KR20120050595A - Automatic left ventricle segmentation method - Google Patents

Abstract

PURPOSE: A method for automatically segmenting a left ventricle is provided to automatically segment the left ventricle by a graph searching and a k-average clustering about an MRI(Magnetic Resonance Imaging). CONSTITUTION: An initial left ventricle region is obtained. Basic left ventricle information is calculated based on the obtained initial left ventricle region. A boundary of a heart is linearized by polar coordinate conversion of a circular heart MRI. A left ventricle region is segmented by applying a k-clustering method based on the brightness of each pixel in a polar coordinate image. A segmentation error is corrected by erasing an excessively segmented region through a graph search based algorithm.

Description

Automatic left ventricle segmentation method

The present invention relates to a left ventricular segmentation method, and more particularly, to a method of automatically segmenting a left ventricular region in a cardiac short magnetic resonance image.

Due to the development of medical science and technology, the mortality rate from diseases and accidents is decreasing, but the mortality rate from heart related diseases is continuously increasing. According to the National Statistical Office, the number of deaths from heart-related diseases increased by 20.1% over the five years from 16,892 in 2003 to 21,102 in 2008, as shown in FIG.

In order to prevent heart disease, it is important to analyze and observe the function of the heart through regular checkups along with ongoing care. Analysis of cardiac function is made by assessing motor performance through calculation of blood volume and ejection fraction between the diastolic and systolic phases. Diastolic and systolic images for cardiac function analysis can be taken using computed tomography (CT), ultrasound, and X-rays, but magnetic resonance imaging is often used in clinical practice because they utilize radio frequencies and magnetic fields that are harmless to the human body. do. However, the analysis of the heart function using the captured images is time consuming because most of the work is performed by hand, and the variability of the result according to the observer becomes a problem. Therefore, research on automation of cardiac function evaluation through computer algorithm has been continuously conducted.

The automatic segmentation of the left ventricle using a computer includes a typical image segmentation technique, a graph-based segmentation method, a method using a dynamic contour model, and a level set-based algorithm.

Typical image segmentation techniques include region expansion, binarization by threshold determination, and image classification. Traditional image segmentation techniques are useful for determining fibrosus muscles and papillary muscles, but rely on intuition and experience to determine thresholds. In addition, left ventricular segmentation performance is poor in basal or apex images with blurred myocardial and intraventricular boundaries.

The graph-based segmentation algorithm generates a graph by considering each pixel of the image as a node, calculates a visit cost for each node, and performs segmentation by detecting a minimum cost path through a graph search algorithm. Graph-based segmentation algorithms show good overall performance, but are susceptible to interference by cardiac external structures and are unable to reflect complex fibrosus muscles or papillary muscles.

The dynamic contour model performs segmentation by detecting the boundary line by minimizing the energy of the external and internal forces of the object. It uses rigidity and elasticity as internal force. The external force is used to change the brightness value. In the dynamic contour model, problems are likely to occur in low-contrast images, and the performance difference of the algorithm depends on the initial contour setting.

Algorithms using the levelset segmentation technique have recently been studied and are effectively used to segment objects in noisy images. The levelset segmentation technique performs image segmentation through an iterative calculation process, but it is difficult to determine the repetition end condition, a large amount of computation is a problem, and like the dynamic contour model, it requires initialization close to the object to be segmented.

As such, due to the importance of the medical field and the enormous marketability, various studies on the left ventricular segmentation algorithm for cardiac function analysis using a computer have been conducted at home and abroad. However, the algorithms studied to date have problems such as execution time, increase of observer interference rate, or low accuracy due to complex heart structure and irregular surrounding structure, and the solution of this problem is important.

SUMMARY OF THE INVENTION The present invention has been made to improve the above-mentioned conventional problems, and an object thereof is to provide a method for automatically segmenting the left ventricle through K-means clustering and graph search for cardiac magnetic resonance images.

The automatic left ventricular segmentation method according to the present invention uses (i) the left ventricular internal region coordinates as an initial point of the left ventricular segmentation algorithm for the corresponding image in the MRI set of the heart received from the observer. Acquiring an initial left ventricular region through region expansion; (ii) calculating basal left ventricular information based on the acquired initial left ventricular region; (iii) performing polar coordinate transformation on a circular magnetic resonance image to generate a polar image to linearize the boundary of the heart; (iv) dividing the left ventricular region by applying a K-clustering technique to each pixel in the polar coordinate image based on the fact that the pixel brightness of the myocardial region is darker than the pixel brightness inside the ventricle in the magnetic resonance image. ; And (v) correcting the segmentation error by erasing the over-divided region from the base image using a graph search based algorithm using the circularity of the left ventricle as a result of applying the K-clustering technique. .

In the present invention, a computer algorithm for automatically dividing the left ventricular region from a short-term magnetic resonance image of a heart and calculating the blood flow rate and the heart rate rescue rate for each of the diastolic and systolic systems is proposed. In addition, the algorithm of the present invention was compared with manual edge detection by experts and GE MASS commercial software. According to the results, the algorithm of the present invention showed high segmentation accuracy, and in particular, it was possible to reduce user interference, which was a problem in previous studies, while maintaining the segmentation accuracy. (Originally, we could reduce the user interference, which was a problem in previous studies, while maintaining the accuracy of segmentation.)

1 is a graph showing the number of heart-related disease deaths for each year published by the National Statistical Office.
2 is a diagram illustrating an automatic left ventricle segmentation method using a K-means clustering technique and graph search according to an embodiment of the present invention.
3 is a diagram illustrating an image in which an error occurs in setting a polar coordinate origin according to an exemplary embodiment of the present invention.
4 is a diagram illustrating a polar coordinate image generated by correcting an error according to an exemplary embodiment of the present invention.
FIG. 5 is a diagram illustrating a minimum cost path calculated through brightness value variation data, path cost accumulation data, and reverse search according to an exemplary embodiment of the present invention.
6 are graphs showing linear regression plots for blood flow and heart rate rescue rates in the diastolic and systolic stages of the present invention and MASS software.
7 is a Bland-Altman diagram of diastolic and systolic blood flow and heart rate rescue as compared to manual contour detection.
8 and 9 illustrate a result of comparing the image of the left ventricular segment with the image of the left ventricle through the automatic contour detection using the image of the automatic segmentation of the left ventricle using the algorithm of the present invention.

The operation principle of the preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings. In the following description of the present invention, detailed descriptions of well-known functions or configurations will be omitted if it is determined that the detailed description of the present invention may unnecessarily obscure the subject matter of the present invention. The following terms are defined in consideration of the functions of the present invention, and may be changed according to the intentions or customs of the user, the operator, and the like. Therefore, the definition should be based on the contents throughout this specification.

Hereinafter, a method for automatically dividing left ventricle according to an embodiment of the present invention will be described with reference to the accompanying drawings. 2 is a diagram illustrating an automatic left ventricle segmentation method using a K-means clustering technique and graph search according to an embodiment of the present invention.

Algorithm start initial point acquisition (step 1)

In the MRI set of the heart, the left ventricular coordinates of the left ventricle are input to the intermediate image between the diastolic and the systole and used as the seed point for the left ventricular segmentation algorithm.

In order to minimize the user interference rate, the initial point of the algorithm for the non-intermediate image is modified by applying the initial point propagation algorithm of Noel et al. The initial point propagation algorithm sets an 11x11 window from the left ventricular center of gravity of the previous image, and the brightness of the pixels in the window, the average (μ _before ) and standard deviation ( _before σ) of the previous image, and the distance from the center of gravity. In consideration of the following, the energy E (p) of each pixel is calculated using Equation 1 as follows.

P is the coordinate of each pixel existing on the window, _then I (p) is the brightness value of the pixel P. P _CoG means the center of gravity of the left ventricular region divided in the previous image, w is 11 as the set window size. μ and σ _old and _before means the mean and standard deviation of intensity values of the left ventricular area in each of the previous image. A coordinate having the smallest energy value in the matrix E (p) modeled according to Equation 1 is set as an initial point of the next image.

Calculation of basic left ventricle information (second stage)

The initial left ventricular region LV _rgnl is obtained using the coordinates specified through the above method as the region extension initial point P _rginit . The initial left ventricular region is calculated by a region expansion technique, and is included in the region when the rate of decrease in brightness with the adjacent pixel is within 5% in the process of expanding the region starting from the initial point. The 5% drop in brightness did not exceed the myocardial region for 634 images of 38 data sets tested under strong constraints.

Magnetic resonance images inevitably cause distortion of image brightness values according to coil positions. Since this distortion affects the cardiac segmentation, the image distortion according to the coil position is obtained by _leveling the three-dimensional plane calculated by the least-square method using Lee's al. Correct it to obtain the distortion correction image IMG _correct .

The region expansion technique is re-executed on the distortion-corrected image to segment the initial left ventricular region LV _rgnl and calculate information of the ventricular region as follows: (1) left ventricular position and center of gravity in magnetic resonance imaging, (2) left ventricle Approximate shape and radius, (3) statistics such as brightness mean and standard deviation, and (4) pixel volume information to determine whether there is a segmentation error in the base image. The information calculated in this way is used in a region of interest setting, polar coordinate transformation of a left ventricular region, and an initial point propagation algorithm.

Polar coordinate image conversion (3rd step)

In cardiac magnetic resonance imaging, the left ventricular region and the myocardial region are generally circular. In general, a method of recognizing a pattern of linear or directional pattern has an advantage in terms of accuracy or complexity, rather than a method of recognizing a circular object in an image processing algorithm. Therefore, the proposed algorithm generates polar coordinates by performing polar coordinate transformation on a circular cardiac magnetic resonance image to linearize the boundaries of the heart, segment the left ventricle through K-means clustering, and correct segmentation errors based on graph search. It was.

In the polar coordinate image conversion process, normal segmentation and segmentation error are estimated for the initial left ventricular region LV _rgnl , and when the segmentation error occurs, the transformation is performed by correcting the polar origin.

Fixed polar origin error

The basal image of the short-term cardiac MRI includes a portion where the left ventricle is connected to other tissues, and thus, the boundary between the ventricle and the myocardial region is ambiguous or connected to other tissues so that the myocardium does not appear. Therefore, in order to solve this problem, existing studies have inevitably generated user interference.

The initial left ventricular region calculated in the first step of the present invention also differs from the actual left ventricular region, including other tissues. For this reason, when the center of gravity of the initial left ventricular region is used as the origin of the polar coordinate transformation, the left ventricular region may not be normally divided. Accordingly, the polarity origin error is corrected by determining the segmentation error of the initial left ventricular region using the basic left ventricle information acquired from the previous image.

The size of the initial left ventricular region of the image in which the polar coordinate origin error appears is significantly increased compared to the size of the initial left ventricular region of the previous image. Experimental results show that the average increase in pixel volume is 1.1 and the standard deviation is 0.18 in the case of normal segmentation, whereas the average increase in pixel volume is 1.8 to 4.0 in the segmentation error. Therefore, in the method of the present invention, when the increase rate of the pixel volume is 1.5 or more, it is regarded as an image in which segmentation error occurs and is replaced by the left ventricular region center of gravity coordinates of the previous segment.

3 and 4 show examples of an image in which an error has occurred in the polar origin setting and a polar image generated by correcting the error. If the polar coordinate origin is not corrected for the image showing the segmentation error, it can be confirmed that the coordinate of the origin is not located inside the left ventricle on the Cartesian coordinate system or the linearity of the converted polar coordinate system is inferior, which is related to the segmentation performance of the algorithm.

Polar coordinate image generation through polar coordinate transformation

In order to perform the polar transformation, the origin of the polar coordinate system and the polar transformation range are determined. The center of gravity P _CoG of the basal left ventricle information calculated in the second step was used as the origin position of the polar coordinate system. If it is determined as a division error as described above, the modified polar coordinate origin is used. In polar coordinate transformation, a polar coordinate image is obtained by converting a coordinate P (x, y) of a Cartesian coordinate system into a coordinate P (r, θ) of a polar coordinate system. The radius at which the Cartesian coordinate system is converted to the polar coordinate system uses three times the radius value of the basal left ventricle information to include the myocardium and part of the surrounding structure. The relation between the coordinates of the Cartesian coordinate system and the coordinates of the polar coordinate system is shown in Equation 2 below.

Left Ventricular Segmentation Based on K-Means Clustering (Stage 4)

In magnetic resonance imaging, the pixel brightness of the myocardial region is darker than the pixel brightness inside the ventricles. Classified. The K-means clustering algorithm is a technique mainly used for clustering of data mining and can classify data without prior knowledge of each class to be classified. Each record consisting of N attributes is represented as a vector and displayed in an N-dimensional data space, and the straight distances in the data space are calculated to classify the records at close distances as having similar characteristics.

If K, the number of clusters and the number of clusters to be classified, is input as a parameter, in the first step of performing the K-means clustering algorithm, K records are randomly selected from the x records and designated as the center value of each cluster. Then, the cluster to which the remaining records belong is determined, where the cluster to which the records belong is the cluster to which the center value closest to the position of the record in the data space belongs. When clustering records, the median value of each cluster is calculated and set as the center value, and the cluster to which the records belong is determined again. This process is repeated until there are no changes in the clusters being classified.

In magnetic resonance images, shadows are generated due to distortion according to coil positions, and distortion characteristics remain even when distortion correction is performed. The effects of this distortion result in partial over-segmentation or under-segmentation. In addition, if clustering is performed on a line basis, an empty cluster or an unbalanced cluster may occur due to a randomly determined initial center value, and the processing takes a long time because the clustering algorithm must be performed by the number of lines. To alleviate this problem, clustering was performed by dividing the polar coordinates into groups of 30 degrees.

Fix graph navigation based segmentation errors (step 5)

The left ventricular region segmented by performing the K-means clustering algorithm on the polar coordinate image may include an external ventricular region. Such segmentation error occurs mainly in the basal image, and the external ventricular region included in the segmentation region is mainly connected to other tissues in the left ventricle. Therefore, by using the circularity of the left ventricle, an oversegmented region is erased from the base image through a graph search based segmentation error correction algorithm. Segmentation error correction is handled in the following way.

Generation of brightness data change in left ventricle

An image representing a change in pixel brightness value in the horizontal direction in the left ventricular region divided in the fourth step is calculated. Brightness value variation data IMG _diff (x, y) is generated through the following equation (3). Since the segmented left ventricle image is a binary image, three values of -1, 0, and 1 appear as pixel values of the variation data.

Vertical least cost route navigation

The minimum cost path connecting the top and bottom of the brightness change data image is calculated by applying a dynamic search technique in the vertical direction from the generated brightness change data. Each pixel of the image is defined as a node, and the change amount of the node is defined as a path cost, and the distance between pixels is defined as a cost of a unit path connecting each node. In order to find the minimum cost path, the path cost matrix Pmap (x, y), which is modeled by the following equation, is obtained using the brightness value variation data.

로 결정하였다. Pmap(x,y)의 한 픽셀의 경로 비용 누적치는 IMG_diff에서 동일한 좌표에 위치하는 픽셀의 밝기값과 Pmap(x,y) 상에서 바로 위까지 계산된 3개의 픽셀 중 최소값을 가지는 픽셀을 합한 값이 된다.Pmap (x, y) is a matrix representing the minimum cost to reach each node, and IMG _diff is a brightness value variation data generated in the fifth step. ω is used to weight the path between nodes connected at the vertical top pixel, depending on the diagonal to vertical ratio

Determined. The path cost cumulative value of one pixel of Pmap (x, y) is the sum of the brightness of the pixels at the same coordinates in IMG _diff plus the pixel with the minimum of the three pixels computed just above on Pmap (x, y). Becomes

FIG. 5 illustrates an example of a brightness value change amount image generated in a segmented image, an image of calculating a path cost to a corresponding pixel, and a minimum cost path calculated by the above method.

At the bottom of the path cost matrix, the nodes that were visited to reach the coordinates from the minimum value were searched backward to extract the path having the minimum cost, and the extracted minimum cost path was estimated as the ventricular inner wall information in the case of normal division. Finally, the image obtained by correcting the segmentation error is obtained through the AND operation of the data generated by applying the estimated data and the clustering algorithm.

Segmentation Image Inversion and Left Ventricular Volume Computation

After performing left ventricular segmentation on polar coordinates, the volume of the left ventricle was calculated by inverting the segmentation result into a Cartesian coordinate system. The coordinates of the Cartesian coordinate system are estimated to be mapped to the coordinate system of the polar coordinates through the polar coordinate transformation, and the left ventricle shape is restored from the original image through the inverse transformation process of returning the pixels in the polar coordinate coordinates to the Cartesian coordinate system. The inverse transform is shown in Equation 5 below.

After reconstructing the shape of the left ventricle on the Cartesian coordinate system through the inverse transformation process, the volume of the left ventricular region is obtained.

Experimental data

The cardiac magnetic resonance images used to verify the proposed algorithm were taken by SSFP scan of 38 volunteers using GE Medical Systems' SIGNA 1.5T scanner with the consent of the volunteers. The parameters used for magnetic resonance imaging were TR 3.3-4.5ms, TE 1.1-2.0ms, flip angle 55-60, image size 256 ㅧ 256, receiver bandwidth 125kHz, FOV 290-400 ㅧ 240-360, slice thickness 6- 8, the slice gap was set to 204mm. The left ventricle of each subject was photographed with 6 to 10 slices each of systolic and diastolic phases for 20 to 28 heart phases.

The proposed algorithm is compared with 38 data sets (total of 634 images), MASS 6.0 commercial software from General Electronics, and manual contour detection by experts.

Blood flow and heart rate

The left ventricle was segmented using the proposed algorithm and GE MASS software for diastolic and systolic images, and blood flow and heart rate were calculated. The calculated results were compared with the result of segmentation through manual contour detection by an expert.

In the manual contour detection, the left ventricular blood flow was 145.0 mL 8,6 and 61.9 mL 4, respectively, and the mean heart rate was 60.5% 13.8. On the average, the left ventricular blood volume divided by the proposed algorithm was 146.7 mL 49.4 and 61.8 mL 43.6, respectively, in the diastolic and systolic stages, and the average heart rate was 60.9% ㅁ 13.4. In the MASS software, the mean blood flow during the diastolic and systolic periods was 164.5 mL and 55.1 and 73.2 mL and 51.5, and the heart rate was 58.8% and 14.3. In the case of the MASS software, the error is larger because the algorithm cannot precisely divide complex heart structures such as fibrous muscles and papillary muscles.

Table 1 and FIG. 6 show a comparison of blood flow rate and heart rate rescue rate calculated by manual contour detection, proposed algorithm, and MASS software for 38 data. 6 are graphs showing linear regression plots for blood flow and heart rate rescue of diastolic and systolic according to the present invention (left view) and MASS software (right view). As a result of calculating the volume of the left ventricle by the method of the present invention, it can be confirmed that the result of the manual contour detection is similar to the result of calculating the volume of the left ventricular region using MASS software. The method of the present invention shows higher accuracy than MASS software in segmenting the left ventricular region to reflect myocardial morphology due to fibromyalgia, papillary muscle or heart disease. High correlation coefficients of R ² = 0.99 and R ² = 0.98 were found in blood flow and heart rate of diastolic and systolic.

FIG. 7 represents the relative error of the values of the two data compared with the Bland-Altman scheme (left figure: present invention, right figure: MASS software) for blood flow and heart rate. As shown in the figure, the proposed algorithm shows high similarity with the manual contour detection data. 8 and 9 compare and compare the image of the left ventricular inner wall (first drawing) by using the method of the present invention (second drawing) with the image of the left ventricle divided by manual contour detection (third drawing). It was. Overall, the results are similar to the manual contour detection results.

User interference rate

There should be no significant errors in the segmented left ventricle to calculate blood flow and heart rate using cardiac MRI. However, in the existing studies and MASS software, if the basal image or the heart shape that the ventricles are connected to other tissues is abnormal, segmentation errors occur, which inevitably requires user intervention.

In this experiment, we measured the user interference rate to correct the errors in the segmented results using the proposed algorithm and MASS software for 634 images of 38 data sets, summarized in Table 2. As shown in the table, the MASS software was measured to require user intervention on 28 data for 38 volunteers. However, in the case of the software of the present invention, only two pieces of data are required for user interference, thereby maintaining high accuracy of segmentation and minimizing user interference rate.

Although the present invention has been described as a specific preferred embodiment, the present invention is not limited to the above-described embodiments, and the present invention is not limited to the above-described embodiments without departing from the gist of the present invention as claimed in the claims. Anyone with a variety of variations will be possible.

Claims (10)

(i) The initial left ventricular region was obtained by region expansion using the left ventricular internal coordinate as the initial point of the left ventricular segmentation algorithm for the image in the middle image between the diastolic and the systole in the magnetic resonance image set from the observer. Obtaining;
(ii) calculating basal left ventricular information based on the acquired initial left ventricular region;
(iii) performing polar coordinate transformation on a circular magnetic resonance image to generate a polar image to linearize the boundary of the heart;
(iv) dividing the left ventricular region by applying a K-clustering technique to each pixel in the polar coordinate image based on the fact that the pixel brightness of the myocardial region is darker than the pixel brightness inside the ventricle in the magnetic resonance image. ; And
and (v) correcting the segmentation error by erasing the over-divided region in the base image through the graph search based algorithm using the circularity of the left ventricle as a result of applying the K-clustering technique. The method according to claim 1, wherein step (i)
Setting a 11 × 11 window from the left ventricle region center of gravity of the previous image by an initial point propagation algorithm;
Calculating the energy E (p) of each pixel in consideration of the brightness of the pixels in the window, the average value of the brightness values of the previous image (μ _Prev ) and the standard deviation (σ _Prev ), and the distance from the center of gravity; And
And setting the coordinate having the smallest energy value in the modeled matrix E (p) as an initial point of the next image. The method according to claim 2,
The energy E (p) of each pixel is

Calculated by, and the coordinates of each pixel, _{and then} I (p) to the above equation, P is present in the window is the brightness value of the pixel P. P _CoG is the center of gravity of the left ventricular region segmented from the previous image, w is the set window size of 11, μ _before and σ _before are the left ventricle representing the mean and standard deviation of the brightness values of the left ventricular region in the previous image, respectively. Auto split method. The process according to claim 3, wherein in step (ii),
_Acquiring a distortion correction image IMG _correct by correcting an image distortion according to a coil position through a process of _leveling a three-dimensional plane calculated by a least square method using the divided initial left ventricular region LV _rgnl ; And
The region expansion technique is re-executed on the distortion-corrected image to segment the initial left ventricular region LV _rgnl , (1) the left ventricle position and center of gravity in the magnetic resonance image, (2) the approximate shape and radius of the left ventricle, and (3) the brightness. And calculating information of a ventricular region including value averages, statistics of standard deviations, and (4) pixel volume information for determining whether there is a segmentation error in the base image. The process according to claim 1, wherein in step (iii),
The polar transformation is obtained by converting the coordinate P (x, y) of the Cartesian coordinate system to the coordinate P (r, θ) of the polar coordinate system based on the origin, and obtaining the polar coordinate image.
The radius at which the Cartesian coordinate system is converted to the polar coordinate system includes a part of the myocardium and the surrounding structure using three times the radius value of the basic left ventricle information. The relation between the coordinates of the Cartesian coordinate system and the coordinates of the polar coordinate system is expressed by the following equation.

Left ventricular auto-segmentation method. The method according to claim 1, wherein step (iii) compares the size of the initial left ventricular region of the image in which the polar origin error appears with the size of the initial left ventricular region of the previous image, and based on the sharp increase in the pixel volume. Determining whether there is a polar coordinate origin error for the initial left ventricular region; And
And correcting the polar coordinate error of the initial left ventricle when there is a polar coordinate error of the initial left ventricle. The method of claim 1, wherein step (v)
(v-1) The brightness value change amount data IMG _diff (x, y), which is image data indicating a change in pixel brightness value in the horizontal direction in the left ventricular region divided in step (iv), is obtained by the following equation: IMG _diff (x, y) Generating via IMG (x, y) -IMG (x−1, y); And
and (v-2) calculating a minimum cost path that connects the top and bottom of the brightness value variation data image by applying a dynamic search technique to the generated brightness value variation amount data in the vertical direction. The method of claim 7, wherein step (v-2) is
Defining each pixel of the image as a node, a change value of the node as a path cost, and defining a distance between pixels as a cost of a unit path connecting each node;
To calculate the minimum cost path, the path cost matrix Pmap (x, y) modeled by the following equation using the brightness value variation data is expressed as the following equation:

Pmap (x, y) is a matrix representing the minimum cost to reach each node, and IMG _diff is the generated brightness value variation data. ω is the weight for the path between nodes connected at the vertical top pixel, and the path cost cumulative value of one pixel of Pmap (x, y) is the brightness value and Pmap (x, y) of the pixel located at the same coordinate in the IMG _diff . Adding pixels having a minimum value among the three pixels calculated up to just above the image; And
Inversely searching the nodes visited to reach the coordinates from the coordinates having the minimum value at the bottom of the path cost matrix, extracting the path having the minimum cost, and inferring the extracted minimum cost path as the ventricular inner wall information when it is normally divided. Left ventricle automatic segmentation method comprising a. The method of claim 1, further comprising: (vi) calculating the blood flow rate of the left ventricle after restoring the shape of the left ventricle in the polar coordinate image by inverse transformation of the divided region. 10. The method of claim 9, wherein the inverse transformation in step (vi) is made by the following equation: (x, y) = (rsinθ, rcosθ), where x and y are Cartesian coordinates, and r and θ of left ventricular information Automated segmentation of the left ventricle with radius and angle.

Images