Coronary vessel registration method, system and medium for heart two-dimensional image and three-dimensional image sets

Abstract

The invention relates to a coronary vessel registration method and system of a heart two-dimensional image set and a heart three-dimensional image set and a medium. The method comprises the steps that the heart two-dimensional image set and the heart three-dimensional image set are acquired; matching the two-dimensional image with different view angle images of the three-dimensional image set to obtain a target view angle image matched with the two-dimensional image; identifying coronary blood vessels in the two-dimensional image and the target view angle image; and aligning the coronary blood vessels in the two-dimensional image and the target view angle image, and fusing to obtain a coronary blood vessel registration result. The registration method provided by the invention is a multi-stage coronary blood vessel registration method, and local coronary blood vessel registration can be well carried out through view angle matching, recognition, alignment and fusion.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a method, system and medium for coronary vessel registration of a two-dimensional cardiac image and a three-dimensional image set. Background Art

[0002] Most current methods for multimodal image registration are single-stage. This means they use only a single model or step to register two images. For example, traditional methods utilize global or local image features (such as mutual information, SIFT feature points, etc.) for registration, directly maximizing the statistical dependence of the two images. Deep learning methods use an end-to-end network as input, taking two images (the source image and the target image) and outputting the registered image.

[0003] There are currently some multi-stage image registration methods, but they are all aimed at general image registration. Deep learning-based methods can handle some relatively complex image registration scenarios, but they generally improve the accuracy of overall image registration rather than the registration of important local features, such as coronary arteries. Therefore, they are not well applied to the registration of specific regional features. In addition, they are less effective for complex deformations, such as those with large displacements and nonlinear deformations; they may also fall into local optimal solutions; or they focus more on the overall degree of image registration.

[0004] Therefore, how to deal with the registration problem in relatively complex scenes without overfitting or underfitting and achieve ideal registration results is a technical problem that needs to be solved urgently in this field. Summary of the Invention

[0005] To solve at least one of the above technical problems, the present invention provides a method for coronary vessel registration of a two-dimensional cardiac image and a three-dimensional image set, comprising:

[0006] S1: Acquire a set of two-dimensional and three-dimensional cardiac images;

[0007] S2: Match the two-dimensional image with images from different perspectives of the three-dimensional image set to obtain a target perspective image that matches the two-dimensional image;

[0008] S3: Identify coronary vessels in the 2D image and the target view image;

[0009] S4: Align the coronary vessels in the two-dimensional image and the target view image, and then fuse them to obtain the coronary vessel registration result.

[0010] Furthermore, the matching step includes:

[0011] The target view images corresponding to the 2D image and 3D image sets are used as positive samples; the other view images in the 3D image set are used as negative samples to construct a training dataset.

[0012] Extract features of two-dimensional images and three-dimensional image sets respectively;

[0013] Based on the training dataset, the model is optimized by using the contrast loss function to determine the target perspective image that matches the two-dimensional image.

[0014] Furthermore, the comparison loss function adopts formula (1):

[0015]

[0016] in, is a pair of positive samples of the i-th perspective; C is the number of perspectives; f i ,g i is x i ,y i The embedded feature vector of ; τ is the temperature parameter.

[0017] Furthermore, the coronary vessel identification step includes:

[0018] Based on the neural network, the coronary artery pixel set is segmented;

[0019] According to the density of discrete points in the pixel set, the discrete points in the high-density area are extended by quadratic curves, while the areas in the low-density area are connected by linear connection to form a complete coronary artery.

[0020] Furthermore, the step of coronary artery alignment includes:

[0021] Point sampling is performed on the coronary arteries in the identified two-dimensional image and the target view image to obtain a number of corresponding key points. Based on the location of the coronary arteries, more key points are collected proximally than distally, and proximal key points are assigned higher weights than distal key points.

[0022] The distance between corresponding key points is minimized according to the weights to obtain the aligned two-dimensional image and the coronary vessels in the target perspective image.

[0023] Furthermore, the distance T between corresponding key points is minimized according to the weight * , using formula (2):

[0024]

[0025] Where T(·) is the affine transformation from the target view image to the two-dimensional image; N is the number of key point pairs; w i represents the weight of the i-th key point.

[0026] Furthermore, the two aligned images are weighted fused using formula (3):

[0027] fusion=α*image+(1-α)*cta(3)

[0028] Among them, fusion is the fused image, image is the two-dimensional image, cta is the target perspective image, and α∈[0,1] is the weighting coefficient.

[0029] Furthermore, it also includes:

[0030] S5: Obtain the updated two-dimensional image and calculate the offset of the key points based on weighted superposition;

[0031] S6: Determine whether the offset exceeds a set threshold; if not, do not process; if yes, re-execute steps S2-S4 with the updated two-dimensional image.

[0032] On the other hand, the present invention also provides a coronary vessel registration system for a two-dimensional cardiac image and a three-dimensional image set, which is used to execute any of the above registration methods.

[0033] On the other hand, the present invention further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned registration methods.

[0034] The coronary artery registration method for a two-dimensional cardiac image and a three-dimensional image set provided by the present invention is a multi-stage coronary artery registration method that can better perform local coronary artery registration through view matching, recognition, alignment and fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flowchart of an embodiment of a coronary vessel registration method according to the present invention;

[0036] Figure 2 A flowchart of another embodiment of the coronary vessel registration method of the present invention;

[0037] Figure 3 This is an example of the matching result of the coronary vessel registration method of the present invention;

[0038] Figure 4 This is an example of the recognition result of the coronary vessel registration method of the present invention;

[0039] Figure 5 is an example of the original image;

[0040] Figure 6 is an example of a set of predicted pixels;

[0041] Figure 7An example of an image with a prediction;

[0042] Figure 8 An example of an image with a comment;

[0043] Figure 9 An example of the fusion result of the coronary vessel registration method of the present application;

[0044] Figure 10 A schematic diagram of the initial state of the interface of the heart CTA alignment software;

[0045] Figure 11 A schematic diagram of the result state of the interface of the heart CTA alignment software. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0047] It should be noted that if the embodiments of the present application involve directional indications, such as up, down, left, right, front, back, etc., the directional indications are only used to explain the relative position relationship, movement condition, etc. between components in a certain posture, and if the certain posture changes, the directional indications also change accordingly. In addition, if the embodiments of the present application involve descriptions such as “first, second”, “S1, S2”, “step one, step two” and the like, such descriptions are only for description purposes, and cannot be understood as indicating or implying the relative importance of the technical features indicated or the number of technical features indicated, or indicating the execution order of the method, etc. Those skilled in the art can understand that any modification within the technical concept of the present application without deviating from the technical points of the present application should be included in the protection scope of the present application.

[0048] Term explanation:

[0049] CTA (Computed Tomography Angiography) is a combination of CT enhancement technology and thin-layer, large-range, fast scanning technology, which can clearly display the details of blood vessels in each part of the body through reasonable post-processing.

[0050] Fast SAM (Fast Segment Anything) model can be used for object segmentation. The Fast SAM model is pre-trained through a large-scale data set, and can be used in various scenes through zero-shot.

[0051] UNet is a medical image segmentation model composed of convolutional neural networks, which is used to automatically segment regions of interest (organs, lesions, etc.) in medical images.

[0052] like Figure 1 、 2 As shown, the present invention provides a method for coronary artery registration of a two-dimensional cardiac image and a three-dimensional image set. This is an image processing method that focuses on image processing technology to segment, match, align, and fuse the two images to improve the accuracy of coronary artery extraction. For example, the improvement in image extraction accuracy can be used, but is not limited to, to assist doctors in more clearly viewing coronary arteries in clinical medical scenarios to perform more accurate heart bypass surgery. Specifically, the method includes:

[0053] S1: Acquire a set of two-dimensional and three-dimensional cardiac images;

[0054] Specifically, two-dimensional and three-dimensional cardiac images can be acquired through X-rays (digital radiography (DR), computed radiography (CR)), echocardiography, digital subtraction angiography (DSA), optical coherence tomography (OCT), computed tomography (CT), and CTA magnetic resonance imaging (MRI). For example, this image processing technology can be used to improve the accuracy of key region extraction in heart bypass surgery to assist doctors in clearly identifying coronary arteries.

[0055] More specifically, the two-dimensional cardiac image may be, but is not limited to, an intraoperative cardiac image. A data preprocessing module may be optionally configured to cut the intraoperative cardiac video stream data into image data, and filter out some of the more blurred images based on some image processing algorithms (such as the Laplacian operator, etc.) to ensure the clarity and completeness of subsequent images, thereby further improving the registration accuracy.

[0056] More specifically, the three-dimensional image set may be, but is not limited to, multiple preoperative CTA images, that is, a set of multiple projection images of three-dimensional CTA, including images from various angles of the heart.

[0057] More preferably, it is also optional but not limited to configuring a heart segmentation module. Based on the captured two-dimensional images and three-dimensional image sets, the FastSAM algorithm can be optionally used to segment the heart. By providing 2 to 3 points (prompts) to indicate the approximate position of the heart, the heart is then segmented and located at the center of the entire image to obtain a two-dimensional image and a three-dimensional image set of the heart, thereby eliminating the influence of the surrounding environment on the subsequent registration accuracy.

[0058] S2: Match the two-dimensional image with images from different perspectives of the three-dimensional image set to obtain a target perspective image that matches the two-dimensional image;

[0059] Specifically, several feature points can be extracted from images of different viewpoints in the 2D image and 3D image sets, and the target viewpoint image closest to the 2D image in the 3D image set can be obtained by comparing the feature points.

[0060] Taking the above application scenario as an example, the intraoperative cardiac image is matched with the projection images of the 3D CTA image at different viewing angles, and the CTA viewing angle image closest to the intraoperative cardiac image is found, that is, one image is selected to be registered with the intraoperative cardiac image, such as Figure 3 As shown, this can improve the accuracy of subsequent coronary vessel alignment.

[0061] Preferably, step S2 may optionally include but is not limited to:

[0062] S21: Build a training dataset using the corresponding target view images in the 2D image and 3D image set as positive samples and the other view images in the 3D image set as negative samples;

[0063] Taking the above application scenario as an example, the corresponding projection images in the intraoperative cardiac image and CTA image of the same patient can be selected as positive samples, and the other projection images can be selected as negative samples.

[0064] S22: extract features of two-dimensional image and three-dimensional image sets respectively;

[0065] For example, optionally but only limited to using a visual encoder to extract features of the intraoperative cardiac image and the CTA image respectively.

[0066] S23: Based on the training data set, the model is optimized by contrasting the loss function to determine the target perspective image that matches the two-dimensional image; that is, by contrasting the loss function, the model is optimized so that the positive sample features are as close as possible and the negative sample features are as far away as possible, and the loss is continuously reduced to reach the preset iteration termination condition.

[0067] Specifically, the contrast loss function can be selected as formula (1):

[0068]

[0069] in, is a pair of positive samples of the i-th view; for example, an intraoperative cardiac image and the corresponding projection image in 3D CTA; C is the number of viewpoints (categories); f i ,g i is x i ,y i The embedded feature vector of ; τ is the temperature parameter.

[0070] S3: Identify the coronary vessels in the 2D image and the target view image, e.g. Figure 4 As shown;

[0071] Specifically, any image recognition method can be used to identify the target object - the coronary blood vessels. More specifically, since the blood vessel features are weak and often covered by fat, it is challenging to identify the coronary blood vessels. In addition, since the target perspective image is a perspective image extracted from a three-dimensional image set, it is relatively clear, so it is relatively easy to identify the target object - the coronary blood vessels in it. However, two-dimensional images are relatively blurry, so it is relatively difficult to detect and identify coronary blood vessels from intraoperative cardiac images. Therefore, as another key to the present invention, based on the actual application of coronary blood vessels, a method for identifying coronary blood vessels is proposed, and the coronary blood vessel identification step preferably includes:

[0072] S31: Based on the neural network, the coronary artery pixel set is segmented;

[0073] Specifically, a coronary vessel recognition model can be constructed based on U-Net and a multimodal cross-attention mechanism. Images including coronary vessels are collected and the frames where the coronary vessels are located are marked to train the coronary vessel recognition model. Finally, through data cleaning and model fine-tuning, coronary vessels with obvious features (located on the surface of the heart and mostly not covered by fat) are identified to obtain a coronary vessel pixel set, such as Figure 5-8 As shown in the figure. Due to the characteristics of coronary arteries and the influence of the surrounding environment, the model is often unable to segment the entire coronary artery, resulting in a discontinuous collection of pixels. More specifically, since there are many coronary arteries, but only one lesion may be present, during the recognition process, it is possible to manually frame the coronary artery region of interest (the vessel to be observed / registered), and then use the model to identify the pixel collection of the coronary artery of interest.

[0074] S32: According to the density of the discrete points in the pixel set, the discrete points in the high-density area are extended by quadratic curves, and the areas in the low-density area are connected by linear connection to form a complete coronary artery.

[0075] This embodiment provides an optimal method for identifying coronary arteries. First, based on a trained neural network, coronary vessel pixels are segmented. However, due to the characteristics of coronary vessels and the surrounding environment, it is found that the trained neural network cannot fully segment the coronary vessels, resulting in only discontinuous pixel sets, i.e., discrete points. Based on this, the model identifies key inflection points of coronary vessels in sections that are curved and have inflection points, while sections are straight. To address this characteristic, the density of discrete points in the pixel set is considered. If the model identifies a dense area of ​​discrete points, these discrete sets are preferably connected using quadratic curve extension to form a relatively complete coronary vessel. If the model identifies a sparse area of ​​discrete points, these discrete sets are preferably connected using linear connections to form a relatively complete coronary vessel, thereby facilitating subsequent coronary vessel alignment. This method integrates a neural network model with a numerical algorithm, incorporating the inherent characteristics of coronary vessels to identify coronary vessels, thereby facilitating subsequent coronary vessel alignment.

[0076] S4: Align the coronary vessels in the two-dimensional image and the target view image, and then fuse them to obtain the coronary vessel registration result, such as Figure 9 shown.

[0077] Specifically, several sampling points can be collected as key points in the coronary vessels in the 2D image and the target view image identified in step S3. Registration and alignment are then performed based on the key points, and then fused to obtain a coronary vessel registration result. Specifically, using the above application scenario as an example, a coronary vessel alignment module can be optionally configured to align the intraoperative coronary vessels with the coronary vessels in the CTA projection image at a certain viewpoint, thereby more conveniently providing the physician with CTA information on these coronary vessels.

[0078] More preferably, the steps of coronary artery alignment may optionally include but are not limited to:

[0079] S41: performing point sampling on the coronary arteries in the identified two-dimensional image and the target view image to obtain a number of corresponding key points; based on the position of the coronary arteries, more key points are collected proximal to the coronary arteries than distal to the coronary arteries; and the proximal key points are assigned a higher weight than the distal key points;

[0080] S42: Minimize the distance between corresponding key points according to the weights to obtain the aligned two-dimensional image and the coronary artery in the target view image.

[0081] In this embodiment, a preferred embodiment of coronary vessel alignment is provided. Considering that the coronary vessels provided by the coronary vessel detection and identification module are in the form of masks (i.e., lines) rather than key points, key point registration cannot be performed directly. Therefore, it is necessary to convert the line alignment task into a point alignment task. First, point sampling is performed on the coronary vessels of each image, and then the sampled point sets are aligned. At the same time, since the proximal end of the coronary vessels (close to the aorta) has a larger blood flow and is more important than the distal end, during sampling, more key points need to be collected at the proximal end of the coronary vessels than at the distal end; and the proximal key points are assigned higher weights than the distal key points; so that the importance of the different weights of the points can be aligned, making it more accurate to match the proximal part of the vessels without neglecting the distal part.

[0082] More preferably, the distance T between corresponding key points is minimized according to the weight * , optionally using formula (2):

[0083]

[0084] Where T(·) is the affine transformation from the target view image to the two-dimensional image, for example: the affine transformation from a certain view projection image in CTA to the intraoperative cardiac image; N is the number of key point pairs; w i represents the weight of the i-th key point.

[0085] More preferably, the fusion of coronary vessels, taking the above application scenario as an example, is to fuse the aligned intraoperative cardiac coronary vessel images and CTA coronary vessel images to provide doctors with clearer coronary vessel alignment information, such as Figure 4 As shown. The vascular fusion module can be optionally configured to perform weighted fusion of the two aligned images using the formula (3):

[0086] fusion=α*image+(1-α)*cta(3)

[0087] Where fusion is the fused image, image is a two-dimensional image, such as an intraoperative cardiac image, cta is the target view image, such as the corresponding view image of a CTA image, and α∈[0,1] is the weighting coefficient.

[0088] More preferably, using the aforementioned application scenario as an example, in order to align and fuse intraoperative cardiac images and CTA images in real time, consider tracking vascular features within the video. Because the model's recognition of coronary vessels is less stable (due to the task itself), and some video frames appear blurry, resulting in even poorer predictions, consideration is given to using tracking based on results from clearer video frames. Furthermore, considering that coronary vessel tracking only requires alignment of a few key points, tracking a specific set of key points can be selected, significantly reducing computational effort without compromising performance.

[0089] Therefore, the registration method also includes:

[0090] S5: Obtain the updated two-dimensional image and calculate the offset of the key points based on weighted superposition;

[0091] S6: Determine whether the offset exceeds a set threshold; if not, do not process; if yes, re-execute steps S2-S4 with the updated two-dimensional image.

[0092] In this embodiment, a video tracking step is given. Taking the above-mentioned application scenario as an example, if the intraoperative cardiac image is continuously updated, the updated two-dimensional image needs to be superimposed and the offset of the key points needs to be calculated according to the weights. If the offset exceeds the set threshold, it means that the current alignment, registration, and fusion results are no longer suitable for the new intraoperative cardiac image, and steps S2-S4 need to be repeated for more accurate alignment, registration, and fusion.

[0093] The coronary artery registration method for a set of two-dimensional and three-dimensional cardiac images provided by the present invention is a multi-stage coronary artery registration method that can effectively perform local coronary artery registration through perspective matching, recognition, alignment, and fusion. An optional application scenario is to assist doctors in performing heart bypass surgery in clinical medical settings. In this scenario, the key points of pre-protection are:

[0094] A view matching module is used to roughly align the intraoperative cardiac image with the CTA image. In a preferred embodiment, this module uses multimodal contrastive learning to roughly align the cardiac image with the CTA image.

[0095] The coronary artery detection and recognition module follows the viewpoint matching module. This module uses image segmentation networks such as UNet, trained and fine-tuned on a proprietary cardiac image dataset, to achieve the ability to identify coronary arteries.

[0096] Coronary artery alignment and fusion module: In this module, the key point registration algorithm is improved according to the task characteristics, the distance between corresponding key points is minimized, and accurate proximal alignment is achieved while considering distal alignment.

[0097] The video tracking module tracks the heart features to determine whether the current registration result is consistent with the current actual situation.

[0098] On the other hand, the present invention also provides a coronary vessel registration system for a set of two-dimensional cardiac images and three-dimensional images, for implementing any of the above registration methods.

[0099] Preferably, the system primarily includes: a data preprocessing module, a heart segmentation module, a viewpoint matching module, a coronary vessel detection and identification module, a coronary vessel alignment and fusion module, and a video tracking module, to respectively implement the corresponding steps of the above-mentioned registration method, which will not be described in detail here. More preferably, the system includes: an input module for inputting intraoperative cardiac video data and preoperative CTA images, and achieving registration and fusion of intraoperative cardiac images captured from the intraoperative video data with multiple projection images of preoperative 3D CTA using the above-mentioned registration method, thereby improving the clarity of coronary vessel identification.

[0100] More preferably, the registration method and registration system can be optionally applied to Figure 10 The cardiac CTA alignment software shown in the figure can upload the intraoperative cardiac video and CTA image set by clicking the button on the upper left. Clicking the "Start / Pause" button will start the registration and alignment process of the intraoperative cardiac image and CTA image. After each frame is aligned and fused, the result is as follows: Figure 11 As shown in the figure, the image displayed in the right area is the result image, the upper left corner is the current heart video frame, the upper right corner is the CTA image aligned with it, the lower right corner is the result image after blood vessel segmentation, and the lower right corner is the result image after final fusion.

[0101] On the other hand, the present invention further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned registration methods.

[0102] On the other hand, the present invention further provides a terminal device, comprising a memory and a processor; the memory stores program code executable by the processor; the program code is used to execute any of the above-mentioned registration methods.

[0103] Exemplarily, the program code may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the program code in the terminal device.

[0104] The terminal device may be a computing device such as a desktop computer, laptop, PDA, or cloud server. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the terminal device may also include input / output devices, network access devices, buses, and the like.

[0105] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0106] The memory may be an internal storage unit of the terminal device, such as a hard disk or memory. The memory may also be an external storage device of the terminal device, such as a plug-in hard disk equipped on the terminal device, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Furthermore, the memory may include both an internal storage unit of the terminal device and an external storage device. The memory is used to store the program code and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is about to be output.

[0107] The above-mentioned registration system, computer storage medium and terminal device are created based on the above-mentioned registration method. Their technical functions and beneficial effects are not repeated here. The various technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0108] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for coronary artery registration of a two-dimensional cardiac image and a three-dimensional cardiac image set, characterized in that: include: S1: Acquire a set of two-dimensional and three-dimensional cardiac images; S2: Match the two-dimensional image with images from different perspectives of the three-dimensional image set to obtain a target perspective image that matches the two-dimensional image; S3: Identify coronary vessels in the 2D image and the target view image; S4: Align the coronary vessels in the two-dimensional image and the target view image, and then fuse them to obtain the coronary vessel registration result.

2. The method according to claim 1, characterized in that The matching steps include: The target view images corresponding to the 2D image and 3D image sets are used as positive samples; the other view images in the 3D image set are used as negative samples to construct a training dataset. Extract features of two-dimensional images and three-dimensional image sets respectively; Based on the training dataset, the model is optimized by using the contrast loss function to determine the target perspective image that matches the two-dimensional image.

3. The method according to claim 2, characterized in that The contrast loss function uses formula (1): in, is a pair of positive samples of the i-th perspective; C is the number of perspectives; f i ,g i is x i ,y i The embedded feature vector of ; τ is the temperature parameter.

4. The method according to claim 1, wherein Coronary vessel identification steps include: Based on the neural network, the coronary artery pixel set is segmented; According to the density of discrete points in the pixel set, the discrete points in the high-density area are extended by quadratic curves, while the areas in the low-density area are connected by linear connection to form a complete coronary artery.

5. The method according to claim 1, wherein The steps for coronary artery alignment include: Point sampling is performed on the coronary arteries in the identified two-dimensional image and the target view image to obtain a number of corresponding key points. Based on the location of the coronary arteries, more key points are collected proximally than distally, and proximal key points are assigned higher weights than distal key points. The distance between corresponding key points is minimized according to the weights to obtain the aligned two-dimensional image and the coronary vessels in the target perspective image.

6. The method according to claim 5, characterized in that Minimize the distance T between corresponding key points according to the weight * , using formula (2): Where T(·) is the affine transformation from the target view image to the two-dimensional image; N is the number of key point pairs; w i represents the weight of the i-th key point.

7. The method according to claim 5, characterized in that Perform weighted fusion of the two aligned images using formula (3): fusion=α*image+(1-α)*cta(3) Among them, fusion is the fused image, image is the two-dimensional image, cta is the target perspective image, and α∈[0,1] is the weighting coefficient.

8. The method according to any one of claims 1 to 7, characterized in that Also includes: S5: Obtain the updated two-dimensional image and calculate the offset of the key points based on weighted superposition; S6: Determine whether the offset exceeds the set threshold; If not, no processing is performed. If yes, steps S2 to S4 are re-executed with the updated two-dimensional image.

9. A coronary artery registration system for a cardiac two-dimensional image and a three-dimensional image set, characterized in that: Used to perform the method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that An executable program code is stored; the executable program code is used to execute the registration method according to any one of claims 1 to 8.

Images