KR102219202B1 - How to Provide Artificial Intelligence-Based Stroke Diagnostic Information - Google Patents

Abstract

According to one aspect of the present invention, a method for providing stroke diagnosis information based on artificial intelligence may comprise: a first step of obtaining, by an image obtaining unit, a non-contrast CT image related to the brain of at least one patient; a second step of pre-processing, by a pre-processing unit, the non-contrast CT image and determining whether the at least one patient is in a non-hemorrhage state or in a hemorrhage state based on the pre-processed image; a third step of normalizing, by an image processing unit, the pre-processed image and dividing and extracting a region of interest (ROI) by using a preset standard mask template; a fourth step of determining, by a determination unit, whether the cerebral large vessel of the at least one patient is abnormal by using the divided and extracted ROI; a fifth step of estimating, by the determination unit, an Alberta stroke program early CT score (ASPECTS) of the at least one patient by using the divided and extracted ROI when the brain large vessel of the at least one patient is abnormal; and a sixth step of determining, by the determination unit, that the at least one patient is a patient to whom mechanical thrombectomy can be applied only when the estimated ASPECTS is greater than or equal to a preset value. According to the present invention, it is possible to dramatically reduce costs required for an analysis process.

Description

How to Provide Artificial Intelligence-Based Stroke Diagnostic Information

The present invention relates to a method of providing stroke diagnosis information based on artificial intelligence. More specifically, it relates to a stroke diagnosis apparatus and method using an artificial intelligence algorithm capable of classifying a stroke patient based on a non-contrast CT image and selecting an emergency macrovascular obstruction patient.

Brain disease, that is, cerebrovascular disease, includes cerebral hemorrhage in which cerebrovascular blood vessels burst, cerebral infarction in which cerebrovascular blood vessels are blocked by blood clots, and cerebral aneurysms in which cerebrovascular blood vessels are abnormally swollen, and is called'stroke' in combination with cerebral hemorrhage and cerebral infarction.

In order to diagnose such brain diseases, non-invasive techniques such as ultrasound diagnosis, brain CT, and brain MRI (Magnetic Resonance Imaging) are used.

The ultrasound diagnosis method can easily diagnose atherosclerotic lesions of the carotid artery, non-invasively, through carotid ultrasound diagnosis. In addition, the transcranial Doppler test is used to measure cerebral blood flow in the cranial cavity and is applied to clinical applications.

The diagnostic method using brain CT is good for diagnosing hemorrhagic diseases, and is a recently developed technology that is very helpful in treating stroke patients by performing cerebral blood flow and cerebrovascular imaging.

The diagnostic method using brain MRI does not have the effect of artificial shading by the skull compared to brain CT, so that lesions in the brainstem, cerebellum, and temporal lobe can be diagnosed in small detail, early detection of cerebral infarction, and micro-diagnosis of cerebral perfusion status. It can be said that it is the best method for diagnosing the condition of the brain tissue as it can be diagnosed closely with the cerebrovascular condition.

In particular, stroke is a disease that causes brain damage due to blockage or rupture of blood vessels supplying the brain, resulting in physical disability, and is the most important cause of death worldwide, and permanent disability even if it does not lead to death. It is classified as a high-risk disease causing.

Conventionally, stroke has been mainly recognized as a disease for the elderly, but recently, as strokes are common even in their 30s and 40s, it is recognized as a very dangerous disease that occurs not only in old age but also in young adults.

These strokes can be divided into two types:'ischemic stroke', which occurs when blood vessels supplying blood to the brain are blocked, and'cerebral hemorrhage,' where blood vessels to the brain burst and bleed. .

The ischemic stroke accounts for about 80% of all strokes, and most of the ischemic strokes are caused by clotting blood clots, thrombosis, blocking blood vessels that supply oxygen and nutrients to the brain.

Various tests have been developed to diagnose stroke, and among them, the method using computed tomography (CT) can be performed in a relatively short time.

Accordingly, the examination method using the CT has the advantage of being able to perform a quick examination, and thus it is considered to be an examination method suitable for the characteristics of a stroke disease in which rapid response is essential.

In addition, in the case of hemorrhagic stroke, since the bleeding is immediately observed on CT after the occurrence of bleeding, the CT is useful as a tool for discriminating cerebral hemorrhage before using a thrombolytic agent that dissolves blood clots and opens up clogged blood vessels in order to treat ischemic stroke. Is used.

In addition, the CT is also important to observe the progress of cerebral hemorrhage after the use of thrombolytic agents.

On the other hand, ASPECTS (Alberta Stroke Program Early CT Score) is used as a representative index that can differentiate stroke.

The ASPECTS divides the middle cerebral artery (MCA, middle cerebral artery) region into 10 predefined anatomical regions and evaluates the presence of an early infarct marker for non-contrast CT (see Non-Patent Document 1 below).

These ASPECTS proved to be a powerful predictor for diagnosing the condition of stroke patients.

On the other hand, when scoring the severity of cerebral infarction based on non-contrast CT images through ASPECTS, etc., it is difficult to maintain the consistency of the diagnosis results as differences in diagnosis results occur according to the clinical doctor's area determination and interpretation. There was this.

In addition, since the brain disease analysis apparatus according to the prior art applies a method of normalizing after detecting only brain lesions in medical images such as CT or MRI obtained from the subject of diagnosis, the area extracted from the medical images of each patient is not consistent. There is a problem that the probability of error occurrence is high when the severity is scored because they are different from each other.

Therefore, in the brain images obtained from the subject of diagnosis, brain regions susceptible to cerebral artery damage are specified, the severity is determined based on the proportion of brain lesions in the brain region, and the prognosis of brain damage is predicted accordingly. Technology development is required.

1.Korean Patent Registration No. 10-1992057 (announced on June 24, 2019) 2. Korean Patent Registration No. 10-1754291 (announced on July 6, 2017)

1. Barber PA, Demchuk AM, Zhang J, et al. 'Validity and reliability of a quantitative computed tomography score in predicting outcome of hyperacute stroke before thrombolytic therapy: ASPECTS Study Group―Alberta Stroke Program Early CT Score.' Lancet 2000; 355:1670-74 CrossRef Medline

On the other hand, determination of early signs of ischemic stroke and conversion to ASPECTS have significant interrater variability, which is influenced by post-experience, among other factors.

That is, the ASPECTS evaluates the disease by calculating the degree of stroke progression of the user with a score of 0 to 10, and through this, the medical staff uses it as a key index used in determining the user's treatment method and predicting the prognosis.

However, in the process of estimating ASPECTS by a clinical specialist, the estimate may vary depending on the experience and experience of each clinical specialist due to the user's initial signs and image complexity.

Such scoring variability has a problem that negatively affects the decision-making process for patient clinical outcomes.

Therefore, there is a need to develop a technology that can objectify and automate ASPECTS based on image processing and deep learning technology to estimate.

That is, an object of the present invention is to solve the above-described problems, and to provide an ASPECTS estimation system and method for estimating ASPECTS, a factor that discriminates stroke disease in a non-contrast CT image of the brain.

Another object of the present invention is to provide an ASPECTS estimation system and method capable of reducing score variability when estimating ASPECTS based on image processing and deep learning techniques.

Another object of the present invention is to provide an ASPECTS estimation system and method capable of improving the reliability of the estimated ASPECTS.

Further, an object of the present invention is to provide a medical image processing apparatus and method using a template capable of performing a normalization process using a template for consistent segmentation of brain regions susceptible to cerebral artery damage in medical images of all subjects.

Another object of the present invention is to provide a medical image processing apparatus and method using a template capable of consistently segmenting and extracting brain regions susceptible to cerebral artery damage in a normalized medical image.

In conclusion, according to the present invention, a non-contrast CT image related to the brain of at least one patient is obtained, the non-contrast CT image is pre-processed, and the pre-processing Based on the image, it is determined whether the at least one patient is in a no hemorrhage state or a hemorrhage state, the pre-processed image is normalized, and a preset standard mask template is used. The region of interest (ROI) is divided and extracted, and, using the divided and extracted regions of interest, whether there is a possibility of large vessel occlusion in the brain of the at least one patient I can judge.

In particular, if there is a possibility of large vessel occlusion in the brain of at least one patient, the determination unit estimates the ASPECTS of the at least one patient using the segmented and extracted regions of interest, and the estimation The object of the present invention is to determine that the at least one patient is a patient who can apply Mechanical Thrombectomy only when the ASPECTS is greater than or equal to a predetermined value.

In addition, it is an object of the present invention to provide an apparatus, system, and method for increasing the probability of success in a clinical trial by utilizing it for screening a patient group and a normal group through a stroke diagnosis method using artificial intelligence.

Meanwhile, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems that are not mentioned are clearly to those of ordinary skill in the technical field to which the present invention belongs from the following It will be understandable.

A method of providing stroke diagnosis information based on artificial intelligence, which is an aspect of the present invention for achieving the above technical problem, is in which an image acquisition unit acquires a non-contrast CT image related to the brain of at least one patient. The first step; The pre-processing unit pre-processes the non-contrast CT image and, based on the pre-processed image, whether the at least one patient is in a No hemorrhage state or a Hemorrhage state A second step of determining whether or not; A third step of normalizing the pre-processed image by an image processing unit, and dividing and extracting a region of interest (ROI) using a preset standard mask template; A fourth step of determining, by a determination unit, whether a cerebral large vessel of the at least one patient is abnormal using the divided and extracted regions of interest; A fifth step of estimating ASPECTS of the at least one patient by using the divided and extracted regions of interest by the determination unit when there is an abnormality in the cerebral large vessel of the at least one patient; And a sixth step of determining that the at least one patient is a patient to which Mechanical Thrombectomy can be applied only when the estimated ASPECTS is greater than or equal to a predetermined value. It may include.

In addition, the second step may include removing noise from the non-contrast CT image by a noise filter of the preprocessor; Performing a co-registration of spatially aligning images within an object or between a plurality of objects existing in the non-contrast CT image from which the noise is removed, by the registration unit of the preprocessor; Performing a skull-stripping in which the skull stripping unit of the preprocessor removes a portion of the CT image on which the co-registration is performed, which is not the brain structure of the at least one patient; And determining whether the at least one patient is in a no hemorrhage state or a hemorrhage state by using the scull-striping CT image by the bleeding classification unit of the preprocessor. have.

In addition, the bleeding classification unit, based on the artificial intelligence model learning hemorrhage-related cases, using the skull-striping CT image, whether the at least one patient is in a non-bleeding (No hemorrhage) state or It is determined whether there is a hemorrhage state, and the artificial intelligence model of the bleeding classification unit may be learned using non-contrast CT data of the at least one patient.

In addition, the third step may include pre-setting the standard mask template by a template setting unit of the image processing unit; Normalizing the pre-processed image by an image normalization unit of the image processing unit; And dividing and extracting a region of interest (ROI) by applying the preset standard mask template to the normalized image by a processing unit of the image processing unit, wherein the at least one patient has no hemorrhage. ), it can only be operated.

In addition, the fourth step may include determining whether ischemia exists in the brain of the at least one patient by using the segmented and extracted region of interest by the Ischemia classification unit of the determination unit; And determining whether a Cerebral Large Vessel of the at least one patient is abnormal when the Large Vessel Occlusion determination unit of the determination unit has the ischemia. Including, in the fifth step, the ASPECTS determination unit of the determination unit, if there is an abnormality in the cerebral large vessel (Cerebral Large Vessel), using the segmented and extracted region of interest, of the at least one patient ASPECTS is estimated, and in the sixth step, the thrombectomy determination unit of the determination unit, the at least one patient is a patient to which mechanical thrombectomy can be applied only when the estimated ASPECTS is greater than or equal to a predetermined value. Can be determined as.

In addition, the ASPECTS determination unit estimates the ASPECTS based on an artificial intelligence model that has learned whether the segmented and extracted region of interest is lesioned and classified, and whether the region of interest is lesioned and classified is, Old infarct ( OI), Recent infarct (RI), Frank hypodensity (FH) and Early ischemic sign (EIS).

In addition, when the estimated ASPECTS is equal to or greater than a predetermined value. The thrombectomy determination unit performs mechanical thrombectomy only when the criteria determined through the information obtained from the Ischemia classification unit and the information obtained from the Large Vessel Occlusion determination unit are equal to or greater than a predetermined standard. It can be determined that the patient is applicable.

In addition, in the sixth step, when it is determined that the at least one patient is a patient to which Mechanical Thrombectomy can be applied, after the sixth step, the communication unit provides information on the at least one patient. It may further include a seventh step of transmitting to the outside designated in advance.

According to the ASPECTS estimation system and method according to the present invention, it is possible to estimate ASPECTS, which is an objective index for diagnosing a stroke patient's condition, using a patient's brain CT image.

In addition, according to the present invention, due to the nature of stroke disease that requires rapid prescription, it can be used as a reliable index that can prevent problems due to score fluctuations among experts and facilitate patient treatment decisions in the medical field. The effect is obtained.

In addition, according to the present invention, it is possible to overcome the inaccuracy of the score value according to the career of a specialist caused by the complexity of the stroke ASPECTS estimation method through brain CT image analysis and the demand for expertise.

In addition, according to the present invention, it is possible to significantly reduce manpower, time, and economic costs required for the analysis process by automating the entire process of stroke analysis based on CT images.

In addition, according to the present invention, it is possible to consistently segment and extract brain regions susceptible to cerebral artery damage by normalizing medical images taken of the brain such as CT and MRI, and applying a preset standard mask template to the normalized medical images. The effect of being able to be obtained is obtained.

In addition, according to the present invention, after normalizing a medical image to be diagnosed through a template, it is possible to consistently divide and extract a region of interest from medical images of all subjects based on the normalized image.

As described above, according to the present invention, by analyzing and diagnosing a brain disease (stroke) through the region of interest divided and extracted based on the standard mask template, an effect of maximizing the precision and accuracy of the diagnosis result is obtained. .

In conclusion, according to the present invention, a non-contrast CT image related to the brain of at least one patient is obtained, the non-contrast CT image is pre-processed, and the pre-processing Based on the image, it is determined whether the at least one patient is in a no hemorrhage state or a hemorrhage state, the pre-processed image is normalized, and a preset standard mask template is used. Thus, it is possible to divide and extract a region of interest (ROI), and determine whether there is a possibility of large vessel occlusion in the brain of the at least one patient using the segmented and extracted region of interest. The effect is provided.

In particular, in the present invention, when there is a possibility of large vessel occlusion in the brain of at least one patient, the determination unit estimates the ASPECTS of the at least one patient using the divided and extracted regions of interest. , Only when the estimated ASPECTS is greater than or equal to a predetermined value, it can be determined that the at least one patient is a patient to which mechanical thrombectomy can be applied.

In addition, in clinical trials for demonstrating drug efficacy, results are determined by showing statistical significance whether the predicted expected effect is achieved for clinical trial participants. When the stroke diagnosis method and apparatus according to the present invention are applied, the new drug is accurately By including only stroke patients targeted for this target as clinical trial subjects, the probability of success in clinical trials can be increased as much as possible.

That is, through the method for diagnosing stroke using artificial intelligence according to the present invention, it can be utilized to increase the probability of success in clinical trials by using it for selecting patient groups and normal groups.

On the other hand, the effects obtainable in the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those of ordinary skill in the art from the following description. I will be able to.

1 shows an example of a block diagram of an artificial intelligence based stroke diagnosis apparatus in relation to the present invention.
FIG. 2 shows an example of a block diagram of a preprocessor described in FIG. 1.
3 shows an example of a block diagram of an image processing unit described in FIG. 1.
4 illustrates an example of a block diagram of a determination unit described in FIG. 1.
5 shows an example of a flow chart for explaining the entire process of a stroke diagnosis method based on artificial intelligence proposed by the present invention.
6 is a flowchart illustrating a process of pre-processing and bleeding classification based on a non-contrast CT image among the processes described in FIG. 5.
7 is a flowchart illustrating a spatial normalization and segmentation process among the processes described in FIG. 5.
FIG. 8 is a diagram illustrating a process of segmenting a normalized image by generating the standard mask template in FIG. 7.
9 is a diagram illustrating a normalized medical image using the template in FIG. 7.
10 is a flowchart illustrating a process of classifying Ischemia among the processes described in FIG. 5.
11 is a flowchart illustrating a process of determining Large Vessel Occlusion among the processes described in FIG. 5.
12 is a flow chart illustrating step-by-step an ASPECTS estimation method in the process described in FIG. 5.
FIG. 13 is a flowchart illustrating an exemplary embodiment capable of selecting a patient with emergency great vessel obstruction based on an artificial intelligence algorithm during the process described in FIG. 5.
FIG. 14 is a flow chart illustrating another embodiment in which emergency large vessel occlusion patients can be selected based on an artificial intelligence algorithm during the process described in FIG. 5.
FIG. 15 is a flowchart illustrating another embodiment of selecting a patient with emergency great vessel obstruction based on an artificial intelligence algorithm during the process described in FIG. 5.
FIG. 16 is a diagram illustrating a method of increasing the probability of success in a clinical trial by using it for screening a patient group and a normal group through a stroke diagnosis method using artificial intelligence.

Hereinafter, a preferred embodiment of the present invention will be described with reference to the drawings. In addition, the examples described below do not unreasonably limit the content of the present invention described in the claims, and the entire configuration described in the present embodiment cannot be said to be essential as a solution to the present invention.

Hereinafter, an artificial intelligence based stroke diagnosis apparatus and method according to an exemplary embodiment of the present invention will be described in detail with reference to the accompanying drawings.

AI-based stroke diagnosis device

1 shows an example of a block diagram of a stroke diagnosis apparatus based on artificial intelligence in relation to the present invention.

The apparatus for diagnosing stroke based on artificial intelligence according to the present invention (1) may include an image acquisition unit (10), a preprocessor (20), an image processing unit (30), and a determination unit (40).

First, the image acquisition unit 10 is a device that acquires a medical image by photographing the brain of an object to be diagnosed.

Here, the image acquisition unit 10 may acquire images from imaging equipment that photographs various medical images such as brain CT and MRI.

Typically, the image acquisition unit 10 of the present invention may acquire a brain CT image.

Next, the preprocessor 20 provides a function of distinguishing between no hemorrhage and hemorrhage based on the non-contrast CT image according to an artificial intelligence algorithm.

In addition, the image processing unit 30 normalizes the medical image obtained from the preprocessor 20 and provides a function of dividing and extracting a region of interest using a preset standard mask template.

In addition, the determination unit 40 may diagnose a brain disease due to a cerebral artery injury by analyzing the divided and extracted regions of interest.

The determination unit 40 according to the present invention makes it easy for the medical staff to directly score the severity of the brain disease due to the cerebral artery injury based on the divided and extracted medical images of the region of interest, so that the medical staff can diagnose and predict the prognosis of the brain disease Can help.

Hereinafter, specific technical features of the pre-processing unit 20, the image processing unit 30, and the determination unit 40, which are components of the stroke diagnosis apparatus 1, will be described with reference to the drawings.

Pretreatment

The preprocessor 20 according to the present invention provides a function of distinguishing between no hemorrhage and hemorrhage according to an artificial intelligence algorithm based on a non-contrast CT image.

FIG. 2 shows an example of a block diagram of a preprocessor described in FIG. 1.

Referring to FIG. 2, the preprocessing unit 20 according to the present invention may include a noise filter unit 21, a registration unit 22, a skull stripping unit 23, and a bleeding classification unit 24.

First, the noise filter unit 21 performs an operation of removing noise from an image collected from the image acquisition unit 10.

Typically, the noise filter unit 21 performs a tilt correction function related to a gantry, which is an element of a CT imaging device.

In some Head CT scans, the gantry is tilted and the image slice is frequently taken at an angle. When the gantry is tilted, the voxel distance between the image slices is not accurate, especially for 3D visualization. May occur.

Therefore, in the noise filter unit 21 according to the present invention, in the case of an image photographed with an inclined gantry, by using'Gantry/Detector Tilt header' information among information stored together with the original DICOM image, Errors can be recovered through resampling.

By performing the Gantry tilt correction at this initial stage, it is expected to improve the performance of the analysis.

Next, the registration unit 22 provides a co-registration function.

The co-registration function of the present invention is to align images for alignment of anatomical structures, and spatially align images within or between objects according to inclination due to movement of a subject or differences in brain shape during CT scan.

In addition, the skull stripping unit 23 provides a skull-stripping function that provides a function for removing non-brain structures from a CT image.

Since the skull has a relatively high Hounsfield unit (HU) value compared to the brain tissue in the CT image, in the present invention, the brain tissue is analyzed after removing the known part of the brain structure through the skull stripping unit 23. It increases the ease of disease analysis.

In addition, the bleeding classification unit 24 provides a function of learning and classifying hemorrhage types based on artificial intelligence (AI).

To learn the form of hemorrhage of the bleeding classification unit 24, voxel information of the patient non-contrast CT data can be used.

In other words, the bleeding classification unit 24 constructs a convolutional neural network (CNN) neural network for feature extraction for individual NCCT slices, and synthesizes a long short-term memory (LSTM) neural network to consider the entire patient serial slice. The AI model architecture of can be built and used.

Based on this, the bleeding classification unit 24 calculates an output for the hemorrhage status of each patient and the hemorrhage type classification of the positive patient.

Alternatively, the bleeding classification unit 24 according to the present invention may use a method of classifying a hemorrhage patient based on the hemorrhage reading over a certain area of the image regardless of the type of hemorrhage.

As another method, the bleeding classification unit 24 according to the present invention may use Intraparenchymal, Intraventricular, Subarachnoid, Subdural, Epidural, etc. to detect the hemorrhage estimated lesion in the CT image and classify the shape.

In addition, in the present invention, by transferring the ID information list of the Hemorrhage negative patient to the Ischemia classification AI model of the determination unit 40, pre-processing and data input of the Hemorrhage and Ischemia classification steps may be independently performed.

Thereafter, a no hemorrhage state and a hemorrhage state can be classified according to the artificial intelligence algorithm of the bleeding classification unit 24.

Image processing unit

The image processing unit 30 according to the present invention refers to a medical image processing apparatus that normalizes the medical image obtained from the preprocessor 20 and divides and extracts an ROI using a preset standard mask template.

3 shows an example of a block diagram of an image processing unit described in FIG. 1.

Referring to FIG. 3, the image processing unit 30 according to the present invention may include a template setting unit 31, an image normalizing unit 32, and an ROI extracting unit 33.

First, the template setting unit 31 sets a standard mask template for segmenting and extracting an ROI from a medical image.

The template setting unit 31 collects a plurality of medical images obtained from a plurality of normal persons and patients with brain diseases from the preprocessor 20, generates 2D and 3D normalized images, and generates voxels based on the 3D normalized image. According to the (Voxel) setting, a 2D normalized image is generated by slicing based on a specific axis.

The present invention extracts a region of interest from a plurality of medical images obtained from a plurality of normal persons and patients with brain disease prior to dividing and extracting the region of interest from the medical image of the diagnosis target, and interest for diagnosing the brain disease of the diagnosis target Create a standard mask template for the area.

Here, the template setting unit 31 may generate a 2D normalized image by setting voxels in units set in advance based on the X-axis, Y-axis, and Z-axis of the 3D normalized image, for example, in mm.

In addition, the template setting unit 31 divides the ROI from the generated 2D normalized image, and generates a standard mask template based on the ROI divided from the plurality of normalized images.

Next, the image normalization unit 32 provides a function of normalizing a medical image to be diagnosed.

For example, the image normalization unit 32 corrects an irregular bias in the original medical image to be diagnosed (Non-uniform bias correction), registers it through spatial alignment (Co-registration), and performs standard stereotactic space. An image can be normalized through the process of performing spatial normalization by applying (Standard stereotaxic space).

Of course, the present invention is not necessarily limited thereto, and medical images may be normalized in various ways.

In addition, the ROI extractor 33 provides a function of segmenting and extracting the ROI by applying a standard mask template to the normalized medical image.

The region of interest extraction unit 33 selects an anterior/middle/posterior cerebral artery (ACA, MCA, PCA) related territory image, and divides the region of interest by dividing the brain structures related to ACA, MCA, and PCA territory in each hemisphere. And extract.

The medical image including the region of interest divided and extracted as described above is transmitted to the determination unit 40.

Judgment

The determination unit 40 may diagnose a brain disease due to a cerebral artery injury by analyzing the divided and extracted regions of interest. That is, the determination unit 40 according to the present invention makes it easy for the medical staff to directly score the severity of the brain disease due to the cerebral artery injury based on the divided and extracted medical images of the region of interest, so that the medical staff can diagnose and It can help predict prognosis.

4 illustrates an example of a block diagram of a determination unit described in FIG. 1.

Referring to FIG. 4, the determination unit 40 according to the present invention may include an Ischemia classification unit 41, a large vessel occlusion determination unit 43, an ASPECTS determination unit 44, and a mechanical thrombectomy determination unit 45. have.

First, the Ischemia classification unit 41 receives the segmented and extracted region of interest by applying a standard mask template to the normalized medical image by the region of interest extracting unit 33, and the patient with no hemorrhage is ischemia. It determines whether (Ischemia) exists.

The Ischemia classification unit 41 according to the present invention may use a method of determining whether or not a lesion in a segmented region is lesioned using artificial intelligence and classifying a shape.

That is, the Ischemia classification unit 41 according to the present invention, in relation to the Old infarct (OI) / Recent infarct (RI) / Frank hypodensity (FH) / Early ischemic sign (EIS), the presence and type of lesions in each divided area. Classification can be performed independently.

Thereafter, the Ischemia classification unit 41 identifies only FH/EIS as Acute lesions in the divided regions of the left and right hemispheres of the brain, and if there is an Acute lesion in even one area, the patient with no hemorrhage will develop ischemia. It is determined to exist.

When a patient with no hemorrhage has ischemia, the Large Vessel Occlusion determination unit 43 determines whether or not a patient with Large Vessel Occlusion.

Whether or not there is a possibility of large vessel occlusion can be determined by whether a Dense MCA sign is detected at the infra-ganglionic level.

On the other hand, despite the occurrence of large vessel occlusion, a case in which the Dense MCA sign is not detected intermittently may occur.

Therefore, in the present invention, even when the Dense MCA sign is not detected, the Large Vessel Occlusion determination unit 43 derives the frequency difference between the left and right hemisphere Hounsfield unit (HU) values in sequential slice images of the Infra-ganglionic level. By doing so, the above problems can be compensated.

That is, the Large Vessel Occlusion determination unit 43 according to the present invention includes the frequency of at least one HU value of the detected two hemispheres being equal to or greater than a predetermined reference value, or the difference in the frequency of the detected HU values of both hemispheres. If the value is higher than the value, it can be determined that the large vessel occlusion is present.

If the patient is not a large vessel occlusion patient, a conservative treatment process is in progress, but in the case of a large vessel occlusion patient, it is determined whether the patient is eligible for Mechanical Thrombectomy. Judgment is required.

Determination of whether a patient can be subjected to mechanical thrombectomy may be performed through the ASPECTS determination unit 44 and the Mechanical Thrombectomy determination unit 45.

First, the ASPECTS determination unit 44 calculates ASPECTS by applying the standard mask template to the normalized medical image by the region of interest extraction unit 33 to receive the segmented and extracted regions of interest.

The ASPECTS determination unit 44 may calculate ASPECTS through lesion detection and type classification.

Representative lesion detection and morphology classification include Old infarct (OI)/Recent infarct (RI)/Frank hypodensity (FH)/Early ischemic sign (EIS), and the like.

At this time, by applying an artificial intelligence (AI) algorithm, if any one of OI/RI/FH/EIS is detected, a conventional ASPECTS calculation method that is recognized as a lesion may be applied.

At this time, the score can be subtracted by reflecting the value recognized as the lesion to the estimation of ASPECTS.

In addition, in the present invention, by applying an artificial intelligence (AI) algorithm, a modified ASPECTS method in which only FH/EIS detection is recognized as a lesion can be applied.

At this time, the score can be subtracted by reflecting the value recognized as the lesion to the estimation of ASPECTS.

In addition, in the present invention, by applying an artificial intelligence (AI) algorithm, ACA, PCA, and ICA-related regions may be added, and an extended ASPECTS calculation method in which only FH/EIS detection is recognized as a lesion may be applied.

In general, the area for calculating ASPECTS is the MCA (Middle cerebral artery) Territory, which in the present invention extends to the ACA (Anterior cerebral artery), PCA (Posterior cerebral artery), and ICA (Internal carotid artery) areas, The ASPECTS determination unit 44 includes Frank hypodensity (FH) among the Old infarct (OI), Recent infarct (RI), Frank hypodensity (FH) and Early ischemic sign (EIS) on the MCA, ACA, PCA and ICA regions. If an early ischemic sign (EIS) is detected, it is recognized as a lesion, and the value recognized as a lesion can be reflected in the estimation of ASPECTS.

In addition, the Mechanical Thrombectomy determination unit 45 determines whether or not a patient can apply mechanical thrombectomy using the ASPECTS transmitted from the ASPECTS determination unit 44.

In the case of a patient with a possibility of large vessel occlusion, the ASPECTS delivered from the ASPECTS determination unit 44 is used, and if the score is lower than a specific criterion, the recovery probability is slim, so the Mechanical Thrombectomy determination unit 45 ), it is judged that Mechanical Thrombectomy cannot be applied, and conservative treatment is proceeded.

However, even in the case of a large vessel occlusion patient, if the ASPECTS delivered from the ASPECTS determination unit 44 is higher than a predetermined level, there is a possibility of recovery, and thus a patient who can apply Mechanical Thrombectomy. Judged by

When a patient is identified that can apply Mechanical Thrombectomy, various information related to the patient can be transferred to a tertiary hospital.

Although not shown, information transfer to the tertiary hospital may be performed through a communication unit, and the communication unit may transmit the corresponding information to a predetermined outside (eg, a hospital, etc.) through short-distance communication or long-distance communication.

As a telecommunication technology that can be used here, WLAN (Wireless LAN) (Wi-Fi), Wibro (Wireless broadband), Wimax (World Interoperability for Microwave Access), HSDPA (High Speed Downlink Packet Access), etc. may be used. .

In addition, as a technology of short range communication, Bluetooth, Radio Frequency Identification (RFID), infrared data association (IrDA), Ultra-Wideband (UWB), ZigBee, and the like may be used.

Artificial intelligence-based stroke diagnosis and classification method

5 shows an example of a flowchart illustrating the entire process of a stroke diagnosis method based on artificial intelligence proposed by the present invention.

Referring to FIG. 5, first, a step (S10) of obtaining a non-contrast CT image by the image acquisition unit 10 is performed.

Thereafter, the pre-processing unit 20 performs a pre-processing and bleeding classification operation (S20).

In the pre-processing and bleeding classification step (S21), the noise filter unit 21 removes the noise (S21), the registration unit 22, the step of performing the co-registration function (S22), the skull stripping unit ( Step 23) performs the skull-stripping function (S23) and the bleeding classification unit 24 learns the form of hemorrhage based on artificial intelligence (AI), and artificially based on the learning model. It may include a step (S24) of classifying a non-bleeding (No hemorrhage) state (S30) and a bleeding (Hemorrhage) state (S130) according to the intelligent algorithm.

Details of step S20 will be described later with reference to FIG. 6.

After step S20, a no hemorrhage state (S30) and a hemorrhage state (S130) are classified according to an artificial intelligence algorithm based on the learning model, and in the case of a no hemorrhage state (S30), , Spatial normalization and segmentation (S40) are performed.

The spatial normalization and segmentation step (S40) includes a 3D normalized image generation step (S41), a voxel setting, a 2D normalized image generation step (S42), a region of interest segmentation, a mask template generation step (S43), and a plurality of medical treatments. It may be performed through an image collection and normalization step (S44), a mask template application, and an ROI segmentation and extraction step (S45).

Details of step S40 will be described later with reference to FIG. 7.

After the spatial normalization and segmentation step (S40), a process (S50) of classifying Ischemia based on the segmented and extracted ROI by applying a standard mask template to the normalized medical image is performed.

In step S50, the Ischemia classification unit 41 receives the segmented and extracted region of interest by applying a standard mask template to the normalized medical image by the region of interest extracting unit 33 to receive a patient with no hemorrhage. Will determine whether ischemia is present.

Details of step S50 will be described later with reference to FIG. 10.

If the patient with no hemorrhage in the S50 stage does not have ischemia, the conservative treatment stage (S120) proceeds.

However, if it is determined that ischemia is present in the patient with no hemorrhage in the S50 stage (S70), the Large Vessel Occlusion determination unit 43 determines whether the patient is a large vessel occlusion. It is done (S90).

Whether a large vessel occlusion in the S90 stage has occurred can be determined by whether a Dense MCA sign has been detected at the infra-ganglionic level.

On the other hand, despite the occurrence of large vessel occlusion, a case in which the Dense MCA sign is not detected intermittently may occur.

Therefore, in the present invention, even when the Dense MCA sign is not detected, the Large Vessel Occlusion determination unit 43 derives the frequency difference between the left and right hemisphere Hounsfield unit (HU) values in sequential slice images of the Infra-ganglionic level. By doing so, the above problems can be compensated.

That is, the Large Vessel Occlusion determination unit 43 according to the present invention includes the frequency of at least one HU value of the detected two hemispheres being equal to or greater than a predetermined reference value, or the difference in the frequency of the detected HU values of both hemispheres. If the value is higher than the value, it can be determined that the large vessel occlusion is present.

Details of step S90 will be described later with reference to FIG. 11.

If the patient is not a patient with large vessel occlusion, the conservative treatment process proceeds (S120), but in the case of a patient with large vessel occlusion, the patient may be subjected to mechanical thrombectomy. It is required to determine whether or not.

Therefore, in order to determine whether a patient can be subjected to mechanical thrombectomy, after step S90, a step of estimating ASPECTS (S100) is performed.

In connection with step S100, the ASPECTS determination unit 44 calculates ASPECTS by applying the standard mask template to the normalized medical image by the region of interest extraction unit 33 to receive the segmented and extracted regions of interest.

Details of step S100 will be described later with reference to FIG. 12.

After step S100, the mechanical thrombectomy determination unit 45 determines whether or not a patient can apply mechanical thrombectomy using the ASPECTS delivered from the ASPECTS determination unit 44 (S110).

In the case of a patient with large vessel occlusion in step S110, using ASPECTS delivered from the ASPECTS determination unit 44, if the score is lower than a specific criterion, the recovery probability is slim, so the Mechanical Thrombectomy determination unit 45 , It is judged that Mechanical Thrombectomy cannot be applied, and conservative treatment is performed (S120).

However, even in the case of a patient with large vessel occlusion, if the ASPECTS delivered from the ASPECTS determination unit 44 is higher than a certain standard, there is a possibility of recovery. As a patient, mechanical thrombectomy can be applied. When it is determined and determined as a patient to which mechanical thrombectomy can be applied, various information related to the patient may be transferred to a tertiary hospital (S140).

The information transmitted to the tertiary hospital may be at least one of Elapsed time, non-contrast CT image, determination result and Tissue clock information, Conventional ASPECTS information, Modified ASPECTS information, and Extended ASPECTS information.

Hereinafter, each step of the entire process of the artificial intelligence-based stroke diagnosis method described with reference to FIG. 5 will be described in detail with reference to the drawings.

Pretreatment and bleeding classification process

6 is a flowchart illustrating a process of preprocessing and bleeding classification based on a non-contrast CT image among the processes described in FIG. 5.

Referring to FIG. 6, first, a non-contrast CT image is acquired (S10), and based on this, the noise filter unit 21 of the preprocessor 20 reduces noise in the image collected from the image acquisition unit 10. The removing operation is performed (S21).

In step S21, the noise filter unit 21 may perform a Tilt correction function related to a gantry, which is an element of a CT imaging device.

Head CT scans frequently occur when the gantry is tilted and image slices are taken at an angle, and when the gantry is tilted, the voxel distance between image slices is not accurate, and in particular, there is a problem in 3D visualization. Since it may occur, in step S21, in the case of an image photographed with a gantry tilted, the error due to tilting is resampled using the'Gantry/Detector Tilt header' information among the information stored together with the original DICOM image. Can be restored through.

Noise included in the CT image is a factor that degrades the accuracy of each step included in the preprocessing step to acquire the main features of the brain, so it needs to be removed, and in step S21, a Gaussian Blur is applied to the entire image. Noise included in the image can be removed by convolution.

By performing the Gantry tilt correction in step S21, it is expected to improve the performance of the analysis.

Next, the registration unit 22 performs a co-registration function (S22).

In step S22, the registration unit 22 arranges the images for alignment of the anatomical structure, and spatially arranges the images within or between the objects according to the inclination due to the movement of the subject or the difference in brain shape during CT scan.

After step S22, the skull stripping unit 23 performs a skull-stripping function that provides a function for removing portions other than brain structures from the CT image (S23).

In other words, since the skull has a relatively higher Hounsfield unit (HU) value than the brain tissue in the CT image, it is easy to analyze the lesion of the brain tissue by removing the known part of the brain structure through step S23 and proceeding with the analysis. Will increase.

After step S23, the bleeding classification unit 24 learns the form of hemorrhage based on artificial intelligence (AI), and based on the learning model, the no hemorrhage state (S30) and bleeding are performed according to the artificial intelligence algorithm. The (Hemorrhage) state (S130) can be identified (S24).

In step S24, the hemorrhage type learning of the bleeding classification unit 24 is to construct a convolutional neural network (CNN) neural network for feature extraction for non-contrast CT individual slice images, and to consider the entire patient serial slice. An AI model architecture in the form of synthesizing Long Short-term Memory (LSTM) neural networks can be constructed and used.Based on this, the bleeding classification unit 24 provides information on whether hemorrhage for each patient and the hemorrhage type of positive patients. Output is calculated.

Alternatively, the bleeding classification unit 24 according to the present invention may use a method of classifying a hemorrhage patient based on the reading of a hemorrhage over a certain area in the image regardless of the type of hemorrhage.

As another method, the bleeding classification unit 24 according to the present invention may use Intraparenchymal, Intraventricular, Subarachnoid, Subdural, Epidural, etc. to detect and classify the shape of the hemorrhage estimated lesion in the CT image.

Spatial normalization and segmentation process

7 is a flowchart illustrating a spatial normalization and segmentation process among the processes described in FIG. 5.

Referring to FIG. 7, after it is determined as no hemorrhage (S30), spatial normalization and segmentation (S40) are performed.

The spatial normalization and segmentation step (S40) will be described in detail based on FIG. 7.

First of all, the template setting unit 31 collects a plurality of medical images obtained from a plurality of normal persons and patients with brain diseases from the preprocessor 20, and generates a 3D normalized image (S41).

Thereafter, the template setting unit 31 generates a 2D normalized image by setting a voxel based on the 3D normalized image (S42).

In step S42, the template setting unit 31 may generate a 2D normalized image by setting voxels in units set in advance based on the X, Y, and Z axes of the 3D normalized image, for example, in mm.

That is, it is possible to generate a 2D normalized image sliced based on one of the X-axis, Y-axis, and Z-axis according to the 3D voxel setting, which is a predetermined unit unit, of the 3D normalized image.

The normalized image is 3D, and if the normalized image is expressed using voxels by setting the voxel to 3D or one of the X-axis, Y-axis and Z-axis as a reference axis, a normalized image converted to 2D is generated. I can.

After step S42, the template setting unit 31 divides the ROI from the generated 2D normalized image, and generates a standard mask template based on the ROI divided from the plurality of normalized images (S43).

FIG. 8 is a diagram illustrating a process of segmenting a normalized image by generating the standard mask template in FIG. 7.

A plurality of medical images are illustrated in (a) of FIG. 8, and an ROI extracted from the medical image and a set standard mask template are illustrated in (b) and (c) of FIG. 8, respectively. In (d) and (e), the normalized medical image input to the medical image processing apparatus and the segmented and extracted ROI are illustrated, respectively.

The present invention extracts the region of interest from a plurality of medical images obtained from a plurality of normal persons and patients with brain diseases, as shown in Fig. 8(a), prior to dividing and extracting the region of interest from the medical image of the diagnosis target. (Fig. 8(b)), a standard mask template of a region of interest for diagnosing a brain disease to be diagnosed is generated (Fig. 8(c)).

Here, the template setting unit 31 may generate a 2D normalized image by setting voxels in units set in advance based on the X-axis, Y-axis, and Z-axis of the 3D normalized image, for example, in mm.

In addition, the template setting unit 31 divides the ROI from the generated 2D normalized image, and generates a standard mask template based on the ROI divided from the plurality of normalized images.

Returning to FIG. 7 again, the image normalization unit 32 normalizes the medical image to be diagnosed obtained from the preprocessor 20 (S44).

Regarding step S44, the original medical image input to the medical image processing apparatus is illustrated in (a) of FIG. 9, and the image normalized through the template is illustrated in (b) of FIG. 9.

The image normalization unit 30 corrects an irregular bias in the original medical image to be diagnosed as shown in FIG. 9A (Non-uniform bias correction) and registers it through spatial alignment (Co-registration). The medical image may be normalized as shown in (b) of FIG. 9 through a process of performing spatial normalization by applying it to a standard stereotaxic space.

After step S44, the ROI extractor 33 divides and extracts only the ROI by applying a standard mask template to the normalized medical image (S45).

That is, the ROI extractor 33 applies the standard mask template to the normalized medical image shown in FIG. 8D to divide and extract only the ROI as shown in FIG. 8E.

In step S45, the region of interest extraction unit 33 selects Territory images related to Anterior/Middle/Posterior cerebral artery (ACA, MCA, PCA), and divides ACA, MCA, PCA Territory related brain structures in each hemisphere, The region of interest can be segmented and extracted.

The medical image including the region of interest divided and extracted as described above is transmitted to the brain disease analyzer 13.

Accordingly, the determination unit 40 can precisely observe and score brain regions susceptible to cerebral artery damage based on the segmented and extracted medical images of the region of interest, and thereby predict the severity and prognosis of brain diseases. have.

Through the above-described spatial normalization and segmentation step (S40), the present invention normalizes a medical image photographed of the brain, and applies a preset standard mask template to the normalized medical image to a region necessary for brain disease diagnosis, that is, By segmenting and extracting the region of interest related to cerebral artery injury, it is possible to consistently segment the region of interest in the medical images of all subjects based on the image normalized through the template, and analyze the brain disease through the segmented and extracted regions of interest. According to the diagnosis, precision and accuracy can be maximized.

After the spatial normalization and segmentation step (S40), a process (S50) of classifying Ischemia based on the segmented and extracted ROI by applying a standard mask template to the normalized medical image is performed.

The process of classifying Ischemia

The Ischemia classification unit 41 receives the region of interest divided and extracted by applying a standard mask template to the normalized medical image by the region of interest extracting unit 33, and the patient with no hemorrhage receives ischemia (Ischemia). ) Is present or not.

10 is a flow chart illustrating a process (S50) of classifying Ischemia among the processes described in FIG. 5.

Referring to FIG. 10, the Ischemia classification unit 41 according to the present invention may use a method of determining whether a lesion in a segmented region is a lesion and classifying a shape using artificial intelligence.

First, in relation to the Ischemia classification unit 41 according to the present invention, in relation to the Old infarct (OI)/Recent infarct (RI)/Frank hypodensity (FH)/Early ischemic sign (EIS), the lesion status and shape of each divided region Classification can be performed independently (S51).

Thereafter, the Ischemia classification unit 41 determines only FH/EIS as Acute lesions in the divided regions of the left and right hemispheres of the brain (S52).

In addition, the Ischemia classification unit 41 determines that a patient with no hemorrhage has ischemia if an Acute lesion exists in even one area (S53).

If, in the step S53, the patient with no hemorrhage does not have ischemia, the conservative treatment step (S120) proceeds.

The process of judging Large Vessel Occlusion

11 is a flowchart illustrating a process of determining Large Vessel Occlusion among the processes described in FIG. 5.

In step S90, first, it is determined whether or not the Dense MCA sign is detected (S91).

If the Dense MCA sign is detected, it is considered that Large Vessel Occlusion has occurred.

On the other hand, despite the occurrence of large vessel occlusion, a case in which the Dense MCA sign is not detected intermittently may occur.

Therefore, in the present invention, even when the Dense MCA sign is not detected, the Large Vessel Occlusion determination unit 43 derives the frequency difference between the left and right hemisphere Hounsfield unit (HU) values in sequential slice images of the Infra-ganglionic level. By doing so, the above problems can be compensated.

That is, referring to FIG. 11, first, the Large Vessel Occlusion determination unit 43 detects a frequency of each of the Hounsfield unit (HU) values of one hemisphere among the left and right hemispheres of sequential slice images of an infra-ganglionic level. (S92).

Thereafter, the Large Vessel Occlusion determination unit 43, when the frequency of at least one HU value among the detected both hemispheres is equal to or greater than a predetermined reference value or the frequency difference between the detected HU values of both hemispheres is equal to or greater than a predetermined difference value, It may be determined that the large vessel occlusion is present (S93).

If the patient is not a patient with large vessel occlusion, the conservative treatment process proceeds (S120), but in the case of a patient with large vessel occlusion, the patient may be subjected to mechanical thrombectomy. It is required to determine whether or not.

In order to determine whether a patient can be subjected to mechanical thrombectomy, after step S90, a step of estimating ASPECTS (S100) is performed.

ASPECTS estimation process

The ASPECTS determination unit 44 calculates ASPECTS by receiving the segmented and extracted ROI by applying a standard mask template to the normalized medical image by the ROI extractor 33.

12 is a flowchart illustrating step-by-step an ASPECTS estimation method in the process described in FIG. 5.

Referring to FIG. 12, the ASPECTS determination unit 44 firstly applies a standard mask template to a normalized medical image, based on the medical image of the region of interest that is segmented and extracted, using artificial intelligence. A step (S101) of determining whether there is a lesion and classifying the shape is performed.

Here, representative lesion detection and morphology classification include Old infarct (OI)/Recent infarct (RI)/Frank hypodensity (FH)/Early ischemic sign (EIS), and the like.

Thereafter, the ASPECTS determination unit 44 performs a step S102 of determining ASPECTS by applying one of a plurality of ASPECTS determination methods using at least one of a lesion and a plurality of forms.

In step S102, the ASPECTS determination unit 44 may calculate ASPECTS through lesion detection and type classification.

At this time, by applying an artificial intelligence (AI) algorithm, if any one of OI/RI/FH/EIS is detected, a conventional ASPECTS calculation method that is recognized as a lesion may be applied.

In addition, in the present invention, the modified ASPECTS method in which only FH/EIS detection is recognized as a lesion by applying an artificial intelligence (AI) algorithm can be applied.

In addition, in the present invention, by applying an artificial intelligence (AI) algorithm, ACA, PCA, and ICA-related regions may be added, and an extended ASPECTS calculation method that recognizes only FH/EIS detection as a lesion may be applied.

After step S102, based on the determined ASPECTS, a step (S103) of determining the severity of the brain disease proceeds.

After step S100, the mechanical thrombectomy determination unit 45 determines whether or not a patient can apply mechanical thrombectomy using the ASPECTS delivered from the ASPECTS determination unit 44 (S110).

Process of screening patients with emergency large vessel obstruction based on artificial intelligence algorithm

Regarding step S110, in the case of a large vessel occlusion patient, using the ASPECTS delivered from the ASPECTS determination unit 44, if the score is lower than a specific criterion, the recovery probability is slim, so the Mechanical Thrombectomy determination unit ( 45), it is determined that mechanical thrombectomy cannot be applied, and conservative treatment is performed (S120).

However, even in the case of a large vessel occlusion patient, if the ASPECTS delivered from the ASPECTS determination unit 44 is higher than a predetermined level, there is a possibility of recovery, and thus a patient who can apply Mechanical Thrombectomy. If it is determined as and is determined to be a patient who can apply mechanical thrombectomy, various information related to the patient may be transferred to a tertiary hospital (S140).

The information transmitted to the tertiary hospital may be at least one of Elapsed time, non-contrast CT image, determination result and Tissue clock information, Conventional ASPECTS information, Modified ASPECTS information, and Extended ASPECTS information.

Hereinafter, a specific method of determining whether the mechanical thrombectomy determination unit 45 is a patient to which mechanical thrombectomy can be applied using ASPECTS will be described.

Embodiment 1

FIG. 13 is a flowchart illustrating an embodiment in which emergency large vessel occlusion patients can be selected based on an artificial intelligence algorithm during the process described in FIG.

Referring to FIG. 13, the mechanical thrombectomy determination unit 45 performs a step (S111) of constructing an AI model of a regression structure based on multi numeric data.

Thereafter, the mechanical thrombectomy determination unit 45, in the case of a patient whose Dense MCA sign was detected at the infra-ganglionic level, based on the AI model, the preceding ELVO (Emergent large vessel occlusion) step (S90) and the Ischemia classification step (S70). It is possible to determine whether the procedure is performed based on the output of) and ASPECTS determined through step S100 (S112).

That is, even if the estimated ASPECTS is equal to or greater than a predetermined value, the Thrombectomy determination unit 45 may determine the criteria determined by the information obtained from the Ischemia classification unit 41 and the information obtained from the Large Vessel Occlusion determination unit in advance. It can be determined that the at least one patient is a patient to which mechanical thrombectomy can be applied only when it exceeds a specified criterion.

Example 2

FIG. 14 is a flowchart illustrating another embodiment in which emergency large vessel occlusion patients can be selected based on an artificial intelligence algorithm during the process described in FIG. 5.

Referring to FIG. 14, the mechanical thrombectomy determination unit 45 performs the step of detecting the volume of the lesion (Early ischemic sign, Frank hypodensity) according to the Emergent large vessel occlusion (ELVO) through artificial intelligence (AI) (S113). Can be done.

Thereafter, the Mechanical Thrombectomy determination unit 45 may perform machine learning (logistic regression) on the volume of the detected lesion (S114).

In addition, the mechanical thrombectomy determination unit 45 determines whether or not a mechanical thrombectomy is performed based on the machine-learned model and ASPECTS (S115).

In other words, based on an artificial intelligence model that has learned the volume value detected by large vessel occlusion, it is determined whether the at least one patient is a patient who can apply mechanical thrombectomy. I can.

Embodiment 3

FIG. 15 is a flow chart illustrating another embodiment in which emergency large vessel occlusion patients can be selected based on an artificial intelligence algorithm during the process described in FIG. 5.

Referring to FIG. 15, the mechanical thrombectomy determination unit 45 performs a step (S116) of collecting the Absolute time, Early ischemic sign, and Frank hypodensity volume from LVO (large vessel occlusion) to non-contrast CT scan. do.

Thereafter, the mechanical thrombectomy determination unit 45 calculates a tissue clock based on the collected information (S117).

In addition, the mechanical thrombectomy determination unit 45 determines whether or not a mechanical thrombectomy is performed and available time based on the tissue clock and ASPECTS (S118).

As described above, when it is determined that a patient can apply mechanical thrombectomy, various information related to the patient may be transmitted to a tertiary hospital (S140).

The information transmitted to the tertiary hospital may be at least one of Elapsed time, non-contrast CT image, determination result and Tissue clock information, Conventional ASPECTS information, Modified ASPECTS information, and Extended ASPECTS information.

A method to increase the probability of success in clinical trials by using artificial intelligence to select patients and normal groups through a stroke diagnosis method

The stroke diagnosis method and apparatus used according to the present invention described above can be used to select a patient group and a normal group to increase the probability of success in a clinical trial.

That is, the present invention can provide an apparatus, system, and method for increasing the probability of success in a clinical trial by using for screening a patient group and a normal group through a stroke diagnosis method using artificial intelligence.

In clinical trials for demonstrating drug efficacy, results are determined by showing statistical significance whether or not the expected effect predicted in advance is achieved for clinical trial participants.When applying the method and apparatus for diagnosing stroke according to the present invention, a new drug is exactly the target. By including only patients with stroke as the target of the clinical trial, the probability of success in the clinical trial can be increased as much as possible.

First, the problems of the existing clinical trials for new drugs will be explained in advance.

In clinical trials for demonstrating drug efficacy, results are judged by showing statistical significance whether or not the expected effect predicted in advance is achieved for clinical trial participants.

On the other hand, in the case of a stroke, it is necessary to measure whether or not symptoms improve due to drug efficacy through a method of questionnaire, which has a problem that the scale of the data is not detailed.

Therefore, in order to prove statistical significance, the value of the evaluation scale must be increased statistically significantly before and after administration or compared to the sham drug group, and the larger the predicted increase value, the smaller the number of target subjects and the probability of achieving statistical significance increases.

At this time, if the predicted increase value is small, the number of target targets increases and the difficulty of verifying statistics increases.

As a result, it is very difficult to increase the stroke rating scale by one level, and thus, the possibility of passing the clinical trial is very low.

In the present invention, in order to solve this problem, it is intended to increase the probability of success in the clinical trial as much as possible by including only stroke patients targeted by the new drug as clinical trial subjects.

One of the important factors of failure in the development of new drugs for the central nervous system is the difficulty of accurately selecting subjects and selecting drug responders.

In the case of central nervous system drugs, particularly, the response rate to placebo is high, reducing the heterogeneity of the subject group, and setting up a biomarker that can predict drug responsiveness is an important strategy to increase the success rate.

In addition, since it takes a long period of time to confirm a stroke (about 3 months), there is a problem in that it is very difficult to include only stroke patients targeted by the new drug as clinical trial subjects because screening is difficult.

Therefore, it can be utilized to increase the probability of success in clinical trials by utilizing the method for diagnosing stroke using artificial intelligence proposed by the present invention to select patient groups and normal groups.

FIG. 16 is a diagram illustrating a method of increasing the probability of success in a clinical trial by using it for screening a patient group and a normal group through a stroke diagnosis method using artificial intelligence.

Referring to FIG. 16, first, a step (S1) of recruiting candidate groups for clinical trials for demonstrating drug efficacy is performed.

Thereafter, the image acquisition unit 10 obtaining a non-contrast CT image (S10), the preprocessing unit 20 performing a pre-processing and bleeding classification operation (S20), a learning model Based on the artificial intelligence algorithm, a no hemorrhage state (S30) and a hemorrhage state (S130) are classified, and in the case of a no hemorrhage state (S30), spatial normalization is performed. ) And segmentation step (S40), the process of classifying Ischemia based on the region of interest segmented and extracted by applying a standard mask template to a normalized medical image (S50), a patient with no hemorrhage ischemia If it is determined that (Ischemia) is present (S70), the Large Vessel Occlusion determination unit 43 determines whether the patient is a large vessel occlusion (S90), and mechanical thrombectomy (Mechanical Thrombectomy) is performed. In order to determine whether the patient is applicable, after step S90, the step of estimating ASPECTS (S100), the mechanical thrombectomy determination unit 45, using the ASPECTS transmitted from the ASPECTS determination unit 44, the mechanical thrombus A step (S110) of determining whether the patient can apply mechanical thrombectomy is performed.

Steps S10 to S110 have been described in detail with reference to FIGS. 5 to 15, and repetitive descriptions are omitted for the sake of simplicity of the specification.

After the diagnosis result is derived through step S110, a step (S210) of dividing a plurality of experimental candidate groups into an actual stroke patient group and a normal patient group may proceed based on the diagnosis result.

At this time, through the step of conducting a clinical trial (S220) and verifying the drug efficacy based on the clinical test result (S230) based only on the subjects classified as the actual stroke patient group, only stroke patients targeted by the new drug By including them as subjects for clinical trials, the probability of successful clinical trials can be increased as much as possible.

As a result, it can be utilized to increase the probability of success in clinical trials by utilizing the method for diagnosing a stroke using artificial intelligence according to the present invention for selecting patient groups and normal groups.

The above-described steps S1 to S230 may be performed independently of the stroke diagnosis device 1, or a separate server (not shown) or a separate central management device (not shown), and the stroke diagnosis device 1 It can also be applied to perform the entire operation together.

Effects according to the invention

According to the ASPECTS estimation system and method according to the present invention, it is possible to estimate ASPECTS, which is an objective index for diagnosing a stroke patient's condition, using a patient's brain CT image.

In addition, according to the present invention, due to the nature of stroke disease that requires rapid prescription, it can be used as a reliable index that can prevent problems due to score fluctuations among experts and facilitate patient treatment decisions in the medical field. The effect is obtained.

In addition, according to the present invention, it is possible to overcome the inaccuracy of the score value according to the career of a specialist caused by the complexity of the stroke ASPECTS estimation method through brain CT image analysis and the demand for expertise.

In addition, according to the present invention, it is possible to significantly reduce manpower, time, and economic costs required for the analysis process by automating the entire process of stroke analysis based on CT images.

In addition, according to the present invention, it is possible to consistently segment and extract brain regions susceptible to cerebral artery damage by normalizing medical images taken of the brain such as CT and MRI, and applying a preset standard mask template to the normalized medical images. The effect of being able to be obtained is obtained.

In addition, according to the present invention, after normalizing a medical image to be diagnosed through a template, it is possible to consistently divide and extract a region of interest from medical images of all subjects based on the normalized image.

As described above, according to the present invention, by analyzing and diagnosing a brain disease (stroke) through the region of interest divided and extracted based on the standard mask template, an effect of maximizing the precision and accuracy of the diagnosis result is obtained. .

In addition, in clinical trials for demonstrating drug efficacy, results are determined by showing statistical significance whether the predicted expected effect is achieved for clinical trial participants. When the stroke diagnosis method and apparatus according to the present invention are applied, the new drug is accurately By including only stroke patients targeted for this target as clinical trial subjects, the probability of success in clinical trials can be increased as much as possible.

That is, through the method for diagnosing stroke using artificial intelligence according to the present invention, it can be utilized to increase the probability of success in clinical trials by using it for selecting patient groups and normal groups.

On the other hand, the effects obtainable in the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those of ordinary skill in the art from the following description. I will be able to.

The embodiments of the present invention described above can be implemented through various means. For example, embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.

In the case of implementation by hardware, the method according to embodiments of the present invention includes 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), processors, controllers, microcontrollers, can be implemented by microprocessors.

In the case of implementation by firmware or software, the method according to the embodiments of the present invention may be implemented in the form of a module, procedure, or function that performs the functions or operations described above. The software code may be stored in a memory unit and driven by a processor. The memory unit may be located inside or outside the processor, and may exchange data with the processor through various known means.

Detailed description of the preferred embodiments of the present invention disclosed as described above has been provided to enable those skilled in the art to implement and implement the present invention. Although the above has been described with reference to preferred embodiments of the present invention, those skilled in the art will appreciate that various modifications and changes can be made to the present invention without departing from the scope of the present invention. For example, a person skilled in the art may use the components described in the above-described embodiments in a manner that combines with each other. Thus, the invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The present invention may be embodied in other specific forms without departing from the spirit and essential features of the present invention. Therefore, the detailed description above should not be construed as restrictive in all respects and should be considered as illustrative. The scope of the present invention should be determined by rational interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention. The invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein. In addition, the embodiments may be configured by combining claims that do not have an explicit citation relationship in the claims, or may be included as new claims by amendment after filing.

Claims (8)

A first step of obtaining, by the image acquisition unit, a non-contrast CT image related to the brain of at least one patient;
The pre-processing unit pre-processes the non-contrast CT image and, based on the pre-processed image, whether the at least one patient is in a No hemorrhage state or a Hemorrhage state A second step of determining whether or not;
A third step of normalizing the pre-processed image by an image processing unit, and dividing and extracting a region of interest (ROI) using a preset standard mask template;
A fourth step of determining, by a determination unit, whether a cerebral large vessel of the at least one patient is abnormal using the divided and extracted regions of interest;
A fifth step of estimating ASPECTS of the at least one patient by using the divided and extracted regions of interest by the determination unit when there is an abnormality in the cerebral large vessel of the at least one patient; And
A sixth step of determining that the at least one patient is a patient to which Mechanical Thrombectomy can be applied only when the estimated ASPECTS is greater than or equal to a predetermined value; Including,

In the sixth step, the determination unit,
When the estimated ASPECTS is less than the predetermined value, it is determined that the at least one patient cannot be subjected to the mechanical thrombectomy and is a patient who needs conservative care,
Method for providing AI-based stroke diagnosis information, characterized in that it is determined that the at least one patient is a patient to which mechanical thrombectomy can be applied, only when the estimated ASPECTS is greater than or equal to a predetermined value.
The method of claim 1,
The second step,
Removing noise from the non-contrast CT image by a noise filter of the preprocessor;
Performing a co-registration of spatially aligning images within an object or between a plurality of objects existing in the non-contrast CT image from which the noise is removed, by the registration unit of the preprocessor;
Performing a skull-stripping in which the skull stripping unit of the preprocessor removes a portion of the CT image on which the co-registration is performed, which is not the brain structure of the at least one patient; And
And determining whether the at least one patient is in a no hemorrhage state or a hemorrhage state by using the scull-striping CT image by the bleeding classification unit of the preprocessor. A method of providing stroke diagnosis information based on artificial intelligence.
The method of claim 2,
The bleeding classification unit is based on an artificial intelligence model that has learned about hemorrhage-related cases, and whether the at least one patient is in a No hemorrhage state or bleeding ( Hemorrhage) state or not,
The artificial intelligence model of the bleeding classification unit is learned by using the non-contrast CT data of the at least one patient.
The method of claim 1,
The third step,
Pre-setting the standard mask template by a template setting unit of the image processing unit;
Normalizing the pre-processed image by an image normalization unit of the image processing unit; And
A step of dividing and extracting a region of interest (ROI) by applying the preset standard mask template to the normalized image by a processing unit of the image processing unit; and
A method of providing stroke diagnosis information based on artificial intelligence, characterized in that it operates only when it is determined that the at least one patient is in a no hemorrhage state.
The method of claim 1,

The fourth step,
Determining whether ischemia exists in the brain of the at least one patient by using the divided and extracted region of interest by the Ischemia classification unit of the determination unit; And
Determining whether a Cerebral Large Vessel of the at least one patient is abnormal when the Large Vessel Occlusion determination unit of the determination unit has the ischemia; Including,

In the fifth step,
The ASPECTS determination unit of the determination unit, when an abnormality exists in the cerebral large vessel, estimates ASPECTS of the at least one patient using the segmented and extracted region of interest,

In the sixth step,
The thrombectomy determination unit of the determination unit determines that the at least one patient is a patient to which Mechanical Thrombectomy can be applied only when the estimated ASPECTS is greater than or equal to a predetermined value. How to provide diagnostic information.
The method of claim 5,
The ASPECTS determination unit,
Estimating the ASPECTS based on an artificial intelligence model that has learned whether the segmented and extracted regions of interest are lesioned and classified,
Classification of the lesion and type of the region of interest,
A method of providing stroke diagnosis information based on artificial intelligence, characterized in that it is performed using Old infarct (OI), Recent infarct (RI), Frank hypodensity (FH) and Early ischemic sign (EIS).
The method of claim 5,
If the estimated ASPECTS is more than a predetermined value,
The thrombectomy determination unit,
A patient to whom the at least one patient can apply Mechanical Thrombectomy only when the criteria determined through the information obtained from the Ischemia classification unit and the information obtained from the Large Vessel Occlusion determination unit are equal to or greater than a predetermined standard Method for providing diagnosis information based on artificial intelligence, characterized in that it is determined to be.
The method of claim 1,
In the sixth step, when it is determined that the at least one patient is a patient to which Mechanical Thrombectomy can be applied,
After the sixth step,
7th step of transmitting, by the communication unit, information on the at least one patient to the outside designated in advance; method for providing stroke diagnosis information based on artificial intelligence, characterized in that it further comprises.

Images