WO2017047819A1

WO2017047819A1 - Blood vessel shape analysis device, method for same, and computer software program for same

Info

Publication number: WO2017047819A1
Application number: PCT/JP2016/077754
Authority: WO
Inventors: 高伸八木; 将彦中本
Original assignee: イービーエム株式会社
Priority date: 2015-09-18
Filing date: 2016-09-20
Publication date: 2017-03-23
Also published as: JPWO2017047819A1

Abstract

[Solution] This device comprises: an acquisition unit for acquiring a medical image; a thinning unit that extracts a blood vessel area contained in at least part of the medical image, and acquires a centerline of a blood vessel in the blood vessel area; a dividing unit that divides the blood vessel area into blood vessel elements on the basis of the centerline of the blood vessel; and a shape measuring unit that generates shape measurement data by measuring the three-dimensional shape of each of the blood vessel elements, the shape measurement data containing at least one of information pertaining to the blood vessel cross section at one or more positions in each of the blood vessel elements, information pertaining to the surface shape of the blood vessel, and information pertaining to the geometrical characteristics of the centerline of the blood vessel.

Description

Blood vessel shape analysis apparatus, method thereof, and computer software program thereof

The present invention relates to a blood vessel shape analysis apparatus having a function of determining and displaying an abnormality of a blood vessel shape, which is a hotbed for the onset and growth of vascular lesions, a method thereof, and a computer software program thereof.

Conventionally, it is known that vascular lesions such as arteriosclerosis and aneurysms have a common site. One cause of vascular lesions is malignant stimulation by blood flow. The fluid force due to the blood flow is closely related to the vascular regulatory function via the vascular endothelial cells, and it is generally considered that the presence of a stimulus exceeding the compensation range of the blood vessel itself triggers the vascular lesion.

In recent years, experimental measurement methods such as phase contrast MRI and computational science methods using computer simulation have entered practical use as methods for analyzing the above-mentioned blood flow state, and attempts have been made to quantify malignant stimulation of blood flow as described above. Has come to be. For example, Japanese Patent No. 5596865 discloses a technique for determining a wall shear stress vector at each position of a blood vessel wall surface by analyzing a blood flow by computer simulation and determining malignant / benign blood flow from the shear stress vector. ing.

Malignant blood flow is closely related to unstable shear stress. Here, “instability” means that the direction of the shear stress is destabilized within one pulsation cycle. For example, flow collisions, vortices, separation, and the like are factors that cause instability of shear stress. This “destabilization” of blood flow is attributed to the shape of the blood vessel as described in detail below. Therefore, by analyzing the blood vessel shape, it is expected that the risk of malignant or unstable blood flow can be predicted without performing the above method.

By the way, in modern medicine, blood vessel lesions such as cerebral aneurysms are analyzed by measuring abnormal blood vessel shapes. However, the measurement is generally performed in a state where a three-dimensional vascular lesion is viewed two-dimensionally from a certain viewpoint direction. In addition, these conventional measurement methods do not fully share and standardize the principles of measurement, and the measurement results depend on the measurer. The problem is that it cannot be built.

Vascular lesions are the main cause of death if the heart and brain are summed up. In particular, in advanced countries that are facing a super-aging society, there is an urgent need to try to prevent vascular lesions. For this reason, it is required to clarify the causal relationship between lesions. In such a situation, it is important to make big data of vascular lesions into a database and statistically analyze the feature values. For example, about 5% of the population of cerebral aneurysms is potentially over 50 years old, and the development and development of cerebral aneurysms will progress if database of cerebral aneurysm big data and computer analysis technology advance It is expected to be able to clarify the causal relationship of rupture.

Japanese Patent No. 5596865

The present invention has been made in view of such a situation, and an object of the present invention is to analyze (diagnose) and display a blood vessel shape serving as a hotbed for the onset and growth of vascular lesions, a method thereof, and computer software thereof. To provide a program. *

In order to solve the above-described problem, according to a first main aspect of the present invention, an acquisition unit that acquires a medical image and a blood vessel region included in at least a part of the medical image are extracted, and the blood vessel region is placed in the blood vessel region. A thinning unit that acquires a blood vessel center line, a dividing unit that divides the blood vessel region into blood vessel elements based on the blood vessel center line, and a three-dimensional shape of each of the blood vessel elements to generate shape measurement data A shape measurement unit, wherein the shape measurement data includes at least one of information relating to a blood vessel cross section at one or more positions of each blood vessel element, information relating to a blood vessel surface shape, and information relating to a geometric characteristic of a blood vessel center line. A blood vessel shape analysis apparatus comprising the shape measurement unit including a plurality of shape measurement units is provided.

Here, according to one embodiment of the present invention, the apparatus further includes a determination unit that determines a risk of developing a lesion for each blood vessel shape based on whether or not the shape measurement data exceeds a threshold value. The threshold value includes the determination unit and an output unit that outputs the determination result. The determination unit is preset based on statistical data on the geometric shape information of the blood vessel. Further, in this case, it is preferable that the output unit displays the determination result together with the blood vessel region.

In addition, the shape measuring unit includes at least one of the diameter, unevenness, bending degree, and twisting degree of the blood vessel cross section at one or more positions of each blood vessel element, and the branching angle at the blood vessel branching portion between the blood vessel elements. Is preferably calculated.

According to another embodiment, the shape measuring unit measures a blood vessel cross-sectional area along the traveling direction of the blood vessel center line and calculates an equivalent diameter thereof. A standard blood vessel generation unit that constructs a standard blood vessel model based on an equivalent diameter, a difference image calculation unit that calculates a difference image between the medical image and the standard blood vessel model and identifies fragments included in the difference image, and Fragment analysis that measures the shape of each fragment, determines the attribute of the fragment from the measured shape of the fragment based on fragment geometric feature information obtained by supervised machine learning, and outputs the determination result This fragment analysis unit determines whether at least a fragment is a blood vessel abnormality or noise. Those having a that fragment analyzer. In this case, it is preferable that the apparatus further includes a blood vessel abnormality dividing unit that divides and displays the blood vessel abnormal portion based on the determination result of the attribute of the fragment.

According to yet another embodiment, the standard blood vessel generation unit generates blood vessel diameter change characteristic data based on an equivalent diameter calculated at a plurality of discrete positions in the blood vessel element, and The standard blood vessel model is constructed by fitting an approximate curve to the blood vessel diameter change characteristic data.

According to another embodiment, the fragment analysis unit discriminates the type of vascular lesion as the attribute of the fragment.

According to another embodiment, the supervised machine learning is to update the fragment geometric feature information by creating a database and analyzing the determination result.

According to the second main aspect of the present invention, the computer acquires the medical image, the computer extracts the blood vessel region included in at least a part of the medical image, and the blood vessel center in the blood vessel region A thinning process to obtain a line;
A dividing step in which the computer divides the blood vessel region into blood vessel elements based on the blood vessel center line; and a shape measuring step in which the computer measures the three-dimensional shape of each of the blood vessel elements to generate shape measurement data. The shape measurement data includes at least one of information on a blood vessel cross-section at one or more positions of each blood vessel element, information on a blood vessel surface shape, and information on a geometric characteristic of a blood vessel center line. There is provided a blood vessel shape analysis method comprising the shape measurement step.

According to a third main aspect of the present invention, there is provided a computer software program for analyzing a blood vessel shape, comprising the following steps: an acquisition step in which a computer acquires a medical image; A thinning step of extracting a blood vessel region included in at least a part and acquiring a blood vessel center line in the blood vessel region; and a dividing step of dividing the blood vessel region into blood vessel elements based on the blood vessel center line; A shape measurement step in which a computer measures the three-dimensional shape of each of the blood vessel elements to generate shape measurement data, the shape measurement data being information relating to a blood vessel cross section at one or more positions of each blood vessel element , Information on the surface shape of the blood vessel, and information on the geometric characteristics of the blood vessel center line. Computer software program, characterized in that it includes instructions for executing the said shape measuring process is provided.

It should be noted that other features of the present invention other than those described above can be easily understood by those skilled in the art by referring to the “Description of Embodiments” and the drawings described below.

FIG. 1 is a schematic configuration diagram of a blood vessel shape analysis apparatus according to an embodiment of the present invention. FIG. 2 is a diagram showing a blood vessel shape analysis flow in one embodiment of the present invention. FIGS. 3A to 3C are diagrams for explaining the thinning of the blood vessel shape. FIG. 4 is a diagram for explaining graphing. FIG. 5 is a diagram for explaining a phenomenon that appears to be adhered. FIG. 6 is a diagram for explaining the graphing of the blood vessel structure and the depth-first search of the graph. FIGS. 7A and 7B are diagrams illustrating the adhesion separation process. FIG. 8 is a diagram for explaining measurement of a blood vessel shape. FIGS. 9A and 9B are diagrams for explaining measurement of irregularities on the surface of a blood vessel. FIG. 10 is a diagram illustrating an example of a table of shape measurement data. FIGS. 11A and 11B are views for explaining determination of blood vessel shape classification. FIG. 12 is a diagram graphically showing the determination result superimposed on the three-dimensional shape of the blood vessel. FIG. 13 is a diagram for explaining a conventional method for measuring a vascular lesion site. FIG. 14 is a schematic configuration diagram of a blood vessel abnormality detection division apparatus according to the second embodiment of the present invention. FIG. 15 is a diagram showing a blood vessel abnormality detection division flow in the second embodiment of the present invention. FIGS. 16A to 16C are views for explaining generation of a standard blood vessel in the second embodiment. FIG. 17 is a diagram for explaining a difference image in the second embodiment. FIG. 18 is a diagram for explaining fragment analysis in the second embodiment. FIG. 19 is a diagram graphically showing a divided abnormal part including a neckline superimposed on a medical image.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

As described above, the first aspect of the present invention is an apparatus that analyzes (diagnose) a blood vessel shape serving as a hotbed for the onset / growth of a vascular lesion and outputs the result. In this invention, focusing on the fact that blood flow instability is caused by the blood vessel shape, calculating the degree of abnormality of the blood vessel shape identifies the blood vessel site that can be a factor that causes the onset and progression of vascular tissue lesions. -Determine the state.

For example, in the industrial field such as piping design, when designing an expansion pipe, it is common to keep the expansion angle of the pipe within a certain angle (eg, 10 degrees). This is because if the flow exceeds the specified angle, the flow is separated, and a vortex is generated in the vicinity of the wall to form a dead zone that does not substantially contribute to mass transport. When the present inventors look at the blood vessel shape from such a viewpoint, most of the normal blood vessels are reduced, but when accompanied by vascular lesions, the blood vessels may be locally enlarged. I learned from experience. That is, in the reduced type pipe, the flow instability is suppressed with respect to the flow direction. On the other hand, as described above, the enlargement type has the property of becoming unstable. This is because the pressure gradient is different with respect to the flow direction, and the fluid is accelerated with a negative value in the reduction type, but the fluid is decelerated with a positive value in the expansion type. This invention has been completed as a result of sincerity verification and experiments based on the above viewpoint.

Hereinafter, an embodiment of the present invention will be specifically described. In the following, a case of a cerebral aneurysm will be described as a preferred example.

FIG. 1 is a schematic configuration diagram of a blood vessel shape analyzer (blood vessel shape diagnostic device) 10 according to an embodiment of the present invention. Take a medical image as input.

In the blood vessel shape analysis apparatus 10, a program storage unit 60 and

data storage units

70 and 71 are connected to a bus 50 to which a CPU 11, a RAM 12, and an input / output IF 13 are connected. In the device 10, the bus 50 may be further connected to other devices and / or communication ports that store various external reference data. Further, an input device (not shown) such as a mouse, a keyboard, or a touch screen, a display (not shown), or another output device can be connected to the input / output IF 13 of the apparatus 10.

The program storage unit 60 includes an input unit 14, a thinning unit 15, a graphing unit 16, a shape measuring unit 17, a determining unit 18, and a display unit 19.

The data storage unit 70 stores the input medical image. The data storage unit 71 stores the blood vessel shape data 20, the blood vessel shape measurement data 21, the risk determination result 22, and the like.

The configuration (input unit 14, thinning unit 15, graphing unit 16, shape measuring unit 17, determining unit 18, and display unit 19) is actually configured by computer software stored in a storage area of the hard disk. By being called by the CPU 11 and expanded and executed on the RAM 12, it is configured and functions as each component of the present embodiment.

The function of each part is described below together with the processes that each takes charge of. FIG. 2 is a diagram showing a blood vessel shape analysis (blood vessel shape diagnosis) flow of the apparatus 10 in the present embodiment.

(Input section)
The input unit 14 reads a medical image or the like (step S0). In this embodiment, the input medical image is a device capable of acquiring a tomographic image of a target blood vessel site such as MRA (magnetic resonance image), CTA (X-ray computed tomography image), DSA (angiographic image), or the like. In addition, it may be obtained by various apparatuses capable of acquiring image data in a target blood vessel site, such as US (ultrasonic image), IVUS (intravascular ultrasonic image), OCT (near infrared image). The target blood vessel site may be, for example, a cerebral artery, a coronary artery, a carotid artery, an aorta, or another target blood vessel site of a subject. The medical image is temporarily recorded in the data storage unit 70 via the input / output IF 13 or by a communication port (not shown) or other transfer means, and then input to the thinning unit 15.

(Thinning section)
The thinning unit 15 binarizes and thins the medical image to acquire the center line of the blood vessel (step S1). Hereinafter, a specific description will be given with reference to FIG. The thinning unit 15 first binarizes the read medical image as shown in FIG. 3A, extracts a target blood vessel region, and generates blood vessel three-dimensional shape data (FIG. 3B). . In this embodiment, the thinning unit 15 automatically sets a binarization threshold so as to extract a characteristic specific to a blood vessel wall based on a histogram of luminance values of the entire image. In other embodiments, the user may select a binarization threshold. Next, the thinning unit 15 performs thinning processing on the binarized blood vessel three-dimensional shape data (hereinafter referred to as “blood vessel shape”) to obtain a blood vessel center line (FIG. 3C). A plurality of algorithms are known for thinning and are not limited to a specific algorithm. In this embodiment, the thinning unit 15 centers the extracted voxel of the blood vessel region from the outer peripheral side (that is, the surface) of the blood vessel region. The center line is obtained by extracting the core of the blood vessel by cutting it toward the center and fitting a spline curve or the like in the blood vessel running direction.

(Graphing part)
The graphing unit 16 performs graphing using the blood vessel center line obtained by the thinning process (step S2). In this embodiment, graphing means labeling center line data corresponding to each blood vessel portion, and the graphed data is referred to as graph data. In the graphing process, the graphing unit 16 first divides the blood vessel center line into elements for each region. As shown in FIG. 4, this element division is performed by specifying the end / branch point (A, B, C,...) In the center line acquired by the thinning process, and at the end / branch point. This is done by dividing the center line. Hereinafter, each blood vessel portion corresponding to the center line between the divided end points and branch points is referred to as a “blood vessel element”. Thereafter, the graphing unit 16 performs labeling (# 1, 2, 3,...) On the center line data of each blood vessel element to generate graph data (graph).

Next, the graphing unit 16 detects and separates adhesions. As shown in FIG. 5, “adhesion” here means that two blood vessels that are running in close proximity are integrated without being separated in the extraction process even though they are not originally adhered. Refers to the phenomenon that is extracted.

Normally, the blood vessel is a loop circuit of a closed circuit at the whole body level, but a loop is not formed in a relatively microscopic region that is a target of blood vessel shape diagnosis. Therefore, in this embodiment, the presence or absence of adhesion is detected according to the presence or absence of a loop. Hereinafter, detection and separation of adhesions will be described in more detail.

First, the adhesion site is detected by executing a depth-first search on the graph. That is, end points and branch points are detected by analyzing the three-dimensional blood vessel shape of the extracted blood vessels, and their connection relationship is examined.

For example, in the graph shown in FIG. 6B, a blood vessel branch structure is expressed. In FIG. 6, circles indicate nodes, which are end points or branch points. When adhesion occurs, this graph appears as a cycle (loop), and this cycle is detected by performing a depth-first search of the graph. This is an operation that follows the edges from the initial node through all nodes. The rules for tracing are as follows.
(1) When a branch point is reached, an edge that has not passed is selected and the process proceeds to the next node.
(2) When the end point is reached, return to the previous branch point.
(3) If there is no side that has not passed when returning to the branch point, the process returns to the previous branch point.
In FIG.6 (c), the number of each node represents the visit order of depth priority search. A solid arrow indicates a forward path, and a dotted arrow indicates a return path. In the example shown in FIG. 6C, a closed circuit is formed between No. 5 and No. 6. When the depth-first search is performed, it proceeds from the 5th node to 6th, 7th, 8th and returns to 6th. Here, since the forward path has passed through the lower side between No. 5 and No. 6, the upper side has not yet passed. However, if you select the upper side, you will arrive at the number 5 that you have already visited. If there is no closed path, only the return path will return to the visited node, but if there is a closed path, the visited node will be reached in the forward path. Therefore, the graphing unit 16 detects the closed circuit by recording the passage conditions of the nodes and edges, and the node at the time point when the closed circuit is determined (for example, in the case of the graph shown in FIG. 6 (c), the number 6). ) Is considered an adhesion site.

When the graphing unit 16 detects the adhesion site, it next separates the adhesion site. The adhesion site is separated based on the running direction and shape of the blood vessel connected to the adhesion site. As a premise for performing the separation process, it is assumed that the blood vessel region and the center line are extracted, and the traveling direction and cross-sectional area of the blood vessel are known.

The flow of separation processing is as follows.
(1) The adhesion section is obtained from the change in the cross-sectional area of the blood vessel.
(2) Estimate the center line of the adhesion section using the center line of the blood vessels before and after the adhesion section.
(3) The shape of each blood vessel is estimated by fitting two tangent ellipses in the blood vessel cross section of the adhesion section to the contour of the blood vessel surface.
(4) Divide the cross section of the adhesion section into two.

Hereinafter, the separation processes (1) to (4) will be described in more detail with reference to FIG. First, the cross-sectional area of the blood vessel before and after the adhesion section is measured. Since the two blood vessels are united in the adhesion part, when the cross-sectional area is plotted along the travel of the blood vessel, the cross-sectional area only in the adhesion section as shown in the lower graph of FIG. Will increase. This change is detected and the adhesion interval is determined. Next, paying attention to the center line of the blood vessel, the two center lines originally shifted to the inside, and finally touched to become a branch point. This is indicated by the one-dot chain line in FIG. However, since it is originally two blood vessels, it should be two curves that do not intersect as shown by the dotted line in FIG. Therefore, the original center line of the adhesion section is estimated by interpolating from the front and back center lines. Next, the outline of each blood vessel is estimated on the cross section of the adhesion section using the estimated center line. FIG. 7B shows a cross section of the adhesion section. Two black dots 51 indicate the estimated center line. Assuming that the cross section of the blood vessel is an ellipse, the two ellipses 52 are fitted to the blood vessel contour 53 of the adhesion section under the condition that the centers of the two ellipses 52 are known and are in contact. Two ellipses 52 fitted by dotted lines in FIG. 7B are shown. Finally, the inside of the adhesion section is divided into two regions according to the ratio of the diameters of the two blood vessels, and the separation process is completed.

(Shape measurement unit)
The shape measuring unit 17 measures the shape of each vascular element obtained by the graphing process (step S3). In this embodiment, the following parameters (1) to (5) are measured.
(1) Diameter (information on blood vessel cross section)
(2) Concavity and convexity (information on blood vessel surface shape)
(3) Bending (Information on geometric features of blood vessel center line)
(4) Twist (Information on geometric features of blood vessel center line)
(5) Bifurcation angle (Information on geometric characteristics of blood vessel center line)
Here, “(1) Diameter” is an equivalent diameter of the blood vessel cross-sectional area in the blood vessel centerline traveling direction. As shown in FIG. 8, this parameter (1) is one or more discrete positions (for example, along the blood vessel centerline running direction in each blood vessel element (# 1, 2, 3,...). The shape measurement unit 17 may determine the minimum value, the maximum value, the average value of the measurement values for each vascular element from the measurement values. A fluctuation value (standard deviation or the like) can be calculated. Similarly, the parameters (2) to (4) described in detail below can also calculate the minimum value, maximum value, average value, variation value (standard deviation, etc.), etc., of each measurement value for each blood vessel element. .

“(2) Concavity and convexity” measures the distance (hereinafter referred to as “blood vessel height”) from the position of the center line on the cross section orthogonal to the center line to the position of the blood vessel surface or the outer contour on the cross section of the center line. To calculate. FIG. 9A is a diagram for explaining the blood vessel surface coordinate system, and FIG. 9B is a three-dimensional information mapping the blood vessel height H measured at each position Q in the blood vessel surface coordinate system. An example is shown. In FIG. 9A,

solid lines

41 and 42 extending in the blood vessel running direction indicate examples of line segments in the blood vessel surface running coordinate system. On the horizontal axis of FIG. 9B, the position on the line segment extending in the blood vessel traveling direction is shown as the blood vessel surface traveling distance L. In addition, on the vertical axis in FIG. 9B, the number of each line segment arranged in the circumferential direction of the blood vessel is shown as a line segment number N.

“(3) Bending” and “(4) Twist” are quantified by approximating the blood vessel center line with a spline function. “(5) Branch angle” is an angle formed by the parent blood vessel and the daughter blood vessel in the blood vessel branch portion. FIG. 10 shows an example of a table of shape measurement data measured by the shape measurement unit 17.

In this embodiment, the above (1) to (5) are measured, but the present invention is not limited to this. For example, one or more parameters of the above (1) to (5) may be measured, and shape parameters other than the above parameters may be measured.

(Judgment part)
The determination unit 18 determines the blood vessel shape classification based on the result output by the shape measurement unit 17 (step S4). The judgment algorithm is based on statistical data. The case of “(1) equivalent diameter” will be described as an example. FIG. 11A is statistical data showing a change ratio of the equivalent diameter. As shown in FIG. 11A, the normal range of the change rate of the equivalent diameter is determined by calculating a change value (change rate) based on the average value of the change ratios. Here, the “variation value” is a standard deviation value or a multiple thereof. In the example shown in FIG. 11A, it is shown that the normal range of the variation value of the equivalent diameter is determined to be −20 deg to +20 deg. The reference information (threshold value) determined based on the statistical data is stored in the storage unit of the apparatus, and the determination unit 18 uses the range based on the read statistical data as a reference for the measured shape. The presence or absence of the risk of vascular lesion occurrence for each vascular shape is determined from the parameters. In the example shown in FIG. 11, it is determined that there is a risk of vascular lesion occurrence when the variation rate of the diameter exceeds a threshold value ± 20 deg. FIG. 11B shows the relationship between the position of each blood vessel element in the blood vessel centerline running direction and the measured equivalent diameter. In this embodiment, as shown in FIG. 11 (b), among the measurement results, there is a disease risk in which a region where the variation rate of the diameter is a positive value and exceeds a threshold value (+20 deg or more) is an area where the blood vessel indicates an enlarged type. Is determined. Further, a region where the fluctuation rate is a negative value and exceeds the threshold value (−20 deg or less) is determined to be a disease risk as a region showing a reduced type with strong locality.

In the case of “(2) Concavity and convexity” as well, it is possible to determine that there is a risk of vascular lesion occurrence when the measured concavity and convexity of the blood vessel exceeds the concavity and convexity threshold based on statistical data. In this embodiment, it is determined that there is a risk of stenosis if it is a negative value and there is a risk of aneurysm if it is a positive value. Similarly, “(3) bending”, “(4) twisting”, and “(5) branching angle” can determine the risk of occurrence of vascular lesion from each measured value. The determination unit 18 can determine the final risk of vascular lesion occurrence using all the measurement / determination results of the parameters (1) to (5). Depending on the type, calculation load, etc., the measurement / determination results of some of the parameters (1) to (5) may be used.

(Display section)
After the determination process, the blood vessel shape analyzer outputs a determination result (step 5). The determination result is displayed to the user in a three-dimensional or two-dimensional manner superimposed on the blood vessel shape. FIG. 12 shows an example of a result output by computer processing. A hatched line in FIG. 12 indicates a portion of the determination result that exceeds the threshold value. In addition, data measured by the shape measuring unit 17 may be output.

As described above, this apparatus can determine the risk of occurrence or occurrence of vascular lesions, the type and extent thereof based on the local shape change of the blood vessel shape, and can visually display the result on the user interface.

(Second Embodiment)
In the above-described embodiment, a technique for analyzing a blood vessel shape serving as a hotbed for the onset and growth of a vascular lesion has been described. In the present embodiment, a case will be described in which automatic division of a vascular lesion part and a normal blood vessel part determined in the analysis of the vascular shape is also performed.

FIG. 13 is a diagram for explaining a conventional method for measuring a vascular lesion performed at a medical site, and shows a case of a cerebral aneurysm as an example. As described above, the conventional measurement is performed in a state in which a three-dimensional cerebral aneurysm is viewed two-dimensionally from a certain viewpoint direction. The neck length (n), maximum length (l), height (h), etc. of the aneurysm as shown by the dotted line in FIG. 13 are measured.

However, it is known that vascular lesions tend to occur in places where the bending and twisting of blood vessels are significant, and the technology for automatically extracting abnormally shaped parts even in such complex vascular parts has not yet been industrially applied. There is nothing that leads to accuracy.

Therefore, in the second embodiment of the present invention, a blood vessel fragment is analyzed based on a standard blood vessel and supervised machine learning, thereby automatically detecting and dividing a blood vessel shape abnormality with high accuracy even in a complicated blood vessel site. A normal blood vessel has a characteristic that the blood vessel diameter changes at a certain rate according to the traveling direction of the blood vessel. In the present specification, a blood vessel artificially constructed based on this fixed ratio is referred to as a “standard blood vessel”. An image obtained by subtracting standard blood vessels from a medical image as an input is referred to as a “difference image”. The difference image includes information and noise in which a part of a blood vessel such as a cerebral aneurysm is specifically changed, and the apparatus according to the present embodiment identifies the information and noise as “fragments”. Then, by classifying the fragments by supervised machine learning, the connection points between the morphological features of the fragments and the vascular lesions are constructed, and vascular abnormalities are analyzed by automatically analyzing the fragments based on this correspondence. Is detected and divided.

Hereinafter, a more specific description will be given with reference to FIGS.

FIG. 14 is a schematic configuration diagram of a blood vessel abnormality detection division apparatus according to the second embodiment of the present invention. The input is a medical image, and the output is a blood vessel shape abnormal portion divided and attributed based on the medical image, shape information of the shape abnormal portion, and the shape abnormal portion superimposed on the input medical image, etc. .

As in the first embodiment, the blood vessel abnormality detection / division device 100 has a program storage unit 160 and

data storage units

170 and 171 connected to a bus 150 to which a CPU 111, a RAM 112, and an input / output IF 113 are connected. .

The program storage unit 160 includes an input unit 114, a thinning unit 115, a graphing unit 116, a shape measurement unit 117, a standard blood vessel generation unit 118, a difference image calculation unit 119, a fragment analysis unit 120, a blood vessel abnormality division unit 121, and a display. Part 122 is provided.

The data storage unit 170 stores the input medical image. The data storage unit 171 stores blood vessel shape data 123, blood vessel shape measurement data 124, standard blood vessel shape data 125, fragment shape data 126, fragment shape measurement data 127, and risk determination result 128.

The above configuration (input unit 114, thinning unit 115, graphing unit 116, shape measuring unit 117, standard blood vessel generating unit 118, difference image calculating unit 119, fragment analyzing unit 120, abnormal blood vessel dividing unit 121, and display unit 122) Is configured by computer software stored in the storage area of the hard disk, and is configured to function as each component of the present embodiment by being called by the CPU 111 and expanded and executed on the RAM 112. It has become.

The function of each part is described below together with the processes that each takes charge of. FIG. 15 is a diagram illustrating a blood vessel shape abnormality detection division flow of the apparatus 100 according to the present embodiment.

(Thinning section, graphing section, shape measuring section)
The apparatus 100 performs thinning and graphing similarly to the apparatus 10 of the first embodiment (steps S10 to S12).

(Shape measurement unit)
The shape measuring unit 117 measures the shape of each vascular element obtained by the graphing process in the same manner as the shape measuring unit 17 of the apparatus 10 of the first embodiment (step S13). The shape measuring unit 117 measures the diameter (equivalent diameter) of each blood vessel. That is, as shown in FIG. 8, the shape measuring unit 117 sets a plurality of vertical cross sections at different positions (p1, p2, p3...) In the blood vessel centerline running direction in each blood vessel element, and the cross sectional areas thereof. From these, the diameter (equivalent diameter) of the blood vessel at each position is calculated.

(Standard blood vessel generator)
The standard blood vessel generation unit 118 generates a standard blood vessel based on the blood vessel diameter in the shape measurement unit 117 (step S14). FIG. 16 schematically shows the processing in the standard blood vessel generation unit 118. First, the standard blood vessel generating unit 118 calculates the diameter of each cross section (P1 to P6) of the blood vessel element as shown in FIG. 16A acquired in the previous step (step S13) as shown in FIG. Then, an image mapped with the center line traveling direction as an axis is generated. Next, as shown in FIG. 16C, the standard blood vessel generation unit 118 generates a graph showing the change characteristic of the blood vessel diameter from the mapped image, with the horizontal axis as the center line position and the vertical axis as the equivalent diameter, By applying an approximate curve to the discrete points, the feature quantity of the blood vessel diameter change is converted into a mathematical expression. Thereafter, a new standard blood vessel image is generated so as to satisfy the feature amount. As described above, the standard blood vessel generation unit 118 calculates a feature amount related to a change in blood vessel shape, and artificially generates a blood vessel model as a standard blood vessel based on the feature amount.

(Difference image calculator)
The difference image calculation unit 119 calculates a difference image between the input medical image and the standard blood vessel image obtained by the standard blood vessel generation unit 118 (step S15). FIG. 17 schematically shows the processing in the difference image calculation unit 119. In FIG. 17, a solid line indicates a medical image as an input, and a broken line indicates a standard blood vessel image. The difference image calculation unit 119 outputs two

locations

43 and 44 indicated by hatching in FIG. 17 as a difference image between the medical image and the standard blood vessel image.

In this embodiment, the standard blood vessel generation unit 118 generates a standard blood vessel image by applying an approximate curve with a limited degree of freedom so that the standard blood vessel image does not follow a local shape change of the input medical image. It has become. For example, the approximate curve may correspond to a low-order polynomial approximation.

The difference image calculation unit 119 further has a cut-off filter function, and appropriately removes a portion that can be recognized as extremely small noise in the difference image. For example, the difference image calculation unit 119 sets a threshold for the volume of each fragment, and removes fragments below the threshold.

(Fragment analysis part)
The fragment analysis unit 120 analyzes the difference image output in the previous process (step S15) (step S16). The difference image shows a deviation of each blood vessel from the standard blood vessel. In this specification, this deviation is referred to as a “fragment”. Specifically, as shown in FIG. 18, the fragment analysis unit 120 analyzes the shape of each fragment, and the component constituting each fragment is either (1) a blood vessel abnormality or (2) noise. It is also determined what kind of abnormality the vascular abnormality is. In this embodiment, the vascular abnormality is a vascular morphology abnormality such as stenosis or aneurysm.

The result obtained from supervised machine learning for a large-scale database is used as the determination algorithm for the fragment analysis process. That is, in this supervised machine learning, for example, in the case of a cerebral aneurysm, the computer determines whether each fragment is a cerebral aneurysm or noise, and tags each of the determined cerebral aneurysm and noise. This will build a database on fragment shape information. In this embodiment, the fragment shape information includes fragment volume, surface area, sphericity, representative length, and blood vessel part information where the fragment is located. After that, a multivariate analysis using a single or a plurality of combinations of these shape information is performed to determine a shape parameter that constitutes a significant difference from the noise. Using this shape parameter, the noise and the blood vessel abnormality to be measured are determined. Get identifying geometric features.

The fragment analysis unit 120 analyzes a new fragment shape based on the geometric feature and determines whether or not it is a blood vessel abnormality and what kind of abnormality is. In machine learning, the determination result of a blood vessel abnormality is improved by creating a database and analyzing the determination result.

In the analysis process, for example, in the initial stage of database creation, a brain surgeon or the like discriminates whether a cerebral aneurysm or noise, and the fragment analysis unit 120 analyzes the discrimination result to obtain the geometric feature. You may do it.

(Abnormal blood vessel division)
The abnormal blood vessel division unit 121 divides the abnormal region into regions for the blood vessel abnormality determined by the fragment analysis unit 120 (step S17). Since the area determination has already been made from the difference image, the area of the abnormal part is divided by tagging the attribute determination. Specifically, those determined as an aneurysm are tagged as an aneurysm, those determined as a stenosis are tagged as stenosis, and those determined as a noise are tagged as noise.

The abnormal blood vessel division unit 121 searches the surroundings of a target divided into regions as an abnormal part (lesioned part) with the surface of the target as a starting point, and calculates the feature amount of the corresponding part in a high dimension, thereby improving the division accuracy. May be raised. More specifically, abnormal blood vessels such as cerebral aneurysms are accompanied by inversion of curvature from the parent blood vessel, and the abnormal blood vessel division unit 121 increases the abnormal blood vessel portion by tracking the inversion of curvature. Can be divided with accuracy. That is, in the case of a cerebral aneurysm, the parent blood vessel that is a normal blood vessel has a convex curvature, while the aneurysm portion has a concave curvature, and the abnormal blood vessel division unit 121 tracks this boundary surface. Thus, high-precision area division is performed. The result of the division is labeled.

(Display section)
As shown in FIG. 19, the display unit 122 superimposes and displays the input medical image (step S18). The thick solid line in FIG. 19 is a neckline specified with high accuracy by high-dimensional processing. Further, the shape information of the abnormal part measured by the shape measuring unit 117 may be displayed together with the superimposed image.
In addition, the configuration of each part of the apparatus in the present invention is not limited to the illustrated configuration example, and various modifications are possible as long as substantially the same operation is achieved.

Claims

An acquisition unit for acquiring a medical image;
Extracting a blood vessel region included in at least a part of the medical image, and obtaining a blood vessel center line in the blood vessel region;
A dividing unit that divides the blood vessel region into blood vessel elements based on the blood vessel center line;
A shape measurement unit that measures shape of each of the blood vessel elements and generates shape measurement data, the shape measurement data including information on a blood vessel cross section at one or more positions of each blood vessel element, blood vessel surface shape The shape measuring unit includes at least one of information related to information and information related to a geometric characteristic of a blood vessel center line.
The blood vessel shape analysis device according to claim 1, wherein the device further comprises:
A determination unit that determines a risk of developing a lesion for each blood vessel shape based on whether or not the shape measurement data exceeds a threshold value, wherein the threshold value is set in advance based on statistical data regarding the geometric shape information of the blood vessel; The determination unit,
And an output unit for outputting the determination result.
3. The blood vessel shape analysis apparatus according to claim 2, wherein the output unit displays the determination result together with the blood vessel region.
2. The blood vessel shape analysis apparatus according to claim 1, wherein the shape measuring unit includes a diameter, an unevenness, a bending degree, and a twisting degree of a blood vessel cross section at one or more positions of each blood vessel element, and between blood vessel elements. An apparatus for calculating at least one of branch angles in a blood vessel branch portion.
The blood vessel shape analysis apparatus according to claim 1, wherein the shape measurement unit measures a blood vessel cross-sectional area along a traveling direction of the blood vessel center line and calculates an equivalent diameter thereof.
This device further
A standard blood vessel generating unit for constructing a standard blood vessel model based on the equivalent diameter;
A difference image calculation unit for calculating a difference image between the medical image and the standard blood vessel model and identifying each fragment included in the difference image;
Fragment analysis that measures the shape of each fragment, determines the attribute of the fragment from the measured shape of the fragment based on fragment geometric feature information obtained by supervised machine learning, and outputs the determination result The fragment analyzer includes at least a fragment analyzer that determines whether a fragment is a blood vessel abnormality portion or noise.
6. The blood vessel shape analyzing apparatus according to claim 5, further comprising a blood vessel abnormality dividing unit that divides and displays a blood vessel abnormal part based on a determination result of the attribute of the fragment.
6. The blood vessel shape analyzing apparatus according to claim 5, wherein the standard blood vessel generating unit generates blood vessel diameter change characteristic data based on an equivalent diameter calculated at a plurality of discrete positions in the blood vessel element, and the blood vessel An apparatus for constructing the standard blood vessel model by fitting an approximate curve to diameter change characteristic data.
6. The blood vessel shape analysis apparatus according to claim 5, wherein the fragment analysis unit discriminates a type of blood vessel lesion as an attribute of the fragment.
6. The blood vessel shape analysis apparatus according to claim 5, wherein the supervised machine learning updates the fragment geometric feature information by creating a database and analyzing the determination result.
An acquisition process in which a computer acquires a medical image;
A thinning step in which a computer extracts a blood vessel region included in at least a part of the medical image and obtains a blood vessel center line in the blood vessel region;
A dividing step in which the computer divides the blood vessel region into blood vessel elements based on the blood vessel center line;
A shape measurement step in which a computer measures a three-dimensional shape of each of the blood vessel elements to generate shape measurement data, the shape measurement data is information on a blood vessel cross section at one or more positions of each blood vessel element; A blood vessel shape analysis method, comprising: the shape measurement step including at least one of information on a blood vessel surface shape and information on a geometric characteristic of a blood vessel center line.
The blood vessel shape analysis method according to claim 10, wherein the method further comprises:
A determination step in which a computer determines a risk of developing a lesion for each blood vessel shape based on whether or not the shape measurement data exceeds a threshold value, the threshold value based on statistical data related to the geometric shape information of the blood vessel; The determination step, which is preset, and
A computer comprising: an output step of outputting the determination result.
12. The blood vessel shape analysis method according to claim 11, wherein the output step displays the determination result together with the blood vessel region.
The blood vessel shape analysis method according to claim 10, wherein the shape measurement step includes a diameter, an unevenness, a bending degree, and a twisting degree of a blood vessel cross section at one or more positions of each blood vessel element, and between blood vessel elements. A method for calculating at least one of branch angles in a blood vessel branch portion.
The blood vessel shape analysis method according to claim 10, wherein the shape measuring step measures a blood vessel cross-sectional area along a traveling direction of the blood vessel center line and calculates an equivalent diameter thereof.
This method further
A standard blood vessel generating step in which a computer constructs a standard blood vessel model based on the equivalent diameter;
A difference image calculation step in which a computer calculates a difference image between the medical image and the standard blood vessel model and specifies each fragment included in the difference image;
The computer measures the shape of each fragment, determines the attribute of the fragment from the measured fragment shape based on fragment geometric feature information obtained by supervised machine learning, and outputs the determination result A fragment analysis step, wherein the fragment analysis step comprises at least a fragment analysis step for discriminating whether a fragment is a blood vessel abnormality or noise.
15. The blood vessel shape analysis method according to claim 14, further comprising: a blood vessel abnormality dividing step in which the computer divides and displays a blood vessel abnormal portion based on a result of discrimination of the attribute of the fragment. ,Method.
15. The blood vessel shape analysis method according to claim 14, wherein the standard blood vessel generation step generates blood vessel diameter change characteristic data based on an equivalent diameter calculated at a plurality of discrete positions in the blood vessel element, and the blood vessel A method for constructing the standard blood vessel model by fitting an approximated curve to diameter change characteristic data.
15. The blood vessel shape analysis method according to claim 14, wherein the fragment analysis step discriminates a type of blood vessel lesion as an attribute of the fragment.
15. The blood vessel shape analysis method according to claim 14, wherein the supervised machine learning updates the fragment geometric feature information by creating a database and analyzing the determination result.
A computer software program for analyzing a blood vessel shape comprising the following steps:
An acquisition process in which a computer acquires a medical image;
A thinning step in which a computer extracts a blood vessel region included in at least a part of the medical image and obtains a blood vessel center line in the blood vessel region;
A dividing step in which the computer divides the blood vessel region into blood vessel elements based on the blood vessel center line;
A shape measurement step in which a computer measures a three-dimensional shape of each of the blood vessel elements to generate shape measurement data, the shape measurement data is information on a blood vessel cross section at one or more positions of each blood vessel element; A computer software program comprising instructions for executing the shape measuring step, wherein the computer program includes at least one of information on a blood vessel surface shape and information on a geometric characteristic of a blood vessel center line.
The computer software program of claim 19, further comprising:
A determination step in which a computer determines a risk of developing a lesion for each blood vessel shape based on whether or not the shape measurement data exceeds a threshold value, the threshold value based on statistical data related to the geometric shape information of the blood vessel; The determination step, which is preset, and
A computer software program comprising: an instruction for causing a computer to execute an output step of outputting the determination result.
21. The computer software program according to claim 20, wherein the output step displays the determination result together with the blood vessel region.
20. The computer software program according to claim 19, wherein the shape measuring step includes a blood vessel cross-sectional diameter, an unevenness degree, a bending degree, a twisting degree, and a blood vessel between the blood vessel elements at one or more positions of each blood vessel element. A computer software program for calculating at least one of branch angles in a branch portion.
The computer software program according to claim 19, wherein the shape measuring step measures a blood vessel cross-sectional area along a running direction of the blood vessel center line and calculates an equivalent diameter thereof.
This computer software program is further
A standard blood vessel generating step in which a computer constructs a standard blood vessel model based on the equivalent diameter;
A difference image calculation step in which a computer calculates a difference image between the medical image and the standard blood vessel model and specifies each fragment included in the difference image;
The computer measures the shape of each fragment, determines the attribute of the fragment from the measured fragment shape based on fragment geometric feature information obtained by supervised machine learning, and outputs the determination result A fragment analysis step, wherein the fragment analysis step includes instructions for executing at least a fragment analysis step for determining whether a fragment is a blood vessel abnormality or noise. Software program.
24. The computer software program according to claim 23, further comprising an instruction for executing a blood vessel abnormality dividing step of dividing and displaying a blood vessel abnormal part based on a determination result of the attribute of the fragment. Computer software program.
24. The computer software program according to claim 23, wherein the standard blood vessel generation step generates blood vessel diameter change characteristic data based on an equivalent diameter calculated at a plurality of discrete positions in the blood vessel element, and the blood vessel diameter A computer software program for constructing the standard blood vessel model by fitting an approximated curve to the change characteristic data of
24. The computer software program according to claim 23, wherein the fragment analysis step discriminates a type of vascular lesion as an attribute of the fragment.
24. The computer software program according to claim 23, wherein the supervised machine learning updates the fragment geometric feature information by creating a database and analyzing the determination result.