KR20200083400A - System for estimating aspect score based on images of stroke patients - Google Patents

Abstract

According to the present invention, an Alberta stroke program early CT (ASPECT) score estimation system comprises: a preprocessing unit which normalizes and standardizes features of an image dataset; an image processing unit which separates each lesion within a CT image classified into the supra ganglion level and the ganglion level; and a determination unit which independently builds a neural network for learning positive/negative images for each lesion to determine whether the lesion has a stroke. Accordingly, score variability can be reduced when estimating the ASPECT score.

Description

System for estimating the ASPECT score based on the image of a stroke patient {SYSTEM FOR ESTIMATING ASPECT SCORE BASED ON IMAGES OF STROKE PATIENTS}

The present invention relates to a system for estimating an ASPECT score based on an image of a stroke patient, and more specifically, estimating an ASPECT score from a brain CT image that estimates an ASPECT score from a brain CT image of a stroke patient based on image processing and neural networks System.

Stroke is a disease that causes brain damage due to clogging or rupture of blood vessels that supply blood to the brain, resulting in physical disability, which is the most important cause of death worldwide, and causes permanent disability even if death does not occur. It is classified as a dangerous disease.

In the past, stroke was mainly recognized as a disease of the elderly, but recently, as stroke frequently occurs in people in their 30s and 40s, it is recognized as a very dangerous disease that occurs widely not only in old age but also in young adults (Non-patent Document 1 and Non Patent Document 2).

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

The ischemic stroke accounts for about 80% of the total stroke, and most of the ischemic stroke is caused by clotting blood clots (thrombosis), which block blood vessels that supply oxygen and nutrients to the brain.

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

Accordingly, as the test method using the CT has the advantage of being able to test quickly, it is considered to be a test method suitable for the characteristics of a stroke disease that requires rapid response.

In addition, in the case of hemorrhagic stroke, since the bleeding is observed immediately after CT, the CT is useful as a tool for differentiating cerebral hemorrhage prior to the use of a thrombolytic agent that melts blood clots and pierces blocked blood vessels to treat ischemic stroke. Is used.

In addition, the CT is also important in observing the progression of brain hemorrhage after thrombolysis.

On the other hand, the ASPECT score (Alberta Stroke Program Early CT Score) is used as a representative indicator for distinguishing stroke.

The ASPECT score divides the MCA (middle cerebral artery) region into 10 predefined anatomical regions and evaluates the presence of early infarction markers with substantial low pollution to NCCT (see Non-Patent Document 3 below).

This ASPECT score was proved to be a powerful predictor of the condition of stroke patients.

Global atlas on cardiovascular disease prevention and control. Geneva, World Health Organization, 2011. WHO Cardiovascular Diseases Fact Sheet No. 317. Updated March 2013, http/www.who.int/mediacentre/factsheets/fs317/en/ 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 PFAFF, J., et al. 'e-ASPECTS Correlates with and Is Predictive of Outcome after Mechanical Thrombectomy.' American Journal of Neuroradiology, 2017.

However, the determination of the initial signs of ischemia and the conversion to the ASPECT score have significant interrater variability, which is influenced by post-experience among other factors (see non-patent document 4 above).

That is, the ASPECT score evaluates the disease by calculating the degree of stroke progression of the user with a score of 0 to 10, and the medical staff uses it as a main index used to determine the treatment method and predict the prognosis of the user.

However, in the process of estimating the ASPECT score by the specialist, the estimated value may vary depending on the experience and experience of each specialist due to the user's initial signs and the complexity of the image.

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

Accordingly, there is a need to develop a technique capable of estimating and estimating ASPECT scores based on image processing and deep learning technologies.

An object of the present invention is to solve the above problems, to provide an ASPECT score estimation system and method for estimating the ASPECT score, which is a factor for differentiating stroke disease in brain CT images.

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

Another object of the present invention is to provide an ASPECT score estimation system and method that can improve the reliability of the estimated ASPECT score.

In order to achieve the above object, the ASPECT score estimation system according to the present invention is a pre-processing unit that normalizes and standardizes features of an image dataset, and separates each lesion within a CT image classified as a Supra ganglion level and a ganglion level. It is characterized by including an image processing unit to divide and a judgment unit to independently determine the stroke of the lesion by independently constructing a neural network for learning positive/negative images for each lesion.

Specifically, a system for estimating an ASPECT score based on a brain CT image of a stroke patient according to the present invention, comprising: a pre-processing unit that normalizes and standardizes features of an image data set; An image processing unit separating each lesion within a CT image classified into a supra ganglion level and a ganglion level; And a judging unit for independently constructing a neural network for learning positive/negative images for each lesion to determine whether or not the lesion is stroked, wherein the pre-processing unit includes a gaussian blur in the entire brain CT image of the patient and includes it in the image. Remove the noise, search the skull ellipse to find the skull in the brain CT image, align the position of the image based on the center point of the searched skull, rotate the image to align the position and rotation angle of the dataset constant, , Performs horizontal invert according to the location of the lesion (Lesion-Side), the pre-processing unit, in the operation of searching for the skull ellipse, searches for an adaptive threshold value considering the pixel value distribution of the segmented image And derives only the pixel information corresponding to the skull, detects an edge based on the thresholded image, acquires ellipse information of the inner and outer edges of the skull by searching the skull ellipse from the image with only the edge information remaining, , According to the operation of the pre-processing unit, the CT image is normalized and normalized to an image aligned in the center such that the left brain is a lesion and has a center vertical line of the image as a symmetry line of the brain. The judging unit cuts each lesion image from the entire image according to the result of separating each lesion from the image processing unit, and performs supervised learning with the neural network configured for each lesion, including information about positive/negative images. , Learning and classifying in an independent neural network for each of the lesions, calculating a final ASPECT score based on the number of neural networks that judged the lesion to be positive, and the determining unit independent of the supra ganglion level and the ganglion level A neural network is constructed, and each of the independently constructed neural networks learns based on a CNN, consists of six hidden layers and two dropout layers to prevent overfitting, and ReLU can be used as an activation function.

As described above, according to the ASPECT score estimation system and method according to the present invention, the effect of being able to estimate the ASPECT score, which is an objective index for diagnosing the condition of a stroke patient, is obtained using a brain CT image of a patient.

And according to the present invention, it can be used as a reliable indicator that can prevent problems due to the fluctuation of scores between experts due to the nature of stroke diseases that require rapid prescription and facilitate the patient's treatment decisions in the medical field. The effect is obtained.

In addition, according to the present invention, the effect of being able to overcome the inaccuracy of the score value according to the career of a specialist, which arises due to the complexity of the stroke ASPECT score estimation method through brain CT image analysis and the need for expertise, is obtained.

In addition, according to the present invention, it is possible to dramatically reduce the manpower, time, and economic cost required for the analysis process by automating the entire process of the CT image-based stroke analysis.

1 is a block diagram of an ASPECT score estimation system according to a preferred embodiment of the present invention,
2 is a flowchart illustrating step-by-step ASPEC score estimation method according to a preferred embodiment of the present invention,
Figure 3 is an exemplary view for explaining the pre-processing step,
4 is a view for explaining the center point alignment process,
5 is a view for explaining the rotation alignment,
6 is a diagram illustrating images before and after fading;
7 is an exemplary diagram illustrating ganglion level division and supra ganglion level division,
8 is a diagram illustrating the results of segmentation of a ganglion level;
9 is a diagram illustrating the neural network structure of each lesion,
10 is a performance evaluation graph.

Hereinafter, an ASPECT score estimation system and method according to a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

Hereinafter, the dataset (DataSeT) used in the algorithm for implementing the ASPECT score estimation method according to the present invention used brain CT images of 287 stroke patients who calculated the ASPECT score according to the judgment of a medical expert, and each patient's brain CT images use a total of 2 images, including 1 Ganglionic Level and 1 Supra Ganglionic Level.

For the ASPECT score calculation, it is judged whether positive/negative for 7 regions at the ganglion level and 3 regions at the supra ganglion level.

The accuracy of the algorithm is judged according to whether the positive/negative true data values of each patient's 10 lesions and the positive/negative values of each lesion calculated through the algorithm match.

1 is a block diagram of an ASPECT score estimation system according to a preferred embodiment of the present invention.

As shown in FIG. 1, the ASPECT score estimation system 10 according to a preferred embodiment of the present invention normalizes and standardizes features of an image dataset to calculate an ASPECT score based on brain CT images of stroke patients Pre-processing unit 20, based on segmentation image processing unit 30 and deep-learning that separates each lesion within CT images classified into supra ganglion level and ganglion level It includes a judging unit 40 to independently determine the stroke of the lesion by independently constructing a neural network that learns positive/negative images for each lesion.

The pre-processing unit 20 is communicatively connected to a CT device or a database that stores and manages CT images captured by the CT equipment, and acquires and preprocesses CT images of a patient who wants to estimate an ASPECT score.

In addition, the ASPECT score estimation method according to a preferred embodiment of the present invention is classified into a pre-processing step, a Supra ganglion level and a ganglion level, normalizing and standardizing the characteristics of the image dataset to calculate the ASPECT score based on the CT image of the stroke patient. It includes a splitting step of separating each lesion within the CT image and a judgment step of determining whether the lesion is stroked by independently constructing a neural network that learns positive/negative images for each lesion based on deep learning.

The patient's ASPECT score is calculated by summing the stroke of each lesion.

2 is a flowchart for explaining the ASPEC score estimation method step by step according to a preferred embodiment of the present invention.

In this embodiment, a step-by-step description of the entire algorithm and a test software result of a diagnostic step of a deep learning-based lesion are examined and verification of the entire algorithm is performed.

(1) Pre-treatment step

In the pre-processing step, the pre-processing unit 20 performs a function to ensure a more accurate result in the subsequent partitioning step and deep learning step through normalization and standardization of the data set.

To this end, in the pre-processing step, the pre-processing unit 20 detects the position of the skull in the brain CT image based on image processing, and aligns it based on a rotation degree and a center point. The Alignment step to be performed, and the Horizontal Invert step according to the Lesion-Side may be performed.

The image that has undergone the pre-processing step is normalized to an image aligned in the center such that the left brain is a lesion and the center vertical line of the image is a Symmetry Line of the brain.

In this pre-processing step, in the deep learning-based determination step, a key feature capable of discriminating whether or not a stroke is optimized is optimized to a specific image.

In FIG. 2, the pre-processing step includes a noise removing step (S10) of removing noise in the image, a search step (S12) of searching for a skull ellipse to find a skull in a patient's brain CT image, and an image based on the center point of the searched skull And an alignment step (S14) of aligning the positions and rotating the image to uniformly align the position and rotation angle of the dataset.

3 is an exemplary view for explaining the pre-processing step.

FIG. 3(a) shows an image in which Gaussian blur is convolution, FIG. 3(b) shows a threshold image, and FIG. 3(c) shows an outline search image, 3(d), the inner and outer ovals of the skull are illustrated.

The noise included in the CT image is a factor that degrades the accuracy of each step included in the pre-processing step to acquire the main features of the brain, and thus needs to be removed.

Therefore, the noise removing step (S10), as shown in Figure 2 (a), removes the noise included in the image by convolution (Gaussian Blur) to the entire image (Gaussian Blur).

The search step (S12) performs an automatic threshold (Auto Thresholding), contour search (Contour Finding) and skull (skull) ellipse search functions.

That is, in the search step (S12), each function can be separated to find the inner and outer ovals of the skull from the noise-removed image.

The automatic thresholding function searches for an adaptive threshold value in consideration of the pixel value distribution of the divided image, as shown in FIG. 3(b), and pixel information corresponding to the skull Induce them to leave only.

The contour search function detects an edge based on a thresholded image as shown in FIG. 3(c), and the skull ellipse search function is shown in FIG. 3(d). As described above, the ellipse of the skull is searched for from the image where only the edge information remains, and the ellipse information of the inner and outer edges of the skull is obtained.

The alignment step (S14) is a pre-processing step for maintaining the consistency of the dataset, and the consistency of the dataset is necessary in order to increase learning and classification accuracy in deep learning-based learning and judgment for determining stroke.

4 is a view for explaining the center point alignment process, and FIG. 5 is a view for explaining rotational alignment.

That is, since the skull position of the patient is different when the specialist takes a brain CT image of a stroke patient, the alignment step is Center-Point Alignment and Rotation Alignment of the skulls shown in FIGS. 4 and 5. ).

Accordingly, the present invention can obtain the consistency of the position and rotation of the dataset.

Here, the position and rotational alignment of the dataset may be calculated based on skull ellipse information obtained in the preprocessing step.

On the other hand, the pre-processing step further includes a fading (Padding) step (S16) for removing the skull area that is unnecessary in determining whether or not stroke, so as to increase the learning and classification accuracy of the deep learning-based learning and judgment step afterwards. Can.

6 is a diagram illustrating images before and after the fading process.

6A and 6B show an image before fading and an image after fading.

That is, the fading step sets a region of interest (ROI) for an area of a certain width from the skull ellipse based on the skull ellipse information obtained in the preprocessing step, as shown in FIG. 6(a), The threshold value corresponding to the skull in the CT image is adaptively calculated, and a fading operation is performed based on the calculated threshold value.

(2) Split stage

The segmentation step is a step for separating a total of 10 regions for ASPECT score estimation from the brain CT image, extracting 7 regions at the ganglion level, and 3 regions at the supra ganglion level (S18).

7 is an exemplary diagram illustrating ganglion level division and supra ganglion level division.

In the ganglion level and supra ganglion level images, there is a feature point that appears at the time of stroke onset, and segmentation of each lesion is performed based on image processing technology using the skull ellipse information obtained in the preprocessing step and the feature point. .

7 shows a reference using a feature point, and in FIG. 7 (a) and (b), a feature point used at the ganglion level and the supra ganglion level is shown as a red dot.

The image processing unit 30 of the ASPECT score estimation system 10 according to the present invention searches for a feature point based on a preset division criterion, and then divides the lesions by geometrically connecting each feature point.

8 is a diagram illustrating the results of segmentation of ganglion levels.

8(a) and 8(b) show images before and after division.

In FIG. 8, it can be confirmed that the 7 lesions extracted at the ganglion level were divided.

Since each lesion has a form that is not clearly distinguished by pixel values, it is optimized to approximate the division criteria established according to the opinion of a specialist.

(3) Diagnosis of deep learning-based lesions

In step S22, the determination unit 40 cuts each lesion image from the entire image according to the division result (image crop), and the cropped image (Cropped image), including information about positive/negative neural networks configured for each lesion. Conduct supervised learning.

Each neural network learns based on CNN (Convolutional Neural Network), consists of 6 hidden layers and 2 dropout layers to prevent overfitting, and ReLU (Rectified Linear) Unit) as the activation function.

Therefore, the determination unit 40 performs learning and classification in an independent neural network for each lesion, and the number of neural networks that judged the lesion to be positive means the number of lesions with stroke disease, so the following equation is based on the numerical value. The final ASPECT score is calculated according to 1 (S24).

[Equation 1]

(4) Verification

In this embodiment, the ASPECT score estimated through the above-described process was developed and verified as a test result for M4 lesions at the Supra ganglion level, and a prototype was developed to predict the accuracy of the neural network for each lesion.

9 is a diagram illustrating the neural network structure of each lesion.

As shown in FIG. 9, in the verification process, a data set of 187 patients for M4 lesions was used, 80% was used as a training data set, and 20% of images were used as a test data set.

Prototype test accuracy for the M4 area was about 70.9%, and the lack of dataset and lack of segmentation accuracy were judged to be the main causes.

Through the verification results of the prototype, it can be seen that the segmentation accuracy for each lesion greatly affects the accuracy of ASPECT score estimation.

Accordingly, the system was changed so as not to consider the effect of the division, and further testing was conducted.

The input data and neural network structure were changed to estimate the ASPECT score for the entire image without dividing the lesions for the ganglion level and the supra ganglion level among brain CT images.

That is, the ASPECT score estimation system has an independent neural network for the ganglion level and the supra ganglion level, and learns the ASPECT score by receiving a brain CT image of a corresponding level that has not been subjected to a pre-processing step.

The training dataset used CT images of 278 stroke patients and ResNet101 for neural network learning.

The performance evaluation algorithm used a Bland-Altman plot, and evaluated how close the expert is to calculate the ASPECT score using the error between the data obtained from the neural network and the ground-truth calculated by the specialist for the dataset. Did.

[Equation 2]

에 대하여 수프라 신경절 레벨과 신경절 레벨에 대해 독립적으로 실측 자료 값(

)과 결과값(

)의 오차를 계산하고, 각 레벨의 오차를 더한 값을 최종 에러값(

)으로 결정하였다.That is, the patient according to equation (2)

For the Supra ganglion level and the ganglion level, independently measured data values (

) And the resulting value (

), and the error value of each level is added to the final error value (

).

Table 1 is a performance evaluation result table, and is a performance evaluation graph obtained with FIG. 10 Bland-Altman plot.

Average Standard Deviation 95% confidence interval 0.1116 2.5080 -0.5453 ~ 0.7684

In Table 1 and FIG. 10, the error average shows a result close to '0', but a 0.95% confidence interval (CI) result of'-0.5453 to 0.7684' as the standard deviation shows a high scatter plot as '2.5080'. Showed.

As described above, the present invention can estimate the ASPECT score, which is an objective index for diagnosing the condition of a stroke patient, using only CT images.

And the present invention is an ASPECT score automation calculation program of stroke disease, due to the nature of stroke disease requiring rapid prescription, to prevent problems due to the fluctuation of scores between experts, and to facilitate the treatment decision of patients in the medical field It can be used as a reliable indicator.

According to the results of the verification, the system described in this embodiment showed somewhat low accuracy due to lack of volume of the dataset of the stroke CT image, reliability problems of the measured data, and ambiguity of features in the image.

However, since neural networks have a characteristic that results are greatly influenced by the quality of the dataset, the reliable results through the increase of the dataset and the scoring of multiple specialists are applied, and the opinions of the specialists are applied in the pre-processing stage of the image. When the neural network is modified to receive only the features specialized for stroke identification through emphasis and selection of the reflected features, high accuracy results can be obtained.

Although the invention made by the present inventors has been described in detail according to the above-described embodiments, the present invention is not limited to the above-described embodiments, and it is needless to say that various modifications can be made without departing from the gist thereof.

10: ASPECT score estimation system
20: pre-processing unit
30: image processing unit
40: judgment unit

Claims (1)

A system for estimating an ASPECT score based on a brain CT image of a stroke patient,
A pre-processing unit to normalize and standardize the features of the image data set;
An image processing unit separating each lesion within a CT image classified into a supra ganglion level and a ganglion level; And
Includes a judgment unit for independently determining the stroke of the lesion by independently constructing a neural network that learns positive/negative images for each lesion.

The pre-processing unit,
Convolution of Gaussian blur across the patient's brain CT image to remove the noise contained in the image,
Searching for a skull ellipse to find a skull in the brain CT image,
Align the position of the image based on the center point of the searched skull and rotate the image to uniformly align the position and rotation angle of the dataset.
Performs Horizontal Invert according to the location of the lesion (Lesion-Side),

The pre-processing unit, in the operation of searching for the skull ellipse,
The adaptive threshold value is searched in consideration of the pixel value distribution of the segmented image, and the pixel information corresponding to the skull is induced to remain.
Edge detection is performed based on the completed image.
By searching for the skull ellipse from the image with only the edge information remaining, we obtain the ellipse information of the inner and outer edges of the skull,

According to the operation of the pre-processing unit, the CT image is normalized and normalized to an image aligned in the center such that the left brain is a lesion, and the center vertical line of the image is a Symmetry Line of the brain,

The determination unit,
According to the result of separating each lesion in the image processing unit, cut each lesion image from the entire image,
Performing supervised learning with the neural network configured for each lesion, including information on positive/negative images of the cut out,
For each of the lesions, learning and classification are performed in an independent neural network, and a final ASPECT score is calculated based on the number of neural networks that judge the lesion to be positive.

The determination unit,
Build an independent neural network with respect to the supra ganglion level and ganglion level,

Each of the independently constructed neural network,
Learning based on CNN,
It consists of 6 hidden layers and 2 dropout layers to prevent overfitting,
ASPECT score estimation system characterized by using ReLU as an activation function.

Images