Magnetic resonance imaging apparatus, image processing apparatus, and image processing method

Abstract

To provide means which emphasizes a small region existing in an image by discriminating the region from another tissue, by using a 2D image.SOLUTION: On the basis of a 2D image visualizing a blood vessel and other organs and minute bleeding and other lesions (generally called a predetermined region), the feature of the shape of the predetermined region and a spatial feature including it are utilized to discriminate the predetermined region from the other regions. The spatial feature includes at least one of spatial tissue distribution of the predetermined region (existence probability of each tissue) and the spatial luminance distribution of a pixel value of the predetermined region.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a magnetic resonance imaging (MRI) apparatus, and more particularly to a technique for enhancing a predetermined region in an MR image of an object to be examined. [Background technology]

[0002] MRI is a technology that processes the nuclear magnetic resonance signals generated from each tissue of the subject to generate images with different tissue contrasts, and is widely used to support diagnostic imaging. MRI can obtain various images with different tissue contrasts by adjusting the conditions for generating the nuclear magnetic resonance signals (imaging conditions), and can be used with a variety of T2 * One or more types of images such as a weighted image, a T1 weighted image, a proton density weighted image, and a diffusion weighted image (susceptibility weighted image) are acquired.

[0003] Among various contrast images, T2 * By extending the echo time TE, the image is obtained by increasing the transverse relaxation time T2 (apparent transverse relaxation time T2 * ) and is useful for diagnosing lesions with high magnetic susceptibility effects (such as hemorrhage) in brain images. Therefore, in many examination protocols, T2 is used as one of the standard imaging types. * Enhanced imaging is included.

[0004] On the other hand, MRI can acquire multi-slice 2D and 3D images by applying a gradient magnetic field when generating nuclear magnetic resonance signals. However, when trying to distinguish blood vessels and microbleeds from MR images, because blood vessels run linearly within tissue, in 2D images, all blood vessels other than those running along the cross section appear as small dots, making them difficult to distinguish. It is particularly difficult to distinguish between normal blood vessels and microbleeds, and conventional algorithms for distinguishing between normal blood vessels and microbleeds have been based on 3D images.

[0005] For example, Patent Document 1 discloses that projection images depicting MB (microbleeds) or calcifications occurring in the brain are obtained by performing projection processing on three-dimensional brain image data in a range from the brain surface to a predetermined depth. Furthermore, Patent Document 2 discloses a technique for distinguishing between blood vessels and microbleeds from multiple images obtained at different times, utilizing the fact that signals from venous blood and microbleeds are similarly affected by magnetic susceptibility but are affected differently by phase due to flow. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent No. 6775944 specification [Patent Document 2] Japanese Patent Publication No. 2020-18695 Summary of the Invention [Problem to be solved by the invention]

[0007] 3D images are suitable for grasping three-dimensional structures, but have issues such as generally requiring longer imaging times than 2D images and being more susceptible to the effects of body movement. Furthermore, the technology described in Patent Document 1 uses projection processing to display images that make it easy to distinguish calcification and MB, but the doctor or examiner must view the images to determine whether or not there is microbleeding, and does not provide a discrimination algorithm. The technology described in Patent Document 2 acquires echoes (nuclear magnetic resonance signals) to obtain multiple images with different phases within the repetition time TR, so imaging separate from that performed in general routine examinations is required.

[0008] The present invention aims to provide a technology that uses multi-slice images, which are widely used in routine examinations, to easily highlight tissues (predetermined regions) such as microvessels that are difficult to distinguish on 2D images. [Means for solving the problem]

[0009] To solve the above problems, the present invention emphasizes a predetermined region by utilizing the shape characteristics of the region and the spatial characteristics of the region. The spatial characteristics include at least one of the distribution of surrounding tissues including the predetermined region (probability of existence of each tissue) and the spatial brightness distribution of pixel values ​​of the predetermined region. The predetermined region includes tissues such as blood vessels and lesions such as microbleeds, and refers to a portion that is identified as a single region based on its characteristics.

[0010] That is, the MRI apparatus of the present invention includes a reconstruction unit that collects magnetic resonance signals from the subject and reconstructs an image, and an image processing unit that processes the image reconstructed by the reconstruction unit and identifies a region of a predetermined contrast (hereinafter referred to as a predetermined region) included in the image. The image processing unit includes an enhancement unit that enhances the predetermined region based on shape information and spatial information of the predetermined region. For example, the image processing unit includes a shape filter unit and a spatial information analysis unit, and the shape filter unit acquires an image of a predetermined shape based on the shape characteristics of the predetermined region as shape information. The spatial information analysis unit uses the image of the predetermined shape to analyze the probability that the predetermined region exists in each tissue of the subject and brightness information of the predetermined shape.

[0011] The present invention also includes an image processing device having some or all of the functions of the image processing unit of the above-mentioned MRI device.

[0012] Furthermore, the image processing method of the present invention is an image processing method that processes an image acquired by MRI and emphasizes a predetermined area contained in the image, and includes a step of acquiring a candidate image of only a predetermined shape contained in the image, and a step of acquiring spatial information of the predetermined shape, wherein the step of acquiring spatial information includes at least one of a step of calculating a tissue distribution of the predetermined shape in the image and a step of calculating a brightness distribution of the image of the predetermined shape.

[0013] The tissue distribution is information indicating how the tissues surrounding a predetermined region are distributed, and the brightness distribution is information indicating changes in brightness values ​​in a predetermined region mainly due to the blooming effect. [Effects of the Invention]

[0014] According to the present invention, by using spatial information such as the spatial distribution of a specific region to be emphasized in addition to shape information obtained from its shape features, it is possible to automatically emphasize and present a specific region in a 2D image. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is a diagram showing the overall configuration of an embodiment of an MRI apparatus according to the present invention; [Figure 2] A diagram showing an overview of the operation of the MRI apparatus of FIG. 1. [Figure 3] Block diagram of an image processing unit according to the first embodiment [Figure 4] FIG. 1 is a diagram showing the flow of image processing according to the first embodiment. [Figure 5] FIG. 10 is a diagram for explaining the processing of the shape filter unit. [Figure 6] A diagram explaining the processing of the spatial information analysis unit [Figure 7] 1A and 1B are diagrams showing examples of displaying the discrimination results of the first embodiment. [Figure 8] 10A and 10B are diagrams showing other examples of displaying the discrimination results of the first embodiment. [Figure 9] FIG. 10 is a diagram for explaining a modified example of the processing of the feature analysis unit of the first embodiment. [Figure 10] Block diagram of an image processing unit according to the second embodiment [Figure 11] FIG. 10 is a diagram for explaining the processing of the discriminator of the second embodiment. [Figure 12] FIG. 10 is a diagram illustrating the configuration of the third embodiment. [Figure 13] FIG. 10 shows a first example of a display screen according to a third embodiment. [Figure 14] FIG. 10 shows a second example of a display screen according to the third embodiment. [Figure 15] FIG. 10 shows a display example 3 of the display screen of the third embodiment. [Figure 16] FIG. 10 shows a fourth example of a display screen according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, an embodiment of the MRI apparatus and image processing method of the present invention will be described.

[0017] First, an overview of the MRI apparatus will be described with reference to Fig. 1. As shown in Fig. 1, the MRI apparatus 1 includes a magnet 11 that generates a uniform static magnetic field in an examination space in which a subject is placed, a gradient coil 12 that provides a magnetic field gradient to the static magnetic field generated by the magnet 11, a probe 13 that includes a transmitter coil that applies a pulsed radio-frequency magnetic field to the subject to cause nuclear magnetic resonance in the nuclei of atoms constituting the subject's tissue and a receiver coil that receives nuclear magnetic resonance signals generated from the subject, a receiver 14 connected to the receiver coil, a radio-frequency magnetic field generator 15 connected to the transmitter coil, a gradient power supply 16 connected to the gradient coil 12, a sequencer 17 that controls the receiver 14, the radio-frequency magnetic field generator 15, and the gradient power supply 16 according to a predetermined pulse sequence, and a computer 20. Here, the above-mentioned elements excluding the computer 20 are collectively referred to as an imaging unit 10.

[0018] The nuclear magnetic resonance signals received by the receiver 14 of the imaging unit 10 are digitized and passed to the computer 20 as measurement data.

[0019] The structure and functions of each component of the imaging unit 10 are similar to those of known MRI devices, and the present invention can be applied to various types of known MRI devices and components, so a detailed description of the imaging unit 10 will be omitted here.

[0020] The computer 20 can be configured as a computer or workstation equipped with a CPU, GPU, and memory, and has a control function for controlling the operation of the imaging unit 10 (control unit 20C) and an image processing function for performing various calculations on measurement data acquired by the imaging unit 10 and images reconstructed from the measurement data (reconstruction unit 20A, image processing unit 20B). Each function of the computer 20 is realized, for example, by a CPU or the like uploading and executing a program for each function. However, some of the functions of the computer 20 can also be realized by hardware such as a programmable IC (ASIC or FPGA). Furthermore, the functions of the image processing unit 20B may be realized by a remote computer connected to the MRI apparatus 1 wirelessly or by a wired connection or by a computer built on the cloud, and such computers (image processing devices) are also encompassed by the present invention.

[0021] The computer 20 includes a storage device 30 that stores data and results (including intermediate results) necessary for control and calculation, and a UI (user interface) unit 40 that displays a GUI and calculation results to the user and receives instructions from the user. The UI unit 40 includes a display device and an input device (not shown).

[0022] In the MRI apparatus of this embodiment, the image processing unit 20B of the computer 20 has a function (enhancement unit 21) of emphasizing specific tissues or regions (hereinafter, "predetermined regions") included in the image using the image reconstructed by the reconstruction unit 20A. At this time, the shape features and spatial information of the predetermined region are used. For this purpose, the enhancement unit 21 includes, for example, a shape filter unit 23 that acquires an image of only the shape of the predetermined region, and a discrimination unit 27 that discriminates specific tissues using the image of only the shape acquired by the shape filter unit.

[0023] The discrimination unit 27 performs discrimination using one or more discrimination methods. In one discrimination method, the discrimination unit 27 uses the results of analyzing the distribution (tissue distribution) of the entire imaged tissue of a predetermined region as spatial information. In another method, the discrimination unit 27 uses the results of analyzing the brightness distribution of pixel values ​​of an image containing only shape as spatial information. In yet another method, the discrimination unit 27 discriminates an image including a predetermined region and a surrounding region (including shape information and spatial information of the predetermined region) using a pre-trained convolutional neural network (CNN). The spatial information analysis unit 25 shown in FIG. 1 is a functional unit including an algorithm and a CNN that executes one or more of the above-mentioned methods.

[0024] The details of the processing of the image processing unit 20B will be described later, and an overview of the processing of the MRI apparatus, including image processing, will be described with reference to FIG.

[0025] First, under the control of the control unit 20C, the imaging unit 10 performs imaging according to imaging conditions set in the examination protocol or imaging conditions set by the user, and collects nuclear magnetic resonance signals to obtain an image of the subject. The pulse sequence used for imaging is not particularly limited, but here, multi-slice 2D imaging is performed, in which a region of a predetermined thickness is divided into multiple cross sections (slices) and images are taken for each slice. In multi-slice 2D imaging, the pulse sequence is repeated while changing the selected slice position, and 2D images of multiple slices are obtained.

[0026] The reconstruction unit 20A performs calculations such as fast Fourier transform using the image data of each slice to obtain an image for each slice (S1). Basically, the multiple cross sections are parallel cross sections, but in addition, images of cross sections perpendicular to the parallel cross sections may be obtained.

[0027] The image processing unit 20B (enhancement unit 21) performs enhancement processing on each of the multiple cross-sectional images to enhance a predetermined region included in the image. To achieve this, a shape filter is first applied based on the shape characteristics of the predetermined region to create an image of only the predetermined shape (S2). For example, if the predetermined region is microbleeding, the shape filter unit 23 applies a shape filter that extracts small circular shapes (granular shapes) to create an image of only the granular shapes. At this time, it is possible to use a combination of multiple shape filters to remove other shapes that may be mixed in with only one shape filter.

[0028] Next, the spatial information analysis unit 25 analyzes the image obtained as a result of filtering to obtain spatial information such as characteristics of tissue distribution of each granular shape (S3). The spatial information includes information on the tissue surrounding the distribution of the granular shapes in the target area (tissue distribution), the distribution of pixel values ​​in each granular shape (brightness distribution), or a combination thereof.

[0029] Using the analysis results of spatial information analysis unit 25, discrimination unit 27 discriminates between the predetermined region that is the object of discrimination and tissue that is similar in shape to the predetermined region but different from the predetermined region, and extracts only the predetermined region (S4). Note that when discrimination unit 27 performs discrimination using a CNN that has learned images that include both the shape and spatial features of the predetermined region, shape filter unit 23 and spatial information analysis unit 25 may be omitted.

[0030] The above process is performed for all slices of the multi-slice image (S5), and the position and size of the predetermined region can be finally specified for the entire region to be imaged. Information on the specified predetermined region is displayed, for example, superimposed on the entire image (S6). The entire image includes, for example, the T2 * It may be a weighted image, or another image acquired in parallel (for example, a susceptibility weighted image or a proton density weighted image).

[0031] By checking the position of the predetermined region displayed on the image, the user can confirm the location of microbleeding or calcification if the predetermined region is affected.

[0032] According to this embodiment, by using shape and spatial information of tissues and lesions that were previously difficult to distinguish from 2D images, it is possible to distinguish and highlight specific regions in 2D images acquired by standard MRI examinations.

[0033] Next, an embodiment of the processing of the image processing unit will be described using an example in which the predetermined region is a microbleed.

[0034] <Embodiment 1> In this embodiment 1, the shape filter unit 23 includes a filter A that extracts granular shapes and a filter B that extracts linear shapes, and removes the shape extracted by filter B from the shape extracted by filter A, resulting in the output of the shape filter unit 23. Furthermore, the spatial information analysis unit 25 uses, as spatial information of the predetermined region, information indicating whether the predetermined region is distributed among multiple organs or regions, and information obtained by analyzing the blooming effect (blurring or enlargement of the lesion outline due to the magnetic susceptibility effect of bleeding) of the granular shapes extracted as the shape of the predetermined region. The discrimination unit 27 discriminates the predetermined region based on the analysis results of the spatial information analysis unit 25.

[0035] Hereinafter, details of the processing of the image processing unit 20B of this embodiment will be described with reference to Fig. 3 and Fig. 4. Fig. 3 is a diagram showing functional blocks of the image processing unit 20B of this embodiment. In Fig. 3, elements that are the same as those in the configuration shown in Fig. 1 are designated by the same reference numerals, and descriptions thereof will be omitted.

[0036] As shown in FIG. 3, the shape filter unit 23 has two types of morphological filters 231 and 232. One is morphological filter A, which extracts granular shapes, and the other is morphological filter B, which extracts linear shapes. A morphological filter is a technique for extracting a desired shape using morphological operations that combine expansion and contraction, and known algorithms can be used. For example, a morphological filter bank may be used as an example of a morphological filter. A morphological filter bank is a process based on morphological operations that extract features from a given image using opening processing and top-hat transformation (see IEICE Technical Report MI2010-101(2011-1) for details). By changing the size of the structuring element used in the morphological operation and repeating the process, granular (circular) components or linear components with a specific size and thickness can be emphasized.

[0037] However, the filter is not limited to a morphology filter as long as it can extract a predetermined shape, and for example, a Hough transform may be used. In addition to the two types of filters, an additional filter may be provided for extracting features other than shape extraction.

[0038] The spatial information analysis unit 25 includes a segmentation unit 251 that divides an image into multiple organs or regions and creates an image for each organ or tissue in order to obtain information on which of multiple organs or regions a predetermined region is distributed in, and a probability calculation unit 252 that calculates the probability of tissue in each segmentation image. It also includes a feature analysis unit 253 that analyzes the features of a predetermined shape and its surrounding tissues in order to analyze the blooming effect.

[0039] Next, the processing of the image processing unit 20B having the above-described configuration will be described with reference to FIGS. 4 to 7. Here, the 2D image to be processed is T2 *4, the processing enclosed by the dotted line is the processing of the shape filter unit 23, the processing enclosed by the dashed line is the processing of the segmentation unit 251 and the probability calculation unit 252, and the processing enclosed by the dashed double-dashed line is the processing of the feature analysis unit 253.

[0040] First, the shape filter unit 23 filters the T2 * An enhanced image (brain image) is input, and as preprocessing, a mask (a mask in which the brain region is set to 0 and the rest is set to 1) is used to remove regions other than the brain (background), and noise is also removed (S21). A known averaging filter or the like can be used for noise removal. Although the preprocessing step S21 is not essential, performing such preprocessing can improve the accuracy of subsequent processing (filtering, etc.). There are various known methods for extracting the brain region mask (for example, A hybrid approach to the skull stripping problem in MRI: Neuroimage, 2004 Jul;22(3):1060-75.), and any of them may be used.

[0041] Next, the shape filter unit 23 applies morphological filters A231 and B232 to the image of the brain region, respectively (S22, S23), to obtain an image (grain image) 503 in which granular shapes (circular or elliptical) are extracted and an image in which linear shapes are extracted. The granular image 503 is subtracted from the linear image (not shown) resulting from process S23, and linear components are removed from the granular image 503 (S24), resulting in an image containing only the granular shapes (grain line difference image) as the candidate image 505. Alternatively, instead of subtraction, the linear image is divided pixel by pixel from the granular image to obtain an image (grain line division image) in which the granular components are emphasized as the candidate image. Alternatively, the grain line difference image or grain line division image may be thresholded with an appropriate threshold to calculate a binary image in which circular regions are 1 and other regions are 0, and this binary image may be used as the candidate image. Combining these two types of filters in this way prevents unwanted shapes from being mixed in and extracts only the shapes to be distinguished.

[0042] Before this subtraction, as shown in Figure 5, it is preferable to perform threshold processing on the filtered granular image 501 to limit the image to images with pixel values ​​equal to or greater than a predetermined value, and to remove tissues other than blood vessels and microhemorrhages that have smaller pixel values ​​and are extracted as granular shapes (S22-2). Threshold processing based on the size of the granular shapes may be performed in addition to or instead of the pixel value threshold processing. Since the target microhemorrhages generally have a diameter of 10 mm or less, circular or elliptical shapes with a diameter exceeding 10 mm are excluded. Alternatively, the number of pixels in each cluster extracted by threshold processing may be calculated, and granular shapes with a predetermined number of pixels (e.g., 10 pixels) or less may be removed.

[0043] Furthermore, after subtracting the granular image 503 after threshold processing from the linear shape image 502, the subtracted image 504 may be subjected to a process of removing granular shapes from areas other than the brain parenchyma using pixel information from the original image 500 (S24-2). This process divides the subtracted image 504 into small regions (small patches), determines whether the median pixel value within the small patch is multiplied by a predetermined coefficient and is smaller than the average value within the mask of the original image 500 (i.e., the brain region). If the value is smaller, the patch (and the granular shapes contained therein) is excluded. By adding this process, granular shapes in areas outside the brain can be reliably excluded. The coefficient multiplied by the median value is an adjustment coefficient to prevent excessive exclusion, and a value such as 0.8 is used. Alternatively, the histogram within the small patch may be analyzed to remove granular shapes with characteristics different from normal blood vessels and microhemorrhages. For example, the T2 * The minimum value of the enhanced image multiplied by a constant (for example, 0.8) (minimum value within the patch) may be compared with the average value within the mask, and granular shapes with a large minimum value within the patch (grain shapes with a light contrast) may be excluded.

[0044] The above is the processing in the shape filter unit 23, and by this series of processing, an image of only the granular shapes present in the brain is obtained as a candidate image 505.

[0045] Next, the spatial information analysis unit 25 analyzes the spatial characteristics of the granular shapes in the candidate image 505. In this embodiment, based on the knowledge that more than half of microbleeds are contained in the brain parenchyma, the brain image is divided into brain parenchyma and cerebrospinal fluid (CSF) regions, and a probability map for each is created as spatial information. For this purpose, the segmentation unit 251 first creates a segmented image for each region from the brain image of the subject. The image used for segmentation is the T2 * Although an example using a weighted image 500 is shown, a pre-processed image 500' may also be used. Furthermore, if the images are brain images acquired from the same subject and depict white matter, gray matter, cerebrospinal fluid (CSF), etc. with different contrasts, T2 * The image is not limited to a weighted image, and for example, a T1 weighted image or a T2 weighted image may be used.

[0046] Segmentation is a technique for generating images (segmented images) divided into tissues based on the characteristics of each tissue appearing in an image, and various algorithms such as the k-means method, region growing method, and nearest neighbor method, as well as methods using CNN, are known, and any of these may be adopted. In this embodiment, which targets brain images, the segmentation unit 251 generates an image 510 of the brain parenchyma and an image 520 of the CSF by segmenting the brain image, as shown in FIG.

[0047] Next, the probability calculation unit 252 calculates the probability that the granular shape of the candidate image is included in the brain parenchyma and the probability that it is included in the CSF (S26). Specifically, as shown in Fig. 6, the candidate image 505 is multiplied by the brain parenchyma image (brain parenchyma probability map) 510 to calculate the probability that the granular shape exists in the brain parenchyma. Similarly, the candidate image 505 is multiplied by the CSF image (CSF probability map) 520 to calculate the probability that the granular shape exists in the CSF.

[0048] By performing segmentation and calculating the probability in this way, it is possible to accurately distinguish the spatial information of microbleeds that are concentrated in a specific area.

[0049] Meanwhile, when the candidate image 505 is input, the feature analysis unit 253 analyzes the microbleed features for each granular shape. Several analysis methods are available, but the example shown in FIG. 4 uses CNN (S27). The CNN is trained to output the probability of blooming (blurring or enlargement of the lesion outline due to the magnetic susceptibility effect of bleeding) for the input data (images) using, for example, numerous combinations of simulated microbleed images and blood vessel images (simulation images 530) as training data. When creating a simulated microbleed, for example, a simple circular image may be used as the simulated blood vessel image, and a circular image with smoothed outlines obtained by applying a Gaussian filter to the circular image may be used as the simulated microbleed image. Alternatively, a spherical model with a constant magnetic susceptibility value may be assumed, and local magnetic field fluctuations may be calculated to create the simulated microbleed image. Note that the images used for CNN training are not limited to simulated images, and actual captured images may also be used. The CNN training may be performed within the image processing unit 20B or by a processing device separate from the image processing unit 20B.

[0050] The feature analysis unit 253 applies CNN to the candidate image 505 and calculates the blooming probability for each granular shape. Note that although the CNN is applied to the candidate image 505 in the above example, the original image (T2 * Small regions corresponding to the granular shapes of the candidate image 505 may be cut out from the original image (enhanced image) 500, and the CNN may be applied to the resulting patch image (dotted arrows in Figure 5). As mentioned above, the candidate image 505 is an image (image of granular shapes) in which only the granular shapes have been extracted by filtering, but the contour information of the original image 500 remains intact, so depending on the training data of the CNN, it may be possible to calculate the blooming probability with greater accuracy.

[0051] In addition to calculating the blooming probability, the feature analysis unit 253 may also calculate statistical values ​​such as the diameter and volume of the identified granular shapes. The diameter is determined, for example, by measuring the lengths of lines that cross the granular shape in parallel in two or more directions, and using the length of the longest line as the diameter. Furthermore, if the granular shape identified as microbleeds is contained in only one slice, the volume may be calculated approximately from the diameter and slice thickness by approximating the microbleeds as a sphere or cylinder. Alternatively, if the granular shape identified as microbleeds is present in approximately the same position in multiple slices, the volume may be calculated approximately from the diameter of the granular shape calculated for each slice and the thickness of the cross section covered by the multiple slices.

[0052] The discrimination unit 27 integrates the results calculated by the probability calculation unit 252 and the feature analysis unit 253, and discriminates whether the granular shape of the candidate image (or a patch image obtained by cutting out a region corresponding to the granular shape from the original image) is a microbleed or a normal blood vessel. For example, the probability of being cerebral parenchyma calculated by the probability calculation unit 252 is subjected to threshold processing to discriminate between two types. For example, if the probability is 50% or higher, it is discriminated as a microbleed, and if the probability is less than 50%, it is discriminated as a normal blood vessel. Similarly, if the blooming probability calculated by the feature analysis unit 253 is equal to or higher than a predetermined threshold, it is discriminated as a microbleed. The discrimination unit 27 integrates both results. For example, the integration may be performed by ANDing both results (the result of the probability calculation unit 252 and the result of the feature analysis unit 253) and determining only those determined to be microbleeds in both cases as microbleeds, or by ORing the two results and including those determined to be microbleeds in either case. Alternatively, the two probabilities may be multiplied.

[0053] Finally, the discrimination result 550, i.e., the information on the microbleeds (number, location, size, etc.), is presented to the user. * Various methods can be used, such as superimposing the microbleeds on the enhanced image to show them in different contrast or color, or presenting information such as the number and size of microbleeds along with the image. Examples are shown in Figures 7 and 8.

[0054] 7A shows an example in which the results of filtering the original image 1500 (granular shapes 1501 discriminated as microbleeds and granular shapes discriminated as normal blood vessels) are displayed in different colors, and FIG. 7B shows an example in which marks 1531, 1532, etc. are added to distinguish between microbleeds and normal blood vessels. When the feature analysis unit 253 calculates statistical values ​​such as the diameter or volume of the shapes, the statistical values ​​may be reflected in the size of the marks 1531, 1532.

[0055] Figure 8 shows an example of displaying statistical values ​​such as the diameter and volume of granular shapes. (A) shows an example where the statistical values ​​of the area where the cursor is placed are displayed in a position that does not overlap with the image (here, at the bottom), and (B) shows an example where the statistical values ​​are displayed directly at the position of the discriminated granular shapes. By displaying statistical values ​​in this way, it is possible to grasp not only the position but also the size of microbleeds, and also to confirm whether the discrimination results are appropriate.

[0056] As described above, the image processing unit 20B of this embodiment is a 2D-T2 * For the enhanced image, the shape filter unit 23 uses filter A, which filters granular shapes, and filter B, which filters linear shapes, to generate an image containing only granular shapes as a candidate image.The spatial information analysis unit 25 uses a segmentation image generated from an image of the same subject to calculate the tissue distribution (brain parenchyma probability, CSF probability) of the candidate image and also calculates the blooming probability of each granular shape.The discrimination unit 27 uses the analysis results of the spatial information analysis unit 25 to threshold the candidate image and discriminate granular shapes that are highly likely to be microvessels.

[0057] In this way, by using the spatial features of the extracted shapes in addition to the shape features that appear in 2D images for discrimination, it is possible to identify and highlight microbleeds, which were previously difficult to distinguish from 2D images. Furthermore, by discriminating each of the multi-slice images, it is possible to grasp three-dimensional features as well.

[0058] <Modification of the First Embodiment> Below, a description will be given of modified examples of the processing of the image processing unit 20B based on embodiment 1. In the following modified examples, redundant descriptions of elements and processing similar to embodiment 1 will be omitted, and differences will be mainly described.

[0059] <Modification of 2D images> In the first embodiment, T2 is used as the source image for generating the candidate image. * However, quantitative susceptibility mapping (QSM) and susceptibility weighted imaging (SWI) are known to be superior for depicting blood in the brain. * It can also be used instead of an image. Imaging methods and calculation methods for acquiring QSM and SWI are known in the art and will not be described here, but in QSM, when the value of the brain parenchyma is set to 0, calcified tissue has a relatively "negative" (diamagnetic) value and microbleeds have a "positive" (paramagnetic) value, making it possible to distinguish not only microbleeds but also calcified tissue.

[0060] It should be noted that QSM and SWI images can also be used as auxiliary images for discrimination, such as when generating segmentation images, rather than as source images for candidate images.

[0061] <Feature Analysis Unit Variation 1> In the first embodiment, a trained CNN is used to determine the blooming effect, but the gradient of brightness change or low-rank approximation may also be used as a tool for analyzing the blooming effect.

[0062] As shown in Figure 9, in blood vessels, the signal values ​​(pixel values) inside are clearly distinguished from the surrounding pixel values ​​due to the effect of blood flow, resulting in a sharp edge in the contour. On the other hand, in microbleeds, the magnetic susceptibility effect causes the edges of the circular contour to become dull, which manifests as a difference in gradient.

[0063] In this case, the feature analysis unit 253 calculates the distribution of pixel values ​​(brightness distribution) of the line L passing through the center of the granular shape, and calculates the gradients of the rising and falling portions from this distribution. If the feature analysis unit 253 calculates the diameter of the granular shape as a statistical value, the line passing through the center of the granular shape may be the line from which the diameter was calculated among multiple lines. Threshold processing is performed on the gradient calculated for each granular shape, and the probability of microbleeding is calculated and output from the feature analysis unit 253.

[0064] Low-rank approximation is a technique that limits the number of singular values ​​and compresses the dimensionality of data using singular value decomposition. By representing an image (matrix) using only a fixed number of basis images, the dimensionality is reduced, and the probability of two types of granular shapes (such as blood vessels or microbleeds) can be calculated robustly against noise and errors.

[0065] Similar to the first embodiment, discrimination is performed using the blooming probability obtained by the above-described modified method, but these methods can be easily implemented in an image processing unit because they do not require CNN training.

[0066] <Feature Analysis Unit Variation 2> This modification is characterized in that a plurality of tools for calculating the blooming probability are prepared according to the imaging conditions.

[0067] The size and shape of blooming depicted in MR images can change depending on imaging conditions such as static magnetic field strength, TE, and static magnetic field application direction (relative to the slice direction). Therefore, a CNN or feature analysis method trained on a single imaging condition may not guarantee the reliability of the analysis results. In this variation, multiple CNNs are prepared according to multiple imaging conditions, and a CNN corresponding to the imaging conditions under which the target image was acquired is selected and used. When the feature analysis unit 253 is not a CNN but a low-rank approximation, multiple base images are prepared, and the base images used are changed depending on the imaging conditions to calculate the probability using low-rank approximation.

[0068] The selection of the CNN or base image may be performed by the image processing unit 20B automatically determining based on information on the imaging conditions attached to the image to be processed, or alternatively, options may be presented to the user via the UI unit 40, allowing the user to make a selection.

[0069] <Modifications of the specified area to be applied> In the first embodiment, the shape filter unit 23 uses two types of filters, a granular filter and a linear filter, for the purpose of discriminating microbleeds occurring in the cerebral parenchyma, but when discriminating linear regions such as epithelial hemoglobin deposits, the shape filter unit 23 uses the linear filter as the main filter. If necessary, a filter that removes shapes other than linear shapes or a filter that limits the length of a linear region may be used, as in the first embodiment.

[0070] Furthermore, when targeting cerebral epithelial hemoglobin deposition, the spatial information analysis unit 25 calculates a pial membrane probability map from a segmentation image of the pial membrane (an image of the area excluding the brain parenchyma and CSF, or the boundary area between the brain parenchyma and CSF) as spatial information. The presence or absence (probability) of the blooming effect is calculated in the same manner as in the first embodiment. Then, the pial membrane probability result and the blooming probability result are integrated for discrimination.

[0071] <Embodiment 2> In the first embodiment, the spatial information analysis unit 25 calculates the probability and blooming probability for each tissue for the candidate image, and the discrimination unit 27 discriminates the candidate image based on the results. However, the present embodiment is characterized in that the discrimination unit 27 uses a CNN trained using training data obtained from annotations of a predetermined region including spatial information.

[0072] 10, the image processing unit 20B of this embodiment omits the spatial information analysis unit 25 including the segmentation unit 251, the probability calculation unit 252, and the feature analysis unit 253 in FIG. 3, and adds the trained CNN 26 that functions as the spatial information analysis unit 25. The other configurations are the same as those of the first embodiment.

[0073] The annotation of the CNN 26 targets, for example, a normal structure image including a normal structure (here, a blood vessel) and its surrounding tissue. Using a large number of such patch images as training data, it becomes possible to train a normal structure image including information about the surrounding tissue. The CNN 26 uses this training data to train to output the probability that the input image is a normal structure or a structure other than the normal structure (for example, a lesion such as a microbleed). Training of the CNN 26 may be performed by the image processing unit 20B or by a computer separate from the image processing unit 20B.

[0074] As shown in FIG. 11 , the discrimination unit 27 inputs the original image 500 together with the candidate image 505 generated by the shape filter unit 23, and creates a patch image 507 including the granular shape and its surrounding tissue from the patch image of the candidate image 505 and the original image. The CNN 26 inputs this patch image and outputs the probability that it is a normal structure or a non-normal structure. Based on this probability, threshold processing is performed to discriminate whether it is a blood vessel or not, and the result is presented. The presentation method is the same as in the first embodiment.

[0075] According to this embodiment, by using a trained CNN, two types of processing in the spatial information analysis unit 25, i.e., tissue distribution (probability) calculation using segmentation and blooming probability calculation, can be omitted, and the processing in the discrimination unit 27 can be simplified.

[0076] <Modification of the second embodiment> In the second embodiment, shape extraction and discrimination are performed in two stages, and candidate images generated by filtering using the shape filter unit 23 are processed using CNN. However, it is also possible to perform processing including shape extraction using CNN.

[0077] In this case, the shape filter unit 23 in FIG. 10 is omitted, and the CNN 26 functions as a shape filter unit and a spatial information analyzer. The training data for the CNN 26 includes patch images of, for example, "circular regions where blooming has occurred" and "circular regions in the brain parenchyma," and the CNN 26 is trained to output the probability that these correspond to the input image. In application, the original 2D image 500 is preprocessed, and then a patch image cut out from the preprocessed image 500' is input to the CNN 26, which outputs the probability that a "circular region where blooming has occurred" or a "circular region in the brain parenchyma" exists in the patch image.

[0078] According to this modification, the work for learning CNN can be performed in a separate image processing unit, and the image processing unit 20B can easily perform discrimination processing.

[0079] <Embodiment 3> This embodiment is characterized by the addition of a means that enables correction of the processing results of the image processing unit 20B from the viewpoint of the user (doctor or examiner). The other configurations are the same as those of the first or second embodiment, and redundant explanations will be omitted, but reference will be made to the drawings used in the explanation of the first and second embodiments as necessary.

[0080] As shown in Fig. 12, a UI unit 40 equipped with a storage device 30, an input device 42, and a display device 41 is connected to the computer 20 of the MRI apparatus 1 or an independent image processing device 2, similar to a general computer. The display control unit 22 of the control unit 20C causes the display device 41 to display the MR images created in the image processing unit 20B and the processing results of the enhancement unit 21 as display images such as those shown in Figs. 7 and 8. The MR images and the processing results are stored in the storage device 30 as needed. They may also be sent to an external database such as a medical image management system (PACS) 50 via communication means.

[0081] The display control unit 22 of this embodiment sets a GUI that allows the user to edit the image displayed on the display device 41. For example, as shown in FIG. 13, a display block 1510 of an image showing the discrimination results and an operation block (GUI) 1520 for "Edit" are displayed. The operation block displays GUIs for accepting editing functions, such as a "Normal" button for changing the discrimination result from normal blood vessels to the discrimination result from lesions, a "Lesion" button for changing the discrimination result from lesions to the discrimination result from normal blood vessels, and a "Delete" button for deleting the two discrimination results. Although not shown, a cursor for selecting an area or the like and a "Select" button for confirming the selection may also be displayed (the selection may be confirmed via an input means such as a mouse). The illustrated example is merely an example, and buttons for recalculating statistical values, updating or confirming records, etc. may also be displayed. The image processing unit 20B accepts changes to the discrimination results via the operation of these GUIs.

[0082] Examples of changes to the discrimination results are shown in Figures 13 to 16. In the example shown in Figure 13, if a doctor or examiner looks at the original T2*-weighted image and determines that a granular shape 1511, which was not identified as a microbleed or a blood vessel in the discrimination results, is a microbleed or a blood vessel, the doctor or examiner, for example, moves cursor 1540 to the position of the granular shape 1511, selects the granular shape by clicking the mouse, and then operates the "Lesion" button 1522. This adds it to the discrimination results and reflects it in the display. If the microbleed is to be displayed in a color different from other tissues to highlight it, the selected granular shape is highlighted by adding a color. If a mark 1531 is to be added, the mark 1531 is added to the selected granular shape. If the granular shape is determined to be a normal blood vessel, the same process is performed by simply changing the "Lesion" button 1522 to the "Normal" button 1521, and a mark indicating a normal blood vessel is added.

[0083] Conversely, if the doctor or examiner determines that the area 1512 that has been identified as a "microbleed" or a "blood vessel" is not a "microbleed" or a "blood vessel," as shown in Figure 14, as in the example of Figure 15, by selecting the granular shape (area 1512) and operating the "Delete" button 1523, the display control unit 22 will delete the color and mark 1531 that had been added to the area 1512 and pass the information to the image processing unit 20B.

[0084] 15 shows an example in which the result of discrimination by discriminator 27 as microbleeding is changed to normal blood vessels. In this case, when area 1513 displayed as microbleeding is selected with cursor 1540 and "normal" button 1521 is pressed, display controller 22 changes mark 1531 indicating microbleeding that has been added to area 1513 to mark 1532 indicating normal blood vessels, and passes this information to image processor 20B. The same applies to changing from "normal blood vessels" to "microbleeding."

[0085] Furthermore, although this is not a change in the discrimination result, changes to the size or location of the marked area may also be accepted. As an example, in the example shown in Fig. 16, a mark 1531 added to an area 1513 displayed as microbleeds is selected, and the size of the mark 1531 is changed (enlarged) by manipulating a cursor 1540 via an input device such as a mouse. In addition, it is also possible to accept filling in an area or partially erasing it using an eraser function, enabling discrimination and editing that utilizes the judgment based on the experience of a doctor or examiner.

[0086] The image processing unit 20B updates the discrimination results based on the results of the user editing. The image processing unit 20B (feature analysis unit 253) may also calculate statistics for the newly added region. If a statistical value (e.g., the number of microbleeds) is changed as a result of deletion, the statistical value may be rewritten.

[0087] If the discrimination results are changed by editing by the user, the results may be updated and registered in the storage device 30 or transferred to the PACS 50. These processes may be performed automatically by the image processing unit 20B or in response to a user instruction.

[0088] According to this embodiment, a function for the user to edit the processing of the image processing unit 20B has been added, thereby making it possible to obtain more reliable discrimination results. Furthermore, reliable discrimination results are useful for diagnosing similar cases, and can be used for training and retraining CNN to improve the accuracy of CNN. [Explanation of symbols]

[0089] 1: MRI device, 2: image processing device, 10: imaging unit, 20: computer, 20A: reconstruction unit, 20B: image processing unit, 20C: control unit, 21: enhancement unit, 22: display control unit, 23: shape filter unit, 231: morphology filter A (granular filter), 232: morphology filter B (linear filter), 25: spatial information analysis unit, 251: segmentation unit, 252: probability calculation unit, 253: feature analysis unit, 27: discrimination unit, 30: storage device, 40: UI unit, 41: display device, 42: input device.

Claims

1. a reconstruction unit that collects magnetic resonance signals of an object to be examined and reconstructs an image; and an image processing unit that processes the image reconstructed by the reconstruction unit and identifies an area of ​​a predetermined contrast (a predetermined area) included in the image, the image processing unit includes an enhancement unit that enhances the predetermined region based on shape information of the predetermined region in the 2D image and spatial information of the predetermined region; The highlighting section is Based on the shape information of the predetermined region, an image of only the predetermined shape is generated as a candidate image, a luminance distribution of the image of only the predetermined shape (candidate image) is obtained, and a probability that the image of only the predetermined shape is the predetermined region is calculated; creating a segmentation image of the organ or tissue of the subject of examination, and calculating the probability that the predetermined region exists for each specific organ or tissue of the subject of examination for the segmentation image created for each specific organ or tissue of the subject of examination to generate a probability map; A magnetic resonance imaging apparatus, characterized in that the probability that the candidate image is the predetermined region calculated from the probability map and the brightness distribution of the candidate image is used as spatial information of the predetermined region.

2. 2. The magnetic resonance imaging apparatus according to claim 1, the highlighting unit includes a shape filter unit that generates an image of only a predetermined shape as a candidate image based on shape information of the predetermined region, and a discrimination unit that discriminates the predetermined region from other regions; The magnetic resonance imaging apparatus according to claim 1, wherein the discrimination unit discriminates the predetermined region based on spatial information of the candidate image created by the shape filter unit.

3. 3. The magnetic resonance imaging apparatus according to claim 2, a second filter that extracts a second shape feature different from the first shape feature; and a magnetic resonance imaging apparatus characterized in that the shape filter unit includes a first filter that extracts a shape feature of the specified region, and a second filter that extracts a second shape feature different from the first shape feature, and the first filter removes the second shape feature extracted by the second filter from the shape features to generate the candidate image.

4. 4. The magnetic resonance imaging apparatus according to claim 3, a first filter configured to extract a circular shape; a second filter configured to extract a linear shape; and a shape filter unit configured to extract a circular shape.

5. 2. The magnetic resonance imaging apparatus according to claim 1, A magnetic resonance imaging apparatus, wherein the image of the subject of examination is a brain image, and the enhancement unit creates a brain parenchyma image and a cerebrospinal fluid image as the segmentation image.

6. 2. The magnetic resonance imaging apparatus according to claim 1, the highlighting unit includes a discrimination unit that discriminates between the predetermined region and other tissues; The magnetic resonance imaging apparatus is characterized in that the discrimination unit discriminates the specified region using a CNN that has learned features of an image including the specified region and its surrounding region.

7. 2. The magnetic resonance imaging apparatus according to claim 1, The magnetic resonance imaging apparatus is characterized in that the enhancement unit further includes a CNN that has been trained using a plurality of types of images with different luminance distributions in a predetermined shape, and the luminance distribution information of the predetermined region is acquired using the CNN.

8. 8. The magnetic resonance imaging apparatus according to claim 7, The image processing unit includes a plurality of CNNs trained under a plurality of imaging conditions as the CNN, The magnetic resonance imaging apparatus is characterized in that the enhancement unit selects and applies one of the plurality of CNNs depending on the imaging conditions when the image to be processed by the image processing unit was acquired.

9. 2. The magnetic resonance imaging apparatus according to claim 1, The magnetic resonance imaging apparatus according to claim 1, wherein the enhancement unit uses a brightness gradient at the contour of the predetermined region as the brightness distribution.

10. 2. The magnetic resonance imaging apparatus according to claim 1, A magnetic resonance imaging apparatus in which the image processed by the image processing unit is a two-dimensional image.

11. 2. The magnetic resonance imaging apparatus according to claim 1, The image processed by the image processing unit is T2 ＊ A magnetic resonance imaging device that produces at least one of a weighted image and a susceptibility weighted image.

12. 2. The magnetic resonance imaging apparatus according to claim 1, The magnetic resonance imaging apparatus further comprises a display control unit that displays the processing result of the enhancement unit together with the image on a display device.

13. 13. The magnetic resonance imaging apparatus according to claim 12, A magnetic resonance imaging apparatus characterized in that the display control unit displays a GUI that accepts user changes to the results displayed on the display device, and passes the changes made via the GUI to the image processing unit.

14. 1. An image processing device for processing 2D images of an object acquired by magnetic resonance imaging, comprising: a shape filter unit for acquiring an image of only a predetermined shape included in the 2D image; an enhancement unit that enhances a predetermined region having the predetermined shape using a tissue distribution of the image of only the predetermined shape in the 2D image and a brightness distribution of the image of only the predetermined shape, The highlighting section is Based on the shape information of the predetermined region, an image having only the predetermined shape is generated as a candidate image, a luminance distribution of the image having only the predetermined shape (candidate image) is obtained, and a probability that the image having only the predetermined shape is the predetermined region is calculated; creating a segmentation image of the organ or tissue of the subject of examination, and calculating the probability that the predetermined region exists for each specific organ or tissue of the subject of examination for the segmentation image created for each specific organ or tissue of the subject of examination to generate a probability map; An image processing device characterized in that the predetermined region is emphasized by integrating the probability map and the probability that the candidate image is the predetermined region, calculated from the luminance distribution of the candidate image.

15. 1. An image processing method for processing a 2D image of an object acquired by magnetic resonance imaging, and enhancing a predetermined region included in the 2D image, comprising: acquiring candidate images of only a predetermined shape included in the 2D image; acquiring spatial information of the predetermined shape; and highlighting the predetermined region based on the spatial information, the step of acquiring spatial information includes a step of creating a segmentation image of the organ or tissue of the subject of examination, and calculating the probability that the predetermined region exists for each specific organ or tissue of the subject of examination for the created segmentation image for each specific organ or tissue of the subject of examination to generate a probability map, and a step of acquiring a luminance distribution of an image (candidate image) of only the predetermined shape, and calculating the probability that the image of only the predetermined shape is the predetermined region; An image processing method characterized in that the emphasizing step emphasizes the specified region by integrating the probability map as the spatial information and the probability that the candidate image is the specified region calculated from the brightness distribution of the candidate image.

16. 16. The image processing method according to claim 15, The images obtained by magnetic resonance imaging are two-dimensional T2 ＊ An image processing method characterized in that the image is an enhanced image.

Images