CN105741310A - Heart's left ventricle image segmentation system and method - Google Patents

Abstract

The invention provides a heart left ventricle image segmentation system and method. The system includes: an image conversion unit; a left ventricle rough extraction unit; a left ventricle fine extraction unit. The method includes: segmenting the nuclear magnetic resonance image to obtain a binary image, converting the binary image into a grayscale image through Euclidean distance transformation; separating each connected area in the binary image, and roughly extracting the outline of the left ventricle ; Determine whether the contour of the left ventricle is connected to the contour of the aorta. If the roughly extracted contour of the left ventricle is connected to the contour of the aorta, remove the aorta and divide it, repair the contour of the left ventricle, and obtain the image segmentation result of the left ventricle of the heart; otherwise, roughly extract The left ventricle contour of is taken as the left ventricle contour. The invention can eliminate the influence of the connection between the left ventricle and the aorta in terms of images, accurately segment the bottom layer image of the left ventricle, and overcome the influence of the edge leakage caused by the aorta on the left ventricle, thereby obtaining accurate segmentation results of the left ventricle.

Description

A system and method for image segmentation of cardiac left ventricle

technical field

The invention relates to the field of image processing, in particular to a system and method for segmenting images of left ventricle of the heart.

Background technique

Among the many image segmentation methods, there are many methods suitable for segmenting short-axis cardiac magnetic resonance images, such as region growing method, threshold method and level set method. For example, the Mumford-Shah model segmentation method combined with shape statistics is used to segment the left ventricle of the heart in the paper "MR image left ventricle contour segmentation of Mumford-Shah model of shape statistics", and the situation of weak boundary and boundary fracture is considered . However, it ignores the absence of part of the left ventricular boundary at the bottom layer of the heart due to the aorta. The paper "Left Ventricle Segmentation Algorithm Combining Texture and Shape with Tagged MR Images" introduces texture classification information and prior knowledge of shape statistics into the Mumford-Shah model, and proposes an improved segmentation of the inner and outer contours of the left ventricle in MRI images with marker lines The same method does not take into account the connection between the bottom layer of the heart and the aorta, it is difficult to distinguish the texture, and it does not specifically introduce the solution in terms of shape statistics. In the paper "Left Ventricle Segmentation Method of Cardiac MR Image Based on Improved Snake Model", it mainly focuses on the segmentation of the inner and outer layers of the left ventricle in the middle of the heart, and does not consider the leakage of the bottom edge of the heart.

Because the bottom layer of the heart is connected to the aorta, the edge leaks, so it is difficult to segment, and most segmentation methods are difficult to deal with such problems. Most scholars mainly focus on the segmentation of the middle layer when segmenting the left ventricle of the heart, ignoring the processing of the bottom layer of the left ventricle. However, the bottom layer of the left ventricle of the heart is connected to the aorta, and the processing of the bottom layer of the left ventricle of the heart is to obtain various information of the left ventricle of the heart and necessary for 3D reconstruction of the left ventricle of the heart.

Contents of the invention

Aiming at the deficiencies in the prior art, the present invention provides a system and method for segmenting images of the left ventricle of the heart.

Technical scheme of the present invention is:

A heart left ventricle image segmentation system, comprising:

Image conversion unit: select a seed point on the left ventricle of the nuclear magnetic resonance image, use the average value of the maximum gray value and the minimum gray value of the nuclear magnetic resonance image as the initial segmentation threshold, segment the nuclear magnetic resonance image into foreground and background, and divide the foreground The average gray value of the background and the average gray value of the background are taken as the new segmentation threshold. If the difference between the new segmentation threshold and the segmentation threshold obtained in the previous iteration is less than the set allowable range, the current segmentation threshold is the final segmentation Threshold, otherwise continue iterative calculation, and finally get a binarized image, and then convert the binarized image into a grayscale image through Euclidean distance transformation;

Left ventricle outline rough extraction unit: sort the pixels of the grayscale image in the order of increasing grayscale, and use a FIFO queue to recursively assign each grayscale minimum area in a width-first manner to form a new grayscale minimum area , and when assigning, mark the pixels in the first gray-scale minimum area as 0, and the marks in the subsequent gray-scale minimum areas are incremented by 1, and distinguish each connected area in the gray-scale image according to the mark, The outline of the left ventricle is roughly extracted through the selected seed point;

Left ventricle contour fine extraction unit: if the roughly extracted left ventricle contour is connected to the aortic contour, the aorta is removed, the left ventricle contour is repaired, and the image segmentation result of the left ventricle of the heart is obtained; if the roughly extracted left ventricle contour is connected to the aortic contour If the contours are not connected, the roughly extracted left ventricle contour is used as the left ventricle contour, and the image segmentation result of the left ventricle of the heart is obtained.

The left ventricular contour fine extraction unit includes:

Analysis and judgment unit: Calculate the distance from the center of mass of the roughly extracted left ventricle contour to each edge point, find out the maximum distance Ma and the minimum distance Mi, and analyze and judge whether the left ventricle is connected to the aorta according to the maximum distance and the minimum distance Connected: If Ma<30 pixel unit length and (Ma-Mi)/Ma>0.55, or Ma>30 pixel unit length and (Ma-Mi)/Ma>0.38, the left ventricle is connected to the aorta, Morphological processing is required, otherwise the roughly extracted left ventricle contour is taken as the left ventricle contour.

The left ventricular contour fine extraction unit also includes:

Morphological processing unit: find out all the minimum values from the center of mass to the edge points in the roughly extracted left ventricle contour, if the position of the maximum distance Ma is between two adjacent minimum value points, then use these two The minimum value point is the dividing point and draw a straight line to separate the contour of the left ventricle from the contour of the aorta, re-determine the centroid of the contour of the left ventricle, and take two dividing points in the direction of the midline of the angle formed by the connection between the two dividing points and the centroid to The average distance of the center of mass is used as the third segmentation point, and these three segmentation points are used to make a curve to complement the edge of the left ventricle contour to obtain the left ventricle contour.

The minimum point Pi satisfies (Pi-Mi)/(Ma-Mi)<0.3.

The present invention also provides a method for image segmentation of the left ventricle of the heart, comprising:

Image conversion: Select the seed point on the left ventricle of the MRI image, use the average value of the maximum gray value and the minimum gray value of the MRI image as the initial segmentation threshold, segment the MRI image into foreground and background, and divide the foreground The average gray value of the background and the average gray value of the background are taken as the new segmentation threshold. If the difference between the new segmentation threshold and the segmentation threshold obtained in the previous iteration is less than the set allowable range, the current segmentation threshold is the final segmentation threshold. , otherwise continue the iterative calculation, and finally get the binarized image, and then convert the binarized image into a grayscale image through Euclidean distance transformation;

Roughly extract the left ventricle contour: sort the pixels of the grayscale image in the order of increasing grayscale, and use a FIFO queue to recursively assign each grayscale minimum area to form a new grayscale minimum area in a width-first manner. And when allocating, the pixels in the first gray-scale area are marked as 0, and the marks of the subsequent gray-scale areas are incremented by 1, and the connected areas in the gray-scale image are distinguished according to the marks. The selected seed point roughly extracts the contour of the left ventricle;

Fine extraction of the left ventricle contour: If the roughly extracted left ventricle contour is connected to the aortic contour, the aorta is removed, the left ventricle contour is repaired, and the image segmentation result of the left ventricle of the heart is obtained; if the roughly extracted left ventricle contour is connected to the aortic contour If they are not connected, the roughly extracted left ventricle contour is used as the left ventricle contour, and the image segmentation result of the left ventricle of the heart is obtained.

The specific steps of finely extracting the contour of the left ventricle include:

Calculate the distance from the center of mass of the roughly extracted left ventricle contour to each edge point;

Find the maximum distance Ma and the minimum distance Mi;

Analyze and judge whether the left ventricle is connected to the aorta according to the maximum and minimum distances: if Ma<30 pixel unit length and (Ma-Mi)/Ma>0.55, or Ma>30 pixel unit length and ( Ma-Mi)/Ma>0.38, the left ventricle is connected to the aorta, and the contour of the left ventricle needs to be obtained after morphological processing, otherwise the roughly extracted contour of the left ventricle will be used as the contour of the left ventricle.

The concrete steps of described morphology processing comprise:

Find all the minimum values from the center of mass to the edge points in the roughly extracted left ventricle contour;

If the position of the maximum distance Ma is between two adjacent minimum value points, then use these two minimum value points as the dividing point to draw a straight line to separate the left ventricle contour from the aortic contour;

Re-determine the centroid of the left ventricle contour;

Take the average distance from the two segmentation points to the center of mass in the direction of the midline of the angle formed by the connection between the two segmentation points and the centroid as the third segmentation point;

The contour edge of the left ventricle is supplemented by curves based on these three segmentation points to obtain the contour of the left ventricle.

Beneficial effect:

The left ventricle image segmentation system and method of the present invention can eliminate the influence of the connection between the left ventricle and the aorta in terms of images, and accurately segment the bottom layer image of the left ventricle. Determine whether the left ventricle is connected to the aorta and cause edge leakage. If edge leakage occurs, the aortic edge will be removed morphologically and the missing edge will be compensated with a curve similar to the left ventricular edge to overcome the aorta’s impact on the left The edge leakage caused by the ventricle can be used to obtain accurate segmentation results of the left ventricle.

Description of drawings

Fig. 1 is the block diagram of the heart left ventricle image segmentation system of embodiment 1 of the present invention;

Fig. 2 is a flow chart of a method for segmenting images of the left ventricle of the heart according to Embodiment 2 of the present invention;

Fig. 3 is a flow chart of finely extracting the contour of the left ventricle according to Embodiment 2 of the present invention;

Fig. 4 is the flow chart of the morphology processing of embodiment 2 of the present invention;

Fig. 5 is the binary image of embodiment 2 of the present invention;

Fig. 6 is the transformation result of the Euclidean distance transformation of embodiment 2 of the present invention;

Fig. 7 is a diagram showing the rough extraction result of the left ventricle contour in Example 2 of the present invention;

Fig. 8 is a curve diagram of the distance from the center of mass of the contour of the left ventricle to each edge point in Example 2 of the present invention;

Fig. 9 is a group of left ventricular bottom layer (Ma-Mi)/MA ratio curves of Example 2 of the present invention;

Fig. 10 is a result diagram of the edge of left ventricle contour complemented by curves using three segmentation points in Example 2 of the present invention;

Fig. 11 is a diagram of the contour results of the left ventricle obtained by fine extraction in Example 2 of the present invention.

detailed description

The specific implementation of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

Example 1

A heart left ventricle image segmentation system, as shown in Figure 1, comprising:

Image conversion unit: select a seed point on the left ventricle of the nuclear magnetic resonance image, obtain the maximum gray value Z0 and the minimum gray value Z1 of the nuclear magnetic resonance image, and average the maximum gray value and the minimum gray value of the nuclear magnetic resonance image Value is used as the initial segmentation threshold T=(Z0+Z1)/2, the MRI image is divided into foreground and background, the average gray value T0 of the foreground and the average gray value T1 of the background are obtained respectively, and the average gray value of the foreground Value and the average gray value of the background are taken as the new segmentation threshold. If the difference between the new segmentation threshold TT=(T0+T1)/2 and the segmentation threshold obtained in the previous iteration is less than the set allowable range, the current segmentation The threshold is the final segmentation threshold, otherwise iterative calculation is continued to finally obtain a binary image, and then the binary image is converted into a grayscale image through Euclidean distance transformation; the image conversion unit extracts the main organs and tissues in the MRI image, Useless areas are filtered out, and the complex grayscale image is converted into a simple grayscale image with the smallest grayscale value in the center of the area and increasing towards the edge.

Left ventricle outline rough extraction unit: sort the pixels of the grayscale image in the order of increasing grayscale, and use a FIFO queue to recursively assign each grayscale minimum area in a width-first manner to form a new grayscale minimum area , and when assigning, mark the pixels in the first gray-scale minimum area as 0, and the marks in the subsequent gray-scale minimum areas are incremented by 1, and distinguish each connected area in the gray-scale image according to the mark, The outline of the left ventricle was roughly extracted through the selected seed points.

Left ventricle contour fine extraction unit: if the roughly extracted left ventricle contour is connected to the aortic contour, the aorta is removed, the left ventricle contour is repaired, and the image segmentation result of the left ventricle of the heart is obtained; if the roughly extracted left ventricle contour is connected to the aortic contour If the contours are not connected, the roughly extracted left ventricle contour is used as the left ventricle contour, and the image segmentation result of the left ventricle of the heart is obtained.

Left ventricular contour extraction unit, including:

Analysis and Judgment Unit: Obtain empirical values by processing 400 groups of left ventricular bottom layer data of 10 patients, calculate the distance from the roughly extracted left ventricular contour centroid to each edge point, find out the maximum distance Ma and the minimum distance Mi, and Analyze and judge whether the left ventricle is connected to the aorta according to the maximum and minimum distances: if Ma<30 pixel unit length and (Ma-Mi)/Ma>0.55, or Ma>30 pixel unit length and ( Ma-Mi)/Ma>0.38, the left ventricle is connected to the aorta, and morphological processing is required, otherwise the roughly extracted left ventricle contour will be used as the left ventricle contour.

Morphological processing unit: find out all the minimum points Pi satisfying (Pi-Mi)/(Ma-Mi)<0.3 from the center of mass to the edge point in the roughly extracted left ventricle contour, if the position of the distance from the maximum value Ma is within Between two adjacent minimum points, a straight line is drawn to separate the left ventricle contour from the aortic contour with these two minimum value points as the dividing point, and the centroid of the left ventricle contour is re-determined. In the direction of the midline of the angle formed between the point and the centroid, take the average distance from the two segmentation points to the centroid as the third segmentation point, use these three segmentation points as a curve to complement the edge of the left ventricle contour, and obtain the left ventricle contour.

By using the system provided by the present invention, the segmentation point between the left ventricle and the aorta can be found more accurately, thereby avoiding the influence of the aorta on the segmentation of the left ventricle, and providing accurate left ventricle segmentation results for the three-dimensional reconstruction of the left ventricle of the heart, so that The observation of the left ventricle is more accurate and effective.

Example 2

The present invention also provides a method for image segmentation of the left ventricle of the heart using the system described in Embodiment 1, as shown in Figure 2, comprising:

Step 201, select a seed point on the left ventricle of the nuclear magnetic resonance image, obtain the maximum gray value Z0 and the minimum gray value Z1 of the nuclear magnetic resonance image, and calculate the average value of the maximum gray value and the minimum gray value of the nuclear magnetic resonance image As the initial segmentation threshold T=(Z0+Z1)/2, the MRI image is divided into foreground and background, the average gray value T0 of the foreground and the average gray value T1 of the background are obtained respectively, and the average gray value of the foreground Take the average of the average gray value of the background as the new segmentation threshold. If the difference between the new segmentation threshold TT=(T0+T1)/2 and the segmentation threshold obtained in the previous iteration is less than the set allowable range, the current segmentation threshold It is the final segmentation threshold, otherwise iterative calculation is continued to finally obtain the binarized image as shown in Figure 5, and then the binarized image is converted into a grayscale image as shown in Figure 6 through Euclidean distance transformation; the image conversion unit Extract the main organs and tissues in the MRI image, filter out useless areas, and convert the complex grayscale image into a simple grayscale image with the smallest grayscale value in the center of the area and increasing towards the edge.

Step 202, sort the pixels of the grayscale image in the order of increasing grayscale, use a FIFO queue to recursively allocate to each extremely small grayscale area in a width-first manner to form new extremely small grayscale areas, and At the same time, the pixels in the first grayscale area are all marked as 0, and the marks in the following grayscale areas are incremented by 1 in turn, and the connected areas in the grayscale image are distinguished according to the marks, and the selected seed The outline of the left ventricle as shown in Figure 7 is roughly extracted by point;

Step 203, fine extraction of the left ventricle contour: if the roughly extracted left ventricle contour is connected to the aortic contour, go to step 204; if the roughly extracted left ventricle contour is not connected to the aortic contour, then the roughly The ventricle contour is used as the left ventricle contour, go to step 205;

Step 204, removing the aorta and repairing the contour of the left ventricle;

Step 205, obtaining the image segmentation result of the left ventricle of the heart.

As shown in Figure 3, the specific steps of finely extracting the contour of the left ventricle are as follows:

Step 301, calculate the distance from the center of mass of the roughly extracted left ventricle contour to each edge point, and obtain the graph as shown in Figure 8;

Step 302, find out the maximum distance Ma and the minimum distance Mi;

Step 303, if Ma<30 pixel point unit length and (Ma-Mi)/Ma>0.55, then go to step 305, otherwise go to step 304;

Step 304, if Ma>30 pixel point unit length and (Ma-Mi)/Ma>0.38, then go to step 305, otherwise go to step 306;

Step 305, the left ventricle is connected to the aorta, and the outline of the left ventricle is obtained after morphological processing;

Step 306, the roughly extracted left ventricle contour is used as the left ventricle contour.

Figure 9 is a set of curves of left ventricular bottom layer (Ma-Mi)/MA ratio. In this set of data, Ma>30 pixel unit length, it can be seen from the figure that (Ma-Mi)/Ma are all greater than 0.38, and there is left ventricle connection with the aorta.

As shown in Figure 4, the specific steps of morphological processing are as follows:

Step 401, find out all the minimum values from the center of mass to the edge points in the roughly extracted left ventricle contour;

Step 402, if the position of the distance from the maximum value Ma is between two adjacent minimum value points, then go to step 403, otherwise return to step 401;

Step 403, using the two minimum points as dividing points to draw a straight line to separate the contour of the left ventricle from the contour of the aorta;

Step 404, re-determining the centroid of the left ventricle contour;

Step 405, taking the average of the distances from the two segmentation points to the centroid in the direction of the midline of the angle formed by the connection between the two segmentation points and the centroid as the third segmentation point;

Step 406 , use these three segmentation points to make a curve to complement the contour edge of the left ventricle, as shown in FIG. 10 . The connected area is obtained by the selected seed points, and the contour of the left ventricle is extracted to obtain a more accurate contour of the left ventricle as shown in FIG. 11 .

It can be seen that the method provided by the present invention can not only separate the bottom layer of the left ventricle from the aorta, but also make up for the missing contour of the left ventricle with a curve that approximates the contour of the left ventricle, thereby obtaining an accurate contour of the left ventricle.

Claims (7)

1. a heart left ventricle image segmentation system, is characterized in that, comprises: Image conversion unit: select a seed point on the left ventricle of the nuclear magnetic resonance image, use the average value of the maximum gray value and the minimum gray value of the nuclear magnetic resonance image as the initial segmentation threshold, segment the nuclear magnetic resonance image into foreground and background, and divide the foreground The average gray value of the background and the average gray value of the background are taken as the new segmentation threshold. If the difference between the new segmentation threshold and the segmentation threshold obtained in the previous iteration is less than the set allowable range, the current segmentation threshold is the final segmentation Threshold, otherwise continue iterative calculation, and finally get a binarized image, and then convert the binarized image into a grayscale image through Euclidean distance transformation; Left ventricle outline rough extraction unit: sort the pixels of the grayscale image in the order of increasing grayscale, and use a FIFO queue to recursively assign each grayscale minimum area in a width-first manner to form a new grayscale minimum area , and when assigning, mark the pixels in the first gray-scale minimum area as 0, and the marks in the subsequent gray-scale minimum areas are incremented by 1, and distinguish each connected area in the gray-scale image according to the mark, The outline of the left ventricle is roughly extracted through the selected seed point; Left ventricle contour fine extraction unit: if the roughly extracted left ventricle contour is connected to the aortic contour, the aorta is removed, the left ventricle contour is repaired, and the image segmentation result of the left ventricle of the heart is obtained; if the roughly extracted left ventricle contour is connected to the aortic contour If the contours are not connected, the roughly extracted left ventricle contour is used as the left ventricle contour, and the image segmentation result of the left ventricle of the heart is obtained. 2. The heart left ventricle image segmentation system according to claim 1, wherein the fine extraction unit of the left ventricle contour comprises: Analysis and judgment unit: Calculate the distance from the center of mass of the roughly extracted left ventricle contour to each edge point, find out the maximum distance Ma and the minimum distance Mi, and analyze and judge whether the left ventricle is connected to the aorta according to the maximum distance and the minimum distance Connected: If Ma<30 pixel unit length and (Ma-Mi)/Ma>0.55, or Ma>30 pixel unit length and (Ma-Mi)/Ma>0.38, the left ventricle is connected to the aorta, Morphological processing is required, otherwise the roughly extracted left ventricle contour is taken as the left ventricle contour. 3. The heart left ventricle image segmentation system according to claim 2, wherein the fine extraction unit of the left ventricle contour also includes: Morphological processing unit: find out all the minimum values from the center of mass to the edge points in the roughly extracted left ventricle contour, if the position of the maximum distance Ma is between two adjacent minimum value points, then use these two The minimum value point is the dividing point and draw a straight line to separate the contour of the left ventricle from the contour of the aorta, re-determine the centroid of the contour of the left ventricle, and take two dividing points in the direction of the midline of the angle formed by the connection between the two dividing points and the centroid to The average distance of the center of mass is used as the third segmentation point, and these three segmentation points are used to make a curve to complement the edge of the left ventricle contour to obtain the left ventricle contour. 4. The heart left ventricle image segmentation system according to claim 3, wherein the minimum point Pi satisfies (Pi-Mi)/(Ma-Mi)<0.3. 5. A heart left ventricle image segmentation method, is characterized in that, comprises: Image conversion: Select the seed point on the left ventricle of the MRI image, use the average value of the maximum gray value and the minimum gray value of the MRI image as the initial segmentation threshold, segment the MRI image into foreground and background, and divide the foreground The average gray value of the background and the average gray value of the background are taken as the new segmentation threshold. If the difference between the new segmentation threshold and the segmentation threshold obtained in the previous iteration is less than the set allowable range, the current segmentation threshold is the final segmentation threshold. , otherwise continue the iterative calculation, and finally get the binarized image, and then convert the binarized image into a grayscale image through Euclidean distance transformation; Roughly extract the left ventricle contour: sort the pixels of the grayscale image in the order of increasing grayscale, and use a FIFO queue to recursively assign each grayscale minimum area to form a new grayscale minimum area in a width-first manner. And when allocating, the pixels in the first gray-scale area are marked as 0, and the marks of the subsequent gray-scale areas are incremented by 1, and the connected areas in the gray-scale image are distinguished according to the marks. The selected seed point roughly extracts the contour of the left ventricle; Fine extraction of the left ventricle contour: If the roughly extracted left ventricle contour is connected to the aortic contour, the aorta is removed, the left ventricle contour is repaired, and the image segmentation result of the left ventricle of the heart is obtained; if the roughly extracted left ventricle contour is connected to the aortic contour If they are not connected, the roughly extracted left ventricle contour is used as the left ventricle contour, and the image segmentation result of the left ventricle of the heart is obtained. 6. The method according to claim 5, characterized in that the specific step of finely extracting the contour of the left ventricle comprises: Calculate the distance from the center of mass of the roughly extracted left ventricle contour to each edge point; Find the maximum distance Ma and the minimum distance Mi; Analyze and judge whether the left ventricle is connected to the aorta according to the maximum and minimum distances: if Ma<30 pixel unit length and (Ma-Mi)/Ma>0.55, or Ma>30 pixel unit length and ( Ma-Mi)/Ma>0.38, the left ventricle is connected to the aorta, and the contour of the left ventricle needs to be obtained after morphological processing, otherwise the roughly extracted contour of the left ventricle will be used as the contour of the left ventricle. 7. method according to claim 6, is characterized in that, the concrete step of described morphology processing comprises: Find all the minimum values from the center of mass to the edge points in the roughly extracted left ventricle contour; If the position of the maximum distance Ma is between two adjacent minimum value points, then use these two minimum value points as the dividing point to draw a straight line to separate the left ventricle contour from the aortic contour; Re-determine the centroid of the left ventricle contour; Take the average distance from the two segmentation points to the center of mass in the direction of the midline of the angle formed by the connection between the two segmentation points and the centroid as the third segmentation point; The contour edge of the left ventricle is supplemented by curves based on these three segmentation points to obtain the contour of the left ventricle.