KR101991037B1

KR101991037B1 - Method, Device and Program for determining clot type

Info

Publication number: KR101991037B1
Application number: KR1020170122880A
Authority: KR
Inventors: 방오영; 김윤철; 정종원; 서우근
Original assignee: 사회복지법인 삼성생명공익재단
Priority date: 2017-09-22
Filing date: 2017-09-22
Publication date: 2019-06-19
Also published as: KR20190034022A; WO2019059448A1

Abstract

A thrombus type discriminating method according to an embodiment of the present invention includes: a thrombus region selecting step of selecting, from magnetic resonance imaging (MRI) data of a specific subject, a thrombus region estimated to have a clot of the specific subject; An input data acquiring step of generating input data to be input to the discrimination model for estimating the type of thrombus from the selected thrombus area; An input step of inputting the input data into a discrimination model for estimating the type of the thrombus; And an output step of discriminating and outputting the type of the thrombus on the basis of the output value of the discrimination model, wherein the discrimination model comprises correlation data of learning input data obtained from magnetic resonance imaging data of a plurality of data providers, Output function.

Description

[0001] The present invention relates to a method, device, and program for determining a clot type,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method, an apparatus, and a program for determining the type of a clot using magnetic resonance image (MRI) data.

Cerebral infarction is a disease in which blood vessels in the brain block and some cells in the brain die. Cerebral infarction occurs in the heart due to thrombosis and atrial fibrillation caused by arteriosclerosis of the cerebral blood vessels, and the cerebral blood vessels are occluded by the cardioembolic clots that migrate to the brain. The blood supply to the brain tissue is blocked and the blood supply to the brain tissue is blocked by the blocked blood vessels, resulting in neurological symptoms such as consciousness disorder or paralysis.

In the case of cerebral infarction, the blood vessels are resumed early and the cerebral blood flow is restored before the cerebral tissue is completely inflated. How quickly the occluded vessels are pierced is the most important factor affecting the prognosis of cerebral infarction. The time to first treatment (Golden Time) is known to be about 4.5-8 hours to minimize the aftereffects of cerebral infarction. The faster the time of revascularization after symptom onset, the greater the probability of recovering neurological symptoms. Diagnosis and treatment are required.

When a patient arrives in the emergency room, a CT scan or magnetic resonance imaging (MRI) is performed to determine the cause and cause of the cerebral infarction and determine the treatment plan. At this time, chemical thrombolysis using a thrombolytic agent for dissolving a thrombus or mechanical thrombolysis for disrupting thrombus directly is performed.

It is known that the thrombus caused by atrial fibrillation is more easily removed by mechanical thrombectomy than the case where the artery is occluded by hardening of the artery. Therefore, if the type of thrombus can be determined objectively, it is very helpful to determine the treatment strategy of mechanical thrombectomy.

Recently, studies have been actively conducted to investigate the correlation between thrombus components and subtype of cerebral infarction through histological examination of thrombus obtained through mechanical thrombectomy.

Kajo van der Marel et al., "Quantitative assessment of device-clot interaction for stent retriever thrombectomy ", J NeuroIntervent Surg. 2016; 0: 1-6 (release date 2016.02.01)

When stroke patients are hospitalized with acute cerebral infarction, it is common to perform etiopathological work ups. However, since the standardized test items are not specified, various tests are carried out according to the clinician or the institution, which is considerable cost. On the other hand, since it takes several days to perform the test to check the characteristics of the thrombus in a cerebral infarction patient, obtaining information on the thrombus in the acute phase is limited. In other words, it takes a considerable amount of time to grasp the components of thrombus through histology of thrombus, so it is difficult to use it as a tool for setting treatment policy in an emergency situation. Therefore, in actuality, diagnosis / prognosis prediction has been made through the clinical judgment of the specialist who has obtained CT or MRI. However, these types of blood clot types are not qualitative and objective, and intra- and inter-observer variability is high.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method, apparatus, and program for determining the type of thrombus in an objectively quantitative manner using magnetic resonance imaging data, in order to solve various problems including the above problems. However, these problems are illustrative and do not limit the scope of the present invention.

A thrombus type discriminating apparatus according to an embodiment of the present invention receives magnetic resonance image data of a specific subject and, when an area in which the clot of the specific subject is located is selected from the magnetic resonance image data, A control unit for generating input data to be input to a discrimination model for estimating the type of the thrombus and inputting the input data to the discrimination model; And an output unit for outputting the type of the thrombus output from the discrimination model, wherein the discrimination model includes an input / output function for expressing a correlation between learning input data and thrombus classification data obtained from magnetic resonance imaging data of a plurality of data providers to be.

One embodiment of the present invention discloses a computer program stored in a medium for executing the above-described clot type determination method.

Other aspects, features, and advantages will become apparent from the following drawings, claims, and detailed description of the invention.

According to the method, device, and program for determining the type of thrombus using machine learning according to an embodiment of the present invention, it is possible to determine the type of thrombus in an objectively quantitative manner. When the thrombus type is determined by a qualitative method, And inter- and inter-observer variability are high. In addition, non-invasive methods can quickly identify the type of thrombus in minutes, which can help determine the treatment plan of the subject quickly. This can improve the therapeutic effect by rapidly treating the cerebral vascular occlusion symptoms that may irreversibly damage the brain such as cerebral infarction. Of course, the scope of the present invention is not limited by these effects.

FIG. 1 is a view schematically showing the configuration of a thrombus type discriminating apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating a method of determining a thrombus type according to an embodiment of the present invention.
3 is a flowchart showing a method of discriminating a thrombus type further including a step of training a discrimination model.
4 is a diagram showing a method of generating learning data.
FIG. 5 is a diagram showing an embodiment for obtaining learning input data.
6 is a diagram showing another embodiment for obtaining learning input data.
FIGS. 7 and 8 are diagrams showing respective steps of the thrombus region selection step. FIG.
9 to 11 are diagrams illustrating an embodiment of the input data acquisition step, respectively.
12 is a diagram illustrating an input step and an output step.
FIG. 13 is a diagram showing magnetic resonance imaging data and signal intensity profiles of a subject having a thrombus caused by atrial fibrillation and a thrombus not caused by atrial fibrillation, respectively.
FIG. 14 is a graph showing a result of a blood clot type discrimination using a clot type discriminating apparatus 1000 according to an embodiment of the present invention.
FIG. 15 is a receiver operating characteristic (ROC) graph of a discrimination model according to an embodiment of the present invention.
FIG. 16 is a diagram illustrating a signal intensity profile in a section of a magnetic resonance image and a thrombus region before and after a thrombus of a specific subject is removed.
17 is a block diagram of a control unit according to an embodiment.
18 is a block diagram of a data learning unit according to an embodiment.
19 is a block diagram of a data recognition unit according to an embodiment.

The present invention is capable of various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in the detailed description. The effects and features of the present invention and methods of achieving them will be apparent with reference to the embodiments described in detail below with reference to the drawings. However, the present invention is not limited to the embodiments described below, but may be implemented in various forms.

Some embodiments of the present disclosure may be represented by functional block configurations and various processing steps. Some or all of these functional blocks may be implemented with various numbers of hardware and / or software configurations that perform particular functions. For example, the functional blocks of the present disclosure may be implemented by one or more microprocessors, or by circuit configurations for a given function. Also, for example, the functional blocks of the present disclosure may be implemented in various programming or scripting languages. The functional blocks may be implemented with algorithms running on one or more processors. In addition, the present disclosure may employ conventional techniques for electronic configuration, signal processing, and / or data processing, and the like. Terms such as "mechanism "," element ", "means ", and" configuration "and the like are widely used and are not limited to mechanical and physical configurations.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only the case of being "directly connected" but also the case of being "electrically connected" with another part in between. Also, when an element is referred to as "comprising ", it means that it can include other elements as well, without departing from the other elements unless specifically stated otherwise.

Also, the connection lines or connection members between the components shown in the figures are merely illustrative of functional connections and / or physical or circuit connections. In practical devices, connections between components can be represented by various functional connections, physical connections, or circuit connections that can be replaced or added.

Also, as used herein, terms including ordinals such as "first" or "second" can be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

If certain embodiments are otherwise feasible, the specific steps may be performed differently from the described order. For example, two successively described steps may be performed substantially concurrently, or may be performed in a reverse order to that described.

The term "magnetic resonance image (MRI)" used herein means a diagnostic technique for generating an image or a photograph of a body structure using a magnetic field, and an image obtained through the diagnosis technique.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings, wherein like reference numerals refer to like or corresponding components throughout the drawings, and a duplicate description thereof will be omitted .

FIG. 1 is a view schematically showing a configuration of a thrombus type discriminating apparatus 1000 according to an embodiment of the present invention.

The thrombus type discriminating apparatus 1000 shown in FIG. 1 shows only the components related to the present embodiment in order to prevent the characteristic of the present embodiment from being blurred. Therefore, it will be understood by those skilled in the art that other general-purpose components other than the components shown in FIG. 1 may be further included.

The thrombus type discriminating apparatus 1000 according to one embodiment may correspond to at least one processor or may include at least one or more processors. Accordingly, the thrombus type discriminating apparatus 1000 can be driven in the form contained in another hardware apparatus such as a microprocessor or a general-purpose computer system.

The thrombus type discriminating apparatus 1000 according to an embodiment is a device that processes magnetic resonance image data of a subject and processes it to discriminate the type of clot generated in the subject's body, for example, the brain. Referring to FIG. 1, a thrombus type discriminating apparatus 1000 includes a control unit 1100 and an output unit 1200.

The control unit 1100 may include any kind of device capable of processing data, such as a processor. Herein, the term " processor " may refer to a data processing apparatus embedded in hardware, for example, having a circuit physically structured to perform a function represented by a code or an instruction contained in the program. As an example of the data processing device embedded in the hardware, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC) ), FPGA (Field Programmable Gate Array), and the like, but the scope of the present invention is not limited thereto.

First, the control unit 1100 receives medical data of a subject, for example, magnetic resonance imaging (MRI) data. The subject may be a patient suffering from cerebral infarction due to clot, but the subject is not limited thereto. MRI image data may include T1-weighted image data, T2-weighted image data, fluid attenuated inversion recovery (FLAIR) image data, diffusion weighted imaging (DWI) Data, and susceptibility weighted imaging (SWI) data, but the type of magnetic resonance imaging data is not limited thereto. The thrombus type discriminating apparatus 1000 can receive magnetic resonance imaging data from an external magnetic resonance imaging apparatus or a previously stored medical database (DB), but the present invention is not limited thereto.

According to one embodiment, the magnetic resonance image data acquired by the control unit 1100 may include gradient echo (GRE) image data. GRE images are a type of T2 * -weighted image, which can be taken in a short time because the 180-degree pulse is not used and the repetition time can be shortened. Since the blood flow is a high signal, diagnosis of the blood vessel, MR angiography ). When a cardioembolic clot due to atherosclerosis is seen through a GRE image, thrombosis is clearly observed due to the signal characteristic that the surrounding area is black due to susceptibility. Therefore, GRE image data can be useful for learning the signal characteristics of thrombus. In addition to GRE image data, various types of data that can identify blooming artifacts caused by susceptibility differences can be used. For example, susceptibility weighted imaging (SWI) data can also be used to learn the signal characteristics of the thrombus.

Thereafter, the control unit 1100 may process the acquired magnetic resonance image data, obtain input data for estimating the type of thrombus, and input the input data to the discrimination model. The 'discriminant model' may refer to an input / output function expressing the correlation between the input data for learning and the thrombus type obtained from the magnetic resonance image data of a plurality of data providers.

The discrimination model according to one embodiment may be generated by a machine learning technique. The thrombus type discriminating apparatus 1000 can use various techniques for training and / or learning such a discrimination model. For example, the thrombus type discriminating apparatus 1000 may include a logic regression, a decision tree, a nearest-neighbor classifier, a kernel discriminant analysis, a neural network, a support vector machine, a random vector forest, Boosted tree, and / or the like. However, the present invention is not limited thereto.

The discrimination model according to an embodiment may be a matrix data set that maps training input data to thrombus classification data that can be expressed in a vector form. Here, the "training" or "learning" of the discrimination model means that an error back-propagation technique is used so that the output value when inputting learning input data approaches the labeled thrombus classification data Or the like, to adjust the value of the discriminant model parameter set.

According to one embodiment, the discrimination model may be a model for determining whether a thrombus of a specific subject includes a red clot. The thrombus is divided into red thrombosis and white thrombosis. Among them, red thrombosis appears red with a lot of red blood cells attached to fibrin. Red blood cells contain a lot of deoxyhemoglobin, In contrast, Blooming artifacts appear more strongly in GRE images. Therefore, the signal characteristics of red and white blood clots are different from each other. In one embodiment of the present invention, the type of the thrombus can be determined by learning the signal characteristic of the red thrombus, and the cause of the thrombus can be estimated.

According to one embodiment, the discrimination model may be a model for determining whether a thrombus of a specific subject is thrombotic due to atrial fibrillation (Afib). Atrial fibrillation is a kind of arrhythmia, which means that the cardiac output decreases because the atrial contraction does not occur. When the atrial fibrillation occurs, the blood does not circulate smoothly but stagnates and blood clots are produced. The thrombus then migrates and blocks the middle cerebral artery of the brain, resulting in cerebral infarction. At this time, thrombus caused by atrial fibrillation mainly includes red blood clots having a high proportion of red blood cells, and thrombosis caused by atherosclerosis mainly includes platelet-rich white clots. In other words, if the subject's thrombus is identified as red thrombosis, it can be assumed that the subject's thrombus is more likely to be caused by atrial fibrillation. In one embodiment of the present invention, the signal characteristic of the thrombus caused by atrial fibrillation is learned based on the fact that the signal characteristics of the red thrombus and the white thrombus are different from each other as described above.

The output unit 1200 displays and outputs the information processed by the blood clot type discriminating apparatus 1000. The output unit 1200 may include a display unit for displaying a blood clot type determination result and a user interface for receiving magnetic resonance image data. The user can determine the treatment method of the subject by referring to the blood clot discrimination result displayed on the output unit 1200.

2 is a flowchart illustrating a method of determining a thrombus type according to an embodiment of the present invention. The thrombus type discriminating method according to an embodiment includes a thrombus region selection step S210, an input data acquisition step S220, an input step S230, and an output step S240.

Referring to FIG. 2, a step (S210) of selecting a thrombus region estimated to have a thrombus in the magnetic resonance imaging data of the subject is performed. The step S210 may include selecting a slice in which a thrombus is visible from a plurality of magnetic resonance imaging slices and selecting a region estimated to be thrombus from the slice, which will be described below with reference to FIGS. 7 and 8 .

Thereafter, the control unit 1100 performs step S220 of obtaining input data to be input to the discrimination model from the thrombus area. In step S220, a process of extracting, generating, or pre-processing input data to be input to the discriminant model among the magnetic resonance image data may be performed. At this time, the input data may be a signal intensity profile, which will be described below with reference to Figs. 9 to 11.

Thereafter, a step of inputting the input data to the discrimination model (S230) and an outputting step S240 of discriminating the type of thrombus based on the output value of the discrimination model and outputting the result are performed. This will be described below with reference to Figs. 11 and 12. Fig.

In other words, in the method of determining the type of thrombus according to an embodiment of the present invention, magnetic resonance imaging data is processed to obtain input data, and then a value derived from an input to a discrimination model that is preliminarily learned / .

At this time, the discrimination model may be learned / generated in an external device, but may be learned / generated in the control unit 1100 of the thrombus type discriminating apparatus 1000 according to an embodiment of the present invention. When the discrimination model is trained in the control unit 1100, the discrimination model must be trained prior to discrimination of the blood type of the specific subject.

3 is a flowchart showing a method of discriminating a thrombus type further including a step of training a discrimination model. The thrombus type discriminating method according to an embodiment may further include a step S310 of training a discrimination model before the thrombus region selecting step S210. At this time, the discrimination model training step (S310) may include learning data from the magnetic resonance imaging data and the thrombus classification data of a plurality of data providers, and machine learning the discrimination model using the training data.

At this time, learning input data is required to machine the discrimination model.

Particularly, in order to machine the discrimination model through a supervised learning method, thrombus type data matching with each learning input data and indicating the actual thrombus type of the data provider is also needed. A method for acquiring learning data for learning / creating a discrimination model will be described below with reference to Figs. 4 to 6. Fig.

4 is a diagram showing a method of generating learning data. The learning data may include 'learning input data' input to the discrimination model and 'thrombus classification data' corresponding thereto.

4, there is shown a step of generating learning input data for learning a discrimination model for estimating the type of thrombus from the thrombus region selected for the magnetic resonance imaging data of a plurality of data providers. The training input data to train the discrimination model may be a graph of the signal intensity profile of the subject's thrombus area. At this time, the signal intensity profile may indicate intensity of the magnetic resonance signal according to the pixel position or intensity of the brightness signal in the image. Learning input data can be generated automatically or semi-automatically, which will be described later with reference to Figs. 7 to 10.

At this time, the learning input data including the signal intensity profile obtained from the magnetic resonance imaging data of each data provider and the thrombus classification data are matched with each other. For example, if the thrombus of the data provider P1 is known to be atrial fibrillation thrombus due to a histology result or the like, the control unit 1100 may map the signal intensity profile graph in the thrombus region of the data provider P1 to 1 . Conversely, if it is already known that the thrombus of the data provider P2 is not atrial fibrillation thrombus as a result of the biopsy, the control unit 1100 may map the signal intensity profile graph in the thrombus region of the data provider P2 to zero. In other words, the control unit 1100 can generate 'training data' by matching the thrombus classification data obtained from the medical data and the clinical data of the data provider with the input data for learning.

At this time, the controller 1100 can learn the characteristics of the signal intensity profile graph from the learning data through the various algorithms described above. For example, in the discriminant model training step (S310), the controller 1100 may learn the characteristics of the signal intensity profile graph labeled as atrial fibrillation.

FIG. 5 is a diagram showing an embodiment for obtaining learning input data.

According to one embodiment, the learning input data is stored at a predetermined angle with a line segment L ₁₂ connecting the first point (P ₁ ) and the second point (P ₂ ) in the thrombus region of the magnetic resonance imaging data of a plurality of data providers May comprise a signal strength profile (IP) obtained from a plurality of sections (S ₁ , S ₂ , S _n ).

5, a first point (P ₁ ), a second point (P ₂ ), a first point (P ₁ ), and a second point (P ₁ ) in the thrombi area CA shown in the magnetic resonance imaging data of the data provider ₂₎ a connecting line (L _12), said segments (L ₁₂₎ and a plurality of section lines having a predetermined angle (S _1, S _2,, there is a marked S _n). The first point P ₁ and the second point P ₂ may be designated by the user or automatically selected by the control unit 1100. For example, the controller 1100 may select two points that are the farthest from the selected thrombus area CA as the first point P ₁ and the second point P ₂ . According to one embodiment, the first point P ₁ and the second point P ₂ selected by the controller 1100 may be changed to another point by the user. At this time, it may be preferable to select the first point P ₁ and the second point P ₂ so that the line segment L ₁₂ between each point is parallel to the direction of the blood vessel or the direction in which the blood clot is elongated.

The plurality of sectional lines S ₁ , S ₂ , and S _n have a certain angle θ with the line segment L ₁₂ connecting the first point P ₁ and the second point P ₂ . The angle? May be a predetermined value or may be set by a user's input. The angle [theta] may be, for example, 90 degrees. That is, the sectional lines S ₁ , S ₂ , ..., and S _n may be perpendicular to the direction of the blood vessel or the direction in which the blood clot is elongated. When the cross-sectional lines S ₁ , S ₂ , ..., S _n are perpendicular to the direction of the blood vessel, the characteristics of the signal strength of the thrombus in relation to the cross-sectional area of the blood vessel, that is, the distance from the center of the blood vessel, .

Since the thrombus caused by arteriosclerosis is formed by solidification of one wall of the blood vessel and the thrombus caused by atrial fibrillation is formed by the vortex formation in the atrium due to irregular heartbeat, The characteristics are different. Therefore, when the cross-sectional lines S ₁ , S ₂ , ..., S _n are perpendicular to the direction of the blood vessel or the elongated direction of the thrombus to obtain a signal intensity profile and mechanically learn it, .

At this time, since a plurality of learning input data or signal strength profiles (IP) are obtained from one data provider, all of them match with the same thrombus classification data. For example, when there is n learning input data obtained from a data provider having an atrial fibrillation thrombus, the thrombus classification data corresponding to n learning input data can be all 1s. That is, if a plurality of signal intensity profiles (IP) are obtained from one data provider, a plurality of signal intensity profiles (IP) labeled with the same value can be obtained, and the input data for learning can be easily generated. In particular, when each section is separated by one pixel, it is possible to generate as many learning input data as possible from one data provider.

6 is a diagram showing another embodiment for obtaining learning input data.

According to one embodiment, the learning input data may further include a signal strength profile (IP) and an inverse signal strength profile (IP ') which is inversion.

Referring to FIG. 6, a total of 2n graphs including n signal strength profiles (IP) and n inverted signal strength profiles (IP ') obtained from one data provider are shown similarly to FIG. Since the mechanism by which the thrombus is generated does not depend on the left and right positions of the blood vessel, the 'inverted signal strength profile (IP') ', which is a graph inversion of each signal strength profile (IP) And the like. Therefore, the inverse signal intensity profile (IP ') can also be used as learning input data. In this case, when the number of signal strength profile (IP) graphs is n, the number of learning input data that can be obtained in one thrombus region is 2n. That is, the total signal strength profile can be doubled, and as many input data for learning as possible can be generated from one data provider. In one embodiment of the present invention, a total of 1624 learning input data are generated from magnetic resonance imaging data obtained from 52 data providers through the above-described method.

In FIGS. 4 to 6, the thrombus classification data is binary classified as 0 or 1, but the present invention is not limited thereto. For example, the discrimination model may be learned to discriminate three or more thrombus types, and in this case, the thrombus classification data may be a vector in which the number of rows or columns is three or more.

Meanwhile, the learning input data generated by the above-described method can be divided into training data (test data) and test data (test data). For example, after the discrimination model is learned using the training data, the accuracy of the discrimination model can be verified through the test data.

In the above, a method of training the discriminant model and a method of generating learning data for learning the discriminant model have been described with reference to FIGS. Hereinafter, a method of determining the type of blood clots of a specific subject using the discriminant model trained with reference to FIGS. 7 to 12 will be described.

FIGS. 7 and 8 illustrate respective steps of the thrombus region selection step S210.

Referring to FIG. 7, a number of slices of magnetic resonance imaging data are shown. That is, the magnetic resonance image data may include a plurality of slices. In FIG. 7, it is illustrated that each slide is a magnetic resonance image appearing on the transverse plane of the subject, but the present invention is not limited thereto. An image of each slide may be displayed on the output unit 1200. At this time, in the selecting step S210, a step of selecting a slide estimated to have 'clots' among a plurality of slides may be performed. Such selection may be made by the user, but may be automatically performed through image processing by the control unit 1100. [

Referring to FIG. 8, a step of selecting a thrombus region CA in which a thrombus is located may be performed on the selected slice. For example, in oblique echo images in magnetic resonance imaging, the thrombus shows a signal characteristic that appears black due to the difference in susceptibility. Therefore, a region that is darker than the periphery may be selected as the thrombus region CA.

The step of selecting the thrombus region CA may be manually performed by the user, but may be performed semi-auto or auto through image processing techniques. For example, when the user selects some of the regions estimated to be thrombus, the controller 1100 semi-automatically selects the pixels whose brightness or signal intensity is lower than or equal to a certain threshold value around the selected pixels through the image processing technique . Alternatively, if the thrombus is formed only on one side of the body with respect to the sagittal plane, the control unit 1100 determines whether the difference in signal intensity between the specific pixel region and the symmetric pixel region is equal to or less than the symmetry axis of the magnetic resonance image passing through the sagittal plane It is possible to automatically select a thrombus area CA by estimating a pixel area having a specific value or more as an area where a 'thrombus' has occurred. At this time, if there are a plurality of 'thrombus areas CA' estimated by the control part 1100, the user can finally select some of them.

The above-described steps can be similarly applied not only to acquiring input data from magnetic resonance imaging data of a specific subject, but also to acquiring learning input data from magnetic resonance imaging data of a plurality of data providers.

9 to 11 are diagrams showing an embodiment of the input data acquiring step (S220).

In the input data acquiring step S220, the control unit 1100 acquires input data to be input to the discrimination model from the selected thrombus area CA. For example, the input data to be input to the discrimination model may be the signal intensity profile of the thrombus area CA. At this time, the signal intensity profile may indicate intensity of the magnetic resonance signal according to the pixel position or intensity of the brightness signal in the image.

In an embodiment of the present invention, a graph of a signal intensity profile at a section line S of the thrombus region is input to the discrimination model to estimate the type of thrombus. In other words, referring to FIG. 9, the controller 1100 may input a signal intensity profile obtained through an arbitrary section line S passing through the thrombus region selected in the thrombus region selection step S210 to the discrimination model. In this case, the process of obtaining the signal intensity profile can be performed by auto.

Alternatively, the process of obtaining the signal strength profile may be semi-auto. According to one embodiment, the input data is a sectional line S having a certain angle? With a line segment L ₁₂ connecting a first point P ₁ and a second point P ₂ in the thrombus region CA, Lt; RTI ID = 0.0 > a < / RTI >

Referring to FIG. 10, a first point P ₁ and a second point P ₂ are designated in the thrombus region CA. The first point P ₁ and the second point P ₂ may be designated by the user or automatically by the control unit 1100. For example, the control unit 1100 may select two points that are the farthest from the selected thrombus area CA as the first point P ₁ and the second point P ₂ , respectively. According to one embodiment, the first point P ₁ and the second point P ₂ selected by the controller 1100 may be changed to another point by the user. At this time, it may be preferable to select the first point P ₁ and the second point P ₂ so that the line segment following each point is parallel to the direction of the blood vessel or the direction in which the thrombus is elongated.

Thereafter, a line segment L ₁₂ connecting the first point P ₁ and the second point P ₂ and a section line S having a certain angle θ can be obtained. At this time, the section line S may pass an arbitrary position (for example, center point) of the line segment L ₁₂ connecting the first point P ₁ and the second point P ₂ . The section line S has a constant angle? With a line connecting the first point P ₁ and the second point P ₂ . The angle? May be a predetermined value or may be set by a user's input. The angle [theta] may be, for example, 90 degrees. That is, the section line S may be perpendicular to the direction of the blood vessel or the direction in which the blood clot is elongated. When the cross-sectional line S is perpendicular to the direction of the blood vessel, the signal intensity of the thrombus with respect to the cross section of the blood vessel, that is, the characteristic of the signal intensity of the thrombus along the distance from the center of the blood vessel, can be grasped.

According to one embodiment, the input data may comprise a plurality of signal strength profiles. 11, the specified first point (P ₁₎ and the second point (P ₂₎ a connecting line (L ₁₂₎ and a plurality of cross-section having a predetermined angle _{(θ) (S 1, S} 2,, S 5 A plurality of signal strength profiles may be obtained. In FIG. 11, five signal intensity profiles are obtained from five cross sections having the same interval, but the present invention is not limited thereto.

When a plurality of signal intensity profiles are acquired from one subject, each signal intensity profile can be input to the discrimination model. The discrimination model can output an output value for each input. Then, the controller 1100 can determine the type of blood clot based on the average of the output values, and can transmit the determined result to the output unit 1200. [ When a plurality of signal intensity profiles are used for a subject, erroneous discrimination of the type of thrombus by an outlier among the input data can be prevented, and the accuracy of discrimination can be improved.

12 is a diagram illustrating an input step and an output step. After the input data acquisition step (S220), an input step (S230) for inputting the input data into the learned discrimination model and an output step (S240) for discriminating and outputting the type of thrombus based on the final output value of the discrimination model . The final output value may mean the average value described in FIG. 11, but the present invention is not limited thereto.

Referring to FIG. 12, the discrimination model may be a binary classification model that is learned to determine whether a thrombus of a subject is a specific type of thrombus (for example, a thrombus due to atrial fibrillation). In this case, the discrimination model can receive a signal intensity profile and output a scalar value. At this time, when the output scalar value is equal to or greater than a specific value (T, threshold), the control unit 1100 determines whether the thrombus of the subject is a specific type of thrombus A (non-A) . At this time, the specific value T may be set to a value that maximizes the prediction accuracy of the discrimination model in the step of learning the discrimination model.

When the output step S240 of outputting the type of thrombus by the output unit 1200 is completed, the user can determine the treatment plan of the subject based on the output thrombus type discrimination result.

< MATLAB Used Example >

Hereinafter, the constitution and effects of the present invention will be described in more detail with reference to examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the present invention.

In one embodiment of the present invention, a program for determining the type of thrombus using MATLAB (Mathworks, MA, USA) has been developed. A random forest model was used for machine learning. The parameters of the random forest model are set to the number of trees = 11, the depth of the tree = 15, and the k-fold cross-validation k = 10, respectively.

In one embodiment of the present invention, a total of 1624 training input data were extracted from 52 data providers of cerebral infarction patients and external validation of the discrimination model was performed through data obtained from 15 data providers.

FIG. 13 is a diagram showing magnetic resonance imaging data and signal intensity profiles of a subject having a thrombus caused by atrial fibrillation and a thrombus not caused by atrial fibrillation, respectively. The subject of FIG. 13 (a) had thrombosis due to atrial fibrillation, and the subject of (b) had thrombosis due to other causes. In an embodiment of the present invention, five signal intensity profiles obtained from five cross-sectional lines passing through the thrombus region are input to the discriminant model to discriminate the type of thrombus of the subject.

At this time, in the signal intensity profile graph, it can be confirmed that the signal intensity is large and the small region is repeated twice in the middle region (region around pixels 10-20) corresponding to the thrombus region in the case of the atrial fibrillation thrombus (a). On the other hand, in the case of thrombosis (b) due to other causes, only one peak is found in the middle region. Therefore, for example, the discrimination model can 'learn' whether the signal intensity profile has two peaks in the thrombus area, and this characteristic can be used as a criterion for determining whether thrombus is caused by atrial fibrillation. On the other hand, the discriminant model can 'learn' other features of the signal intensity profile of atrial fibrillation as well as the above.

FIG. 14 is a graph showing a result of a blood clot type discrimination using a clot type discriminating apparatus 1000 according to an embodiment of the present invention. In one experimental example of the present invention, the signal intensity profile graphs of five sections perpendicular to the line segment passing through the first point and the second point in the thrombus area designated by the user are input to the discrimination model, and the five output values are averaged, &Lt; / RTI > At this time, if the final value is greater than a specific value (T: 0.27), it is set to be 'thrombus due to atrial fibrillation.' Referring to FIG. 14, the average value of the subjects confirmed to be thrombus due to atrial fibrillation was large, and the average value of the subjects not confirmed to be atrial fibrillation was small.

FIG. 15 is a receiver operating characteristic (ROC) graph of a discrimination model according to an embodiment of the present invention. In one embodiment of the present invention, the area under the curve (AUC) value of the ROC graph was obtained through external verification data (n = 15) to confirm the performance of the discriminant model. At this time, the first point and the second point designated by the user are selected twice differently. In the first test (a), the area under the ROC graph, the AUC, was 0.93, and in the second test (b), the AUC was 0.97. Both AUC values were higher in the two assays, confirming the superiority of the method and apparatus for determining thrombus type.

FIG. 16 is a diagram illustrating a signal intensity profile in a section of a magnetic resonance image and a thrombus region before and after a thrombus of a specific subject is removed. (a), before the intraarterial thrombolysis (IAT) is performed on a specific subject who has had cerebral infarction due to thrombosis due to atrial fibrillation, the signal of the middle region of the signal intensity profile (pixels 5 to 13) Showed a pattern in which two peaks appeared. (b), the signal intensity profile of the thrombus after performing the IAT can be confirmed to have a reduced signal as a whole without such a pattern. In other words, it can be seen that the thrombus is well removed by the IAT. As described above, the thrombus type discrimination method of the present invention can also be used to predict the treatment outcome and prognosis after treatment.

Hereinafter, an embodiment of the control unit 1100 will be described in detail.

17 is a block diagram of the controller 1100 according to an embodiment.

Referring to FIG. 17, the controller 1100 may include a data learning unit 1110 and a data recognizer 1120.

The data learning unit 1110 can learn a criterion for discriminating the blood clot type. In one embodiment, the data learning unit 1110 can learn a predetermined criterion by using the input data for learning input to the data learning unit 1110. The data learning unit 1110 can learn how the shape of the signal intensity profile differs for each type of thrombus.

The data recognition unit 1120 can determine the type of blood clot based on the data. The data recognition unit 1120 can determine the type of thrombus from the predetermined data using the learned data recognition model. The data recognizing unit 1120 can determine the type of thrombus based on predetermined data by acquiring predetermined data according to a predetermined reference by learning and using the obtained data as an input value and using a data recognition model have. Further, the resultant value output by the data recognition model with the obtained data as an input value can be used to update the data recognition model.

At least one of the data learning unit 1110 and the data recognizing unit 1120 may be manufactured in the form of at least one hardware chip and mounted on the electronic device. For example, at least one of the data learning unit 1110 and the data recognition unit 1120 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or may be a conventional general-purpose processor Or application processor) or a graphics-only processor (e.g., a GPU), and may be mounted on various electronic devices as described above.

In this case, the data learning unit 1110 and the data recognizing unit 1120 may be mounted on one electronic device or on separate electronic devices, respectively. For example, one of the data learning unit 1110 and the data recognizing unit 1120 may be included in the electronic device, and the other may be included in the server. The data learning unit 1110 and the data recognizing unit 1120 may provide the model information constructed by the data learning unit 1110 to the data recognizing unit 1120 via the wired or wireless network, 1120 may be provided to the data learning unit 1110 as additional learning data.

At least one of the data learning unit 1110 and the data recognizing unit 1120 may be implemented as a software module. When at least one of the data learning unit 1110 and the data recognition unit 1120 is implemented as a software module (or a program module including an instruction), the software module may be a computer-readable, And may be stored in non-transitory computer readable media. Also, in this case, the at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an Operating System (OS), and some of the software modules may be provided by a predetermined application.

18 is a block diagram of a data learning unit 1110 according to an embodiment.

18, a data learning unit 1110 according to some embodiments includes a data obtaining unit 1111, a preprocessing unit 1112, a learning data selecting unit 1113, a model learning unit 1114, 1115).

The data acquisition unit 1111 can receive magnetic resonance image data from an external magnetic resonance imaging apparatus or a previously stored medical database (DB).

The preprocessing unit 1112 can preprocess the acquired data so that the acquired data can be used for learning / training to determine the type of thrombus. The preprocessing unit 1112 can process the acquired data into a predetermined format so that the model learning unit 1114, which will be described later, can use data acquired for learning to determine the type of blood clots. For example, the preprocessing unit 1112 may extract / generate learning input data and input data from the input magnetic resonance image, and normalize the signal intensity profile graph.

The learning data selection unit 1113 can select data necessary for learning from the preprocessed data. The selected data may be provided to the model learning unit 1114. The learning data selection unit 1113 can select data necessary for learning from among the preprocessed data according to a predetermined criterion. The learning data selection unit 1113 can also select data according to a predetermined reference by learning by the model learning unit 1114, which will be described later.

The model learning unit 1114 is a component that can correspond to the above-described " discrimination model ", and can learn a criterion on how to determine the type of blood clot based on the learning data. In addition, the model learning unit 1114 may learn a criterion about which learning data should be used to determine the type of blood clots.

Further, the model learning unit 1114 can learn the data recognition model used for discriminating the type of blood clots by using learning data. In this case, the data recognition model may be a pre-built model.

The data recognition model can be constructed considering the application field of the recognition model, the purpose of learning, or the computer performance of the device. The data recognition model may be, for example, a model based on a neural network. For example, models such as DNN (Deep Neural Network), CNN (Convolutional Neural Network), RNN (Recurrent Neural Network) and BRDNN (Bidirectional Recurrent Deep Neural Network) may be used as a data recognition model, no.

According to various embodiments, when there are a plurality of data recognition models built in advance, the model learning unit 1114 can determine a data recognition model in which the input learning data and the basic learning data are highly relevant, have. In this case, the basic learning data may be pre-classified according to the type of data, and the data recognition model may be pre-built for each data type. For example, the basic learning data may be pre-classified by various criteria such as an area where the learning data is generated, a time at which the learning data is generated, a size of the learning data, a genre of the learning data, a creator of the learning data, .

Also, the model learning unit 1114 can learn a data recognition model using, for example, a learning algorithm including an error back-propagation method or a gradient descent method.

Further, the model learning unit 1114 can learn the data recognition model through supervised learning using the learning data as an input value, for example. In addition, the model learning unit 1114 learns, for example, the type of data necessary for discriminating the type of thrombus without any further guidance, thereby learning unsupervised learning that finds a criterion for discriminating the type of thrombus ), The data recognition model can be learned. Further, the model learning unit 1114 can learn the data recognition model through reinforcement learning using, for example, feedback as to whether the result of the blood type determination based on learning is correct.

Further, when the data recognition model is learned, the model learning unit 1114 can store the learned data recognition model. In this case, the model learning unit 1114 can store the learned data recognition model in the memory of the electronic device including the data recognition unit 1120. [ Alternatively, the model learning unit 1114 may store the learned data recognition model in a memory of the electronic device including the data recognition unit 1120 to be described later. Alternatively, the model learning unit 1114 may store the learned data recognition model in the memory of the server connected to the electronic device and the wired or wireless network.

In this case, the memory in which the learned data recognition model is stored may also store, for example, instructions or data associated with at least one other component of the electronic device. The memory may also store software and / or programs.

The model evaluation unit 1115 inputs the evaluation data to the data recognition model, and can cause the model learning unit 1114 to learn again when the recognition result output from the evaluation data does not satisfy the predetermined criterion. In this case, the evaluation data may be predetermined data for evaluating the data recognition model.

For example, if the number or the ratio of the evaluation data whose recognition result is not correct out of the recognition results of the learned data recognition model for the evaluation data exceeds a predetermined threshold value, the model evaluation unit 1115 does not satisfy the predetermined criterion It can be evaluated as not successful. For example, when a predetermined criterion is defined as a ratio of 2%, and the learned data recognition model outputs an incorrect recognition result for evaluation data exceeding 20 out of a total of 1000 evaluation data, It is possible to evaluate that the data recognition model is not suitable.

On the other hand, when there are a plurality of learned data recognition models, the model evaluation unit 1115 evaluates whether each of the learned data recognition models satisfies a predetermined criterion, and if the model satisfying the predetermined criterion is a final data recognition model You can decide. In this case, when there are a plurality of models satisfying a predetermined criterion, the model evaluating unit 1115 can determine any one or a predetermined number of models previously set in descending order of the evaluation score as a final data recognition model.

At least one of the data acquisition unit 1111, the preprocessing unit 1112, the learning data selection unit 1113, the model learning unit 1114, and the model evaluation unit 1115 in the data learning unit 1110 includes at least one And can be mounted on an electronic device. For example, at least one of the data acquisition unit 1111, the preprocessor 1112, the learning data selection unit 1113, the model learning unit 1114, and the model evaluation unit 1115 may be an artificial intelligence (AI) Or may be implemented as part of a conventional general-purpose processor (e.g., a CPU or an application processor) or a graphics-only processor (e.g., a GPU) and mounted on the various electronic devices described above.

The data acquisition unit 1111, the preprocessor 1112, the learning data selection unit 1113, the model learning unit 1114, and the model evaluation unit 1115 may be mounted on one electronic device, Electronic devices, respectively. For example, some of the data acquisition unit 1111, the preprocessor 1112, the learning data selection unit 1113, the model learning unit 1114, and the model evaluation unit 1115 are included in the electronic device, May be included in the server.

At least one of the data acquisition unit 1111, the preprocessing unit 1112, the learning data selection unit 1113, the model learning unit 1114, and the model evaluation unit 1115 may be implemented as a software module. At least one of the data acquisition unit 1111, the preprocessor 1112, the learning data selection unit 1113, the model learning unit 1114, and the model evaluation unit 1115 is a software module Program modules), the software modules may be stored in a computer-readable, readable non-transitory computer readable media. Also, in this case, the at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an Operating System (OS), and some of the software modules may be provided by a predetermined application.

FIG. 19 is a block diagram of a data recognition unit 1120 according to an embodiment.

19, a data recognition unit 1120 according to some embodiments includes a data acquisition unit 1121, a preprocessing unit 1122, a recognition data selection unit 1123, a recognition result providing unit 1124, 1125 < / RTI >

The data acquisition unit 1121 may acquire data necessary for determining the type of blood clot, and the preprocessing unit 1122 may preprocess acquired data so that the acquired data can be used. The preprocessing unit 1122 can process the acquired data into a predetermined format so that the recognition result providing unit 1124, which will be described later, can use the acquired data for predicting the type of blood clot.

The recognition data selection unit 1123 can select data necessary for determining the type of blood clot from among the preprocessed data. The selected data may be provided to the recognition result providing unit 1124. The recognition data selection unit 1123 can select some or all of the preprocessed data according to a predetermined criterion for determining the type of blood clot. The recognition data selection unit 1123 can also select data according to a predetermined criterion by learning by the model learning unit 1114, which will be described later.

The recognition result providing unit 1124 can apply the selected data to the data recognition model to predict the type of thrombus. The recognition result providing unit 1124 can provide the recognition result according to the data recognition purpose. The recognition result providing unit 1124 can apply the selected data to the data recognition model by using the data selected by the recognition data selecting unit 1123 as an input value. In addition, the recognition result can be determined by the data recognition model.

The model updating unit 1125 can update the data recognition model based on the evaluation of the recognition result provided by the recognition result providing unit 1124. [ For example, the model updating unit 1125 may provide the model learning unit 1114 with the recognition result provided by the recognition result providing unit 1124 so that the model learning unit 1114 can update the data recognition model have.

At least one of the data acquiring unit 1121, the preprocessing unit 1122, the recognition data selection unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 in the data recognizing unit 1120, It can be manufactured in the form of one hardware chip and mounted on the electronic device. For example, at least one of the data acquisition unit 1121, the preprocessor 1122, the recognition data selection unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 may be an artificial intelligence Or may be mounted on a variety of electronic devices as described above and manufactured as part of a conventional general purpose processor (e.g., a CPU or an application processor) or a graphics dedicated processor (e.g., a GPU).

The data acquisition unit 1121, the preprocessing unit 1122, the recognition data selection unit 1123, the recognition result providing unit 1124 and the model updating unit 1125 may be mounted on one electronic device, Lt; RTI ID = 0.0 > electronic devices, respectively. For example, some of the data acquisition unit 1121, the preprocessing unit 1122, the recognition data selection unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 are included in the electronic device, May be included in the server.

At least one of the data acquisition unit 1121, the preprocessing unit 1122, the recognition data selection unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 may be implemented as a software module. At least one of the data acquisition unit 1121, the preprocessing unit 1122, the recognition data selection unit 1123, the recognition result providing unit 1124 and the model updating unit 1125 includes a software module (or an instruction) , The software module may be stored in a computer-readable, non-transitory computer readable medium. Also, in this case, the at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an Operating System (OS), and some of the software modules may be provided by a predetermined application.

Meanwhile, the thrombus type discriminating method according to an embodiment of the present invention can be implemented in a general-purpose digital computer that can be created as a program that can be executed in a computer and operates the program using a computer-readable recording medium. The computer-readable recording medium includes a storage medium such as a magnetic storage medium (e.g., ROM, floppy disk, hard disk, etc.), optical reading medium (e.g., CD ROM,

According to the method, device, and program for determining the type of thrombus using machine learning according to an embodiment of the present invention, it is possible to determine the type of thrombus in an objectively quantitative manner. When the thrombus type is determined by a qualitative method, And inter- and inter-observer variability are high. In addition, non-invasive methods can quickly identify the type of thrombus in minutes, which can help determine the treatment plan of the subject quickly. In this way, cerebral vascular occlusion symptoms, which may irreversibly damage the brain, such as cerebral infarction, can be treated promptly and the probability of reopening can be increased to increase the therapeutic effect. In addition to this, selective screening can be carried out to reduce the burden of medical expenses.

While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

1000: Thrombus type discrimination device 1100:
1200: output part CA: thrombus area
IP: Signal strength profile IP ': Inverse signal strength profile

Claims

A method for identifying a thrombus type performed by a thrombus type discriminating device,
A thrombus region selection step of selecting, in magnetic resonance imaging (MRI) data of a specific subject, a thrombus region estimated to have a clot of the specific subject;
An input data acquiring step of generating input data to be input to the discrimination model for estimating the type of thrombus from the selected thrombus area;
An input step of inputting the input data into a discrimination model for estimating the type of the thrombus; And
And an output step of discriminating and outputting the type of the blood clot based on the output value of the discrimination model,
In the discrimination model,
Output function that expresses correlation between learning input data and thrombus types obtained from magnetic resonance imaging data of a plurality of data providers.

The method according to claim 1,
Wherein the magnetic resonance imaging data includes gradient echo (GRE) image data.

The method according to claim 1,
Wherein the magnetic resonance image data includes susceptibility weighted imaging (SWI) data.

The method according to claim 1,
In the discrimination model,
And determining whether the thrombus of the specific subject includes a red clot.

The method according to claim 1,
In the discrimination model,
And determining whether the thrombus of the specific subject is thrombotic due to atrial fibrillation (Afib).

The method according to claim 1,
Before the thrombus area selection step,
Further comprising the step of training the discrimination model.

The method according to claim 1,
Wherein the learning input data includes:
And a signal intensity profile obtained from a plurality of sections having a certain angle with a line segment connecting a first point and a second point in the thrombus region of the magnetic resonance imaging data of the plurality of data providers.

8. The method of claim 7,
Wherein the learning input data includes:
Further comprising a signal strength profile and an inverse signal strength profile which is inversion.

The method according to claim 1,
Wherein the input data includes:
And a signal intensity profile obtained from a section having a certain angle with a line segment connecting the first point and the second point in the thrombus region.

10. The method of claim 9,
Wherein the input data comprises a plurality of signal intensity profiles.

11. The method of claim 10,
Wherein the output step of discriminating and outputting the type of the thrombus comprises:
Determining a type of the thrombus based on a value obtained by averaging each output value of the discrimination model when each of the plurality of signal intensity profiles is input, and outputting the determined thrombus type.

An input to be input to a discrimination model for estimating the type of the thrombus from the selected region when the region of the magnetic resonance imaging data where the clot of the specific subject is located is received, A control unit for generating the data and inputting the input data to the discrimination model;
And an output unit for outputting the type of the thrombus output by the discrimination model,
In the discrimination model,
Output function that expresses a correlation between learning input data and thrombus classification data obtained from magnetic resonance imaging data of a plurality of data providers.

12. A computer program stored on a medium for carrying out the method of any one of claims 1 to 11 using a computer.