Noise filtering device for three-dimensional velocity field data and noise filtering method for three-dimensional velocity field data using same

Abstract

Disclosed is a noise filtering method for three-dimensional velocity field data. A noise filtering method for three-dimensional velocity field data comprises: a data collection step of receiving three-dimensional velocity field data; a step of dividing the three-dimensional velocity field data into a first sub-domain including M voxels; a step of calculating an abnormal component in the voxels included in the first sub-domain; and a step of correcting the abnormal component by applying singular value decomposition to a voxel including the abnormal component, and generating first three-dimensional velocity field correction data including the voxel including the abnormal component.

Description

Noise filtering device for three-dimensional velocity field data and noise filtering method for three-dimensional velocity field data using the same

[0001] The present invention relates to a noise filtering device for three-dimensional velocity field data and a noise filtering method for three-dimensional velocity field data using the same. More specifically, the invention relates to a noise filtering device for three-dimensional velocity field data that calculates anomaly components in three-dimensional velocity field data and effectively removes noise from voxels containing the anomaly components through a singular value decomposition technique, and a noise filtering method for three-dimensional velocity field data using the same.

[0002] A Magnetic Resonance Imaging (MRI) device is a medical imaging machine that uses powerful magnetic fields and high-frequency signals to non-invasively image internal human tissues in detail. When a patient lies down and enters a tube-shaped scanner, a magnetic field is generated to align hydrogen nuclei within the body, and high-frequency signals are applied to detect the position and movement of these nuclei.

[0003] Computers analyze this data to generate cross-sectional images of the body taken from various angles, which are used to diagnose lesions or formulate treatment plans. MRI provides excellent resolution, particularly in soft tissues, the nervous system, the brain, and joints. For example, blood flow reconstruction using magnetic resonance imaging has the advantage of measuring fluid flow within the human blood vessels.

[0004] Conventional magnetic resonance devices and blood flow image reconstruction methods using them measure magnetic signals to identify the image and velocity field of the fluid in the measurement area. This process involves noise generated by electronic components used to apply, measure, or amplify the signal, and the noise becomes more significant as the measurement time is shorter or the measurement resolution is increased.

[0005] When measuring a patient's blood flow movement using the above magnetic resonance device, the measurement time is limited due to the patient's movement, and high resolution is required due to the small cross-sectional area of ​​the blood vessels; as a result, significant noise is generated, which may reduce data reliability. In addition, accurate measurement requires a very long time, and large errors may occur for complex shapes.

[0006] The present invention provides a noise filtering device for three-dimensional velocity field data capable of efficiently removing noise generated in three-dimensional velocity field data measured by a magnetic resonance device and improving measurement accuracy, and a noise filtering method for three-dimensional velocity field data using the same.

[0007] A noise filtering method for three-dimensional velocity field data using a noise filtering device for three-dimensional velocity field data according to the present invention comprises: a data collection step of receiving three-dimensional velocity field data; a step of dividing the three-dimensional velocity field data into a first subdomain including M voxels; a step of calculating an abnormal component among the voxels included in the first subdomain; and a step of applying a singular value decomposition to the voxels containing the abnormal component to correct the abnormal component, and generating a first three-dimensional velocity field correction data including the voxels corrected for the abnormal component.

[0008] In addition, the step of calculating the above-mentioned abnormal component may calculate the abnormal component by multiplying the velocity vector of each voxel included in the first subdomain by a weight and comparing the magnitude of the velocity vector with the magnitude of the average velocity vector of adjacent voxels.

[0009] In addition, the first subdomain may have the M voxels arranged in three dimensions.

[0010] In addition, the first subdomain may have the M voxels arranged in a rectangular shape.

[0011] Additionally, the voxels included in the first subdomain include voxels containing the anomalous component and voxels not containing the anomalous component, and the step of generating the first three-dimensional velocity field correction data may be reconstructed by applying singular value decomposition to the voxels not containing the anomalous component, cutting off the remaining bases excluding the first basis obtained from the singular value decomposition, and synthesizing and averaging only the vectors of the first basis.

[0012] Additionally, the method may further include the steps of: dividing the first three-dimensional velocity field correction data into a second subdomain containing N voxels, which is fewer than M; calculating an abnormal component among the voxels included in the second subdomain; and applying a singular value decomposition to the voxels containing the abnormal component to correct the abnormal component, and generating second three-dimensional correction data including the voxels with the abnormal component corrected.

[0013] Additionally, the noise filtering device for three-dimensional velocity field data according to the present invention comprises: a data collection unit for receiving three-dimensional velocity field data; a subdomain generation unit for dividing the three-dimensional velocity field data into a first subdomain including M voxels; an abnormal component calculation unit for calculating abnormal components among the voxels included in the first subdomain; and a data processing unit for correcting the abnormal components by applying a singular value decomposition to the voxels containing the abnormal components, and generating first three-dimensional velocity field correction data including the voxels in which the abnormal components have been corrected.

[0014] In addition, the above-mentioned abnormal component calculation unit can calculate the abnormal component by multiplying the velocity vector of each voxel included in the first subdomain by a weight and comparing the magnitude of the velocity vector with the magnitude of the average velocity vector of adjacent voxels.

[0015] Additionally, the voxels included in the first subdomain include voxels containing the anomalous component and voxels not containing the anomalous component, and the data processing unit can apply singular value decomposition to the voxels not containing the anomalous component, cut off the remaining bases excluding the first base obtained from the singular value decomposition, and reconstruct by synthesizing and averaging only the vector of the first base.

[0016] Additionally, the sub-domain generation unit divides the first three-dimensional velocity field correction data into a second sub-domain containing N voxels, which is fewer than M voxels, the anomaly component calculation unit calculates an anomaly component among the voxels included in the second sub-domain, the data processing unit applies a singular value decomposition to the voxels containing the anomaly component to correct the anomaly component, and can generate second three-dimensional correction data including the voxels with the anomaly component corrected.

[0017] A noise filtering device for three-dimensional velocity field data according to the present invention calculates an abnormal component of voxels in a first subdomain divided from three-dimensional velocity field data, applies a quiescent singular value decomposition to voxels containing the abnormal component to correct the abnormal component, generates first three-dimensional velocity field correction data including voxels with the abnormal component corrected, calculates an abnormal component of voxels in a second subdomain divided from the first three-dimensional velocity field correction data, applies a quiescent singular value decomposition to voxels containing the abnormal component to correct the abnormal component, and generates second three-dimensional velocity field correction data including voxels with the abnormal component corrected.

[0018] In addition, the number of voxels included in the second subdomain may be less than the number of voxels included in the first subdomain.

[0019] According to the present invention, high-resolution, accurate three-dimensional velocity field data can be obtained by selectively correcting abnormal components, thereby increasing the precision of diagnosis and research in medical imaging such as magnetic resonance imaging (MRI), and enabling rapid noise filtering by maximizing data processing efficiency.

[0020] In addition, the reliability and accuracy of the data can be significantly improved by effectively removing noise generated in three-dimensional velocity field data.

[0021] FIG. 1 is a block diagram showing a noise filtering device for three-dimensional velocity field data according to an embodiment of the present invention.

[0022] Figure 2 is a diagram for explaining the subdomain generation unit of Figure 1.

[0023] Figure 3 is a diagram illustrating the process of calculating an abnormal component in the abnormal component calculation unit of Figure 1.

[0024] FIG. 4 is a flowchart illustrating a noise filtering method for three-dimensional velocity field data according to an embodiment of the present invention.

[0025] FIG. 5 is a flowchart illustrating the process of generating three-dimensional velocity field correction data according to an embodiment of the present invention.

[0026] FIG. 6 is a diagram showing an image of the thoracoabdominal aorta of an actual patient and an image processed according to the noise filtering method of three-dimensional velocity field data according to an embodiment of the present invention.

[0027] FIG. 7 is a diagram showing the number of abnormal components calculated in a thoracoabdominal aortic image taken during maximum ventricular systole and the number of abnormal components calculated in an image to which a noise filtering method for three-dimensional velocity field data according to an embodiment of the present invention is applied.

[0028] Figure 8 is a graph showing the change in the number of abnormal components over time according to the noise filtering method of the three-dimensional velocity field data of Figure 7.

[0029] A noise filtering method for three-dimensional velocity field data according to an embodiment of the present invention comprises: a data collection step of receiving three-dimensional velocity field data; a step of dividing the three-dimensional velocity field data into a first subdomain including M voxels; a step of calculating an abnormal component among the voxels included in the first subdomain; and a step of applying a singular value decomposition to the voxels containing the abnormal component to correct the abnormal component, and generating a first three-dimensional velocity field correction data including the voxels corrected for the abnormal component.

[0030] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. However, the technical concept of the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided to ensure that the disclosed content is thorough and complete and to ensure that the concept of the present invention is sufficiently conveyed to those skilled in the art.

[0031] In this specification, when a component is described as being on another component, it means that it may be formed directly on the other component or that a third component may be interposed between them. Additionally, in the drawings, the thicknesses of the films and regions are exaggerated for the effective description of the technical content.

[0032] Additionally, although terms such as first, second, third, etc., have been used to describe various components in the various embodiments of this specification, these components should not be limited by such terms. These terms are used merely to distinguish one component from another. Accordingly, what is referred to as the first component in one embodiment may be referred to as the second component in another embodiment. Each embodiment described and illustrated herein also includes its complementary embodiment. Furthermore, in this specification, "and / or" is used to mean including at least one of the components listed before and after it.

[0033] In the specification, singular expressions include plural expressions unless the context clearly indicates otherwise. Furthermore, terms such as "include" or "have" are intended to specify the existence of the features, numbers, steps, components, or combinations thereof described in the specification, and should not be understood as excluding the existence or addition of one or more other features, numbers, steps, components, or combinations thereof. Additionally, in this specification, "connection" is used to include both indirectly connecting multiple components and directly connecting them.

[0034] In addition, in describing the present invention below, if it is determined that a detailed description of related known functions or configurations could unnecessarily obscure the essence of the invention, such detailed description will be omitted.

[0035]

[0036] FIG. 1 is a block diagram showing a noise filtering device for three-dimensional velocity field data according to an embodiment of the present invention, FIG. 2 is a diagram for explaining the sub-domain generation unit of FIG. 1, and FIG. 3 is a diagram for explaining the process of calculating an abnormal component in the abnormal component calculation unit of FIG. 1.

[0037] First, referring to FIG. 1, a noise filtering device (10) for three-dimensional velocity field data can calculate and correct abnormal components to provide accurate and reliable three-dimensional velocity field data. The noise filtering device (10) for three-dimensional velocity field data includes a data collection unit (100), a sub-domain generation unit (200), an abnormal component calculation unit (300), and a data processing unit (400).

[0038] The data collection unit (100) collects three-dimensional velocity field data. According to an embodiment, the three-dimensional velocity field data is data that represents a velocity vector changing over time at a specific point in space in three dimensions. The data includes Magnetic Resonance Velocimetry (MRV) data and Direct Numerical Simulation (DNS) data obtained through a flow visualization technique using a Magnetic Resonance Imaging (MRI) device. The MRV data and DNS data represent the results of measuring velocity, temperature, concentration, and kinetic energy in complex shapes or opaque environments.

[0039] The above sub-domain generation unit (200) can receive the three-dimensional velocity field data and generate the three-dimensional velocity field data by dividing it into a first sub-domain. The size of the first sub-domain can be adjusted according to the user's settings. The first sub-domain can be divided into multiple parts.

[0040] Each of the first subdomains described above includes at least one voxel. The voxel is the smallest unit element constituting three-dimensional space and is primarily used in medical imaging, such as MRI, or in three-dimensional data analysis.

[0041] According to an embodiment, the first subdomain includes a plurality of voxels. The voxels are provided in a plurality in the x-axis, y-axis, and z-axis directions, respectively, in three-dimensional space. According to an embodiment, the voxels may be provided in 5 to 30 each in the x-axis, y-axis, and z-axis directions. By combining these voxels, the first subdomain is provided as a hexahedron. According to an embodiment, the first subdomain is provided in the shape of a rectangular prism. The voxels can be arranged in three dimensions to determine spatial resolution.

[0042] Referring to FIG. 2, the subdomain generating unit (200) creates a first region (A1) by cutting one side of the first subdomain (S) in three-dimensional space by 1 / n each in the x-axis, y-axis, and z-axis directions, and creates a second region (A2) that is superimposed by attaching the first region (A1) to another side of the first subdomain (S) so that the voxels (V) overlap.

[0043] In this process, the first subdomain (S) is overlapped for each voxel (V), and the first subdomain (S) is generated in a form with uniform overlap. As a result, the first subdomains (S) are arranged overlappingly, enabling consistent analysis in the second region (A2). According to an embodiment, the number of the first subdomains (S) overlapped in the x-axis, y-axis, and z-axis directions of the three-dimensional space can be provided as 3 × (n - 1) + 1. Here, n is a parameter that controls the size of each first subdomain (S) and represents the subdivision unit for dividing the first subdomain (S).

[0044]

[0045] The above-mentioned abnormal component calculation unit (300) calculates abnormal components among the voxels included in the first subdomain. Here, the abnormal component refers to a data component with severe noise among the three-dimensional velocity field data. According to an embodiment, the above-mentioned abnormal component calculation unit (300) calculates the abnormal component by multiplying the velocity vector of each voxel included in the first subdomain by a weight, and then comparing the magnitude of the velocity vector with the magnitude of the average velocity vector of adjacent voxels. The weight may be provided as 0.5 to 1.5. The above-mentioned abnormal component calculation unit (300) determines it as an abnormal component if the magnitude of the velocity vector multiplied by the weight is greater than the magnitude of the average velocity vector of adjacent voxels.

[0046]

[0047] Referring to FIG. 3, the above-mentioned abnormal component calculation unit (300) can calculate the distance radius (R) between the velocity vector (V1) of each voxel included in the first subdomain and the average velocity vector (V2) of the adjacent voxels through [Equation 1].

[0048]

[0049] [Formula 1]

[0050]

[0051]

[0052] Here, the above is a parameter that determines the ideal component, and the above represents the magnitude of the velocity vector (V1) of the voxel included in the first subdomain with respect to the x-axis and y-axis of three-dimensional space.

[0053] The average velocity vector (V2) of the adjacent voxels can be calculated through [Equation 2].

[0054]

[0055] [Equation 2]

[0056]

[0057]

[0058] Here, the above represents the magnitude of the average velocity vector (V2) of adjacent voxels, and the above represents the magnitude of the velocity vector of the reference voxel (V1), and the above represents the magnitude of the velocity vector of a voxel located one step ahead in the x-axis direction from the reference voxel (V1), and the represents the magnitude of the velocity vector of a voxel located one step behind the reference voxel (V1) in the x-axis direction, and the represents the magnitude of the velocity vector of a voxel located one step ahead in the y-axis direction from the reference voxel (V1), and the represents the magnitude of the velocity vector of a voxel located one step behind the reference voxel (V1) in the y-axis direction, and the represents the magnitude of the velocity vector of a voxel located one step ahead in the z-axis direction from the reference voxel (V1), and the represents the magnitude of the velocity vector of a voxel located one step behind the reference voxel (V1) in the z-axis direction.

[0059] If the distance between the velocity vector (V1) and the average velocity vector (V2) of the voxel included in the first subdomain is less than or equal to the distance radius calculated through [Equation 1], the voxel included in the first subdomain is determined to be a voxel that does not contain the above abnormal component through [Equation 3].

[0060]

[0061] [Equation 3]

[0062]

[0063]

[0064] Here, the above represents the distance between the velocity vector (V1) and the average velocity vector (V2) of the voxel included in the first subdomain, and the represents the distance radius (R) calculated through [Equation 1].

[0065] On the other hand, if the distance between the velocity vector (V1) and the average velocity vector (V2) of the voxel included in the first subdomain is greater than the distance radius (R) calculated through [Equation 1], the voxel included in the first subdomain is determined to be a voxel containing the abnormal component through [Equation 4].

[0066]

[0067] [Equation 4]

[0068]

[0069]

[0070] Here, the above represents the distance between the velocity vector (V1) and the average velocity vector (V2) of the voxel included in the first subdomain, and the represents the distance radius (R) calculated through [Equation 1].

[0071]

[0072] When the above-described abnormal component is detected through the process described above, the first subdomain includes a voxel region containing the abnormal component and a voxel region not containing the abnormal component.

[0073]

[0074] Referring again to FIG. 1, the data processing unit (400) applies singular value decomposition to a voxel that does not contain the above-mentioned abnormal component, cuts off the remaining basis excluding the first basis obtained from the singular value decomposition, synthesizes and averages only the first basis vector, and then reconstructs it.

[0075] The above singular value decomposition is a technique for decomposing matrix data of voxels that does not contain the above-mentioned abnormal components according to mutually orthogonal bases and weights corresponding to the bases. The above singular value decomposition can extract singular values ​​for the voxels by decomposing the data matrix of the voxels into the product of a first orthogonal matrix, a diagonal matrix, and a second orthogonal matrix. According to an embodiment, the above singular value decomposition can extract singular values ​​through [Equation 5]. The weights may be proportional to the covariance of the voxels.

[0076]

[0077] [Formula 5]

[0078]

[0079]

[0080] Here, A represents a two-dimensional matrix having a predetermined number of rows and columns, U represents a first orthogonal matrix, and the represents a diagonal matrix, and the above represents the second orthogonal matrix.

[0081]

[0082] The data processing unit (400) applies a singular value decomposition to a voxel containing the abnormal component to correct the abnormal component, thereby generating first three-dimensional velocity field correction data including a voxel with the abnormal component corrected and a voxel that does not contain the abnormal component.

[0083] Specifically, the data processing unit (400) corrects the abnormal component by utilizing only the basis vector with the highest weight among the singular values ​​extracted through [Equation 5]. As a result, the first three-dimensional velocity field correction data is reconstructed with reduced noise.

[0084] When the correction of the voxel containing the above-mentioned abnormal component is first completed through the process described above, the calculation of the abnormal component for the voxel included in the first three-dimensional velocity field correction data and the correction thereof are additionally carried out.

[0085] The subdomain generation unit (200) divides the first three-dimensional velocity field correction data into a second subdomain containing fewer voxels than the first subdomain. According to an embodiment, the number of voxels included in the second subdomain is provided to be less than the number of voxels in the first subdomain according to the x-axis, y-axis, and z-axis directions of the three-dimensional space.

[0086] The above abnormal component calculation unit (300) calculates abnormal components among the voxels included in the second subdomain. The above abnormal component calculation unit (300) calculates the abnormal components in the same way as described in FIG. 2.

[0087] If the above abnormal component is detected, the second subdomain includes a voxel region containing the abnormal component and a voxel region not containing the abnormal component.

[0088] The data processing unit (400) applies singular value decomposition to voxels that do not contain the above-mentioned abnormal component, cuts off the remaining basis excluding the first basis obtained from the singular value decomposition, synthesizes and averages only the first basis vector, and then reconstructs it. Then, it applies singular value decomposition to voxels containing the above-mentioned abnormal component to correct the above-mentioned abnormal component, thereby generating second three-dimensional velocity field correction data including voxels with the above-mentioned abnormal component corrected and voxels that do not contain the above-mentioned abnormal component.

[0089] When the generation of the second three-dimensional velocity field correction data is completed, the sub-domain generation unit (200) included in the second three-dimensional velocity field correction data divides the second three-dimensional velocity field correction data into a third sub-domain containing fewer voxels than the voxels of the second sub-domain.

[0090] The above abnormal component calculation unit (300) calculates abnormal components among the voxels included in the third sub-domain. Then, the data processing unit (400) corrects the abnormal components to generate third three-dimensional velocity field correction data.

[0091] This process is repeated until a user-set condition is satisfied. According to an embodiment, the process may be repeated until the number of voxels included in a subdomain reaches a set number. According to one example, the process may be repeated until the number of voxels included in a subdomain is provided as four in the x-axis, y-axis, and z-axis directions in three-dimensional space.

[0092] Through this process, the aforementioned abnormal components included in the three-dimensional velocity field data can be removed as much as possible.

[0093]

[0094] Hereinafter, a noise filtering method (S10) for three-dimensional velocity field data using the noise filtering device (10) for three-dimensional velocity field data described above will be explained.

[0095] FIG. 4 is a flowchart illustrating a noise filtering method for three-dimensional velocity field data according to an embodiment of the present invention.

[0096] Referring to FIG. 4, the noise filtering method (S10) for the three-dimensional velocity field data includes a data collection step (S100) and a three-dimensional velocity field correction data generation step (S200).

[0097] The above data collection step (S100) collects three-dimensional velocity field data. According to an embodiment, the three-dimensional velocity field data includes magnetic resonance velocimetry (MRV) data obtained by a flow visualization technique through a magnetic resonance imaging (MRI) device and direct numerical simulation (DNS) data.

[0098] The above three-dimensional velocity field correction data generation step (S200) divides the three-dimensional velocity field data into a first subdomain containing M voxels, calculates an abnormal component from the voxels included in the first subdomain, and corrects the abnormal component to generate the first three-dimensional velocity field correction data.

[0099] Then, a second subdomain containing N (M>N) voxels is partitioned from the first three-dimensional velocity field correction data, an abnormal component is calculated from the voxels included in the second subdomain, and the abnormal component is corrected to generate second three-dimensional velocity field correction data.

[0100] The generation of the aforementioned three-dimensional velocity field correction data is repeated until a condition pre-set by the user is satisfied, and each time the generation of the three-dimensional velocity field correction data is repeated, the number of voxels included in the sub-domains divided from the three-dimensional velocity field correction data decreases by a pre-set unit. When the number of voxels included in the sub-domains corresponds to the number set by the user, the generation of the three-dimensional velocity field correction data is terminated.

[0101]

[0102] FIG. 5 is a flowchart illustrating the process of generating three-dimensional velocity field correction data according to an embodiment of the present invention.

[0103] Referring to FIG. 5, the three-dimensional velocity field correction data generation step (S200) includes the step of dividing into sub-domains (S210), the step of calculating an abnormal component (S220), and the step of generating three-dimensional velocity field correction data (S230).

[0104] The step of dividing into subdomains (S210) divides the three-dimensional velocity field data into subdomains containing a plurality of voxels. The plurality of voxels are arranged along the x-axis, y-axis, and z-axis in three-dimensional space. According to an embodiment, the voxels may be provided in groups of 5 to 30 in the x-axis, y-axis, and z-axis directions. The subdomain is provided as a hexahedron through a combination of these voxels. According to an embodiment, the subdomain is provided in the form of a rectangular prism.

[0105] The step of dividing into the above subdomains (S210) creates a first region by cutting one side of the above subdomain in three-dimensional space by 1 / n each in the x-axis, y-axis, and z-axis directions, and creates a second region that is superimposed by attaching the first region to another side of the above subdomain so that the voxels overlap. In this process, the above subdomains are superimposed for each voxel, and a subdomain having a uniform overlap is created. According to an embodiment, the number of the above subdomains superimposed in the x-axis, y-axis, and z-axis directions of the above three-dimensional space may be provided as 3 × (n - 1) + 1. Here, n is a parameter that controls the size of each above subdomain and represents the subdivision unit for dividing the above subdomains.

[0106] The step of calculating the abnormal component (S220) calculates the abnormal component among the voxels included in the subdomain. The step of calculating the abnormal component (S220) calculates the abnormal component by multiplying the velocity vector of each voxel included in the subdomain by a weight, and then comparing the magnitude of the velocity vector with the magnitude of the average velocity vector of adjacent voxels. According to an embodiment, the weight may be provided as 0.5 to 1.5.

[0107] The step of calculating the above abnormal component (S220) determines it as an abnormal component if the magnitude of the velocity vector multiplied by the weights is greater than the magnitude of the average velocity vector of adjacent voxels. Specifically, the step of calculating the above abnormal component (S220) can calculate the distance radius between the velocity vector of each voxel included in the subdomain and the average velocity vector of adjacent voxels through [Equation 1]. The average velocity vector of adjacent voxels can be calculated through [Equation 2].

[0108] If the distance between the velocity vector and the average velocity vector of a voxel included in the above subdomain is less than or equal to the distance radius calculated through [Equation 1], the voxel included in the above subdomain is determined to be a voxel that does not contain the above abnormal component through [Equation 3]. On the other hand, if the distance between the velocity vector and the average velocity vector of a voxel included in the above subdomain is greater than the distance radius calculated through [Equation 1], the voxel included in the above subdomain is determined to be a voxel that contains the above abnormal component through [Equation 4].

[0109] The step (S230) of generating the above three-dimensional velocity field correction data applies singular value decomposition to voxels that do not contain the above-mentioned abnormal components, cuts off the remaining bases excluding the first basis obtained from the singular value decomposition, and reconstructs by synthesizing and averaging only the vectors of the first basis. The singular value decomposition can extract singular values ​​for the voxels by decomposing the data matrix of the voxels into the product of a first orthogonal matrix, a diagonal matrix, and a second orthogonal matrix. According to an embodiment, the singular value decomposition can extract singular values ​​through [Equation 5].

[0110] The step (S230) of generating the three-dimensional velocity field correction data applies a singular value decomposition to a voxel containing the abnormal component to correct the abnormal component, thereby generating three-dimensional velocity field correction data including a voxel with the abnormal component corrected and a voxel that does not contain the abnormal component. Specifically, the step (S230) of generating the three-dimensional velocity field correction data corrects the abnormal component by utilizing only the basis vector with the highest weight among the singular values ​​extracted through [Equation 5]. As a result, the three-dimensional velocity field correction data is reconstructed with reduced noise.

[0111]

[0112] In this way, the abnormal components included in the three-dimensional velocity field data are removed as much as possible through the noise filtering method (S10) of the three-dimensional velocity field data.

[0113]

[0114] FIG. 6 is a diagram showing an image of the thoracoabdominal aorta of an actual patient and an image processed according to the noise filtering method of three-dimensional velocity field data according to an embodiment of the present invention.

[0115] Referring to FIG. 6, the raw image (R) and the filtered image (F) are shown, which are processed through a noise filtering method on the three-dimensional velocity field data of blood flow in the thoracoabdominal aorta of an actual patient during the peak systole and diastole states. Here, peak systole is the stage in which the heart contracts maximally to send blood from the aorta to the pulmonary artery, and diastole is the stage in which the heart relaxes to refill with blood.

[0116] The first image (a) is three-dimensional velocity field data representing blood flow velocity from the feet toward the head, the second image (b) is three-dimensional velocity field data representing blood flow velocity from the abdomen toward the back, the third image (c) is three-dimensional velocity field data representing blood flow velocity from the right side of the body toward the left side, and the fourth image (d) is three-dimensional velocity field data representing divergence values ​​representing the inflow and outflow of blood flow.

[0117] Upon examination, it can be seen that the blood flow velocity distribution appearing in the filtered image (F) is more clearly visible than in the raw image (R). In the first to third images (a to c), an improved blood flow velocity distribution in each direction can be observed. In the fourth image (d), it can be seen that the divergence value has decreased.

[0118]

[0119] FIG. 7 is a figure showing the number of abnormal components calculated in a thoracoabdominal aortic image captured during maximum ventricular systole and the number of abnormal components calculated in an image to which the noise filtering method for three-dimensional velocity field data according to an embodiment of the present invention is applied, and FIG. 8 is a graph showing the change in the number of abnormal components over time according to the noise filtering method for three-dimensional velocity field data of FIG. 7. Here, the X-axis represents the flow of time according to an interval of 43.2 ms, and the Y-axis represents the number of abnormal components.

[0120] First, referring to Fig. 7, the raw image (R) shows 52 abnormal components during peak systole. On the other hand, the filtered image (F) shows 21 abnormal components.

[0121] In this way, the number of abnormal components is reduced as the raw image (R) is converted into a filtered image (F) through the noise filtering method (S10) of the three-dimensional velocity field data.

[0122] Referring to FIG. 8, in the raw image (R), there are about 80 to 100 abnormal components in the range of 5 to 10 time frames. However, in the filtered image (F), the number of abnormal components in the range of 5 to 10 is reduced to about 10 to 30.

[0123] As a result, the noise filtering method (S10) for three-dimensional velocity field data can provide accurate and reliable three-dimensional velocity field data by correcting abnormal components and reducing the number of abnormal components.

[0124]

[0125] Although the present invention has been described in detail using preferred embodiments, the scope of the invention is not limited to specific embodiments and should be interpreted by the appended claims. Furthermore, those skilled in the art will understand that many modifications and variations are possible without departing from the scope of the invention.

[0126] The present invention can be used to improve the precision of diagnosis and research in medical imaging, such as magnetic resonance imaging (MRI).

Claims

1. A data collection step for receiving three-dimensional velocity field data; A step of dividing the above three-dimensional velocity field data into a first subdomain containing M voxels; A step of calculating an anomalous component among the voxels included in the first subdomain; and A method for noise filtering of three-dimensional velocity field data comprising the step of applying a singular value decomposition to a voxel containing the above-mentioned abnormal component to correct the above-mentioned abnormal component, and generating first three-dimensional velocity field correction data including the voxel with the above-mentioned abnormal component corrected.

2. In Paragraph 1, The step of calculating the above-mentioned abnormal components is, A noise filtering method for three-dimensional velocity field data, wherein the velocity vector of each voxel included in the first subdomain is multiplied by a weight, and the magnitude of the velocity vector is compared with the magnitude of the average velocity vector of adjacent voxels to calculate the abnormal component.

3. In Paragraph 1, The above first subdomain is, A noise filtering method for three-dimensional velocity field data in which the above M voxels are arranged in three dimensions.

4. In Paragraph 3, The above first subdomain is, A noise filtering method for three-dimensional velocity field data in which the above M voxels are arranged in a rectangular parallelepiped shape.

5. In Paragraph 1, The voxels included in the first subdomain above include voxels containing the above-mentioned abnormal component and voxels not containing the above-mentioned abnormal component, and The step of generating the above-mentioned first three-dimensional velocity field correction data is A noise filtering method for three-dimensional velocity field data, wherein singular value decomposition is applied to voxels that do not contain the above-mentioned abnormal components, the remaining bases excluding the first basis obtained from the singular value decomposition are cut off, and only the vectors of the first basis are synthesized and averaged to reconstruct.

6. In Paragraph 1, A step of dividing the first three-dimensional velocity field correction data into a second subdomain comprising N voxels, which is fewer than M; A step of calculating an anomalous component among the voxels included in the second subdomain; and A noise filtering method for three-dimensional velocity field data, further comprising the step of applying a singular value decomposition to a voxel containing the above-mentioned abnormal component to correct the above-mentioned abnormal component, and generating second three-dimensional corrected data including the voxel with the above-mentioned abnormal component corrected.

7. Data acquisition unit receiving three-dimensional velocity field data; A subdomain generation unit that divides the above three-dimensional velocity field data into a first subdomain including M voxels; An abnormal component calculation unit that calculates abnormal components among the voxels included in the first subdomain; and A noise filtering device for three-dimensional velocity field data comprising a data processing unit that applies a singular value decomposition to a voxel containing the above-mentioned abnormal component to correct the above-mentioned abnormal component and generates first three-dimensional velocity field correction data including the voxel with the above-mentioned abnormal component corrected.

8. In Paragraph 7, The above-mentioned abnormal component calculation unit A noise filtering device for three-dimensional velocity field data that calculates the abnormal component by multiplying the velocity vector of each voxel included in the first subdomain by a weight and comparing the magnitude of the velocity vector with the magnitude of the average velocity vector of adjacent voxels.

9. In Paragraph 7, The voxels included in the first subdomain above include voxels containing the above-mentioned abnormal component and voxels not containing the above-mentioned abnormal component, and The above data processing unit is, A noise filtering device for three-dimensional velocity field data that applies singular value decomposition to voxels that do not contain the above-mentioned abnormal components, cuts off the remaining bases excluding the first basis obtained from the singular value decomposition, and reconstructs by synthesizing and averaging only the vector of the first basis.

10. In Paragraph 7, The above subdomain generation unit divides the first three-dimensional velocity field correction data into a second subdomain containing N voxels, which is fewer than M, and The above abnormal component calculation unit calculates abnormal components among the voxels included in the above second subdomain, and A noise filtering device for three-dimensional velocity field data, wherein the data processing unit applies a singular value decomposition to a voxel containing the above-mentioned abnormal component to correct the above-mentioned abnormal component, and generates second three-dimensional corrected data including the voxel in which the above-mentioned abnormal component has been corrected.

11. Calculate the abnormal components of voxels in a first subdomain partitioned from three-dimensional velocity field data, apply a singular value decomposition to the voxels containing the abnormal components to correct the abnormal components, and generate first three-dimensional velocity field correction data including the voxels corrected for the abnormal components. A noise filtering device for three-dimensional velocity field data that calculates abnormal components of voxels in a second subdomain divided from the first three-dimensional velocity field correction data, applies a singular value decomposition to voxels containing abnormal components to correct the abnormal components, and generates second three-dimensional velocity field correction data including voxels with corrected abnormal components.

12. In Paragraph 11, A noise filtering device for three-dimensional velocity field data in which the number of voxels included in the second subdomain is less than the number of voxels included in the first subdomain.

Images