Medical image automatic matching method and system based on double-layer neural network model

Abstract

The invention discloses a medical image automatic matching method and system based on a double-layer neural network model, and relates to the field of image processing and deep learning. The method comprises the following steps: firstly, constructing a double-layer neural network model, the first-layer model being used for blood vessel segmentation, and the second-layer model being used for labeling three attribute endpoints of a root node, a bifurcation point and an end point for a segmentation map; then performing blood vessel segmentation and endpoint labeling on two angiography images acquired at different angles by adopting a first layer model and a second layer model of a double-layer neural network model respectively to obtain a first blood vessel segmentation result, a second blood vessel segmentation result, a first endpoint labeling result and a second endpoint labeling result; and finally, matching the blood vessel segments in the two images based on pixel values and endpoint attributes in the segmentation result and the labeling result. According to the invention, the automatic segmentation of the angiography image is realized, the matching of the blood vessel segments is automatically completed after the segmentation task is completed, and the efficiency of the three-dimensional reconstruction of the blood vessel is improved.

Description

Technical Field

[0001] The present invention belongs to the field of image processing and deep learning, and specifically relates to a medical image automatic matching method and system based on a double-layer neural network model. Background Art

[0002] The World Health Organization's research found that the number of people dying from cardiovascular diseases is increasing year by year and the age of the patients is getting younger. The treatment of cardiovascular diseases has become a key issue for life safety and human health.

[0003] At present, the medical equipment widely used in the diagnosis of vascular diseases mainly includes: Digital Subtraction Angiography (DSA), Intravenous Ultrasound (IVUS), Magnetic Resonance Angiography (MRA), CT Angiography (CTA), etc. These imaging technologies provide medical researchers with a large amount of high-precision vascular information, thus bringing breakthrough developments to medical research. Among them, the imaging principle of digital subtraction angiography is to inject a harmless contrast agent into the human coronary artery, and then develop it in a specific observation device by sending and receiving X-rays. DSA coronary angiography technology is one of the important means of diagnosing cardiovascular diseases. Whether in terms of safety or imaging accuracy, it is often regarded as the "gold standard" in clinical diagnosis.

[0004] The core task of medical image segmentation is to extract coronary arteries from contrast images so that doctors can observe the tissue structure and diagnose the disease. Contrast image segmentation not only provides a basis for quantitative analysis of the degree of vascular stenosis, but also provides a theoretical basis for the three-dimensional reconstruction of vascular structure and surgical navigation of doctors. High-quality segmentation results are the key to analyzing the content of contrast images.

[0005] The core task of vascular segment matching is to accurately match different vascular segments, calculate the three-dimensional coordinates, and finally stitch them together to reconstruct the complete three-dimensional structure. The accuracy of vascular segment matching determines the similarity between the final reconstruction result and the angiographic image. Accurate and fast completion of vascular segment matching can greatly improve reconstruction efficiency.

[0006] At present, traditional segmentation methods mainly include threshold processing, region growing, matched filtering and mathematical morphology. In recent years, deep learning has shown significant advantages in medical image analysis and has broad application prospects. For example, 3D FCN can be used to learn the three-dimensional semantic features of coronary arteries, and the attention gate mechanism is introduced to enhance blood vessels and suppress irrelevant areas. 3D U-net is used for artery segmentation, and various data transformations are selected and tested to reduce the problem of overfitting.

[0007] There are currently methods based on image registration and feature point matching for vascular segment matching. However, these methods cannot guarantee the correctness of the reconstruction results and require a lot of time to match vascular segments. Summary of the invention

[0008] Purpose of the invention: In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for automatic matching of medical images based on a two-layer neural network model, which can realize automatic matching of blood vessel segments according to the segmentation results after blood vessel segmentation, thereby improving the efficiency of three-dimensional reconstruction.

[0009] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides a medical image automatic matching method based on a two-layer neural network model, comprising the following steps:

[0010] (1) A two-layer neural network model is constructed, in which the first layer model inputs the angiography image and outputs the segmented vascular segments, where vascular segments with different names are displayed in different colors; the second layer model inputs the segmentation map of vascular segments with different colors and outputs the endpoints of the blood vessels, including endpoints with three attributes: root node, bifurcation point, and terminal point, where endpoints with different attributes are displayed in different colors;

[0011] (2) using the first layer model of the two-layer neural network model to segment the blood vessels of the two angiography images acquired at different angles, respectively, to obtain a first blood vessel segmentation result and a second blood vessel segmentation result;

[0012] (3) using the second layer model in the two-layer neural network model to respectively label the root node, bifurcation point and terminal point of the blood vessel using the first blood vessel segmentation result and the second blood vessel segmentation result to obtain a first endpoint labeling result and a second endpoint labeling result;

[0013] (4) Based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result, the blood vessel segments in the two angiography images are matched based on pixel values ​​and endpoint attributes.

[0014] Preferably, the two-layer neural network model is composed of two U-Net models.

[0015] Preferably, the U-Net model is improved by adding a bidirectional attention mechanism to the skip connection module of the traditional U-Net network model to enhance the correlation between feature maps.

[0016] Furthermore, the first layer model and the second layer model are trained based on the first training set and the second training set respectively, and the construction of the first training set includes: collecting the original DSA coronary angiography image and performing data enhancement, dividing the blood vessels in the angiography image into different categories based on the medical naming of the coronary artery blood vessels, and distinguishing and representing them with different colors as the label data of the angiography image, to construct the first training set; the construction of the second training set includes: in the output of the first layer model, setting the root node, bifurcation point and terminal point of the coronary artery, and marking them with circles of different colors, wherein the bifurcation point contains at least two pixel values, and combining the output result of the first layer model and the marked endpoints to construct the second training set.

[0017] Furthermore, matching the blood vessel segments in the two angiography images based on pixel values ​​and endpoint attributes based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result, includes:

[0018] Based on the first blood vessel segmentation result and the first endpoint labeling result, obtaining a first two-dimensional coordinate group corresponding to the endpoints of different blood vessel segments;

[0019] Based on the second blood vessel segmentation result and the second endpoint labeling result, obtaining a second two-dimensional coordinate group corresponding to the endpoints of different blood vessel segments;

[0020] For two angiography images, compare the corresponding pixel values ​​and corresponding endpoint attributes (ps: different endpoint attributes may have the same pixel values) of the first two-dimensional coordinate group and the second two-dimensional coordinate group to see if they are the same. If they are the same, the corresponding two-dimensional coordinate groups are the same endpoint attributes, and the endpoint matching is completed;

[0021] After the endpoint matching is completed, two-dimensional coordinates with the same pixel values ​​are found in the first and second blood vessel segmentation results as matching pairs of the same blood vessel segment. The matching endpoint pair with the same pixel value and the blood vessel segment form a whole matching blood vessel.

[0022] Furthermore, the obtaining of first two-dimensional coordinate groups corresponding to different blood vessel segment endpoints based on the first blood vessel segmentation result and the first endpoint annotation result includes: obtaining two-dimensional coordinates of pixel values ​​at the endpoint attribute annotation in the first endpoint annotation result; searching for corresponding segmentation image pixel values ​​in the first blood vessel segmentation result according to the two-dimensional coordinates, and saving two-dimensional coordinate points with the same pixel values ​​as a group, corresponding to the first two-dimensional coordinate group corresponding to one blood vessel segment endpoint;

[0023] The method of obtaining second two-dimensional coordinate groups corresponding to different blood vessel segments based on the second blood vessel segmentation result and the second endpoint annotation result includes: obtaining the two-dimensional coordinates of the pixel values ​​at the endpoint attribute annotation in the second endpoint annotation result; searching for the corresponding segmentation image pixel values ​​in the second blood vessel segmentation result according to the two-dimensional coordinates, and saving the two-dimensional coordinate points with the same pixel values ​​as a group, which corresponds to the second two-dimensional coordinate group corresponding to a blood vessel segment endpoint.

[0024] Furthermore, after the endpoint matching is completed, two-dimensional coordinates with the same pixel value are searched in the first blood vessel segmentation result and the second blood vessel segmentation result as a matching pair of the same blood vessel segment, including:

[0025] According to the endpoint attributes and their corresponding pixel values, the root node, bifurcation point and terminal point are matched. The bifurcation point has two or more pixel values. Multiple comparisons are performed and the points with the same attributes and pixel values ​​are matched.

[0026] The conjugate point pair of the vascular segment is obtained by using the epipolar line constraint matching method for the vascular segment after endpoint matching. The coordinate points of each segment except the previously marked endpoint attributes are traversed to find a point in a segment with the shortest distance to the epipolar line among all the points in the matching segment. The conjugate point pair is each matched point in the vascular segment.

[0027] Based on the same inventive concept, the present invention provides a medical image automatic matching system based on a double-layer neural network model, comprising:

[0028] The two-layer model module is used to construct a two-layer neural network model, in which the first-layer model inputs the angiography image and outputs the segmented blood vessel segments. Blood vessel segments with different names are displayed in different colors. The second-layer model inputs the segmentation map of blood vessel segments with different colors and outputs the endpoints of the blood vessels, including endpoints with three attributes: root node, bifurcation point and terminal point. Endpoints with different attributes are displayed in different colors.

[0029] The blood vessel segmentation module is used to perform blood vessel segmentation on two angiography images acquired at different angles by using the first layer model in the double-layer neural network model to obtain a first blood vessel segmentation result and a second blood vessel segmentation result;

[0030] An endpoint labeling module is used to label the first blood vessel segmentation result and the second blood vessel segmentation result respectively using the second layer model in the double-layer neural network model to label the blood vessel root node, bifurcation point and terminal point, so as to obtain a first endpoint labeling result and a second endpoint labeling result;

[0031] The blood vessel matching module is used to match the blood vessel segments in the two angiography images based on the pixel values ​​and the endpoint attributes based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result.

[0032] The present invention also provides a computer system, comprising a memory, a processor, and a computer program / instruction stored in the memory and executable on the processor, wherein the computer program / instruction, when executed by the processor, implements the steps of the automatic matching method for medical images based on a two-layer neural network model.

[0033] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the automatic matching method of medical images based on a two-layer neural network model.

[0034] Beneficial effect: The present invention proposes a medical image automatic matching method based on a two-layer neural network model. Through the constructed two-layer neural network model, blood vessel segmentation and endpoint annotation are performed respectively. After the first layer model completes the blood vessel segmentation task, the segmentation result is input into the second layer model to extract the root node, terminal point, and bifurcation point in each blood vessel segment. After two angiography images at different angles are output by the two-layer model, their endpoint attributes and pixel values ​​are compared to automatically match the blood vessel segments. Compared with the prior art, the present invention realizes the automatic segmentation of angiography images, can assist in the rapid and accurate processing and analysis of images, and automatically completes the matching of blood vessel segments after completing the segmentation task, thereby improving the application efficiency in practice, ensuring the continuity of the reconstruction results, and improving the efficiency of three-dimensional reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is an overall flow chart of an embodiment of the present invention.

[0036] Figure 2 Schematic diagram of a network model in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] To make the purpose, technical solution and advantages of the present invention clearer, the following is a further detailed description in conjunction with the accompanying drawings and specific embodiments. This embodiment is only intended to illustrate the principle of the present invention and does not represent any limitation to the present invention.

[0038] like Figure 1 As shown, an automatic matching method for medical images based on a two-layer neural network model disclosed in an embodiment of the present invention comprises the following steps:

[0039] Step 1: Construct a two-layer neural network model, in which the first layer model inputs the angiography image and outputs the segmented vascular segments. The vascular segments with different names are displayed in different colors. The second layer model inputs the segmentation map of vascular segments with different colors and outputs the endpoints of the blood vessels, including endpoints with three attributes: root node, bifurcation point and terminal point. Endpoints with different attributes are displayed in different colors.

[0040] This embodiment is described by taking a two-layer neural network model built based on the U-Net network model as an example.

[0041] Specifically, in step 1, the first layer model and the second layer model are trained based on the first training set and the second training set, respectively. The construction of the first training set includes: collecting the original DSA coronary angiography image and performing data enhancement, dividing the blood vessels in the angiography image into different classes based on the medical naming of the coronary artery vessels, and distinguishing and representing them with different colors as the label data of the angiography image, to construct the first training set. The construction of the second training set includes: in the output of the first layer model, setting the root node, bifurcation point and terminal point of the coronary artery, and marking them with circles of different colors, wherein the bifurcation point contains at least two pixel values, combining the output result of the first layer model and the marked endpoints, and constructing the second training set.

[0042] In this embodiment, CLANE is used to enhance the contrast of the angiography image, a non-mean filtering algorithm is used to remove background noise, and the original image data set is expanded by data enhancement, such as random cropping, random rotation, horizontal mirroring, and vertical mirroring. Since the amount of original coronary angiography vascular data is too small, direct training is prone to overfitting. By cropping image blocks, the amount of data can be increased to suppress overfitting; the training set is used to train a preliminary model, the validation set is used to improve the model, and the test set is used to complete the test of model performance and evaluate the quality.

[0043] The following are examples of dataset annotations:

[0044] First, the dataset is divided into left coronary artery and right coronary artery; then, based on the medical classification of each vascular segment in the left and right coronary arteries, the images are divided into the following 14 labels: "_background_","LM_1","LAD_2","D_3","S_4","LCX_5","OM_6","RCA_7","SA_8","AM_9","PLA_10","PDA_11","CA_12","RAVA_13".

[0045] Different labels are assigned different colors for the final pixel judgment. In order to reduce the amount of calculation during the experiment, the background color is generally set to black, and the remaining blood vessels are filled according to the designed color rules.

[0046] After obtaining the output image of the first-layer model, endpoint labels with three attributes, namely, root node, bifurcation point and terminal point, are marked according to the branching of the coronary arteries; for a bifurcation point, each different pixel value of the bifurcation point is marked with the same bifurcation point label.

[0047] In this embodiment, the schematic diagram of the double-layer neural network model constructed based on the U-Net model is as follows Figure 2 .

[0048] In the U-Net network model, skip connections are used to directly connect the feature maps output by the encoder (downsampling part) and the decoder (upsampling part) to retain high-resolution spatial information. This structure performs well in tasks such as medical image segmentation.

[0049] In some embodiments, the U-Net model can be further improved to further improve the segmentation effect. For example, a bidirectional attention mechanism is added to the jump connection module of the traditional U-Net network model to enhance the correlation between feature maps and further improve the segmentation ability of the model. The bidirectional attention mechanism focuses on important areas in the feature map and makes the network more focused on important contextual information during the segmentation process by adaptively adjusting the features.

[0050] Step 2: The two angiography images acquired at different angles are respectively segmented by the first layer model in the double-layer neural network model to obtain a first blood vessel segmentation result and a second blood vessel segmentation result.

[0051] Step 3: The first blood vessel segmentation result and the second blood vessel segmentation result are respectively labeled with the second layer model in the double-layer neural network model for the blood vessel root node, bifurcation point and terminal point to obtain the first endpoint labeling result and the second endpoint labeling result.

[0052] Step 4: Based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result, the blood vessel segments in the two angiography images are matched based on pixel values ​​and endpoint attributes.

[0053] Specifically, the detailed steps of step 4 are as follows:

[0054] Step 4.1: Based on the first blood vessel segmentation result and the first endpoint labeling result, obtain a first two-dimensional coordinate group corresponding to the endpoints of different blood vessel segments.

[0055] Specifically, the two-dimensional coordinates of the pixel value at the endpoint attribute annotation in the first endpoint annotation result are obtained; the corresponding segmentation map pixel value in the first blood vessel segmentation result is searched according to the two-dimensional coordinates, and the two-dimensional coordinate points with the same pixel value are saved as a group, corresponding to the first two-dimensional coordinate group corresponding to a blood vessel segment endpoint.

[0056] Step 4.2: Based on the second blood vessel segmentation result and the second endpoint labeling result, obtain a second two-dimensional coordinate group corresponding to the endpoints of different blood vessel segments.

[0057] Specifically, the two-dimensional coordinates of the pixel value at the endpoint attribute annotation in the second endpoint annotation result are obtained; the corresponding segmentation map pixel value in the second blood vessel segmentation result is searched according to the two-dimensional coordinates, and the two-dimensional coordinate points with the same pixel value are saved as a group, corresponding to the second two-dimensional coordinate group corresponding to a blood vessel segment endpoint.

[0058] Step 4.3: For the two angiography images, compare the corresponding pixel values ​​and corresponding endpoint attributes (ps: different endpoint attributes may have the same pixel values) of the first two-dimensional coordinate group and the second two-dimensional coordinate group to see if they are the same. If they are the same, the corresponding two-dimensional coordinate groups are the same endpoint attributes, and the endpoint matching is completed.

[0059] Step 4.4: After the endpoint matching is completed, two-dimensional coordinates with the same pixel value are found in the first blood vessel segmentation result and the second blood vessel segmentation result as matching pairs of the same blood vessel segment. The matching endpoint pair with the same pixel value and the blood vessel segment are a whole segment of matched blood vessels. Specifically including:

[0060] Step 4.4.1: Match the root node, bifurcation point and terminal point according to the endpoint attributes and their corresponding pixel values. The bifurcation point has two or more pixel values. Multiple comparisons are performed and the points with the same attributes and pixel values ​​are matched.

[0061] Step 4.4.2: Use the epipolar line constraint matching method to obtain the conjugate point pair of the vascular segment after the endpoint matching is completed. Traverse the coordinate points of each segment except the previously marked endpoint attributes, and find a point in a segment with the shortest distance to the epipolar line among all the points in the matching segment. The conjugate point pair is each matched point in the vascular segment.

[0062] Afterwards, the three-dimensional coordinates are calculated based on the endpoint pairs with successful endpoint attribute matching, and the three-dimensional coordinates calculated after the point pairs of the entire segment are successfully matched are subjected to B-spline fitting with the three-dimensional coordinates of the endpoint attribute matching to form a three-dimensional reconstruction of the blood vessel segment.

[0063] Based on the same inventive concept, an embodiment of the present invention discloses a medical image automatic matching system based on a two-layer neural network model, comprising:

[0064] The two-layer model module is used to construct a two-layer neural network model, in which the first-layer model inputs the angiography image and outputs the segmented blood vessel segments. Blood vessel segments with different names are displayed in different colors. The second-layer model inputs the segmentation map of blood vessel segments with different colors and outputs the endpoints of the blood vessels, including endpoints with three attributes: root node, bifurcation point and terminal point. Endpoints with different attributes are displayed in different colors.

[0065] The blood vessel segmentation module is used to perform blood vessel segmentation on two angiography images acquired at different angles by using the first layer model in the double-layer neural network model to obtain a first blood vessel segmentation result and a second blood vessel segmentation result;

[0066] An endpoint labeling module is used to label the first blood vessel segmentation result and the second blood vessel segmentation result respectively using the second layer model in the double-layer neural network model to label the blood vessel root node, bifurcation point and terminal point, so as to obtain a first endpoint labeling result and a second endpoint labeling result;

[0067] The blood vessel matching module is used to match the blood vessel segments in the two angiography images based on the pixel values ​​and the endpoint attributes based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result.

[0068] An embodiment of the present invention also discloses a computer system, including a memory, a processor, and a computer program / instructions stored in the memory and executable on the processor. When the computer program / instructions are executed by the processor, the steps of the automatic matching method of medical images based on a two-layer neural network model are implemented.

[0069] An embodiment of the present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the automatic matching method of medical images based on a two-layer neural network model.

[0070] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A medical image automatic matching method based on a two-layer neural network model, characterized in that: The following steps are involved: (1) A two-layer neural network model is constructed, in which the first layer model inputs the angiography image and outputs the segmented vascular segments, where vascular segments with different names are displayed in different colors; the second layer model inputs the segmentation map of vascular segments with different colors and outputs the endpoints of the blood vessels, including endpoints with three attributes: root node, bifurcation point, and terminal point, where endpoints with different attributes are displayed in different colors; (2) using the first layer model of the two-layer neural network model to segment the blood vessels of the two angiography images acquired at different angles, respectively, to obtain a first blood vessel segmentation result and a second blood vessel segmentation result; (3) using the second layer model in the two-layer neural network model to respectively label the root node, bifurcation point and terminal point of the blood vessel using the first blood vessel segmentation result and the second blood vessel segmentation result to obtain a first endpoint labeling result and a second endpoint labeling result; (4) Based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result, the blood vessel segments in the two angiography images are matched based on pixel values ​​and endpoint attributes.

2. The automatic matching method for medical images based on a two-layer neural network model according to claim 1, characterized in that: The two-layer neural network model is composed of two U-Net models.

3. The automatic matching method for medical images based on a two-layer neural network model according to claim 2 is characterized in that: The U-Net model is improved by adding a bidirectional attention mechanism to the skip connection module of the traditional U-Net network model to enhance the correlation between feature maps.

4. The automatic matching method for medical images based on a two-layer neural network model according to claim 1, characterized in that: The first layer model and the second layer model are trained based on the first training set and the second training set respectively. The construction of the first training set includes: collecting the original DSA coronary angiography image and performing data enhancement, dividing the blood vessels in the angiography image into different categories based on the medical naming of the coronary artery blood vessels, and distinguishing and representing them with different colors as the label data of the angiography image to construct the first training set; the construction of the second training set includes: in the output of the first layer model, setting the root node, bifurcation point and terminal point of the coronary artery, and marking them with circles of different colors, wherein the bifurcation point contains at least two pixel values, and combining the output result of the first layer model with the marked endpoints to construct the second training set.

5. The automatic matching method for medical images based on a two-layer neural network model according to claim 1, characterized in that: The matching of the blood vessel segments in the two angiography images based on the first blood vessel segmentation result and the second blood vessel segmentation result, and the first endpoint labeling result and the second endpoint labeling result, based on the pixel values ​​and the endpoint attributes, includes: Based on the first blood vessel segmentation result and the first endpoint labeling result, obtaining a first two-dimensional coordinate group corresponding to the endpoints of different blood vessel segments; Based on the second blood vessel segmentation result and the second endpoint labeling result, obtaining a second two-dimensional coordinate group corresponding to the endpoints of different blood vessel segments; For two angiography images, compare the corresponding pixel values ​​and corresponding endpoint attributes of the first two-dimensional coordinate group and the second two-dimensional coordinate group to see if they are the same. If they are the same, the corresponding two-dimensional coordinate groups have the same endpoint attributes, and the endpoint matching is completed; After the endpoint matching is completed, two-dimensional coordinates with the same pixel values ​​are found in the first and second blood vessel segmentation results as matching pairs of the same blood vessel segment. The matching endpoint pair with the same pixel value and the blood vessel segment form a whole matching blood vessel.

6. The automatic matching method for medical images based on a two-layer neural network model according to claim 5, characterized in that: The method of obtaining first two-dimensional coordinate groups corresponding to endpoints of different blood vessel segments based on the first blood vessel segmentation result and the first endpoint annotation result includes: obtaining two-dimensional coordinates of pixel values ​​at the endpoint attribute annotation in the first endpoint annotation result; searching for corresponding segmentation image pixel values ​​in the first blood vessel segmentation result according to the two-dimensional coordinates, and saving two-dimensional coordinate points with the same pixel values ​​as a group, which corresponds to the first two-dimensional coordinate group corresponding to an endpoint of a blood vessel segment; The method of obtaining second two-dimensional coordinate groups corresponding to different blood vessel segments based on the second blood vessel segmentation result and the second endpoint annotation result includes: obtaining the two-dimensional coordinates of the pixel values ​​at the endpoint attribute annotation in the second endpoint annotation result; searching for the corresponding segmentation image pixel values ​​in the second blood vessel segmentation result according to the two-dimensional coordinates, and saving the two-dimensional coordinate points with the same pixel values ​​as a group, which corresponds to the second two-dimensional coordinate group corresponding to a blood vessel segment endpoint.

7. The automatic matching method for medical images based on a two-layer neural network model according to claim 5, characterized in that: After the endpoint matching is completed, two-dimensional coordinates with the same pixel value are searched in the first blood vessel segmentation result and the second blood vessel segmentation result as a matching pair of the same blood vessel segment, including: According to the endpoint attributes and their corresponding pixel values, the root node, bifurcation point and terminal point are matched. The bifurcation point has two or more pixel values. Multiple comparisons are performed and the points with the same attributes and pixel values ​​are matched. The conjugate point pair of the vascular segment is obtained by using the epipolar line constraint matching method for the vascular segment after endpoint matching. The coordinate points of each segment except the previously marked endpoint attributes are traversed to find a point in a segment with the shortest distance to the epipolar line among all the points in the matching segment. The conjugate point pair is each matched point in the vascular segment.

8. A medical image automatic matching system based on a two-layer neural network model, characterized in that: include: The two-layer model module is used to construct a two-layer neural network model, in which the first-layer model inputs the angiography image and outputs the segmented blood vessel segments. Blood vessel segments with different names are displayed in different colors. The second-layer model inputs the segmentation map of blood vessel segments with different colors and outputs the endpoints of the blood vessels, including endpoints with three attributes: root node, bifurcation point and terminal point. Endpoints with different attributes are displayed in different colors. The blood vessel segmentation module is used to perform blood vessel segmentation on two angiography images acquired at different angles by using the first layer model in the double-layer neural network model to obtain a first blood vessel segmentation result and a second blood vessel segmentation result; An endpoint labeling module is used to label the first blood vessel segmentation result and the second blood vessel segmentation result respectively using the second layer model in the double-layer neural network model to label the blood vessel root node, bifurcation point and terminal point, so as to obtain a first endpoint labeling result and a second endpoint labeling result; The blood vessel matching module is used to match the blood vessel segments in the two angiography images based on the pixel values ​​and the endpoint attributes based on the first blood vessel segmentation result and the second blood vessel segmentation result, as well as the first endpoint labeling result and the second endpoint labeling result.

9. A computer system comprising a memory, a processor, and a computer program / instruction stored in the memory and executable on the processor, characterized in that: When the computer program / instructions are executed by the processor, the steps of the automatic matching method of medical images based on a two-layer neural network model according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the steps of the automatic matching method of medical images based on a two-layer neural network model according to any one of claims 1 to 7 are implemented.

Images