Artificial intelligence-based apparatus and method for predicting large vascular occlusion

Abstract

Disclosed are an artificial intelligence-based method and apparatus for predicting large vascular occlusion. The method includes the steps of: inputting a brain scan image acquired by a photographing device; masking the brain scan image with a brain area mask, calculating a characteristic value for each brain area, and calculating a brain lesion characteristic value on the basis of whether a brain lesion area can be divided; and inputting, to a predictive artificial intelligence, the characteristic value for each brain area and the brain lesion characteristic value to output a degree of certainty for large vascular occlusion.

Description

AI-based large vessel occlusion prediction device and method

[0001] The present disclosure relates to an artificial intelligence-based large vessel occlusion prediction device and method, and more particularly, to an artificial intelligence-based large vessel occlusion prediction device and method that extracts various characteristics from an image captured without a contrast agent and predicts large vessel occlusion based on the extracted characteristics.

[0002]

[0003] The artificial intelligence model that predicted large vessel occlusion from NCCT (Non-contrast Computed Tomography), a CT image taken without a conventional contrast agent, had a tendency to be dependent on the clot sign, and thus had limitations in demonstrating the same performance across races.

[0004] Specifically, blood clots are prevalent in Caucasians, but are relatively rare in other ethnicities. Consequently, existing AI models, which tend to be dependent on blood clots, have the problem of performing relatively well only in Caucasians.

[0005] Moreover, existing artificial intelligence models that predict large vessel occlusions have difficulty being applied in environments where accurate clinical data cannot be secured.

[0006]

[0007] The technical problem of the present disclosure is to provide an artificial intelligence-based large vessel occlusion prediction device and method that extracts various characteristics from an image captured without a contrast agent and predicts large vessel occlusion based on the extracted characteristics.

[0008] The technical problems to be achieved in the present disclosure are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present disclosure belongs from the description below.

[0009]

[0010] According to one aspect of the present disclosure, a method for predicting large vessel occlusion based on artificial intelligence is disclosed. The method may include the steps of: inputting a brain scan image acquired by a photographing device; masking the brain scan image with a brain region mask to derive characteristic values ​​for each brain region; and deriving brain lesion characteristic values ​​based on whether or not a brain lesion region can be segmented; and inputting the characteristic values ​​for each brain region and the brain lesion characteristic values ​​into a prediction artificial intelligence to output a certainty for large vessel occlusion.

[0011] According to another aspect of the present disclosure, an artificial intelligence-based large vessel occlusion prediction device is disclosed. The device includes a memory storing at least one instruction and a processor executing the at least one instruction stored in the memory based on data acquired from the memory, wherein the processor inputs the brain scan image acquired by the photographing device, masks the brain region mask onto the brain scan image to calculate the characteristic value for each brain region, and calculates the brain lesion characteristic value based on whether the brain lesion region is segmented, and inputs the characteristic value for each brain region and the brain lesion characteristic value into the prediction artificial intelligence to output the certainty for the large vessel occlusion.

[0012] According to one aspect of the present disclosure, the brain scan image may be a computed tomography image including at least one image taken in a single layer.

[0013] According to one aspect of the present disclosure, the brain region mask may be a reference image in which a brain region is defined by merging at least one specific region of the brain that separates the brain.

[0014] According to one aspect of the present disclosure, the masking may be performed based on a transformation matrix obtained through a process of registering the input brain scan image to common coordinates.

[0015] According to one aspect of the present disclosure, the step of calculating the characteristic value for each brain region may include the step of filtering pixels of a composite image generated by masking the brain region mask on the brain scan image based on density, and the step of calculating the characteristic value for each brain region of the filtered composite image.

[0016] According to one aspect of the present disclosure, the filtering may remove pixels exceeding a threshold value from the center value of the density of the pixels in each brain region of the synthetic image.

[0017] According to one aspect of the present disclosure, the brain region-specific characteristic value can be calculated based on the brain region-specific difference included in the filtered synthetic image.

[0018] According to one aspect of the present disclosure, the characteristic value for each brain region can be calculated based on the statistical results of the difference in the density of the pixels for each brain region of the filtered composite image, the ratio of the average value, the ratio of the standard deviation, and the ratio of the volume for each brain region.

[0019] According to one aspect of the present disclosure, the step of calculating the brain lesion characteristic value may include the step of selecting a slice of the brain scan image including a cerebral artery region among the input brain scan images, and the step of calculating the brain lesion characteristic value from the slice.

[0020] According to one aspect of the present disclosure, the brain lesion characteristic value can be defined depending on whether the brain lesion area of ​​the slice is segmented.

[0021] According to one aspect of the present disclosure, the brain lesion area may mean an area where a blood clot blocking a blood vessel is located.

[0022] The features briefly summarized above regarding the present disclosure are merely exemplary aspects of the detailed description of the present disclosure that follows and do not limit the scope of the present disclosure.

[0023]

[0024]

[0025] According to the present disclosure, an artificial intelligence-based large vessel occlusion prediction device and method can be provided that extracts various characteristics from an image captured without a contrast agent and predicts large vessel occlusion based on the extracted characteristics.

[0026] In addition, according to the present disclosure, the dependence of artificial intelligence on blood clots can be reduced by utilizing various characteristics that can be obtained from brain scan images such as NCCT, as well as blood clots.

[0027] In addition, according to the present disclosure, an artificial intelligence model capable of predicting large blood vessel occlusion with a certain performance or higher can be provided even in an environment where accurate clinical data cannot be secured, taking into account the characteristics of each brain region.

[0028] Additionally, according to the present disclosure, the dependence of an artificial intelligence model on blood clots can be reduced, thereby reducing performance variations according to race.

[0029] In addition, according to the present disclosure, non-brain regions can be more accurately removed by merging brain regions in brain scan images such as NCCT and filtering pixels of brain regions.

[0030] Additionally, according to the present disclosure, by using a filtered brain scan image, the possibility of information loss according to volume or false positives can be reduced.

[0031] The technical effects to be achieved in the present disclosure are not limited to the technical effects mentioned above, and other technical effects not mentioned will be clearly understood by a person having ordinary skill in the technical field to which the present disclosure pertains from the description below.

[0032]

[0033] Figure 1 is a diagram illustrating a server receiving a brain scan image and extracting target data.

[0034] Figure 2 is a configuration diagram of the modules that make up the server.

[0035] Figure 3 is a schematic diagram of the overall process for predicting large vessel occlusion.

[0036] Figure 4 is a flowchart illustrating the process of predicting large vessel occlusion.

[0037] Figure 5 is a flowchart illustrating the process of calculating characteristic values ​​for each brain region.

[0038] Figure 6 is a diagram illustrating a filtered brain scan image.

[0039] Figure 7 is a flowchart illustrating the process of calculating brain lesion characteristic values.

[0040] Figure 8 is a flowchart of a predictive artificial intelligence model that predicts large vessel occlusion based on brain region-specific characteristic values ​​and brain lesion characteristic values.

[0041]

[0042] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present disclosure. However, the present disclosure may be implemented in various different forms and is not limited to the embodiments described herein.

[0043] In describing embodiments of the present disclosure, detailed descriptions of known configurations or functions will be omitted if they are deemed to obscure the gist of the present disclosure. Furthermore, portions unrelated to the description of the present disclosure in the drawings have been omitted, and similar portions have been designated with similar reference numerals.

[0044] In the present disclosure, when a component is said to be "connected," "coupled," or "connected" to another component, this may include not only a direct connection but also an indirect connection in which another component exists in between. Furthermore, when a component is said to "include" or "have" another component, unless otherwise specifically stated, this does not exclude the other component, but rather implies that the other component may be included.

[0045] In this disclosure, terms such as first, second, etc. are used only for the purpose of distinguishing one component from another, and do not limit the order or importance of components, unless specifically stated otherwise. Accordingly, within the scope of this disclosure, a first component in one embodiment may be referred to as a second component in another embodiment, and similarly, a second component in one embodiment may be referred to as a first component in another embodiment.

[0046] In this disclosure, distinct components are used to clearly illustrate their respective characteristics, and do not necessarily imply that the components are separated. That is, multiple components may be integrated into a single hardware or software unit, or a single component may be distributed into multiple hardware or software units. Therefore, even if not specifically mentioned, such integrated or distributed embodiments are also included within the scope of this disclosure.

[0047] In this disclosure, each of the phrases "A or B", "at least one of A and B", "at least one of A or B", "A, B or C", "at least one of A, B and C", and "at least one of A, B, C or combination thereof" may include any one of the items listed together in the corresponding phrase, or all possible combinations thereof.

[0048] In the present disclosure, the components described in various embodiments are not necessarily essential components, and some may be optional components. Therefore, embodiments comprising a subset of the components described in one embodiment are also within the scope of the present disclosure. Furthermore, embodiments including other components in addition to the components described in various embodiments are also within the scope of the present disclosure.

[0049] The advantages and features of the present invention, and the methods for achieving them, will become clearer with reference to the embodiments described in detail below, along with the accompanying drawings. However, the present invention is not limited to the embodiments presented below and may be implemented in various different forms. These embodiments are provided solely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the invention of the scope of the invention.

[0050] Hereinafter, with reference to the attached drawings, an artificial intelligence-based large vessel occlusion prediction device and method for extracting various characteristics from images captured without contrast agents and predicting large vessel occlusion based on the extracted characteristics will be described.

[0051] Figure 1 is a diagram illustrating a server receiving a brain scan image and extracting target data.

[0052] Referring to FIG. 1, the brain scan image may be a computed tomography image captured without using a contrast agent, fluorescent agent, or contrast medium. For example, it may be a Non-Contrast Computed Tomography (NCCT). The brain scan image may be an image captured by dividing the brain into at least one or more single-layer sections, but is not limited thereto, and all brain scan images may be included in the present disclosure.

[0053] The server (100) can receive brain scan images and extract target data. The target data may refer to a value predicting large vessel occlusion. For example, the target data may output a confidence score for large vessel occlusion. The confidence score may include information regarding the probability of a positive or negative result based on the large vessel occlusion prediction result. Furthermore, the target data is not limited thereto and may include any information including the degree of prediction for large vessel occlusion.

[0054] The server (100) receives a brain scan image acquired by a photographing device to extract certainty, masks a brain region mask onto the brain scan image to calculate characteristic values ​​for each brain region, and calculates brain lesion characteristic values ​​based on whether or not the brain lesion region is segmented. Subsequently, the server (100) inputs the characteristic values ​​for each brain region and the brain lesion characteristic values ​​into the predictive artificial intelligence to output certainty for large vessel occlusion. This will be described in detail later with reference to FIGS. 3 to 7.

[0055] Figure 2 is a configuration diagram of the modules that make up the server.

[0056] Referring to FIG. 2, the server (100) may be composed of a communication unit (105), a processor (110), a display (115), and a memory (120). Each component is not an essential component, and additional components may be provided or omitted, and one component may be included in or combined with another component so that a single component may perform multiple functions.

[0057] The communication unit (105) can receive a brain scan image captured by a photographing device. As another example, if the server (100) additionally includes a photographing device (not shown) capable of scanning the brain to directly generate a brain scan image, the communication unit (105) can transmit target data extracted from the brain scan image to another device. In the present disclosure, the server (100) receives a brain scan image stored in a photographing device or another server and extracts target data for large vessel occlusion.

[0058] The processor (110) masks the input brain scan image with a brain region mask to calculate characteristic values ​​for each brain region. The masking may be performed based on a transformation matrix obtained through a process of registering the brain scan image to a common coordinate system. Registration may refer to a process of aligning the image to a common coordinate system. This will be described later with reference to FIG. 5 .

[0059] The processor (110) can segment a brain lesion region from an input brain scan image and derive brain lesion characteristic values. Specifically, the processor (110) can define brain lesion characteristic values ​​based on whether or not the brain lesion region is segmented for each slice of at least one single-layer brain scan image. This will be described later with reference to FIG. 7.

[0060] Next, the processor (110) can output target data for large vessel occlusion using predictive artificial intelligence that inputs characteristic values ​​for each brain region and characteristic values ​​for brain lesions.

[0061] The display (115) can visualize target data output from the processor (110) and provide it to the user, and the memory (120) can be a non-removable memory or a removable memory and can store data required to process the above-described process of the processor (110) and output data.

[0062] Next, the process of extracting target data for large vessel occlusion is described through Fig. 3.

[0063] Figure 3 is a schematic diagram of the overall process for predicting large vessel occlusion.

[0064] The process of predicting large vessel occlusion can be divided into a feature extraction stage and a prediction stage.

[0065] Specifically, the feature extraction process may include a clot sign prediction process (Clot Sign Prediction) that detects a brain lesion region, such as a clot (Clot Sign), to extract a brain lesion feature value (A 3.1 in Figure 3). In addition, the feature extraction process may include a registration process of a brain scan image and a comparison process of the left and right hemispheres of the brain (Left / Right) to extract feature values ​​for each brain region (A 3.2 in Figure 3).

[0066] Specifically, the characteristic values ​​for each brain region can be calculated based on the ratio of pixel density or volume for each brain region, which will be described later. According to one embodiment of the present disclosure, the processor (110) can compare not only the brain regions of the left and right hemispheres, but also, according to another embodiment, specific regions of the brain to calculate the characteristic values ​​for each brain region. For example, the processor (110) may calculate a characteristic value for each brain region based on the ratio of pixel density or volume for each brain region that is a specific region of the brain included in the brain scan image, such as the insula, caudate, putamen, internal capsule, anterior middle cerebral artery cortex (hereinafter, M1), lateral middle cerebral artery cortex (MCA Cortex Lateral to Insular Ribbon, hereinafter, M2), posterior middle cerebral artery cortex (hereinafter, M3), lateral middle cerebral artery immediately superior to M2 (hereinafter, M4), lateral middle cerebral artery immediately superior to M2 (hereinafter, M5), posterior middle cerebral artery immediately superior to M3 (hereinafter, M6), or a combination of the above-described regions.

[0067] The specific brain region described above may refer to the Alberta Stroke Program Early CT Score (ASPECTS) Region, a scoring system used to measure stroke severity by evaluating specific areas of the brain on the initial CT scan of stroke patients.

[0068] The processor (110) can use a prediction artificial intelligence model in the prediction process, and can output target data by inputting brain region characteristic values ​​and brain lesion characteristic values ​​derived from a brain scan image.

[0069] Hereinafter, a process of inputting brain region-specific characteristic values ​​and brain lesion characteristic values ​​into a predictive artificial intelligence and outputting target data according to one embodiment of the present disclosure will be described.

[0070] Figure 4 is a flowchart illustrating the process of predicting large vessel occlusion.

[0071] Referring to FIG. 4, the server (100) receives a brain scan image via the communication unit (105) (S410). According to one embodiment of the present disclosure, the brain scan image may be a computed tomography image obtained by dividing the brain into at least one single layer without using a contrast agent, fluorescent agent, or contrast medium. For example, the brain scan image may be a non-contrast computed tomography (NCCT).

[0072] Next, the processor (110) can calculate characteristic values ​​for each brain region and brain lesion characteristic values ​​based on the input brain scan image (S420). According to one embodiment of the present disclosure, the processor (110) can calculate characteristic values ​​for each brain region based on differences in each brain region of the left and right hemispheres. In addition, the processor (110) can detect a brain lesion region, for example, a clot sign, to calculate the brain lesion characteristic values. The process of calculating characteristic values ​​for each brain region and brain lesion characteristic values ​​will be described with reference to FIGS. 5 to 7.

[0073] Next, the processor (110) can input the characteristic values ​​for each brain region and the characteristic values ​​for brain lesions into the predictive artificial intelligence to output target data (S430).

[0074] In one embodiment, the target data may represent a value predicting a large vessel occlusion. For example, a confidence score for a large vessel occlusion may be output. The confidence score may include information regarding the probability of a positive or negative result based on the large vessel occlusion prediction result.

[0075] Hereinafter, a process for extracting characteristic values ​​for each brain region according to one embodiment of the present disclosure will be described with reference to FIG. 5.

[0076] Figure 5 is a flowchart illustrating the process of calculating characteristic values ​​for each brain region.

[0077] First, the processor (110) registers the input brain scan image to common coordinates (S510). Specifically, the processor (110) may use a template image, which is a reference image that serves as a standard for registering to common coordinates. For example, the processor (110) projects at least one brain scan image, which is an image taken in a single layer, onto the template image and aligns the projected brain scan image to common coordinates.

[0078] The processor (110) can extract a transformation matrix during the registration process. The transformation matrix can be obtained during the process of projecting a brain scan image onto a template image and aligning it to a common coordinate system, and specifically, can refer to a matrix that can commonly transform a brain scan image to geometrically match the template image.

[0079] Next, the processor (110) can apply the transformation matrix obtained through registration to the brain region mask to generate a brain region mask that is geometrically consistent with the brain scan image aligned to common coordinates (S520).

[0080] A brain region mask may refer to a reference image in which a brain region is defined by merging at least one specific region of the brain, and the specific region of the brain may refer to an ASPECTS Region, which is a region of a scoring system used to measure the severity of stroke, according to one embodiment. Accordingly, the brain region mask may be a reference image in which a total of four regions are defined, including an insula region among the ASPECTS Regions, a striatocapsular region merging the cerebral nuclei, the outer cerebral layer, and the internal dorsal capsule, a region merging M1 to M3, and a region merging M4 to M6.

[0081] Next, the processor (110) masks the brain region mask onto the brain scan image to generate a composite image (S530). In the present disclosure, masking is performed on the brain scan image by applying a transformation matrix to the brain region mask to separately generate a brain region mask aligned to common coordinates. However, the present disclosure is not limited thereto, and the brain region mask can be directly masked onto the aligned brain scan image using the transformation matrix without generating a separate aligned brain region mask. The composite image generated through this is described with reference to FIG. 6.

[0082] Next, the processor (110) filters the pixels of the composite image (S540). Specifically, the processor (110) removes pixels that exceed a threshold value from the center value of the pixel density for each brain region of the composite image. For example, the processor (110) may remove pixels that exceed 10 HU (Hounsfield Units) from the center value. The threshold value is not limited thereto and may be set differently for each brain region.

[0083] This can reduce the possibility of volumetric information loss or false positives depending on pixel size. For example, if a small structure captured in a brain scan image is contained within a single pixel, the actual size of the structure may not be accurately represented. Ultimately, depending on the resolution, if the boundaries of a brain region overlap a single pixel, the volumetric characteristics of the brain region may become unclear. In addition, the possibility of the brain region masking non-brain regions can be reduced, and old cerebral infarctions can be removed. This is explained in Fig. 6.

[0084] Figure 6 is a diagram illustrating a filtered brain scan image.

[0085] Referring to FIG. 6, according to one embodiment, when a brain region mask is masked on a brain scan image registered with common coordinates, four regions in the left and right hemispheres of the brain may be masked. The composite image generated through this may include non-brain regions masked, volumetric features, and old cerebral infarctions. As described in FIG. 5, when pixels whose pixel density exceeds 10 HU (Hounsfield Units) from the median value are removed for each brain region, the masked portion in non-brain regions may be removed, and unclear volumetric features (Partial Volume) and old cerebral infarctions may be removed.

[0086] Returning to FIG. 5, characteristic values ​​for each brain region are calculated from the filtered composite image (S550). Specifically, the calculation may be based on the ratio of pixel density or volume for each brain region included in the composite image. According to one embodiment, the processor (110) may compare each brain region of the left and right hemispheres to calculate the characteristic values ​​for each brain region. For example, the characteristic values ​​for each brain region may be calculated based on the statistical results of the difference (Non-equivalence Score), the ratio of the average value (NWU), the ratio of the standard deviation (Std ratio), and the ratio of the volume for each brain region (Volume) for the pixel density for each brain region of the left and right hemispheres of the composite image.

[0087] Specifically, the processor (110) can calculate a statistical result of the difference in density of pixels in each brain region masked in the left and right hemisphere regions as a score between 0 and 1 to produce a characteristic value for each brain region.

[0088] In addition, the processor (110) can calculate the ratio of the average density of pixels in each brain region masked in the left and right hemisphere regions to produce characteristic values ​​for each brain region. For example, it can be calculated by the formula 1-mean(left / right), where mean(left / right) means the ratio of the average density of pixels according to the brain region included in the left hemisphere and the average density of pixels according to the brain region included in the right hemisphere.

[0089] In a similar manner, the processor (110) can calculate the standard deviation ratio of pixels in each brain region masked in the left and right hemisphere regions to produce characteristic values ​​for each brain region. For example, it can be calculated by the formula 1-std(left / right), where std(left / right) means the ratio of the standard deviation of the density of pixels according to the brain region included in the left hemisphere and the standard deviation of pixels according to the brain region included in the right hemisphere.

[0090] In addition, the processor (110) can calculate the ratio of the volume of each brain region masked in the left and right hemisphere regions to produce characteristic values ​​for each brain region. For example, it can be calculated by the formula 1-vol(left / right), where vol(left / right) means the ratio of the volume of the brain region included in the left hemisphere to the volume of the brain region included in the right hemisphere.

[0091] The characteristic values ​​for each brain region are not limited to those calculated by the above-described formula, and any method for calculating the statistical results of the difference (Non-equivalence Score), the ratio of the mean value (NWU), the ratio of the standard deviation (Std ratio), and the ratio of the volume for each brain region (Volume) for pixel density through comparison of the brain regions of the left and right hemispheres by a certain method can be included.

[0092] The processor (110) can input the characteristic values ​​for each brain region calculated by the above-described method into the prediction artificial intelligence. According to one embodiment of the present disclosure, characteristic values ​​for each brain region can be calculated for each of the four regions masked in the left and right hemispheres of the brain, and ultimately, a total of 16 characteristic values ​​for each brain region can be input into the prediction artificial intelligence.

[0093] Next, with reference to Fig. 7, the process of calculating brain lesion characteristic values ​​is described.

[0094] Figure 7 is a flowchart illustrating the process of calculating brain lesion characteristic values.

[0095] The processor (110) selects only slices of the brain scan image that include cerebral arteries among the input brain scan images (S710). Specifically, the processor (110) can select only slices that include the middle cerebral artery among the cerebral arteries. The middle cerebral artery is one of the major blood vessels of the brain, and refers to an artery that supplies nutrients to the outer surface of the brain.

[0096] Next, the processor (110) segments the brain lesion area of ​​each slice using artificial intelligence (S720). According to one embodiment, the brain lesion area may mean a blood clot (a blood clot that blocks a blood vessel). That is, the processor (110) may perform a process of detecting a blood clot from each slice. The processor (110) may use an artificial intelligence model to detect a blood clot. For example, the processor (110) may use a convolutional neural network (CNN) structure having an encoder and decoder structure to detect a blood clot. For example, U-Net may be used. In addition, without being limited thereto, the processor (110) may use any artificial intelligence model capable of extracting features of an image through image segmentation.

[0097] Next, the processor (110) calculates brain lesion characteristic values ​​based on the brain lesion region segmentation results (S730). Specifically, the processor (110) can define brain lesion characteristic values ​​based on the results of segmenting brain lesion regions according to whether or not brain lesion regions are detected in slices of the input brain scan image.

[0098] According to one embodiment, the processor (110) may determine that a brain scan image contains a thrombus if a thrombus is detected in any one of the slices of at least one input brain scan image. That is, if a thrombus is detected in any one of the slices of the brain scan image, the processor (110) may define a brain lesion characteristic value determined based on the presence of a thrombus.

[0099] Figure 8 is a flowchart of a predictive artificial intelligence model that predicts large vessel occlusion based on brain region-specific characteristic values ​​and brain lesion characteristic values.

[0100] The brain region characteristic values ​​and brain lesion characteristic values ​​extracted through the above-described process are used as input values ​​of a predictive artificial intelligence model. According to one embodiment of the present disclosure, brain region-specific characteristic values ​​are calculated for each of the four masked regions in a brain scan image, and a total of 16 brain region-specific characteristic values ​​can be input into the predictive artificial intelligence model. For example, brain region-specific characteristic values ​​calculated based on the ratio of the mean value, the ratio of the standard deviation, and the ratio of the volume can be input as a percentage, and brain region-specific characteristic values ​​based on the statistical result of the difference in pixel density can be input as a value between 0 and 1.

[0101] Next, the processor (110) inputs the characteristic values ​​for each brain region and the characteristic values ​​for brain lesions into the predictive artificial intelligence to output the certainty of large blood vessel occlusion.

[0102] Specifically, according to one embodiment of the present disclosure, the processor (110) can output a confidence score for large vessel occlusion, which is a value that predicts large vessel occlusion, and the confidence score can include information about the probability of positive or negative according to the large vessel occlusion prediction result.

[0103] While the exemplary methods of the present disclosure described above are presented as a series of operations for clarity of description, this is not intended to limit the order in which the steps are performed, and individual steps may be performed simultaneously or in different orders, if desired. To implement a method according to the present disclosure, additional steps may be included in addition to the exemplified steps, some steps may be excluded and the remaining steps may be included, or some steps may be excluded and additional steps may be included.

[0104] The various embodiments of the present disclosure are not intended to list all possible combinations but rather to illustrate representative aspects of the present disclosure, and the matters described in the various embodiments may be applied independently or in combinations of two or more.

[0105] Additionally, various embodiments of the present disclosure may be implemented by hardware, firmware, software, or a combination thereof. In the case of hardware implementation, the embodiments may be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), general processors, controllers, microcontrollers, microprocessors, etc.

[0106] The scope of the present disclosure includes software or machine-executable instructions (e.g., operating systems, applications, firmware, programs, etc.) that cause operations according to the methods of various embodiments to be executed on a device or a computer, and a non-transitory computer-readable medium having such software or instructions stored thereon and executable on the device or computer.

[0107]

[0108] The present invention can be used in an artificial intelligence-based large vessel occlusion prediction device that extracts various characteristics from an image captured without a contrast agent and predicts large vessel occlusion based on the extracted characteristics.

Claims

1. In a method for predicting large vessel occlusion based on artificial intelligence, A step of inputting a brain scan image acquired by a photographing device; A step of masking a brain region mask onto the brain scan image to derive characteristic values ​​for each brain region and derive brain lesion characteristic values ​​based on whether or not the brain lesion region is segmented; and A prediction method comprising a step of inputting the brain region-specific characteristic values ​​and the brain lesion characteristic values ​​into a prediction artificial intelligence to output the certainty of large vessel occlusion.

2. In paragraph 1, The above brain scan image is, A method for predicting a computed tomography image comprising at least one image taken in a single layer.

3. In paragraph 1, The above brain region mask is, A prediction method, wherein the reference image is a brain region defined by merging at least one specific region of the brain that separates the brain.

4. In paragraph 1, The above masking is, A prediction method performed based on a transformation matrix obtained through a process of registering the input brain scan image to common coordinates.

5. In paragraph 1, The step of calculating the characteristic values ​​for each brain region is as follows: A step of filtering pixels of a synthetic image generated by masking the brain region mask onto the brain scan image based on density; and A prediction method comprising a step of calculating characteristic values ​​for each brain region of the filtered synthetic image.

6. In paragraph 5 The above filtering is, A prediction method, wherein pixels exceeding a threshold value are removed from the center value of the density of pixels in each brain region of the composite image.

7. In paragraph 5, The characteristic values ​​for each brain region are as follows: A prediction method, wherein the prediction method is calculated based on the difference between the brain regions included in the filtered synthetic image.

8. In paragraph 7, The characteristic values ​​for each brain region are as follows: A prediction method, wherein the prediction is based on the statistical results of the difference, the ratio of the average values, the ratio of the standard deviations and the ratio of the volumes of the brain regions for the density of the pixels of the filtered synthetic image for each brain region.

9. In paragraph 2, The step of calculating the above brain lesion characteristic value is: A step of selecting a slice of the brain scan image including a cerebral artery region among the input brain scan images; and A prediction method comprising a step of calculating a brain lesion characteristic value from the slice.

10. In paragraph 9, The above brain lesion characteristic values ​​are, A prediction method defined according to whether or not the brain lesion area of ​​the above slice can be segmented.

11. In paragraph 10, The above brain lesion area is, A prediction method that refers to the area where a blood clot (Clot Sign) blocking a blood vessel is located.

12. In an AI-based large vessel occlusion prediction device, memory for storing at least one instruction; and A processor for executing at least one instruction stored in the memory based on data acquired from the memory, The above processor, Input the brain scan image acquired by the above photographing equipment, By masking the brain region mask on the brain scan image, the characteristic values ​​for each brain region are calculated, and the brain lesion characteristic values ​​are calculated based on whether the brain lesion region is segmented. A prediction device that inputs the brain region-specific characteristic values ​​and the brain lesion characteristic values ​​into the above prediction artificial intelligence and outputs the certainty for the large vessel occlusion.

13. In paragraph 12, The above brain scan image is, A prediction device, said computed tomography image comprising at least one image captured in a single layer.

14. In paragraph 12, The above brain region mask is, A prediction device, wherein said reference image is defined as said brain region, wherein said brain region is a combination of at least one specific region of said brain that has been segmented.

15. In paragraph 12, The above masking is, A prediction device, which is performed based on the transformation matrix obtained through the process of registering the input brain scan image to the common coordinates.

16. In paragraph 12, The above processor, Filtering the pixels of the synthetic image generated by masking the brain region mask onto the brain scan image based on density, A prediction device that calculates characteristic values ​​for each brain region of the filtered synthetic image.

17. In Article 16 The above filtering is, A prediction device that removes pixels exceeding the threshold value from the center value of the density of pixels in each brain region of the composite image.

18. In paragraph 16, The characteristic values ​​for each brain region are as follows: A prediction device, which is calculated based on the difference between the brain regions included in the filtered synthetic image.

19. In paragraph 18, The characteristic values ​​for each brain region are as follows: A prediction device, wherein the prediction is based on the statistical results of the difference in the density of the pixels for each brain region of the filtered composite image, the ratio of the average values, the ratio of the standard deviations, and the ratio of the volume for each brain region.

20. In paragraph 13, The above processor, Selecting the slice of the brain scan image including the cerebral artery region among the input brain scan images, A prediction device for calculating brain lesion characteristic values ​​from the above slices.

21. In paragraph 20, The above brain lesion characteristic values ​​are, A prediction device defined according to whether or not the brain lesion area of ​​the above slice can be segmented.

22. In paragraph 21, The above brain lesion area is, A prediction device indicating the area where the clot that blocks the blood vessel is located.

Images