Coronary angiography image-based end-of-diastole recognition method and related equipment

Abstract

The invention provides a coronary angiography image-based end-diastolic recognition method and related equipment, and the method comprises the steps: collecting a plurality of frames of coronary angiography images to construct a coronary angiography sequence which at least comprises coronary angiography images of two cardiac cycles; the coronary angiography sequence is input into a trained recognition model for probability prediction of the end diastole to obtain a probability sequence, and the recognition model is used for extracting spatial features and time sequence features of each frame of image, calculating attention between frames and outputting a probability value of each frame of coronary angiography image belonging to the end diastole, the end-diastole image is an image with most dispersed blood vessels in the cardiac cycle, and the probability value of the end-diastole image is maximum; and extracting a wave crest from the probability sequence output by the recognition model by adopting a wave crest extraction algorithm, wherein the coronary angiography image corresponding to the wave crest is the end-diastole image. According to the method, the end-diastole prediction can be carried out on the coronary angiography sequence image, the accuracy is high, the generalization ability is high, and the robustness is good.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and analysis technology, and in particular to a method and related equipment for identifying end-diastolic heart rate based on coronary angiography images. Background Technology

[0002] The coronary arteries (CAs) are the arteries that supply blood to the heart, providing sufficient oxygen and nutrients to the myocardium. This ensures that myocardial cells have enough energy to contract, allowing the heart to continuously and effectively pump blood to all tissues and organs to meet their metabolic needs. When coronary arteries become diseased, it can lead to insufficient blood supply to the myocardium, triggering serious cardiac events and severely threatening human life and health. Therefore, timely and accurate detection of coronary artery disease is crucial. Early identification of high-risk plaques through plaque vulnerability assessment can effectively prevent or avoid acute coronary events such as myocardial infarction and sudden cardiac death.

[0003] Traditional methods for assessing plaque vulnerability, such as optical coherence tomography (OCT) and intravascular ultrasound (IVUS), require additional invasive procedures, increasing patient risk and cost. Radial Wall Strain (RWS) is a new technique for assessing coronary plaque vulnerability. It can be calculated from routine coronary angiography images without additional consumables or invasive procedures, significantly reducing the complexity and cost of assessment and providing a new perspective and method for evaluating coronary lesions. RWS requires analysis of frames before and after end-diastole. Accurate and rapid identification of end-diastole in coronary angiography images can facilitate the promotion and application of RWS technology.

[0004] Chinese invention CN116895077A discloses a cardiac phase recognition method based on dynamic echocardiography, comprising: generating data labels that satisfy a Gaussian probability distribution as targets for the cardiac phase recognition model to learn; sampling a dynamic echocardiography video sequence into several overlapping video segments using a sliding window; constructing and training a cardiac phase recognition model; and using the trained cardiac phase recognition model to identify the dynamic echocardiography video to be identified, obtaining the probability that each frame in the video represents a cardiac key phase ED or ES. In the above method, the cardiac phase recognition model first uses a CNN model to extract spatial features, and then uses the spatial features of the video segment to perform temporal modeling. Although this can meet the requirements for inferring videos containing multiple cardiac cycles to a certain extent, it is difficult to capture the spatiotemporal relationship of image frames, lacks processing for recognition errors, and has poor robustness. Summary of the Invention

[0005] The purpose of this invention is to provide a method and related equipment for identifying end-diastolic heart rate based on coronary angiography images, in order to solve the above-mentioned problems. This method can predict end-diastolic heart rate based on coronary angiography sequence images with high accuracy, strong generalization ability, and good robustness.

[0006] This invention proposes a method for identifying end-diastolic heart rate based on coronary angiography images, comprising the following steps:

[0007] A coronary angiography sequence is constructed by acquiring several frames of coronary angiography images, and the coronary angiography sequence contains coronary angiography images of at least two cardiac cycles;

[0008] The coronary angiography sequence is input into the trained recognition model to predict the probability of end-diastole, and a probability sequence is obtained. The recognition model is used to extract the spatial and temporal features of each frame image, calculate the attention between frames, and output the probability value of each frame of coronary angiography image belonging to end-diastole. Among them, the end-diastole image is the image of the most dispersed blood vessels in the cardiac cycle, and the probability value of the end-diastole image is the highest.

[0009] A peak extraction algorithm is used to extract peaks from the probability sequence output by the recognition model. The coronary angiography image corresponding to the peak is the end-diastolic image.

[0010] In one embodiment, the recognition model includes a 2D convolutional layer, a 3D convolutional layer, a pooling layer, a Transformer layer, and a fully connected layer;

[0011] The 2D convolutional layer is used to extract the spatial features of each frame of coronary angiography image;

[0012] The 3D convolutional layer is used to extract the temporal features of each frame of coronary angiography image;

[0013] The pooling layer is used to reduce the feature dimension;

[0014] The Transformer layer is used to calculate the attention between frames of coronary angiography images;

[0015] The fully connected layer is used to summarize and map the features, and output the probability value of each frame of coronary angiography image belonging to the end-diastolic phase.

[0016] In one embodiment, the coronary angiography images are selected from dynamic coronary angiography images at different frame rates.

[0017] In one embodiment, the process of training the recognition model includes:

[0018] Data annotation for each coronary angiography image: First, the annotation value of each end-diastolic image in the coronary angiography sequence is set to H1, and the annotation value of the remaining non-end-diastolic images is set to H0, where H1 is greater than H0. Then, in each cardiac cycle, several frames of non-end-diastolic images are selected and the annotation value H0 is modified to a value between H0 and H1 that satisfies a Gaussian distribution.

[0019] Set the training parameters for the recognition model and train the recognition model using several labeled coronary angiography sequences.

[0020] In one embodiment, the coronary angiography sequence input to the recognition model falls into one of the following three categories:

[0021] When the number of coronary angiography images in a coronary angiography sequence is a preset number, it is directly input into the recognition model. The recognition model outputs a probability sequence consisting of probability values ​​between H0 and H1, which is the preset number of frames.

[0022] When the number of coronary angiography images in a coronary angiography sequence is less than the preset number of frames, the images are padded to the preset number of frames and then input into the recognition model. The recognition model outputs a probability sequence consisting of probability values ​​between H0 and H1, which are the preset number of frames.

[0023] When a coronary angiography sequence has more than a preset number of coronary angiography images, the coronary angiography images are segmented based on the preset number of frames and then input into the recognition model. The recognition model outputs the probability value of each segment and splices the probability values ​​of each segment to form a probability sequence.

[0024] In one embodiment, the specific operation of extracting peaks from the probability sequence output by the recognition model using the peak extraction algorithm is as follows:

[0025] Step S101: Set the initial size of the window;

[0026] Step S102: Let the window slide on the probability sequence. Whenever the probability value of the center point of the window is greater than the probability values ​​of the positions on both sides of the window, count once, until the window slides to the end of the probability sequence and the total number is obtained.

[0027] Step S103: Starting from the initial size, increase the window size, and repeat step S102 for each window size in turn. The count value gradually increases as the window size increases until the count value no longer increases. Record the window size at this time, which is the approximate period size of the probability sequence.

[0028] Step S104: Slide windows from the initial size to the approximate period size on the probability sequence one by one, and count the number of times that the probability value of the window center point is greater than the probability values ​​of the positions on both sides of the window when each probability value is used as the window center point.

[0029] Step S105: The points with the most probability values ​​are the peaks of the probability sequence, and the coronary angiography images corresponding to each peak are the end-diastolic images.

[0030] This invention also proposes a device for identifying end-diastolic heart rate based on coronary angiography images, used to implement the method for identifying end-diastolic heart rate based on coronary angiography images as described above, comprising:

[0031] The input module is used to input a coronary angiography sequence, which includes multiple frames of coronary angiography images;

[0032] The image detection module is used to predict the probability value of each frame of coronary angiography image belonging to the end-diastolic phase and output a probability sequence. Among them, the end-diastolic image is the image of the most dispersed blood vessels in the cardiac cycle, and the probability value of the end-diastolic image is the highest.

[0033] The peak detection module is used to extract peaks from the probability sequence output by the image detection module.

[0034] The output module is used to output the coronary angiography image corresponding to the peak extracted by the peak detection module. The coronary angiography image corresponding to the peak is the end-diastolic image.

[0035] In one embodiment, the image detection module includes a feature extraction unit, a pooling unit, a self-attention unit, and a probability prediction unit;

[0036] The feature extraction unit is used to extract the spatial and temporal features of each coronary angiography image;

[0037] The pooling unit is used to reduce the dimensionality of the feature sequence output by the feature extraction unit;

[0038] The self-attention unit is used to calculate attention information between frames of coronary angiography images;

[0039] The probability prediction unit is used to predict the probability that each coronary angiography image belongs to the end-diastolic phase.

[0040] The present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method for identifying end-diastolic heart rate based on coronary angiography images as described above.

[0041] The present invention also proposes a computer storage medium storing a computer program, which, when executed by a processor, implements the method for identifying end-diastolic heart rhythm based on coronary angiography images as described above.

[0042] Compared with the prior art, the beneficial effects of the method for end-diastolic cardiac identification based on coronary angiography images of the present invention are as follows:

[0043] 1) The recognition model in this invention integrates 3D convolution and transformer self-attention mechanism on the basis of traditional convolutional neural network. It can extract local spatiotemporal features and analyze the temporal relationship between these features between frames. It can better understand the changes in the image, and is more accurate in recognizing the end-diastolic phase of the heart. It has strong generalization ability and better robustness.

[0044] 2) This invention extracts peak values ​​through peak detection algorithms such as automatic multi-scale peak finding algorithm, which can avoid the occurrence of multiple end-diastolic frames within a period, thereby filtering out some unreasonable end-diastolic points and having better noise resistance. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a method for identifying end-diastolic heart rate based on coronary angiography images according to an embodiment of the present invention.

[0046] Figure 2 This is a coronary angiography image;

[0047] Figure 3a This is a coronary angiography image of a relatively dilated blood vessel;

[0048] Figure 3b For a comparison Figure 3a Coronary angiography images showing vasoconstriction;

[0049] Figure 4 An architecture diagram for identifying the model;

[0050] Figure 5 The waveform of the probability sequence output by a recognition model;

[0051] Figure 6 This is a schematic diagram of the structure of a cardiac end-diastolic recognition device based on coronary angiography images according to an embodiment of the present invention.

[0052] Figure Labels

[0053] 1. Input module, 2. Image detection module, 21. Feature extraction unit, 22. Pooling unit, 23. Self-attention unit, 24. Probability prediction unit, 3. Peak detection module, 4. Output module. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention more readily understood, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that while many specific details are set forth in the following description to provide a thorough understanding of the invention, the invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the invention; therefore, the invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the phrase "an embodiment" or "a particular embodiment" in this application refers to a specific feature, structure, or characteristic that can be included in at least one implementation of the present invention. The phrases "in one embodiment" and "a particular embodiment" appearing in different places in this specification do not all refer to the same embodiment, nor are they embodiments that are mutually exclusive, either alone or selectively, with other embodiments. The terms "comprising" or "including" indicate the presence of the claimed feature but do not exclude the presence of one or more other features. The term "and / or" as used in this application includes any and all combinations of one or more of the related listed items.

[0056] This invention proposes a method for identifying end-diastolic heart rate based on coronary angiography images, see [link to relevant documentation]. Figure 1 It includes the following steps:

[0057] Step S1: Acquire several consecutive frames of coronary angiography images (see...) Figure 2 The black area represents the coronary arteries as visualized by contrast agent (the location of the coronary arteries changes with the heartbeat). A coronary angiography sequence must be constructed, containing images from at least two cardiac cycles. The coronary angiography images are selected from dynamic coronary angiography footage at different frame rates, such as 15fps, 7.5fps, and 30fps, to ensure good model performance across various frame rates. It's important to note that frame rate refers to the number of frames played per second in a video or animation, while the aforementioned acquisition of several consecutive coronary angiography frames (frame selection) refers to the number of image frames acquired.

[0058] Step S2: Input the coronary angiography sequence into the trained recognition model to predict the probability of end-diastolic phase, obtaining a probability sequence. The recognition model extracts spatial and temporal features from each frame, calculates the attention between frames, and outputs the probability value of each coronary angiography image belonging to end-diastole. The recognition model outputs the highest probability value for end-diastolic images. End-diastolic images are the most dispersed images of blood vessels during the cardiac cycle, such as... Figure 3a The blood vessels in Figure 3b The blood vessels in the heart become more dilated, and each cardiac cycle will show a picture of the most dilated blood vessels, which is called end-diastole.

[0059] Specifically, such as Figure 4 As shown, the recognition model includes 2D convolutional layers, 3D convolutional layers, pooling layers, Transformer layers, and fully connected layers. The 2D convolutional layers extract spatial features from each frame of coronary angiography images, the 3D convolutional layers extract temporal features from each frame of coronary angiography images, the pooling layers reduce feature dimensionality, the Transformer layers calculate the attention between frames of coronary angiography images, and the fully connected layers summarize and map the features, outputting the probability value that each frame of coronary angiography image belongs to the end-diastolic phase. In one embodiment of the recognition model, the Transformer layer contains multiple attention heads, such as nhead = 4. An attention head refers to an independent processing unit that, during attention calculation, divides the input data (including query Q, key K, and value V) into multiple parts, each processed independently through a separate attention mechanism.

[0060] Step S3: A peak extraction algorithm is used to extract peaks from the probability sequence output by the recognition model. The coronary angiography image corresponding to the peak is the end-diastolic image. This is because the output of the recognition model contains some noise. For example... Figure 5 As shown, the coronary angiography image corresponding to the upper dot (peak) of the probability sequence waveform is the end-diastolic image. By extracting the peak, multiple end-diastolic frames within one cycle can be avoided, thereby filtering out some unreasonable end-diastolic points and providing better noise resistance. One embodiment of the peak extraction algorithm of this invention uses the Automatic Multiscale-based Peak Detection (AMPD) algorithm, and the specific operation is as follows:

[0061] Step S101: Set the initial size of the window. Preferably, the initial size can be set to 3.

[0062] Step S102: Let the window slide on the probability sequence waveform. Whenever the probability value of the center point of the window is greater than the probability values ​​of the positions on both sides of the window, count once (count = count + 1) until the window slides to the end of the probability sequence and the total number count is obtained.

[0063] Step S103: Starting from the initial size, gradually increase the window size, and repeat step S102 for each size window in turn. The total count value will gradually increase as the window size increases until the count value no longer increases. Record the window size at this time, which is the approximate period size of the probability sequence.

[0064] Step S104: Slide windows from the initial size to the approximate period size on the probability sequence waveform one by one, and count the number of times that the probability value of the window center point is greater than the probability values ​​of the positions on both sides of the window when each probability value is used as the window center point.

[0065] Step S105: The points with the most probability values ​​are the peaks of the probability sequence, and the coronary angiography images corresponding to each peak are the end-diastolic images.

[0066] A similar method involves inverting the waveform of the probability sequence and then using the AMPD algorithm to extract the peaks, which can be used as the troughs of the waveform before inversion. The corresponding coronary angiography image is the end-systolic image.

[0067] In one embodiment of the present invention, 1,500 coronary angiography sequences were constructed. After being randomly shuffled, 80% of them were used for training the recognition model and 20% were used for testing the recognition model.

[0068] In one embodiment of the present invention, the specific process of training the recognition model is as follows:

[0069] 1) Data annotation for each coronary angiography image: First, the annotation value of each end-diastolic image in the coronary angiography sequence is set to H1, and the annotation value of the remaining non-end-diastolic images is set to H0, where H1 is greater than H0. Then, in each cardiac cycle, several frames of non-end-diastolic images are selected, and the annotation value H0 is modified to a value between H0 and H1 that satisfies a Gaussian distribution. In other words, the probability of each coronary angiography image belonging to the end-diastolic phase is set to H0 to H1. This step is to adapt to the smooth output of the recognition model.

[0070] For example, H1 can be set to 1, and H0 can be set to 0. The data annotation process described above is as follows: First, the annotation value of each end-diastolic image in the coronary angiography sequence is set to 1, and the annotation value of the remaining non-end-diastolic images is set to 0. Then, Gaussian smoothing is applied to the coronary angiography images near the end-diastolic images within each cardiac cycle. For instance, in each cardiac cycle, 1-2 frames of non-end-diastolic images are selected and their annotation values ​​are modified. The specific modification rule is that the annotation values ​​for images 1, 2, and 3 frames away from the end-diastolic image are 0.8, 0.6, and 0.4, respectively. This also indicates that the range of the annotation values ​​is 0-1, meaning that the probability of each coronary angiography image belonging to the end-diastolic phase is set to 0-1.

[0071] 2) Set the training parameters of the recognition model and train the recognition model using several labeled coronary angiography sequences.

[0072] When training the recognition model, the RepNet can be modified. The training parameters are set as follows: a total of 20 epochs, with a batch size of 4 in each epoch. An epoch refers to the number of times the entire training dataset is traversed and used to update the model weights. The batch size refers to the number of samples selected from the training dataset before each parameter update. The optimizer uses Adamw, and the loss function is MAE loss. The optimizer parameters include learning rate, momentum, gradient squared decay (beta2), gradient decay rate (beta1), and weight regularization coefficient (weight_decay). The learning rate controls the step size of parameter updates, momentum accelerates convergence, gradient squared decay controls the decay rate of the squared gradient, gradient decay rate adjusts the decay rate of the first moment estimate, and weight regularization coefficient prevents overfitting and enhances the generalization ability of the recognition model.

[0073] The coronary angiography sequences input into the recognition model fall into the following three categories:

[0074] When the number of coronary angiography images in a coronary angiography sequence is a preset number, it is directly input into the recognition model. At this time, the recognition model outputs a probability sequence consisting of probability values ​​between H0 and H1, which is the preset number of frames.

[0075] When the number of coronary angiography images in a coronary angiography sequence is less than the preset number of frames, the preset number of frames is padded with a label value H0 representing the non-end-diastolic phase before being input into the recognition model. At this time, the recognition model outputs a probability sequence consisting of probability values ​​between H0 and H1, which are the preset number of frames.

[0076] When a coronary angiography sequence contains more than a preset number of coronary angiography images, the images are divided into multiple segments based on the preset number of frames and then input into the recognition model. A certain number of frames overlap between adjacent segments to facilitate stitching. The recognition model then outputs the probability value of each segment, and these probability values ​​are then concatenated to form a probability sequence.

[0077] The preset frame count refers to the set frame count for the input recognition model. This frame count is determined based on the size of the recognition model, and the feature dimension is generally a power of 2. In one embodiment of the present invention, the preset frame count is 64. This is because a typical DICOM (Digital Imaging and Communications in Medicine) consists of around tens of frames, most of which can be completed within 64 frames, requiring only one pass through the model, while a small number require two to three passes. Furthermore, a coronary angiography image of a cardiac cycle consists of around 10-15 frames, and 64 frames can cover multiple cycles, resulting in more robust performance.

[0078] The following example illustrates the architecture of the recognition model by predicting the end-diastolic probability of a coronary angiography sequence to be identified from 64 frames of coronary angiography images. (See [link to relevant documentation]). Figure 4 :

[0079] 1) Input the 64-frame coronary angiography sequence to be identified into the identification model;

[0080] 2) First, a 2D convolutional layer (which can use ResNet50) extracts the spatial features of each frame of coronary angiography image. The spatial feature sequence is 64*1024*7*7.

[0081] 3) Subsequently, the 3D convolutional layer further extracts the temporal features of each frame of coronary angiography image, and the resulting feature sequence is 64*512*7*7;

[0082] 4) Then, the feature dimension is reduced by pooling layers to obtain a 64*512 feature map;

[0083] 5) Then, after passing through a Transformer layer with a dimension of 512 and nhead=4, the relevant attention information between frames is obtained, resulting in 64*512 features;

[0084] 6) Finally, the features are summarized and mapped through a fully connected layer, and the probability values ​​of each frame of coronary angiography image belonging to the end-diastolic phase are output, resulting in 64 probability values ​​between 0 and 1.

[0085] This invention also proposes a device for identifying end-diastolic heart rate based on coronary angiography images, used to implement the method for identifying end-diastolic heart rate based on coronary angiography images as described above. The device includes an input module 1, an image detection module 2, a peak detection module 3, and an output module 4. (See [link to relevant documentation]). Figure 6 Input module 1 is used to input the coronary angiography sequence, which includes multiple frames of coronary angiography images. Image detection module 2 is used to predict the probability value of each frame of coronary angiography image belonging to end-diastole, and outputs a probability sequence. Among them, the end-diastolic image is the image with the most dispersed blood vessels in the cardiac cycle, and the probability value of the end-diastolic image is the highest. Peak detection module 3 is used to extract peaks from the probability sequence output by image detection module 2. Output module 4 is used to output the coronary angiography image corresponding to the peak extracted by peak detection module 3. The coronary angiography image corresponding to the peak is the end-diastolic image.

[0086] An image detection module 2 according to one embodiment of the present invention includes a feature extraction unit 21, a pooling unit 22, a self-attention unit 23, and a probability prediction unit 24. The feature extraction unit 21 is used to extract the spatial and temporal features of each coronary angiography image, the pooling unit 22 is used to reduce the dimensionality of the feature sequence output by the feature extraction unit 21, the self-attention unit 23 is used to calculate the attention information between frames of the coronary angiography image, and the probability prediction unit 24 is used to predict the probability value of each coronary angiography image belonging to the end-diastolic phase.

[0087] The present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method for identifying end-diastolic heart rate based on coronary angiography images as described above.

[0088] The present invention also proposes a computer storage medium storing a computer program, which, when executed by a processor, implements the method for identifying end-diastolic heart rate based on coronary angiography images as described above.

[0089] It should be noted that, in this application, unless otherwise explicitly specified and limited, terms such as "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can also be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0090] The present invention has the following beneficial effects:

[0091] 1) The recognition model in this invention integrates 3D convolution and transformer self-attention mechanism on the basis of traditional convolutional neural network. It can extract local spatiotemporal features and analyze the temporal relationship between these features between frames. It can better understand the changes in the image, and is more accurate in recognizing the end-diastolic phase of the heart. It has strong generalization ability and better robustness.

[0092] 2) This invention extracts peak values ​​through peak detection algorithms such as automatic multi-scale peak finding algorithm, which can avoid the occurrence of multiple end-diastolic frames within a period, thereby filtering out some unreasonable end-diastolic points and having better noise resistance.

[0093] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0094] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0095] The method steps of this application illustrated in several different exemplary embodiments are merely illustrative. Although only a few embodiments are described in detail in this disclosure, those who consult this disclosure will readily understand that many modifications (e.g., structures, parameter values, etc.) are possible without substantially departing from the novel teachings and advantages of the subject matter described in this application. Therefore, all such modifications are intended to be included within the scope of this invention. The order or sequence of any process or method steps may be changed or rearranged according to alternative embodiments. Various corresponding modifications and variations can be made by those skilled in the art according to this invention without departing from its spirit and essence, but all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for identifying end-diastolic heart rhythm based on coronary angiography images, characterized in that, Includes the following steps: A coronary angiography sequence is constructed by acquiring several frames of coronary angiography images, and the coronary angiography sequence contains coronary angiography images of at least two cardiac cycles; The coronary angiography sequence is input into the trained recognition model to predict the probability of end-diastole, and a probability sequence is obtained. The recognition model is used to extract the spatial and temporal features of each frame image, calculate the attention between frames, and output the probability value of each frame of coronary angiography image belonging to end-diastole. Among them, the end-diastole image is the image of the most dispersed blood vessels in the cardiac cycle, and the probability value of the end-diastole image is the highest. A peak extraction algorithm is used to extract peaks from the probability sequence output by the recognition model. The coronary angiography image corresponding to the peak is the end-diastolic image.

2. The method for identifying end-diastolic heart rate based on coronary angiography images according to claim 1, characterized in that, The recognition model includes 2D convolutional layers, 3D convolutional layers, pooling layers, Transformer layers, and fully connected layers; The 2D convolutional layer is used to extract the spatial features of each frame of coronary angiography image; The 3D convolutional layer is used to extract the temporal features of each frame of coronary angiography image; The pooling layer is used to reduce the feature dimension; The Transformer layer is used to calculate the attention between frames of coronary angiography images; The fully connected layer is used to summarize and map the features, and output the probability value of each frame of coronary angiography image belonging to the end-diastolic phase.

3. The method for identifying end-diastolic heart rate based on coronary angiography images according to claim 1, characterized in that, The coronary angiography images are selected from dynamic coronary angiography images at different frame rates.

4. The method for identifying end-diastolic heart rate based on coronary angiography images according to claim 1, characterized in that, The process of training the recognition model includes: Data annotation for each coronary angiography image: First, the annotation value of each end-diastolic image in the coronary angiography sequence is set to H1, and the annotation value of the remaining non-end-diastolic images is set to H0, where H1 is greater than H0. Then, in each cardiac cycle, several frames of non-end-diastolic images are selected and the annotation value H0 is modified to a value between H0 and H1 that satisfies a Gaussian distribution. Set the training parameters for the recognition model and train the recognition model using several labeled coronary angiography sequences.

5. The method for identifying end-diastolic heart rate based on coronary angiography images according to claim 4, characterized in that, The coronary angiography sequences input into the recognition model fall into the following three categories: When the number of coronary angiography images in a coronary angiography sequence is a preset number, it is directly input into the recognition model. The recognition model outputs a probability sequence consisting of probability values ​​between H0 and H1, which is the preset number of frames. When the number of coronary angiography images in a coronary angiography sequence is less than the preset number of frames, the images are padded to the preset number of frames and then input into the recognition model. The recognition model outputs a probability sequence consisting of probability values ​​between H0 and H1, which are the preset number of frames. When a coronary angiography sequence has more than a preset number of coronary angiography images, the coronary angiography images are segmented based on the preset number of frames and then input into the recognition model. The recognition model outputs the probability value of each segment and splices the probability values ​​of each segment to form a probability sequence.

6. The method for identifying end-diastolic heart rate based on coronary angiography images according to claim 1, characterized in that, The specific operations for extracting peaks from the probability sequence output by the recognition model using the peak extraction algorithm include: Step S101: Set the initial size of the window; Step S102: Let the window slide on the probability sequence. Whenever the probability value of the center point of the window is greater than the probability values ​​of the positions on both sides of the window, count once, until the window slides to the end of the probability sequence and the total number is obtained. Step S103: Starting from the initial size, increase the window size, and repeat step S102 for each window size in turn. The count value increases as the window size increases until the count value no longer increases. Record the window size at this time, which is the approximate period size of the probability sequence. Step S104: Slide windows from the initial size to the approximate period size on the probability sequence one by one, and count the number of times that the probability value of the window center point is greater than the probability values ​​of the positions on both sides of the window when each probability value is used as the window center point. Step S105: The points with the most probability values ​​are the peaks of the probability sequence, and the coronary angiography images corresponding to each peak are the end-diastolic images.

7. A device for identifying end-diastolic heart rhythm based on coronary angiography images, used to implement the method for identifying end-diastolic heart rhythm based on coronary angiography images as described in any one of claims 1-6, characterized in that, include: The input module is used to input a coronary angiography sequence, which includes multiple frames of coronary angiography images; The image detection module is used to predict the probability value of each frame of coronary angiography image belonging to the end-diastolic phase and output a probability sequence. Among them, the end-diastolic image is the image of the most dispersed blood vessels in the cardiac cycle, and the probability value of the end-diastolic image is the highest. The peak detection module is used to extract peaks from the probability sequence output by the image detection module. The output module is used to output the coronary angiography image corresponding to the peak extracted by the peak detection module. The coronary angiography image corresponding to the peak is the end-diastolic image.

8. The cardiac end-diastolic identification device based on coronary angiography images according to claim 7, characterized in that, The image detection module includes a feature extraction unit, a pooling unit, a self-attention unit, and a probability prediction unit; The feature extraction unit is used to extract the spatial and temporal features of each coronary angiography image; The pooling unit is used to reduce the dimensionality of the feature sequence output by the feature extraction unit; The self-attention unit is used to calculate attention information between frames of coronary angiography images; The probability prediction unit is used to predict the probability that each coronary angiography image belongs to the end-diastolic phase.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method for identifying end-diastolic heart rate based on coronary angiography images as described in any one of claims 1-6.

10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for identifying end-diastolic heart rate based on coronary angiography images as described in any one of claims 1-6.

Images