KR101030169B1 - Method for ventricle segmentation using radial threshold determination - Google Patents

Abstract

The present invention relates to an automatic ventricular segmentation method using a radial threshold determination method, and calculates a left ventricle size based on a brightness value in a cardiac magnetic resonance image, and automatically divides a left ventricular region in the cardiac magnetic resonance image according to the result of the calculation. It provides a method for automatic ventricular segmentation through radial threshold determination, which can minimize user interference and improve accuracy. The present invention detects an initial point according to a brightness value in a cardiac magnetic resonance image and detects a plurality of adjacent critical points radiated from the initial point; A binarization step of setting brightness values of the plurality of threshold points as a plurality of threshold values and repeatedly performing binarization of the cardiac MRI according to the threshold values; And calculating a left ventricle size in the cardiac magnetic resonance image and the binarized image based on the coordinate value of the critical point, and dividing the left ventricle region in the cardiac magnetic resonance image according to the calculation result.

By this configuration, it is possible to minimize the user interference rate and at the same time save the manpower, time and cost for manual contour segmentation. In addition, a person does not have to manually divide the contour, there is an effect that can increase the accuracy.

Magnetic resonance imaging, left ventricle, auto segmentation, binarization, threshold, brightness

Description

Ventricular Automated Segmentation Using Radial Threshold Determination {Method for ventricle segmentation using radial threshold determination}

The present invention relates to an automatic ventricular segmentation method using a radial threshold determination method, and calculates a left ventricle size based on a brightness value in a cardiac magnetic resonance image, and divides the left ventricle region in the cardiac magnetic resonance image according to the calculation result. The present invention relates to an automatic ventricular segmentation method through a radial threshold determination method which can minimize the interference rate and improve accuracy.

As medical technology develops gradually, mortality due to pathological diseases such as pneumonia, malnutrition and hepatitis A decreases, while mortality due to heart disease increases. The number of deaths from heart disease from 1997 to 2007 is shown in <Table 1> .The deaths from heart disease in 1997 was 10,585, while the deaths from heart disease in 2007 was 17,581, 6,998, about You can see the 70% increase.

<Table 1> Deaths due to Heart Disease by Year (Unit: Persons) (Excerpt from Statistics Korea) year 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Death toll 10,583 11,599 12,426 13,347 12,568 14,235 13,827 14,506 15,695 16,806 17,581

Regular screening of heart function is important for early diagnosis, treatment or prevention of heart disease. Magnetic resonance imaging (MRI), computed tomography (CT), and X-ray imaging can be used to diagnose cardiac function in hospitals. Computed tomography and X-rays have the disadvantage of exposing harmful ionizing radiation to patients, but magnetic resonance imaging is suitable for regular cardiac diagnosis because it uses a magnetic field and radio frequency that is harmless to the human body.

In regular clinical practice, blood flow and heart rate are calculated to analyze cardiac function, which is done by the observer (doctor or specialist) manually dividing the contour of the left ventricle from the image. However, manual contour segmentation has a high processing time and cost, and results of segmentation of the same observer are different due to differences in lighting conditions and physical conditions. Especially if the observers are different, there is a problem with the consistency of the results.

The present invention has been proposed to solve the above problems, and calculates the left ventricle size based on the brightness value in the cardiac magnetic resonance image, and divides the left ventricular region in the cardiac magnetic resonance image according to the result of the calculation, thereby minimizing user interference. The purpose of the present invention is to provide an automatic ventricular segmentation method through a radial threshold determination method that can increase accuracy.

In order to achieve the above object, the automatic ventricular segmentation method using the radial threshold determination method according to the present invention detects an initial point according to the brightness value in the cardiac magnetic resonance image, and detects a plurality of adjacent critical points radiated from the initial point. A binarization step of setting the brightness values of the plurality of threshold points as a plurality of threshold values and repeatedly performing the binarization of the cardiac MRI according to each threshold value; And calculating a left ventricle size in the cardiac magnetic resonance image and the binarized image based on the coordinate value of the critical point, and dividing the left ventricle region in the cardiac magnetic resonance image according to the calculation result.

Preferably, the brightness value of the adjacent point is specified in eight directions of the radial on the basis of the arbitrary point tP input in the cardiac magnetic resonance image, and the brightness value of the random point tP is compared with the brightness value of the adjacent point. Calculates the position of the adjacent point having the largest brightness value, repeats the process of moving the arbitrary point tP to the previously calculated position of the adjacent point, and detects the point having the maximum brightness value as the initial point mP. It may include a detection step.

In particular, the method may include a detecting step of detecting, as the critical point, a point at which the brightness values of the adjacent points in the 16-direction radially satisfy the following equation at the initial point mP.

T <MAX _value / 2.0

Here, T is the brightness value of the adjacent points in the 16 direction, MAX _value is the brightness value of the mP before the initial stage.

Preferably, the method may further include a binarization step of additionally calculating a threshold that satisfies the following equation based on the brightness values of the threshold points.

T _{v (17)} = MAX [T _{v (1)} , T _{v (2), ...,} T _{v (16)} ] × 1.1

T _{v (18)} = MAX [T _{v (1)} , T _{v (2), ...,} T _{v (16)} ] × 1.2

Here, T _{v 17} and T _{v 18} are additionally calculated thresholds, and T _{v (1)} , T _{v (2),...,} T _{v 16} are brightness values of the threshold points.

Preferably the first step of calculating the heart size based on the coordinate value of the critical point in the cardiac magnetic resonance image; And a second process of calculating an ellipse size based on coordinate values of a left ventricular corresponding region in the binarized image according to the threshold value, and a third process of selecting a left ventricular image by comparing the heart size with an ellipse size. can do.

In particular, the cardiac magnetic resonance image may include a first process satisfying the following equation based on the coordinate values of the left ventricular corresponding region in the binarized image according to the coordinate values or threshold values of the critical points.

Heart size or ellipse size = ∏ (wr ² + hr ² ) / 2

Here, wr is the horizontal radius of the heart size or ellipse size, | hx-lx | / 2, hr is the vertical radius of the heart size or the ellipse size, | hy-ly | / 2, and hx is the heart In magnetic resonance imaging, the largest value of the x-axis coordinate values of the left ventricular corresponding region in the binarized image according to the thresholds or thresholds, and lx is the left ventricular corresponding region in the binarized image according to the thresholds or thresholds in the cardiac magnetic resonance imaging. hy is the smallest value among the x-axis coordinate values, hy is the largest value among the y-axis coordinate values of the left ventricular corresponding region in the binarized image according to the thresholds or thresholds in the cardiac magnetic resonance image, and ly is the It is the smallest value of the y-axis coordinate value of the left ventricular corresponding region in the binarized image according to the thresholds or thresholds.

In particular, when the calculated ellipse size is the same as or smaller than the calculated heart size, a third process of selecting a binarized image including the corresponding ellipse size as the left ventricle image may be included.

In particular, when all of the ellipse size is more than twice the size of the heart, it may include a step of performing the left ventricle and other tissue separation process for the binarized image including the ellipse size.

Preferably, the left ventricle corresponding area is rotated with respect to the binarized image, the height of the left ventricle is estimated, and the left ventricle and the other tissue separation process for separating the left ventricle and the other tissue from the binarized image are performed based on the estimation result. It may include.

In particular, the coordinates of adjacent points in the radial 16 directions are obtained based on an arbitrary point tP input in the binarized image, and the line with the longest distance between the coordinates coincides with the x-axis of the left ventricular corresponding region in the binarized image. It may include a rotation conversion to.

Preferably, the method may include a segmentation step of dividing the left ventricle region in the binarized image by determining the calculated ellipse size part as the left ventricle.

As described above, there is an effect that can minimize the user interference rate and at the same time save manpower, time and cost for manual contouring.

 In addition, a person does not have to manually divide the contour, there is an effect that can increase the accuracy.

Examples of the automatic ventricular segmentation method through the radial threshold determination method according to the present invention can be applied in various ways, hereinafter with reference to the accompanying drawings will be described a preferred embodiment.

1 is a flowchart of an automatic ventricular segmentation method through a radial threshold determination method according to an embodiment of the present invention.

As shown in FIG. 1, the automatic ventricular segmentation method using the radial threshold determination method divides the initial point detection step (S110), the binarization step (S120), the left ventricular size calculation step (S130), and the left ventricular region of the image by the radial threshold value. It is made, including the step (S140).

Hereinafter, each step of the present invention described above will be described in detail with reference to FIGS. 2 to 11. In the following description, the left ventricular division is described as an example.

First, FIG. 2 is an example of an input image processed according to the present invention.

As shown in Figure 2, the heart is circular, it can be seen that the change in brightness value between the left ventricle and the myocardium is prominent.

The present invention intends to segment the left ventricle based on the change in brightness value between the left ventricle and the myocardium.

Initial point detection step (S110)

In the initial point detection step (S110), first, a noise reduction preprocessing process for the input cardiac magnetic resonance image is performed. Unlike the CT image, the magnetic resonance image has a lot of noise in the captured image, which may cause a large error in the segmentation of the left ventricle. Therefore, the magnetic resonance image is preprocessed through a median filter.

As described above, in order to detect an initial point and a critical point existing on the left ventricle from the cardiac MRI performed by the median filter, an arbitrary point tP in the cardiac MRI is input. The point tP obtains a residual of an image corresponding to the diastolic and systolic phases, and uses a circular Hough transform to use the center point of the detected circle as an arbitrary point tP.

As described above, an initial point detection process using the arbitrary point tP will be described with reference to FIGS. 3A to 3C.

First, a brightness value of an arbitrary point tP previously input and adjacent points in eight directions (east, west, south, north, northwest, northeast, southeast, and southwest) are compared from the arbitrary point tP. The random point tP performs a process of moving the brightness value to the largest position by comparing the brightness value with the adjacent points, and repeats this process to set the point having the maximum brightness value as the initial point mP. Detect.

If the random point tP has no position change, it is determined that the brightness value of the random point is the largest, and the random point tP is detected as the initial point mP.

As shown in Figs. 3 (a) to 3 (c), the arbitrary point tp is represented by (+), and the initial point mP is represented by a pixel point, so that even if seen with the naked eye, It can be seen that the brightness value of the initial point mP is larger than the brightness value.

Binarization step of the image by the radial threshold (S120)

The brightness value of the detected initial point mP is defined as MAX _value , the brightness value T between adjacent points in 16 directions is radially compared with respect to the initial point mP and the initial point mP, and a threshold point is detected. do.

Looking at Equation 1 for the threshold detection as follows.

[Equation 1]

T <MAX _value / 2.0

When the brightness value T with adjacent points in the radial direction is smaller than the brightness value of the initial point mP divided by the MAX _value divided by 2.0, the point corresponding to the T is determined as a threshold point, and the threshold point is detected. do. As described above, 16 threshold points are detected by calculating using Equation 1 through the brightness value T of adjacent points in the radial direction 16 and the MAX _value of the initial point mP.

4 (a) is a diagram illustrating a radial search in 16 directions radially with respect to an initial point, and FIG. 4 (b) shows 16 threshold points detected through FIG. 4 (a).

As shown in FIG. 4, it can be seen that 16 threshold points (small +) are detected based on the initial point mP (pixel point).

The brightness values of the detected 16 critical points are defined as T _{v (1)} , T _{v (2)} , .... T _{v (16)} , and the heart generated by the diversity of the imaging conditions of the magnetic resonance image. Two thresholds T _{v (17)} , T _{v (18)} , are further used to handle the internal brightness value.

At this time, the two threshold values T _{v 17} and T _{v 18 added} are calculated through the following equation (2).

[Equation 2]

T _{v (17)} = MAX [T _{v (1)} , T _{v (2)} .... T _{v (16)} ] × 1.1

T _{v (18)} = MAX [T _{v (1)} , T _{v (2)} .... T _{v (16)} ] × 1.2

Accordingly, the brightness value T of the 16 threshold points previously detected through the radial search in the 16-direction radially with respect to the initial point mP, and the two additional threshold values T _{v (17) and} T _{v (18)} calculated through Equation 2 above _. A total of 18 thresholds are detected, including.

Based on the detected 18 thresholds, the cardiac magnetic resonance images from which the initial point mP was detected are binarized.

5 is a diagram illustrating a binarized cardiac magnetic resonance image.

As shown in FIG. 5, 18 cardiac magnetic resonance images binarized based on the 18 threshold values are displayed. In addition to the left ventricle, various regions are detected during the binarization of the cardiac magnetic resonance image. Among them, only the region including the initial point mP is detected by a labeling technique, and the binarization of the detected region is performed.

Referring to FIG. 5, it can be seen that a region corresponding to the left ventricle is included in the binarized image. However, since the image does not appear at a fixed threshold, it is necessary to estimate the size of the left ventricle by using the 16 threshold points detected through the 16-way radial.

Left ventricle size calculation step (S130)

In order to calculate the size of the left ventricle, an approximate size of the left ventricle may be estimated by using the coordinate values of the 16 threshold points previously detected.

6 (a) and 6 (b) are diagrams showing a method of estimating heart size through a 16-way threshold in cardiac magnetic resonance imaging.

As shown in FIGS. 6 (a) and 6 (b), the largest value hx and the smallest value lx of the x-axis coordinate values of the 16 critical points are obtained, and the y-axis coordinate values of the 16 critical points are obtained. Find the largest value hy and the smallest value ly. Using these values and the following Equation 3, the horizontal radius wr and the vertical radius hr of the heart are calculated.

[Equation 3]

wr = hx-lx

hr = | hy-ly | / 2

The cardiac size Cardiac size is calculated using the horizontal radius wr, the vertical radius hr, and [Equation 4].

[Equation 4]

Cardiac size = ∏ (wr ² + hr ² ) / 2

In addition, the ellipse size may be calculated by using the above-described equations [3] and [4] for the corresponding region of the left ventricle among the 18 binarized images shown in FIG. 5.

6 (c) and 6 (d) are diagrams illustrating an elliptic size estimation method from binarized cardiac magnetic resonance images according to respective thresholds.

The largest value hx and the smallest value lx among the x-axis coordinate values among the corresponding regions of the left ventricle are obtained for the binarized image, and the largest value hy and the smallest value ly among the y-axis coordinate values among the corresponding regions of the left ventricle. Obtain The horizontal radius wr and the vertical radius hr of the ellipse are calculated using the values obtained as described above and [Equation 3].

The ellipse size using the binarized image can be calculated using the above [Equation 4] by using the horizontal radius wr and the vertical radius hr of the ellipse calculated as described above. The ellipse size calculation process is performed on 18 binarized images respectively binned according to different threshold values shown in FIG. 5.

Thereafter, the 18 elliptic sizes calculated as described above are compared with the previously calculated heart size. At this time, an image corresponding to an elliptic size smaller or similar to the heart size is adopted as the left ventricular image.

However, if the ellipse size exceeds 2 times the heart size, it is determined as an image recognition error, and the left ventricle and other tissues are separated from the image.

7 (a1) to 7 (a3) are diagrams showing an original image of a cardiac basal image, a diagram relating to a threshold point, and a binarized image through 18 threshold points, and FIGS. 7 (b1) to 7 ( b3) is a diagram showing an original cardiac MRI including a small left ventricle, a drawing about a critical point and a binarized image through 18 threshold points.

As shown in Figure 7 (a1) to Figure 7 (a3), in the case of the basal part of the heart in the magnetic resonance imaging, the left ventricular region is connected to other tissues and represented long, so that the myocardium does not exist left ventricle An error occurs in the partition.

Similarly, as shown in FIGS. 7B1 to 7B3, when the size of the left ventricle in the cardiac MRI is significantly small, segmentation errors occur.

However, even in this case, the largest threshold value is selected from the 18 threshold brightness values previously obtained, and the binarization of the cardiac magnetic resonance image is performed again based on the threshold value. The rotation transformation and the height of the left ventricle in the image are calculated on the binarized image, and the left ventricle in the image is separated from other tissues based on the calculation result.

Separation stage between left ventricle and other tissues

First, in order to separate between the left ventricle and other tissues in the cardiac magnetic resonance image, rotation transformation of the image is performed.

FIG. 8 is a diagram illustrating a radial shape of a 16-direction critical point for rotation transformation of a cardiac magnetic resonance image.

As shown in FIG. 8, 16-direction radiation is performed based on any point tP previously used for initial point mP detection to obtain coordinates of an adjacent critical point. By using the coordinates of the critical point, two critical points of directional symmetry are detected in the 16 directions. As the two critical points detected as described above are linearly connected to each other, a total of eight straight lines are generated.

The longest of the eight straight lines is selected, the corresponding straight direction is set in a direction in which the left ventricle is connected to another tissue, and the left ventricle is aligned so that the x-axis direction of the corresponding left ventricular region in the binarized image coincides. Rotate the corresponding area.

9 (a) and 9 (b) are diagrams showing an image before rotation conversion of the cardiac magnetic resonance image and an image after rotation conversion of 45 degrees.

9 (a) and 9 (b), when the left ventricular corresponding region is rotated through the rotation conversion process described with reference to FIG. The image as shown in b) appears.

In addition, a process of calculating the height of the left ventricle is performed based on the random point tP in the rotated image.

9 (c) to 9 (e) are views illustrating images in which the height of the left ventricle is calculated and images in which the left ventricle is separated from other tissues.

As shown in FIG. 9 (c), the height h of the upper and lower ends is measured about the arbitrary point tP in the left ventricular corresponding region. The arbitrary point tP is moved to the left by a quarter of the height h based on the measured height h.

Subsequently, the height of the left ventricular corresponding region is continuously estimated from the moved position, and the arbitrary point tP is moved to the right.

The gray part shown in Fig. 9 (d) is a region in which the arbitrary point tP is moved to the right. Thus, the largest height of the estimated computed left ventricular corresponding area is defined as the radius of a circle for separating from the left ventricle and other tissues.

Accordingly, when the arbitrary point tP is the center of the circle and a circle is formed based on the defined radius, as shown in FIG. 9 (d), an image including the left ventricular region is generated in the circle.

Left ventricular zone division step (S140)

FIG. 10 is a view showing the left ventricular region divided when the left ventricle is not connected to another tissue.

As shown in FIG. 10, the image having an ellipse size calculated before or smaller than or equal to the heart size is divided into the left ventricle by determining that the left ventricle is not connected to other tissues.

11 is a view showing a magnetic resonance image of the left ventricle separated from other tissues.

As shown in FIG. 11, when the left ventricle is connected to another tissue, the left ventricle is divided through a separation process between the left ventricle and the other tissue in the image.

In this way, blood flow can be calculated using the segmented left ventricle region in the cardiac MRI.

10 to 11, the blood flow rate may be calculated by summing the number of pixels included in the left ventricular region and multiplying the weight of the volume per pixel present in the header of each cardiac MRI. . In the blood flow calculation, the volume per pixel specifies a shooting interval during cardiac imaging, and thus may have different values for each of the MR images.

Example of Experiment Results

SSFP scanned data for a total of 36 subjects were used as experimental data using a General Electronics Sigma 1.5T scanner. The age ranged from 14 to 77 years with a mean age of 48 years. Shooting parameters are TR 3.3-4.5ms, TE 1.1-2.0ms, flip angle 55-60, image size 256 × 256, receiver bandwidth 125kHz, FOV 290-400 × 240-360, slice thickness 6-8mm, and slice gap 2 -4mm. The left ventricle of each subject was taken with 6-10 images for 20-28 heart phases. In this experiment, 600 images were used for each subject.

In order to quantitatively verify the performance of the proposed segmentation method, we compared and compared the manual contour segmentation data of the observer considered as a standard with the general electronics MASS 6.0 commercial software. Manual contour detection was performed by experts in the 8th and 3rd year of CMR, and was divided to not include papillary and fibrous muscles. We also compared the MASS software with user interference rates.

Comparative analysis of blood flow rate, heart rate and user interference rate

_Blood flow rate was calculated for the subject's diastolic and systolic images, and the results of heart rate (EF) calculated as (D _blood -S _blood ) / D _blood × 100 can be found in [Table 2]. The results of comparing the proposed segmentation method and MASS software with manual contour segmentation data can be found in [Table 3].

[Table 2] Results of blood flow and heart rate using manual contour segmentation, proposed algorithm, MASS software

Manual Contour Split (mL) Proposed Algorithm (mL) MASS Software (mL) Case Diastolic
Blood flow Systolic
Blood flow Heart rate Diastolic
Blood flow Systolic
Blood flow Heart rate Diastolic
Blood flow Systolic
Blood flow Heart rate One 146.5 65.2 55.5 148.95 68.28 54.16 173.40 83.10 52.08 2 116.3 42.5 63.5 111.70 37.98 66.00 125.48 50.52 59.74 3 190.4 76.4 59.9 194.13 85.22 56.10 218.34 102.27 53.16 4 215.2 93.2 56.7 214.81 105.63 50.83 241.94 119.78 50.49 5 93.3 31.8 65.9 97.28 33.50 65.56 106.14 38.89 63.36 6 162.8 97.6 40.0 158.92 91.10 42.68 178.47 108.85 39.01 7-29 ~ 30 127.2 49.4 61.2 127.99 55.43 56.69 147.49 62.08 57.91 31 277.4 215.7 22.2 289.52 229.54 20.72 314.66 250.35 20.44 32 130.3 54.3 58.3 128.41 55.55 56.74 151.69 72.60 52.14 33 231.4 170.5 26.3 213.16 153.41 28.03 250.61 185.26 26.08 34 72.2 24.7 65.8 68.89 21.91 68.20 84.05 25.57 69.58 35 130.7 83.3 36.3 127.41 86.00 32.50 150.87 100.46 33.41 36 177.0 88.0 50.3 146.99 75.95 48.33 189.51 92.86 51.00

In the proposed segmentation method, the absolute errors of diastolic, systolic, and heart rate were 6.20 mL, 5.87 mL, and 3.13%, respectively, and the margins of error were -3.19 to 3.00 mL, -1.58 to 4.29 mL, and -2.56 to 0.57%, respectively. Appeared. In the MASS software, the absolute errors of diastolic, systolic, and heart rate were 19.55 mL, 12.21 mL, and 3.23%, respectively, and the error ranges were 0.00 ~ 19.55mL, 0.00 ~ 12.21mL, and -2.98 ~ 0.24%, respectively. The largest error in the results of the MASS software was Case 31 diastolic blood flow +37.26 mL, and the largest error in the proposed segmentation method was Case 31 diastolic blood flow -30.01 mL.

Overall, the error of the MASS software in diastolic and systolic was large because the MASS software lacked the ability to segment in consideration of papillary and fibrous muscles. Considering the papillary and fibrous muscles, the error may be reduced, but it is difficult to obtain better results than the proposed segmentation method. This is because the proposed segmentation method is similar to the manual contour segmentation data than MASS software.

[Table 3] Proposed segmentation method and error with manual contour segmentation data of MASS software

Proposed Algorithm (mL)
(Proposal Algorithm-Manual Contour Split) MASS Software (mL)
(MASS Software-Manual) Case Diastolic blood flow and error Systolic Blood Flow Error Heart rate
Rescue rate error Diastolic blood flow and error Systolic Blood Flow Error Heart rate
Rescue rate error One 2.45 3.08 -1.34 26.90 17.90 -3.42 2 -4.60 -4.52 2.54 9.18 8.02 -3.72 3 3.73 8.82 -3.77 27.94 25.87 -6.71 4 -0.39 12.43 -5.86 26.74 26.58 -6.20 5 3.98 1.70 -0.35 12.84 7.09 -2.56 6 -3.88 -6.50 2.63 15.67 11.25 -1.04 7-29 ~ 30 0.79 6.03 -4.47 20.29 12.68 -3.25 31 12.12 13.84 -1.52 37.26 34.65 -1.80 32 -1.89 1.25 -1.59 21.39 18.30 -6.19 33 -18.24 -17.09 1.71 19.21 14.76 -0.24 34 -3.31 -2.79 2.41 11.85 0.87 3.79 35 -3.29 2.70 -3.77 20.17 17.16 -2.85 36 -30.01 -12.05 -1.96 12.51 4.86 0.72

[Table 4] Analysis of user interference rate

Proposed algorithm MASS Software Base image 0 out of 72 videos (0%) 31 of 72 videos (43.1%) Other video 0 of 563 videos (0%) 63 of 563 videos (11.2%)

In the proposed segmentation method and MASS software, if the segmentation error occurs, user intervention is indispensable for the diagnosis of heart disease. In other words, if the division is wrong, the user must divide manually to correct it. The results of the user interference rate are summarized in [Table 4]. As shown in the table, the user interference rate of MASS software is high. In addition to the MASS software, Lee et al. Existing studies, such as [1], also require most user intervention on the basal image between the left ventricle and other tissues. On the other hand, the proposed segmentation method improves automation performance by minimizing user interference through the separation process of the left ventricle from other organizations.

12 is a graph showing diastolic blood flow, systolic blood flow, and heart rate for 36 subjects.

As shown in FIG. 12, it can be seen that the proposed segmentation method has a profile that is more similar to the manual contour segmentation data than the MASS software.

In some cases, the performance of the MASS software is better, because the error is further reduced because of the user's intervention due to the segmentation error of the MASS software. Therefore, it can be seen that the proposed segmentation method is superior to MASS software in consideration of blood flow, heart rate and user interference rate.

As described above, there is an effect that can minimize the user interference rate and at the same time save manpower, time and cost for manual contouring.

In addition, a person does not have to manually divide the contour, there is an effect that can increase the accuracy.

The automatic ventricular segmentation method through the radial threshold determination method according to the present invention has been described above. Such technical configuration of the present invention can be understood by those skilled in the art that the present invention can be implemented in other specific forms without changing the technical spirit or essential features of the present invention.

Therefore, the embodiments described above are intended to be illustrative in all respects and not to be considered as limiting, and the scope of the present invention is indicated by the following claims rather than the foregoing description, and the meaning of the claims And all changes or modifications derived from the scope and equivalent concept thereof should be construed as being included in the scope of the present invention.

1 is a flowchart of an automatic ventricular segmentation method through a radial threshold determination method according to an embodiment of the present invention.

2 is a view showing an original image of a cardiac magnetic resonance image.

3 is a diagram illustrating an initial point detection process of an automatic ventricular segmentation method through a radial threshold determination method according to an embodiment of the present invention.

4 is a diagram illustrating a threshold detection process of an automatic ventricular segmentation method through a radial threshold determination method according to an embodiment of the present invention.

FIG. 5 is a diagram illustrating binarized cardiac magnetic resonance (ARM) images of an automatic ventricular segmentation method through a radial threshold determination method according to an exemplary embodiment of the present invention.

6 (a) and 6 (b) show a method for estimating heart size through a 16-way critical point, and FIGS. 6 (c) and 6 (d) show a method for estimating heart size from binarized cardiac MR images. to be.

7 (a1) to FIG. 7 (a3) are diagrams illustrating an original image of a cardiac basal image, a diagram relating to a threshold point, and a diagram illustrating binarized images through 18 threshold points.

7 (b1) to 7 (b3) are views showing an original cardiac MRI including a small left ventricle, a diagram relating to a threshold point, and a diagram showing binarized images through 18 threshold points.

8 is a diagram illustrating a radial form of a 16-direction threshold point for rotation transformation of a cardiac magnetic resonance image.

9 (a) to 9 (e) illustrate images before and after rotational conversion of cardiac magnetic resonance images, images after 45 degree rotational transformation, images in which the height of the left ventricle is calculated, and images in which the left ventricle is separated from other tissues. .

10 is a diagram illustrating cardiac magnetic resonance images binarized by a critical point.

11 is a view showing a magnetic resonance image of the left ventricle separated from other tissues.

12 (a) to 12 (b) are graphs showing the diastolic blood flow, systolic blood flow, and heart rate, respectively, using the present invention, manual contouring, and MASS software, respectively.

Claims (11)

Detecting an initial point according to a brightness value in a cardiac magnetic resonance image and detecting a plurality of adjacent critical points radiated from the initial point; A binarization step of setting brightness values of the plurality of threshold points as a plurality of threshold values and repeatedly performing binarization of the cardiac MRI according to the threshold values; Calculating a left ventricle size in the cardiac magnetic resonance image and the binarized image based on the coordinate value of the critical point; And dividing the left ventricle region in the cardiac magnetic resonance image according to the calculation result. The method of claim 1, The detecting step Calculating brightness values of adjacent points in eight radial directions on the basis of an arbitrary point tP input in the cardiac magnetic resonance image, Comparing the brightness value of the random point tP with the brightness value of the adjacent point to calculate the position of the adjacent point having the largest brightness value, and moving the random point tP to the calculated position of the adjacent point. Iteratively perform, automatic ventricular segmentation method using a radial threshold determination method characterized in that for detecting the point having the maximum brightness value as the initial point mP. The method of claim 2, The detecting step Ventricular automatic segmentation method using a radial threshold determination method characterized in that for detecting the point where the brightness value of the adjacent points in the radial direction 16 in the radial direction at the initial point mP satisfies the following equation. T <MAX _value / 2.0 Here, T is the brightness value of the adjacent points in the 16 direction MAX _value is the brightness _value of the initial point mP The method of claim 1, The binarization step is On the basis of the brightness value of the threshold points, ventricular automatic segmentation method through a radial threshold determination method, characterized in that additionally calculating a threshold that satisfies the following equation. T _{v (17)} = MAX [T _{v (1)} , T _{v (2), ...,} T _{v (16)} ] × 1.1 T _{v (18)} = MAX [T _{v (1)} , T _{v (2), ...,} T _{v (16)} ] × 1.2 Here, T _{v (17)} and T _{v (18)} is a threshold value that is additionally calculated T _{v (1)} , T _{v (2),...,} T _{v (16)} are brightness values of the threshold points. The method of claim 1, The operation step Calculating a heart size based on a coordinate value of a critical point in the cardiac magnetic resonance image; A second process of calculating an ellipse size based on a coordinate value of a left ventricular corresponding region in the binarized image according to the threshold value; And a third process of selecting a left ventricular image by comparing the heart size with an ellipse size. The method of claim 5, The first process The following equation is satisfied based on the coordinate values of the left ventricular corresponding region in the binarized image according to the coordinate values of the critical points or the threshold values in the cardiac magnetic resonance image. Heart size or ellipse size = ∏ (wr ² + hr ² ) / 2 Here, wr is the horizontal radius of the heart size or the ellipse size, | hx-lx | / 2, hr is the vertical radius of the heart size or ellipse size, | hy-ly | / 2, hx is the largest value of the x-axis coordinate value of the left ventricular corresponding region in the binarized image according to thresholds or thresholds in the cardiac magnetic resonance image, lx is the smallest value of the x-axis coordinates of the left ventricular corresponding region in the binarized image according to thresholds or thresholds in the cardiac magnetic resonance image, hy is the largest value of the y-axis coordinate values of the left ventricular corresponding region in the binarized image according to thresholds or thresholds in the cardiac magnetic resonance image,  ly calculates the smallest value of the y-axis coordinates of the left ventricular corresponding region in the binarized image according to thresholds or thresholds in the cardiac magnetic resonance image. Automated ventricular segmentation method using radial threshold determination method characterized in that the calculation of the heart size or ellipse size. The method of claim 5, The third process is When the calculated ellipse size is equal to or smaller than the calculated heart size, the ventricle automatic segmentation method using the radial threshold determination method, characterized in that for selecting the binarized image including the ellipse size as a left ventricle image. The method of claim 5, The operation step When the elliptic size is more than twice the size of the heart, the ventricle automatic segmentation method using a radial threshold determination method characterized in that the left ventricle and other tissue separation process for the binarized image including the ellipse size. The method of claim 8, The left ventricle and other tissue separation process Rotating the left ventricular corresponding region with respect to the binarized image, estimating the height of the left ventricle, and then separating the left ventricle and the other tissue from the binarized image based on the estimation result. Automatic segmentation method. 10. The method of claim 9, The rotation conversion Acquiring coordinates of adjacent points in a radial 16 direction based on an arbitrary point tP input in the binarized image, and matching the x axis of the left ventricular corresponding region in the binarized image with a straight line having the longest distance between the coordinates. Ventricular automatic segmentation method through the radial threshold determination method characterized in that. The method of claim 7, wherein The dividing step And dividing the left ventricle region in the binarized image by determining the calculated ellipse size as the left ventricle.

Images