the imaging conditions at the time of imaging, such as the imaging parameters such as the measurement matrix size, can also be included in the elements at the time of classification.

[0056] In the case where the measurement matrix size is added as an element of classification, it is also possible to learn for each measurement matrix size to create a processor for each measurement matrix, but it is also possible to generate a processor for an image with a measurement matrix size different from that used in the learning data set by adjusting the kernel size and the coefficients of the kernel of the processor learned using an image data set with a given measurement matrix size.

[0057] [Noise / Artifact Removal Processing]

[0058] After learning a plurality of processors corresponding to the noise patterns, the medical image processing apparatus 2 performs noise processing on an actually input medical image. An outline of the noise processing is shown in Figure 5 As shown, if the image processing apparatus 2 is input with a medical image (S501), the analysis section 21 analyzes the noise of the input image (S502), and based on the analysis result, classifies the input image into one of a plurality of noise patterns decided in advance (S503). The analysis and classification performed by the analysis section 21 are the same as those performed at the time of creating the learning model, and are performed based on the characteristic values extracted from the features of the image.

[0059] The processor selection section 22 selects the processor 23 corresponding to the noise pattern to which the input image is classified (S505). In a case where the analysis result of the analysis section 21 is, for example, a characteristic value for a plurality of patterns, a case where it is a value between a plurality of measurement matrix sizes used in classification, and the like, a processor for a plurality of patterns can be used in combination, or a processor corresponding to a group for which a characteristic value is closest to a central value in the classified group can be used.

[0060] Further, in a case where the result of the analysis of the analysis section 21 is, for example, a case where the noise amount greatly exceeds the range of the noise amount of the learning data used for learning of the processor 23, or a case where artifacts are large and cannot be applied to classification, and the like, the processor selection section 22 regards that there is no optimal processor (S504), and can display it on the display device 3 (S507), whereby it is possible to prevent inappropriate processing.

[0061] The processor 23 selected by the processor selection section 22 performs the quality improvement processing on the input image (S506). In the CNN-based processing, the reliability of the processing can be obtained as, for example, a probability of the output result, and thus it is also possible to judge the reliability in accordance with the probability and the like, and display it on the display device 3 (S507).

[0062] Next, the processing of the analysis section 21 will be described. Here, as an example of images different in noise pattern, an example in which a characteristic value is calculated for two images different in measurement matrix size, which are images acquired by an MRI apparatus, will be described. In this example, the measurement matrix size of the image of (A) is 512 x 512, and the measurement matrix size of the image of (B) is 256 x 256. Figure 6 The flow of the analysis processing is shown.

[0063] The noise analysis section 21 analyzes the pattern of noise in the input image as a noise characteristic.

[0064] The size of the noise (noise level) can be calculated, for example, from a histogram of a noisy image. The analysis section 21 calculates the noise level from the histogram, and classifies (large classification) based on the size of the noise by setting a threshold value thereto (S511).

[0065] Next, the analysis of the noise pattern is performed for each group different in noise level (S512).

[0066] Figure 7 An example of images different in noise pattern is shown. Figure 7 is an MR image acquired by making the measurement matrix size different, Figure 7 (A) of is an image in which the measurement matrix size is 512 x 512, Figure 7 (B) of is an image in which the measurement matrix size is 256 x 256, and the reconstruction matrix is 512 x 512. In a case where observation is made in a histogram of the entire image, the same distribution is shown in (A) and (B), and the noise level looks the same.

[0067] Figure 8 Indicating targeting Figure 7 The regions within a given range of histograms in images (A) and (B) were extracted and visualized. Figure 8 The measurement matrix size of (A) is 512×512, and that of (B) is 256×256. As can be seen from the magnified images, the noise patterns are different. Specifically, compared to (A), the pattern in (B) is coarser.

[0068] In order to classify the pattern, the analysis unit 21 first transforms the image into data in a space other than the image space (e.g., sparse space) (S513), and calculates characteristic values ​​based on the characteristics of the transformed spatial data (S514).

[0069] exist Figure 9 In China, regarding Figure 7 Images (A) and (B) illustrate data transformed from noisy images into spatial data outside the image space. Here, the data after Discrete Cosine Transform (DCT) is represented (frequency-space data).

[0070] When comparing data within the same spatial dataset, the characteristics of each data point differ due to variations in noise patterns (in this case, differences in the spatial distribution of noise caused by the size of the measurement matrix). Compared to (A), the value in the lower right corner of graph (B) is smaller, indicating that the noisy image contains fewer high-frequency components. This indicates a coarser noise pattern.

[0071] Therefore, in this embodiment, a discrete cosine transform is first performed on the input image (S513), and the transformed data, such as... Figure 9 As shown, the average value of the four regions of the transformed data is calculated (S514), and the pattern is classified according to a threshold (S515). Alternatively, the ratio of signal values ​​of multiple regions can be used. For example, a threshold of 0.5, 1, 1.5, etc., can be set as a ratio, and the data can be classified into groups with ratios less than 0.5, 0.5 to less than 1.0, 1.0 to less than 1.5, and 1.5 or more.

[0072] and then Figure 10 This illustrates the case where the measurement matrix is ​​asymmetric (in...). Figure 10 In the middle, (A) and (B) are Figure 9 The measurement matrices shown are 512×512 and 256×256. (C) is a measurement matrix of size 512×256, and (D) is 256×512. In case (C), the noise has a small frequency in the longitudinal direction, and in case (D), the noise has a small frequency in the transverse direction. This can also be seen in the DCT results.

[0073] The above, using Figures 6-10 The above, using

[0074] For example, the analysis section 21 can use a Fourier transform, a wavelet transform, or the like, as a transform other than a discrete cosine transform (DCT). Further, instead of transforming into non-image space data, classification can be performed based on characteristics of the image itself. For example, texture analysis of the image can be performed, a GLCM (Gray-Level Co-occurrence Matrix) or a GLZM (Gray-Level Size Zone Matrix) can be calculated, and an index of uniformity reduction due to noise can be calculated from the distribution as a characteristic value. In the case of performing texture analysis of the image, the display device 3 can be caused to display the input image, and the user can be instructed to indicate a region that should be analyzed. Thus, the computational load of the analysis section 21 can be reduced.

[0075] Further, classification can be performed based on information on the spatial distribution of the signal from the living body received by the reception coil. Even in the case of imaging the same part in the same examination, the spatial distribution of the sensitivity differs depending on the size of the patient's body. For example, in the case of abdominal imaging of a patient with a large size, the reception sensitivity easily becomes low in a deep region of the abdomen. Further, depending on the imaged part and the kind of the reception coil used, there are cases where a region where the reception sensitivity locally decreases, a region where the reception sensitivity is high, and the like are generated. For example, in the case of imaging of the spine, the reception sensitivity distribution differs between the case where the reception coil is disposed only on the dorsal side of the spine and the case where the reception coil is disposed on both the dorsal side and the ventral side of the spine. Depending on the reception sensitivity distribution, the spatial distribution of the noise also differs. The spatial distribution of the noise can be regarded as a noise pattern, and the image can be classified. In this case, a processor that has learned a plurality of noise patterns that are the spatial distribution of the noise is prepared.

[0076] Further, for the image classified into the noise pattern of the spatial distribution of the noise described above, a region that should be analyzed is designated or segmented, analysis of a noise pattern other than the noise pattern (for example, a noise pattern based on the size of the measurement matrix, a noise pattern based on the reception sensitivity distribution, a signal pattern related to noise, and the like) is performed, and a processor corresponding to the noise pattern can be selected for each region to perform the noise reduction process.

[0077] The analysis of the noise pattern is basically performed automatically by the analysis section 21, but it can be possible for the user to assistively designate which classification of the spatial distribution of the noise is used or how the analysis of the noise pattern is performed. Figure 11An example of a GUI that accepts a user designation is shown.

[0078] In Figure 11 In the example shown, there are a GUI block 1101 that selects a spatial distribution of noise as a noise pattern, a GUI block 1102 that selects a noise pattern based on a measurement matrix, and a GUI block 1103 that selects a noise pattern based on a signal pattern. Figure 11 The lower side of the GUI block 1101 is a graph that shows the noise pattern selected in these GUI blocks 1101 to 1103. In the GUI block 1101, for example, for the profile of the head, noise patterns 1 to 4 corresponding to the reception sensitivity distribution can be selected, and the user selects a desired noise pattern (in this case, pattern 3 in which the sensitivity in the center of the image is high) based on the reception sensitivity distribution information obtained in advance. In the GUI blocks 1102 and 1103, a noise pattern can be selected based on the measurement matrix size, the signal pattern, for a region of a given size (a partial region of the cross-sectional image).

[0079] By providing such a GUI, the user can perform appropriate adjustment or additional processing, for example, for the de-noising processing performed by the analysis section 21, and the degree of freedom of the user can be improved.

[0080] The above describes the details of the processing performed by the analysis section 21, but the analysis section 21 can also classify based on information of the spatial distribution of the transmission sensitivity generated by the transmission RF coil 151, in addition to the reception sensitivity distribution.

[0081] Further, the case of analyzing the noise pattern is described, but as the pattern analysis, the characteristics of the artifact can also be analyzed and classified in correspondence thereto.

[0082] For example, as an artifact of an MRI image, there is a linear artifact that is caused by an electrical product such as a biological monitor disposed next to the MRI apparatus. An image including the artifact can also be subjected to analysis such as Fourier transform, wavelet transform, discrete cosine transform, texture analysis, and the like.

[0083] In the analysis of the artifact, the display apparatus 3 can also be caused to display the input image, and the user can be instructed to designate the region that should be analyzed. By this, the computational load of the analysis section 21 can be reduced.

[0084] Alternatively, in the analysis of the artifact, the position and width of the line can also be further extracted and classified as a characteristic value, and in this case, in the case where, as a learned model, a different processor 23 has been made for each given noise pattern (for example, measurement matrix size) according to the presence or absence of the artifact, the noise pattern analysis can also be omitted to perform the classification.

[0085] Further, although the linear artifact is described, the same pattern analysis and classification can be performed for other types of artifacts such as ringing artifacts and blood flow artifacts.

[0086] Further, as the imaging parameter, as information other than the measurement matrix size, information such as the slice thickness and the magnetic field strength can be used.

[0087] <Modified example>

[0088] The above describes embodiments of the analysis method of the analysis section 21, but these methods can be combined, whereby appropriate construction of a learning model and quality improvement processing using the learning model can be achieved.

[0089] Further, the application of the processor 23 can not be performed by a single processor 23, but can be performed by a plurality of processors 23 at the same time as shown in (A) of FIG. 10, the results of which are synthesized by the image synthesizing section 25, or can be performed by sequentially applying a plurality of processors as shown in (B) of FIG. 10. For example, the processor for artifact removal can be used first according to the presence or absence of an artifact, and then the processor based on the noise pattern classification can be used, or the processor selected based on the noise pattern classification other than the measurement matrix size can be used after the noise removal processing based on the measurement matrix size is performed. In the case of sequentially applying the processor, the analysis section can also interpose the processing based on 21 as an intermediate processing. Figure 12 Figure 12 <Medical imaging device embodiment>

[0090] As an embodiment of the medical imaging device, the structure of an MRI device is described.

[0091] As an embodiment of the medical imaging device, the structure of an MRI device is described.

[0092] Figure 13 is a diagram showing an embodiment of an MRI device to which the present application is applied. The MRI device 100 is provided with a static magnetic field magnet (static magnetic field generation section) 110, a gradient magnetic field coil (gradient magnetic field generation section) 131, a transmission RF coil 151, a reception RF coil 161, a gradient magnetic field power supply 132, a shim coil 121, a shim power supply 122, an RF magnetic field generator 152, a receiver 162, a computer (image reconstruction section) 170, and a sequence generator 140. In addition, 102 is a stage on which a subject (subject) 103 is placed at an imaging site in an imaging space.

[0093] The static magnetic field magnet 110 generates a static magnetic field in the imaging space. The static magnetic field magnet 110 can be a tunnel magnet that generates a static magnetic field in the horizontal direction by a solenoid coil, or a static magnetic field magnet 110 that generates a static magnetic field in the vertical direction can be used.

[0094] ​The gradient magnetic field coil 131 is connected to the gradient magnetic field power supply 132, and generates a gradient magnetic field in the imaging space. The shim coil 121 is connected to the shim power supply 122, and adjusts the uniformity of the static magnetic field.

[0095] The transmit RF coil 151 is connected to the RF magnetic field generator 152, and irradiates (transmits) an RF magnetic field to the subject 103. The frequency of the RF magnetic field is set to the frequency of the nuclear magnetic of the nucleus (proton, etc.) of the nucleus of the subject 103 which is desired to be imaged. As the transmit RF coil 151, any structure can be used, for example, a birdcage type RF coil can be used.

[0096] The receive RF coil 161 is connected to the receiver 162, and receives a nuclear magnetic resonance signal from the subject 103. The receive RF coil 161 is, for example, a multi-channel RF coil (array coil) composed of a plurality of coil units, and thus, high-speed imaging can be performed by a parallel imaging method.

[0097] The sequence generator 140 transmits a command to the gradient magnetic field power supply 132 and the RF magnetic field generator 152, and causes them to act. Further, a command is transmitted to the shim power supply 122, and the uniformity of the static magnetic field is corrected. The command is sent in accordance with an instruction from the computer 170. Further, the sequence generator 140 sets a magnetic resonance frequency which is a reference for detection in the receiver 162 in accordance with an instruction from the computer 170. Specifically, at the time of imaging, the gradient magnetic field and the RF magnetic field are irradiated to the subject 103 from the gradient magnetic field coil 131 and the transmit RF coil 151 at a given timing in accordance with a command from the sequence generator 140. The nuclear magnetic resonance signal generated by the subject 103 is detected by the receive RF coil 161, and is detected by the receiver 162. Thus, an imaging pulse sequence which realizes a given imaging method is executed.

[0098] The computer 170 performs control of the entire operation of the MRI apparatus 100, and various signal processing. For example, the computer 170 transmits a command to the sequence generator 140, and causes each device to act at a timing and intensity which are programmed in advance. In the case of performing parallel imaging, the execution time of the imaging pulse sequence is shortened by dividing the phase encoding, etc. every 1 line, and high-speed imaging is performed.

[0099] Further, the computer 170 receives the signal demodulated by the receiver 162 via an A / D conversion circuit not shown, and performs signal processing such as image reconstruction. Further, in the case of high-speed imaging such as parallel imaging, the sensitivity map of the array coil is used for image reconstruction, or an operation for removing folding is performed after image reconstruction. Furthermore, the computer 170 performs processing such as correction including noise reduction on the reconstructed image. These processes are performed by an image processing section 200 built in the computer 170. Further, the computer 170 is connected to a display device that displays the processing result, and an input / output device 180 that has an input device for a user to input a desired imaging condition and the like.

[0100] The structure of the image processing section 200 is the same as that of the image processing device 2 shown in Figure 1 The computer 170 reconstructs the image, and the image processing section 200 performs quality improvement processing such as noise reduction and artifact removal on the image as an input image. At this time, the imaging parameters at the time of imaging the input image are used as necessary, and analysis / classification of the input image is performed.

[0101] The image subjected to the high-quality processing by the image processing section 200 is displayed on the input / output device (display device). At this time, the image before the high-quality processing and the like can also be displayed together, or the reliability of the high-quality processing and the like can be displayed, whereby the user can confirm whether the processing has been appropriately performed.

[0102] According to the MRI apparatus of the present embodiment, for various images taken at different imaging conditions and at different rates of high-speeding, appropriate noise processing can be performed without performing inappropriate noise processing.

Claims

1. A medical image processing device, wherein an image acquired by a medical imaging device is input, and an image with reduced noise is output, characterized in that, Possessing: a plurality of processors for noise reduction; an analysis section that analyzes a signal of an input image and a pattern of noise to classify the input image; and a processor selection section that selects one or more processors from the plurality of processors based on a classification result of the analysis section, and causes the selected processor to start up, the analysis section classifies the input image based on any one of a characteristic value calculated from data of a space other than an image space into which the input image is transformed, a characteristic value calculated from a texture analysis of the input image, a pattern of artifacts included in the input image, information on a spatial distribution of signals from a living body received by a reception coil in the case where the input image is an image taken by a magnetic resonance imaging apparatus, and an imaging parameter in the case where the input image is taken.

2. The medical image processing apparatus according to claim 1, wherein the analysis section transforms the input image into sparse space data, and classifies the input image using signal values of a plurality of regions of the sparse space data or a ratio thereof.

3. The medical image processing apparatus according to claim 1, wherein the analysis section transforms the input image into frequency space data, and classifies the input image using a signal pattern of the frequency space data.

4. The medical image processing apparatus according to claim 1, wherein the analysis section classifies the input image based on the presence or absence of a linear artifact included in the input image and a pattern of lines.

5. The medical image processing apparatus according to claim 1, wherein the input image is an image taken by a magnetic resonance imaging apparatus, and the imaging parameter is a measurement matrix size and a reconstruction matrix size.

6. A medical imaging apparatus, possessing: an imaging section that takes an image of an object and takes a medical image; and an image processing section that processes the medical image taken by the imaging section, the medical imaging apparatus being characterized in that as the image processing section, the medical image processing apparatus according to claim 1 is possessed, the analysis section classifies the input image based on any one of a characteristic value calculated from data of a space other than an image space into which the input image is transformed, a characteristic value calculated from a texture analysis of the input image, a pattern of artifacts included in the input image, information on a spatial distribution of signals from a living body received by a reception coil in the case where the input image is an image taken by a magnetic resonance imaging apparatus, and an imaging parameter in the case where the input image is taken.

7. A noise reduction method of reducing noise contained in an image taken by a medical camera device, characterized by, comprising the steps of: analyzing a pattern of noise of an image to classify an input image; selecting a given processor from a plurality of processors for noise reduction prepared for each classification based on a classification result of the step of classifying; and performing noise reduction processing of the input image by the selected processor, The input image is classified based on any one of the following parameters: a characteristic value calculated from data transformed from the input image into a space other than an image space; a characteristic value calculated from a texture analysis of the input image; a pattern of an artifact included in the input image; information on a spatial distribution of a signal from a living body received by a reception coil in the case where the input image is an image obtained by a magnetic resonance imaging apparatus; and an imaging parameter used when the input image is obtained.

8. The noise reduction method according to claim 7, wherein The analysis of the pattern of the noise includes any one of a histogram analysis of the image, a texture analysis, and an analysis of data after a frequency space or a sparse space transformation, and the classification of the input image is performed based on a result of the one or more analyses.

9. The noise reduction method according to claim 7, wherein In the classification step, a high-speed imaging condition at the time when the input image is obtained is used together with the analysis result in the classification of the input image.

10. The noise reduction method according to claim 7, wherein The noise reduction method further includes the steps of: after the step of classifying, not selecting a processor; and causing a display device to display that a processor is not selected.

Images