Apparatus and method for extracting vascular function from brain-related information

Abstract

To provide an apparatus and method for extracting a vascular function from brain-related information.SOLUTION: An apparatus for extracting a vascular function includes: an information reception unit configured to extract an original CT image from brain-related information; an NIFTI image transformation unit configured to transform the original CT image into an NIFTI file format image to acquire time sequence data; a time interpolation unit configured to apply time interpolation to the original CT image through the time sequence data to transform the original CT image into a 3D CT image; a vessel segmentation unit configured to predict a vessel segmentation mask by passing the 3D CT image through a deep-learning model and generate a 4D vessel mask image by stacking the 3D CT images; and a vascular function extraction unit configured to extract a vascular function from a vessel region of the 4D vessel mask image and calculate a blood flow parameter using an artery function which is one of the vascular functions.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an apparatus and method for extracting vascular function from information associated with the brain, and relates to an apparatus and method that can accurately extract vascular function from information associated with the brain. [Background technology]

[0002] A phase contrast magnetic resonance imaging technique using a magnetic resonance imager is widely used as a method for non-invasively measuring fluid flows in the human body, such as blood flow. In this technique, a reference image that does not reflect velocity is taken using a gradient magnetic field pulse that makes the blood flow velocity image zero immediately after the R-wave of the electrocardiogram, and a velocity-encoded image that reflects the blood flow velocity is taken using a velocity-encoding gradient magnetic field. Then, the reference image and the velocity-encoded image are taken. These are taken continuously in pairs during one cycle of the electrocardiogram, and the images are taken with different phase encoding sizes until the entire k-space is filled. The taken images are then reconstructed into velocity images to measure the blood flow velocity during one cycle of the electrocardiogram.

[0003] However, it has been pointed out that such a technique is problematic in that it only measures the blood flow velocity when analyzing blood flow, and is insufficient in terms of high time resolution and accurate representation of blood flow.

[0004] For this purpose, U.S. Patent Publication No. 8,837,800 and other publications disclose a technique for analyzing arterial and venous flow by selecting a portion of the pixels constituting a medical image in order to derive arterial and venous functions using the image, generalizing the signal of each pixel, and applying a Gaussian transform to derive the arterial and venous functions for the blood vessels that each pixel constitutes.

[0005] However, even with these conventional methods, there are problems in that the arterial function and venous function cannot be accurately derived, and errors occur. In particular, it has been pointed out that the inaccurate arterial function makes it impossible to accurately analyze blood flow from blood vessels recognized from medical images.

[0006] Therefore, the present invention provides a user with an apparatus and method that can accurately extract vascular functions so that blood flow from blood vessels can be accurately analyzed.

[0007] Specifically, the present invention provides a device and method for extracting a vascular function from information related to the brain, which can extract a vascular function from information related to the brain generated based on an original CT image in order to accurately calculate blood flow parameters such as cerebral blood flow (CBF), cerebral blood volume (CBV), mean transit time (MTT), etc., and then correct inaccurate arterial function to obtain an accurate arterial function. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Korean Patent Publication No. 10-1992057 (Announced on June 24, 2019) [Patent Document 2] Korean Patent Registration No. 10-1754291 (Announced on July 6, 2017) [Patent Document 3] U.S. Patent Publication No. 8,837,800 (registered September 16, 2014) Summary of the Invention [Problem to be solved by the invention]

[0009] It is an object of the present invention to provide a user with an apparatus and method for extracting vascular function from information associated with the brain.

[0010] Specifically, the present invention provides a user with an apparatus and method for extracting a vascular function from information related to the brain, which can extract a vascular function from information related to the brain generated based on an original CT image in order to accurately calculate blood flow parameters such as CBV, MTT, and CBF, and then correct inaccurate arterial function to obtain an accurate arterial function.

[0011] On the other hand, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by a person having ordinary skill in the art to which the present invention pertains from the following description. [Means for solving the problem]

[0012] A vascular function extraction device, which is one aspect of the present invention for achieving the above technical objectives, may include an information receiving unit that extracts an original CT image from brain-related information received from an external source; a NIFTI image conversion unit that converts the original CT image into an image in a NIFTI file format to obtain time sequence data; a time interpolation unit that applies time interpolation to the original CT image through the time sequence data to convert the original CT image into a 3D CT image for each time period; a vascular segmentation unit that passes each of the 3D CT images for each time period through a deep learning-based vascular segmentation deep learning model 141 to predict a vascular segmentation mask and generate a 4D vascular mask image by stacking the 3D CT image based on a time axis; and a vascular function extraction unit that extracts a vascular function from a vascular region of the 4D vascular mask image and calculates blood flow parameters using an arterial function, which is one of the vascular functions.

[0013] In addition, the vascular function extraction device may further include a motion compensation unit that compensates for motion of each of the time-specific 3D CT images based on a rigid transformation between the step of predicting a vascular segmentation mask by the vascular segmentation unit and the step of generating the 4D vascular mask image.

[0014] In addition, the blood vessel segmentation unit may acquire a blood vessel segmentation mask of the four-dimensional blood vessel mask image, the mask predicted by the blood vessel segmentation deep learning model including at least two of a bone mask, a brain mask, a tissue mask, a vein mask, and an artery mask.

[0015] Also, the arterial mask can be used when the vascular function extraction unit extracts the arterial function.

[0016] Also, the vein mask can be used when the vascular function extraction unit extracts the venous function.

[0017] In addition, the vascular function extraction unit may generate a first graph based on the highest intensity value at a time-phase corresponding to a pixel point of each blood vessel corresponding to a slice coordinate, height coordinate, and width coordinate in the blood vessel region of the 4D vascular mask image.

[0018] In addition, the vascular function extracting unit may remove spike noise in the process of generating the first graph.

[0019] Also, the vascular function extraction unit may generate a second graph by sorting the intensities of the first graph based on the order in which the intensities reach a peak earliest in the first graph.

[0020] Also, the second graph may be a top 70 to 90% graph based on the order in which the intensity reaches a peak earliest among the first graphs.

[0021] Also, the vascular function extracting unit may generate a third graph based on the intensity values ​​of vertices in the second graph in descending order.

[0022] The third graph may be a top 10 to 20% graph based on the order of the intensity values ​​of the vertices of the second graph.

[0023] Also, the vascular function extracting unit may generate a fourth graph based on a sum of intensity changes for each time point of the third graph.

[0024] The fourth graph may be a graph in the bottom 20 to 40% of the third graph, based on the order of the largest sum of the amount of change in intensity for each time series point.

[0025] In addition, the vascular function extraction unit may calculate an average of Intensity based on the time axis in the fourth graph, and remove points outside a specific range as outliers to generate a fifth graph.

[0026] The fifth graph may be a graph in which points outside the range of the average ±1 to 3σ in the fourth graph are removed as outliers.

[0027] Also, the vascular function extraction unit may generate a sixth graph by performing Gaussian fitting on the remaining points of the fifth graph.

[0028] In addition, the vascular function extraction unit may extract a plurality of graphs from the fourth graphs that have a distance closest to the sixth graph, simultaneously with the Gaussian fitting process, and generate a candidate graph based on the plurality of graphs and the sixth graph.

[0029] In addition, the candidate graphs can be added or deleted by the user.

[0030] In addition, the vascular function extraction unit extracts an arterial function graph based on an arterial function based on a process of generating the candidate graph from the first graph, and the arterial function graph may be corrected based on a graph addition or deletion process of the candidate graph.

[0031] In addition, the vascular function extraction unit can calculate blood flow parameters, such as erebral blood volume (CBV), mean transit time (MTT), and capillary blood flow (CBF), based on the corrected arterial function.

[0032] The apparatus for extracting vascular function from information related to a brain, wherein the time interpolation unit applies time interpolation in units of 1 second to the original CT image to convert the original CT image into 3D CT images for each time period in units of 1 second.

[0033] Meanwhile, a vascular function extraction method, which is another aspect of the present invention for achieving the above technical objective, may include: (a) an information receiving unit extracting an original CT image from brain-related information received from an external source; (b) a NIFTI image conversion unit converting the original CT image into an image in a NIFTI file format to obtain time sequence data; (c) a time interpolation unit applying time interpolation to the original CT image through the time sequence data to convert the original CT image into a 3D CT image for each time; (d) a vascular segmentation unit passing the 3D CT image for each time through a deep learning-based vascular segmentation deep learning model 141 to predict a vascular segmentation mask and generating a 4D vascular mask image by stacking the 3D CT image based on a time axis; and (e) a vascular function extraction unit extracting a vascular function from a vascular region of the 4D vascular mask image and calculating blood flow parameters using an arterial function, which is one of the vascular functions. Effect of the Invention

[0034] In accordance with the present invention, a user may be provided with an apparatus and method for extracting vascular function from information associated with the brain.

[0035] Specifically, the present invention can provide a user with an apparatus and method for extracting a vascular function from information related to the brain, which can extract a vascular function from information related to the brain generated based on an original CT image in order to accurately calculate blood flow parameters such as CBV, MTT, and CBF, and then correct an inaccurate arterial function to obtain an accurate arterial function.

[0036] On the other hand, the effects obtained in the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those having ordinary skill in the art to which the present invention pertains from the following description. [Brief description of the drawings]

[0037] [Figure 1] 1 illustrates an example of a block diagram of an apparatus for extracting vascular function from brain-related information in accordance with an embodiment of the present invention. [Diagram 2] 2 illustrates an example of a vascular function extraction method performed by the vascular function device described in FIG. 1. [Diagram 3] 3 is a diagram illustrating a blood vessel division method in the blood vessel division step described in FIG. 2. [Figure 4] 4 is a diagram illustrating a motion compensation concept of a motion compensation unit according to an embodiment of the present invention; [Diagram 5] 3 illustrates an example of the motion compensation step described in FIG. 2. [Figure 6a] 3 is a diagram for explaining an arterial and venous function extraction method in the arterial and venous function extraction step described in FIG. 2. [Figure 6b] 3 is a diagram for explaining an arterial and venous function extraction method in the arterial and venous function extraction step described in FIG. 2. [Figure 6c] 3 is a diagram for explaining an arterial and venous function extraction method in the arterial and venous function extraction step described in FIG. 2. [Figure 7] 7 illustrates an example of points and images corresponding to the candidate graph described in FIG. 6. [Figure 8] 7 is a diagram illustrating an example of arterial and venous functions extracted based on the graph extraction method in the vascular function extraction step described in FIG. 6. [Figure 9] 1 illustrates an example of a block configuration degree of a head division portion. [Figure 10] This shows the operation of the head division unit described in FIG. 9 over time. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0038] A preferred embodiment of the present invention will be described below with reference to the drawings. The embodiment described below does not unduly limit the content of the present invention described in the claims, and the entire configuration described in this embodiment is not essential to the solution of the present invention.

[0039] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The apparatus and method according to the preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0040] Problems with the prior art A phase contrast magnetic resonance imaging technique using a magnetic resonance imager is widely used as a method for non-invasively measuring fluid flows in the human body, such as blood flow. In this technique, immediately after the R-wave of an electrocardiogram, a reference image that does not reflect velocity is taken using a gradient magnetic field pulse that makes the blood flow velocity image zero, and a velocity-encoded image that reflects the blood flow velocity is taken using a velocity-encoding gradient magnetic field. Then, the reference image and the velocity-encoded image are taken. These are taken in pairs continuously during one cycle of an electrocardiogram, and the sizes of the phase encoding are varied until the entire k-space is filled. The taken images are then reconstructed into velocity images to measure the blood flow velocity during one cycle of an electrocardiogram.

[0041] However, it has been pointed out that such a technique is problematic in that it only measures the blood flow velocity when analyzing blood flow, and is insufficient in terms of high time resolution and accurate representation of blood flow.

[0042] For this purpose, U.S. Patent Publication No. 8,837,800 and other publications disclose a technique for analyzing arterial and venous flow by selecting a portion of the pixels constituting a medical image in order to derive arterial and venous functions using the image, generalizing the signal of each pixel, and applying a Gaussian transform to derive the arterial and venous functions for the blood vessels that each pixel constitutes.

[0043] However, even with these conventional methods, there are problems in that the arterial function and venous function cannot be accurately derived, and errors occur. In particular, it has been pointed out that the inaccurate arterial function makes it impossible to accurately analyze blood flow from blood vessels recognized from medical images.

[0044] Therefore, the present invention provides a user with an apparatus and method that can accurately extract vascular functions so that blood flow from blood vessels can be accurately analyzed.

[0045] Specifically, the present invention provides a user with an apparatus and method for extracting a vascular function from information related to the brain, which can extract a vascular function from information related to the brain generated based on an original CT image in order to accurately calculate blood flow parameters such as CBV, MTT, and CBF, and then correct inaccurate arterial function to obtain an accurate arterial function.

[0046] Vascular function extraction device and method The vascular function extraction device 100 accurately extracts a vascular function based on information related to the brain generated based on a CT image and a cerebral vascular map, and components for this purpose are as shown in FIG.

[0047] FIG. 1 illustrates an example block diagram of an apparatus for extracting vascular function from brain-related information according to one embodiment of the present invention.

[0048] Referring to FIG. 1, the vascular function extraction device 100 proposed in the present invention may include an information receiving unit 110, a NIFTI image conversion unit 120, a time interpolation unit 130, a vascular division unit 140, a motion compensation unit 150 and a vascular function extraction unit 160 to extract a vascular function.

[0049] In this specification, the components of the vascular function extracting device 100 will be described in detail through the process of the vascular function extracting method S100 performed by the vascular function extracting device 100.

[0050] The process of the vascular function extraction method S100, which is a method for accurately extracting the vascular function using the vascular function extraction device 100, is shown in FIG.

[0051] FIG. 2 illustrates an example of a vascular function extraction method performed by the vascular function device described in FIG.

[0052] Referring to FIG. 2, the information receiving unit 110 can receive brain-related information from an external source (S110).

[0053] At this time, the brain-related information received by the information receiving unit 110 may include a CT image and a cerebrovascular map as data conforming to the standard format in the medical imaging field (Digital Imaging and Communications in Medicine, DICOM).

[0054] In this specification, the terms brain-related information and Dicom data may be used interchangeably, and the two terms have the same meaning.

[0055] In addition, the CT image included in the Dicom data received by the information receiving unit 110 may be a 3D original (raw) CT image received by the head dividing unit 10 .

[0056] The information receiving unit 110 can extract the original CT image and the cerebral blood vessel map from the received Dicom data, and can provide information such as the cerebral blood vessel map excluding the original CT image to a user.

[0057] That is, since the purpose of the information receiving unit 110 is to extract the original CT image, the process of receiving information related to the brain from the outside can be replaced with a process of receiving a CT image from an external device, similar to the head dividing unit 10 described below.

[0058] Referring to FIG. 2, when the NIFTI image converter 120 receives an original CT image from the information receiver 110, the NIFTI image converter 120 can convert the received original CT image into an image in a Neuroimaging Informatics Technology Initiative (NIFTI) file format (S120).

[0059] Also, the NIFTI image converter 120 converts the original CT image into an image in the NIFTI file format, thereby acquiring a time sequence.

[0060] Referring to FIG. 2, when the time interpolation unit 130 receives the original CT image converted into an image in NIFTI file format and time sequence data from the NIFTI image conversion unit 120, it can apply time interpolation to the received original CT image converted into an image in NIFTI file format (S130).

[0061] At this time, the original CT image can be subjected to time interpolation in units of 1 second by the time interpolation unit 130, and can be converted into 3D CT images (D) for each time period in units of 1 second.

[0062] In this specification, the three-dimensional CT image (D) may be a 3D image of Slice×Height×Width.

[0063] Referring to FIG. 2, when the blood vessel segmentation unit 140 receives a 3D image (D) from the time interpolation unit 130, it converts the received 3D CT image (D) into a 4D CTP (CT Perfusion) image and obtains a blood vessel segmentation mask of the 4D CTP image.

[0064] The steps for the blood vessel segmentation unit 140 to obtain a blood vessel segmentation mask of a 4D CTP image will be described in detail with reference to FIG.

[0065] FIG. 3 is a diagram for explaining a blood vessel division method in the blood vessel division step described in FIG.

[0066] Referring to FIG. 3, when the blood vessel segmentation unit 140 receives a 3D CT image (D) from the time interpolation unit 130 in order to obtain a blood vessel segmentation mask from a 4D CTP (CT Perfusion) image, the blood vessel segmentation unit 140 can predict (or obtain) a blood vessel segmentation mask by passing the 3D CT image (D) for each time through a deep learning-based blood vessel segmentation deep learning model 141 (S140).

[0067] In this case, the deep learning-based vascular segmentation deep learning model 141 may be a deep learning model that has completed learning and verification in the same manner as the Head Segmentation Network configured in the head segmentation unit 10 described later, except for the drawing symbol.

[0068] In addition, the mask predicted by the vascular segmentation deep learning model 141 may be a vascular segmentation mask including at least two or more of a bone mask, a brain mask, a tissue mask, a vein mask, an artery mask, etc.

[0069] The artery mask predicted by the vascular segmentation deep learning model 141 can be used to extract the arterial function, and the vein mask can be used to extract the venous function.

[0070] After predicting (or acquiring) the blood vessel segmentation mask, the blood vessel segmentation unit 140 can generate a 4D blood vessel mask image by stacking the 3D CT image (D) based on the time axis (S150).

[0071] In this specification, the terms 4D vascular mask image and 4D CTP image may be used interchangeably, and the two terms have the same meaning.

[0072] In this specification, a 4D vascular mask image may be a 4D image of time×Slice×Height×Width, generated through a stack of 3D CT images (D).

[0073] In addition, the blood vessel division unit 140 divides the bones, brains, tissues, veins, arteries, etc. for each time axis of the 4D CTP image as described above, and then sets a region of interest in the brain region for the entire Phase to set one Volume of Interest (VOI) for the brain.

[0074] Also, referring to FIG. 2, the motion compensation unit 150 may perform motion compensation of the 3D CT image (D) between a step (S140) of passing the 3D CT image (D) for each time of the blood vessel segmentation unit 140 through a deep learning-based blood vessel segmentation deep learning model 141 to predict a mask, and a step (S150) of stacking the 3D CT image (D) based on the time axis (S145).

[0075] The motion compensation step (S145) of the motion compensation unit 150 is not necessarily limited to being performed, and the step may be omitted depending on the setting made by the user of the apparatus 100.

[0076] In one embodiment, the motion compensation step (S145) of the motion compensation unit 150 is illustrated in FIGS.

[0077] FIG. 4 is a diagram illustrating a motion compensation concept of a motion compensation unit according to an embodiment of the present invention.

[0078] If the patient's head moves even slightly while the patient's 3D CT images (D) are taken based on the time axis, it becomes difficult to detect accurate arterial and venous functions, and therefore, it becomes difficult to accurately calculate blood flow parameters.

[0079] In this case, when viewed from the perspective of the 3D CT image (D) of the 0th frame, the 3D CT image (D) of a specific frame that is shifted due to the patient's movement needs to have a Rigid transformation 145b applied to the 3D CT image (D) of the 0th frame as shown in FIG.

[0080] Here, the rigid transformation is also called Euclidean transformation, and refers to a transformation in which only the position and direction are changed while the shape and size are maintained.

[0081] In this specification, the motion compensation unit 150 predicts a rigid transform 145b for the 3D CT image (D) of the 0th frame for the 3D CT image (D) of a frame other than the 0th frame, and then applies an inverse transform to match the 3D CT image (D) of the 0th frame with the 3D CT image (D) of the remaining frames to compensate for the motion of the 3D CT image (D). An example of the motion compensation step (S145) for compensating for the motion of the 3D CT image (D) is shown in FIG. 5.

[0082] FIG. 5 illustrates an example of the motion compensation step described in FIG.

[0083] Referring to FIG. 5, in the motion compensation step (S145), the motion compensation unit 150 inputs the 3D CT images (D) of frames 1 to N together with the 3D CT image (D) of the 0th frame into an artificial intelligence model, RigidNet (145a), through which it obtains a Rigid transform 145b and calculates an inverse transform. Then, it applies the inverse transform to the 3D CT images (D) of frames 1 to N on a spatial transformation layer 145c, thereby applying motion compensation to obtain 3D CT images (D') of the 1st to Nth compensated frames aligned with the 3D CT image (D) of the 0th frame.

[0084] In one embodiment, the blood vessel segmentation unit 140 can generate a 4D blood vessel mask image by stacking the 3D CT image (D) of the 0th frame and the 3D CT image (D') of the ~Nth corrected frame in the stacking step (S150).

[0085] Further referring to FIG. 2, when the vascular function extraction unit 160 receives a 4D vascular mask image from the vascular division unit 140, it extracts a vascular function from a graph extracted from the vascular region of the received 4D vascular mask image, and then corrects an inaccurate arterial function included in the vascular function to accurately extract the vascular function (S160).

[0086] The arterial and venous function extraction step (S160) of the vascular function extraction unit 160 will be described in detail with reference to FIGS.

[0087] Figures 6a to 6c are drawings explaining an arterial and venous function extraction method in the arterial and venous function extraction step described in Figure 2, Figure 7 illustrates an example of points and images corresponding to the candidate graph described in Figure 6, and Figure 8 illustrates an example of arterial and venous functions extracted based on the graph extraction method in the arterial and venous function extraction step described in Figure 6.

[0088] Referring to FIG. 6a, when the vascular function extraction unit 160 receives a 4D vascular mask image from the vascular division unit 140, it can generate a first graph 161 based on the highest intensity value at a time-specific phase corresponding to the pixel point (Slice, Height, Width coordinates) of each blood vessel in the vascular region of the received 4D vascular mask image.

[0089] In this case, the vascular function extracting unit 160 can remove spike noise, which is noise having a brightness different from that of normal pixels, in the process of generating the first graph 161.

[0090] After generating the first graph 161, the vascular function extraction unit 160 may generate a second graph 162 by sorting the intensities of the first graph 161.

[0091] In this case, the criteria for generating the second graph 162 are not limited, but in this specification, the second graph 162 may be a graph that is in the top 70 to 90% (preferably, 80%) of the first graph 161 based on the order in which the Intensity reaches a peak the earliest.

[0092] Referring to FIG. 6b, after generating the second graph 162, the vascular function extracting unit 160 may generate a third graph 163 based on the intensity values ​​of the peak points in the second graph 162 in descending order.

[0093] In this case, the criteria for generating the third graph 163 are not limited, but in this specification, the third graph 163 may be a graph that is in the top 10 to 20% (preferably 10%) of the second graph 162 based on the order of the intensity values ​​of the peak points.

[0094] After generating the third graph 163, the vascular function extracting unit 160 may generate a fourth graph 164 based on the sum of the changes in intensity for each time point of the third graph 163.

[0095] In this case, the criteria for generating the fourth graph 164 are not limited, but in this specification, the fourth graph 164 may be a graph that is in the bottom 20 to 40% (preferably, 30%) of the third graph 163, based on the order of the largest sum of the amount of change in intensity for each time point.

[0096] Referring to FIG. 6c, after generating the fourth graph 164, the vascular function extraction unit 160 calculates the average intensity based on the time axis in the fourth graph 164, and, for example, assumes that points outside the range of the average ±1 to 3σ (preferably, 1.5σ) are outliers and removes them to generate a fifth graph 165.

[0097] After generating the fifth graph 165, the vascular function extraction unit 160 may generate a sixth graph 166 by performing Gaussian fitting on the remaining points of the fifth graph 165.

[0098] At this time, the method by which the vascular function extraction unit 160 performs Gaussian fitting is not limited, but in this specification, Gaussian fitting may be performed based on a plurality of (e.g., five) Gaussian functions.

[0099] The vascular function extracting unit 160 can extract a plurality of graphs having the shortest distance from the sixth graph 166 from the fourth graph 164 simultaneously with the above-mentioned Gaussian fitting process.

[0100] Thereafter, the vascular function extraction unit 160 may generate a candidate graph 167 based on a plurality of graphs having the shortest distance from the sixth graph 166 among the sixth graph 166 and the fourth graph 164 obtained based on Gaussian fitting.

[0101] At this time, the candidate graph 167 is generated so that the user of the device 100 can delete a graph that he or she wishes to remove from the candidate graph 167 generated from the vascular function extraction unit 160, or can add a graph that is not included in the candidate graph 167.

[0102] An example of the first, second, third, and fourth points (167a, 167b, 167c, and 167d) corresponding to the graph of the candidate graph 167 generated from the vascular function extraction unit 160 and the first, second, third, and fourth images (168a, 168b, 168c, and 168d) corresponding thereto is shown in FIG. 6.

[0103] In this manner, the points (167a to 167d) shown in FIG. 7 may include Slice coordinates, Height coordinates, and Width coordinates.

[0104] Meanwhile, the arterial function graph 169a and the venous function graph 169b can be generated by the vascular function extraction unit 160 by applying the same process of generating the candidate graph 167 from the first graph 161 described above, as shown in FIG. 8.

[0105] At this time, the vascular function extraction unit 160 generates the arterial function graph 169a based on the process of generating the candidate graph 167 from the first graph 161 in order to correct inaccurate arterial functions based on the process of adding or deleting graphs to generate the candidate graph 167.

[0106] In this manner, the vascular function extraction unit 160 can correct inaccurate arterial functions by adding or deleting the arterial function graph 169a, and by correcting the inaccurate arterial function, it can accurately calculate blood flow parameters such as CBV, MTT, and CBF, which are calculated based on the accurately corrected arterial function variables.

[0107] In one embodiment, the vascular function extraction unit 160 can be configured with a blood flow parameter model for accurately calculating blood flow parameters.

[0108] Blood flow parameter model The blood flow parameter model of the vascular function extractor 160 is implemented as follows.

[0109] First, the blood flow parameter model of the vascular function extraction unit 160 can be trained to calculate blood flow parameters through an SVD (Singular Value Decomposition) based approach, which is one of the matrix decomposition methods, and is a technique for separating a high-dimensional matrix into a low-dimensional matrix.

[0110] The blood flow parameter model of the vascular function extraction unit 160 can calculate cerebral blood volume (CBV), which is one of the blood flow parameters, through the following Equation 1.

number

[0111] In the above [Equation 1], C v is the density change of the VOI tissue, C a may be the arterial function 159a.

[0112] In this manner, in order to calculate the CBV, which is one of the blood flow parameters, the arterial function 159a is applied as a variable, and in order to accurately calculate the CBV, a process of correcting the inaccurate arterial function 159a is required.

[0113] The blood flow parameter model of the vascular function extraction unit 160 can calculate MTT (mean transit time), which is one of the blood flow parameters, through the following Equation 2.

number

[0114] In the above [Number 2], F vmay be another blood flow parameter, CBF (capillary blood flow).

[0115] Meanwhile, an example of a formula for the blood flow parameter model of the blood vessel function extraction unit 160 to perform singular value decomposition is given by the following [Equation 3].

number

[0116] In the above [Equation 3], R is the Residue Function, and C a and C. v can be normalized to a value of 0 at t=0.

[0117] The blood flow parameter model of the vascular function extraction unit 160 decomposes the matrix through the SVD-based approach based on the equation (3) as shown in the following equations (4) to (8).

number

number

number

number

number

[0118] The blood flow parameter model of the vascular function extraction unit 160 can find a basis that is robust to noise by applying a threshold to the W term of the above-mentioned [Equation 7], which is a diagonal matrix.

[0119] Blood flow parameter calculation The blood flow parameter model of the vascular function extraction unit 160 trained through the SVD-based approach can calculate the blood flow parameters CBF, CBV, and MTT, and the relationship between the blood flow parameters is as shown in the following [Equation 9] to [Equation 11].

number

number

number

[0120] The blood flow parameter model of the vascular function extraction unit 160 can calculate the blood flow parameters CBF, CBV, and MTT based on an approximation r' value such that the blood flow parameters CBF, CBV, and MTT have mutual relationships.

[0121] In this case, the conditions for calculating the blood flow parameters CBF, CBV, and MTT may be 1.04 g / ml for the average brain density, the unit of CBF may be #ml / 100 g / min, and the unit of CBV may be ml / 100 g.

[0122] Also, 4D CTP images can be reconstructed with 1 mm isotropic voxels.

[0123] When the blood flow parameter model of the vascular function extraction unit 160 matches the units as described above, it applies 100g / 1.04 as the 1 / p value, and when plotting CBF, since it is based on a minute basis, it is possible to display the blood flow parameters calculated by multiplying by 60s.

[0124] Front-end configuration of the vascular function extraction device A head division section 10 may be provided at the front end of the vascular function extraction device 100 of the present invention.

[0125] The head segmentation unit 10 receives a CT image and passes the CT image through a head segmentation network that can separate the skull and the brain, thereby predicting a bone mask of the skull, a brain mask of the brain, and a vessel mask associated with the brain.

[0126] Furthermore, the head segmentation unit 10 can combine the predicted brain mask and the blood vessel mask to extract a brain image including only the brain region from the CT image.

[0127] Head division section FIG. 9 illustrates an example of the block configuration degree of the head division portion.

[0128] Referring to FIG. 9 , the head division unit 10 may include an image receiving unit 11 , a head pre-processing unit 12 , a pre-trained head division network unit 13 , a head post-processing unit 14 and a prediction result calculation unit 15 .

[0129] The region of interest to be analyzed in the present invention is the brain. Since the thickness and shape of the skull vary from person to person, the head segmentation unit 10 passes through a deep learning-based Head Segmentation Network that can separate the skull and brain regions, predicts each bone, brain, and vessel mask, and then combines the brain mask and vessel mask to extract the complete brain region from the original CT image.

[0130] First, the image receiving unit 11 receives a CT image from an external device.

[0131] Next, the head pre-processing unit 12 divides the 3D image of the Raw CT image into 2D axial slices, and then concatenates adjacent slices of the slice to be segmented to form input data.

[0132] Specifically, the input data is subjected to pre-processing in the head pre-processing unit 12 and then input to the head division network unit 13 which has been pre-trained.

[0133] The head pre-processing unit 12 can perform operations such as (1) Resize, (2) HU Windowing, and (3) Normalize.

[0134] (1) Resizing is the process of reducing and / or enlarging an image to a size required by the head division network.

[0135] Next, (2) HU windowing is the task of setting upper and lower limits of the Hounsfield unit (HU).

[0136] In addition, the normalizing operation is an operation of determining HU levels from a set lower limit to an upper limit and mapping the HU levels at the determined intervals of [0, 1]. Typically, a method of mapping HU levels from a lower limit to an upper limit of HU windowing at intervals of [0, 1] may be applied.

[0137] The pre-processed CT image passes through a pre-trained head division network unit 13, and the head post-processing unit 14 performs at least one post-processing operation of a brain VOI crop operation and an outlier cluster removal operation on the CT image that has passed through the head division network 13.

[0138] The pre-trained head segmentation network unit 13 is a deep learning-based segmentation network that is pre-trained on the skull and brain regions so as to be able to extract the brain region, which is the region of interest.

[0139] Here, the Brain VOI Crop operation is a process of removing an area other than the Volume of Interest area for the brain and then re-adjusting the size.

[0140] The outlier cluster removal process is a process for removing small outlier clusters that are not connected in three dimensions and are separated from one another.

[0141] Thereafter, the prediction result calculation unit 15 predicts the bone mask, the brain mask, and the vessel mask based on the post-processed CT image.

[0142] In other words, a bone mask of the skull, a brain mask of the brain, and a vessel mask associated with the brain are predicted, and by combining the predicted brain mask and vessel mask, a brain image containing only the brain region can be extracted from the initially input CT image.

[0143] Head division operation FIG. 10 shows the operation of the head division unit described in FIG. 9 over time.

[0144] Referring to FIG. 10, an image receiving unit 11 receives a CT image from an external device.

[0145] Next, the head pre-processing unit 12 divides the 3D image of the Raw CT image into 2D axial slices, and then concatenates adjacent slices of the slice to be segmented to form input data.

[0146] Specifically, the input data is subjected to pre-processing in the head pre-processing unit 12 and then input to the head division network unit 13 which has been pre-trained.

[0147] The head pre-processing unit 12 can perform the above-mentioned operations such as (1) Resize, (2) HU Windowing, and (3) Normalize.

[0148] The pre-processed CT image passes through a pre-trained head division network unit 13, and the head post-processing unit 14 performs at least one post-processing operation of a brain VOI crop operation and an outlier cluster removal operation on the CT image that has passed through the head division network 13.

[0149] Thereafter, the prediction result calculation unit 15 predicts information including the bone mask (17), brain mask (18) and vessel mask (19) based on the post-processed CT image.

[0150] That is, a bone mask (Bone Mask, 17) of the skull, a brain mask (Brain Mask, 18) of the brain, and a vascular mask (Vessel Mask, 19) associated with the brain are predicted, and by combining the predicted brain mask 18 and the vascular mask 19, a brain image (K) containing only the brain region can be extracted from the initially input CT image.

[0151] Effects of the present invention In accordance with the present invention, a user may be provided with an apparatus and method for extracting vascular function from information associated with the brain.

[0152] Specifically, the present invention can provide a user with an apparatus and method for extracting a vascular function from information related to the brain, which can extract a vascular function from information related to the brain generated based on an original CT image in order to accurately calculate blood flow parameters such as CBV, MTT, and CBF, and then correct an inaccurate arterial function to obtain an accurate arterial function.

[0153] On the other hand, the effects obtained in the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those having ordinary skill in the art to which the present invention pertains from the following description.

[0154] The above-described embodiments of the present invention may be implemented in a variety of ways, for example, in hardware, firmware, software, or a combination thereof.

[0155] In the case of a hardware implementation, the methods of the present invention may be implemented using one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, etc.

[0156] In the case of implementation by firmware or software, the method according to the embodiment of the present invention may be implemented in the form of modules, procedures, or functions that perform the functions or operations described above. The software code may be stored in a memory unit and driven by a processor. The memory unit may be located inside or outside the processor and may exchange data with the processor by various means that are well known in the art.

[0157] The detailed description of the preferred embodiments of the present invention disclosed above is provided to enable those skilled in the art to embody and practice the present invention. Although the present invention has been described above with reference to the preferred embodiments, those skilled in the art will understand that the present invention can be modified and changed in various ways without departing from the scope of the present invention. For example, those skilled in the art can use the configurations described in the above embodiments in a combination with each other. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0158] The present invention may be embodied in other specific forms without departing from the spirit and essential characteristics of the present invention. Therefore, the above detailed description should not be construed as limiting in all respects, but should be considered as illustrative. The scope of the present invention should be determined by a reasonable analysis of the appended claims, and all modifications within the scope of the equivalents of the present invention are included in the scope of the present invention. The present invention is not intended to be limited to the embodiments shown herein, but to be accorded the widest scope consistent with the principles and novel features disclosed herein. In addition, claims that do not have an explicit reference relationship in the claims may be combined to form an embodiment or may be included as a new claim by amendment after filing.

Claims

1. an information receiving unit for extracting original CT images from brain-related information received from an external source; an NIFTI image converter for converting the original CT image into an image in a NIFTI file format to obtain time sequence data; a time interpolation unit that applies time interpolation to the original CT image through the time sequence data to convert the original CT image into a 3D CT image for each time; a blood vessel segmentation unit that predicts a blood vessel segmentation mask by passing each time-based 3D CT image through a deep learning-based blood vessel segmentation deep learning model 141 and generates a 4D blood vessel mask image by stacking the 3D CT images based on a time axis; and An apparatus for extracting a vascular function from information related to the brain, comprising: an arterial and venous function extraction unit that extracts a vascular function from a vascular region of the 4D vascular mask image and calculates blood flow parameters using an arterial function, which is one of the vascular functions.

2. The apparatus for extracting vascular functions from brain-related information as described in claim 1, further comprising: a motion compensation unit that compensates for motion of each of the time-specific 3D CT images based on a rigid transformation between the process of predicting a vascular segmentation mask and the process of generating the 4D vascular mask image.

3. The blood vessel division portion includes:

2. The apparatus for extracting vascular functions from brain-related information according to claim 1, wherein the mask predicted by the vascular segmentation deep learning model is a vascular segmentation mask of the 4-dimensional vascular mask image including at least two of a bone mask, a brain mask, a tissue mask, a vein mask, and an artery mask.

4. The arterial mask comprises:

4. An apparatus for extracting vascular function from information associated with the brain as claimed in claim 3, characterized in that said vascular function extractor is used when extracting said arterial function.

5. The venous mask comprises:

4. An apparatus for extracting a vascular function from information associated with the brain as claimed in claim 3, characterized in that said vascular function extractor is used when extracting said venous function.

6. The vascular function extraction unit The apparatus for extracting vascular functions from information related to the brain as described in claim 1, characterized in that a first graph is generated based on the highest intensity value at a time-phase corresponding to a pixel point of each blood vessel corresponding to a slice coordinate, height coordinate, and width coordinate in a blood vessel region of the 4D vascular mask image.

7. The vascular function extraction unit 7. The apparatus for extracting a vascular function from brain-related information according to claim 6, further comprising removing spike noise in the process of generating the first graph.

8. The vascular function extraction unit 7. The apparatus for extracting vascular functions from brain-related information according to claim 6, wherein the second graph is generated by sorting the intensities of the first graph based on the order in which the intensities of the first graph reach a peak earliest.

9. The second graph is 9. The apparatus for extracting a vascular function from brain-related information according to claim 8, wherein the first graph is a top 70 to 90% graph based on an order in which intensity reaches a peak earliest among the first graphs.

10. The vascular function extraction unit 9. The apparatus for extracting a blood vessel function from information related to a brain according to claim 8, wherein the third graph is generated based on the increasing intensity value of the vertices from the second graph.

11. The third graph is 11. The apparatus for extracting a vascular function from brain-related information according to claim 10, wherein the second graph is a top 10 to 20% graph based on the order of the intensity values ​​of the vertices of the second graph.

12. The vascular function extraction unit 11. The apparatus for extracting a vascular function from brain-related information according to claim 10, wherein a fourth graph is generated based on a sum of intensity changes for each time point of the third graph.

13. The fourth graph is The apparatus for extracting a vascular function from brain-related information according to claim 12, wherein the third graph is a graph in the bottom 20 to 40% based on the order of the largest sum of intensity changes for each time series point among the third graphs.

14. The vascular function extraction unit The apparatus for extracting vascular functions from brain-related information according to claim 12, characterized in that an average of intensity is calculated based on a time axis in the fourth graph, and points outside a specific range are assumed to be outliers and removed to generate a fifth graph.

15. The fifth graph is The apparatus for extracting vascular functions from brain-related information as described in claim 14, characterized in that the fourth graph is a graph in which points outside the range of the mean ±1 to 3σ are removed as outliers.

16. The vascular function extraction unit 15. The apparatus for extracting vascular functions from brain-related information as claimed in claim 14, further comprising: performing Gaussian fitting on the remaining points of the fifth graph to generate a sixth graph.

17. The vascular function extraction unit extracting a plurality of graphs having a closest distance to the sixth graph from the fourth graph simultaneously with the Gaussian fitting process; 17. The apparatus for extracting a vascular function from brain-related information according to claim 16, further comprising: generating a candidate graph based on the plurality of graphs and the sixth graph.

18. The candidate graph is 20. An apparatus for extracting vascular functions from brain-related information as claimed in claim 17, characterized in that graphs can be added or deleted by the user.

19. The vascular function extraction unit Extracting an arterial function graph based on an arterial function based on a process of generating the candidate graph from the first graph; The arterial function graph is 20. The apparatus for extracting a vascular function from information related to a brain according to claim 18, wherein the arterial function is corrected based on a graph addition or deletion process of the candidate graph.

20. The vascular function extraction unit 20. The apparatus for extracting a vascular function from information related to the brain according to claim 19, further comprising: calculating blood flow parameters, cerebral blood volume (CBV), mean transit time (MTT), and capillary blood flow (CBF), based on the corrected arterial function.

21. The time interpolation unit is 2. The apparatus for extracting vascular functions from information related to the brain according to claim 1, further comprising: applying time interpolation in units of 1 second to the original CT image to convert the original CT image into 3D CT images for each time period in units of 1 second.

22. (a) extracting an original CT image from brain-related information received from an external device by an information receiving unit; (b) an NIFTI image converter converts the original CT image into an image in a NIFTI file format to obtain time sequence data; (c) a time interpolation unit applies time interpolation to the original CT image through the time sequence data to convert the original CT image into a time-specific 3D CT image; (d) a blood vessel segmentation unit predicts a blood vessel segmentation mask by passing the 3D CT images for each time period through a deep learning-based blood vessel segmentation deep learning model 141, and generates a 4D blood vessel mask image by stacking the 3D CT images based on a time axis; and (e) a vascular function extraction unit extracts vascular functions from vascular regions of the 4D vascular mask image, and calculates blood flow parameters using an arterial function, which is one of the vascular functions.

Images