Coronary artery segmentation method, equipment, and media based on improved U-Net neural network
By combining an improved U-Net neural network with multiple modules and loss functions, the problems of information loss and boundary ambiguity in coronary artery segmentation are solved, achieving efficient and accurate coronary artery segmentation and improving the diagnostic accuracy of coronary artery diseases.
Patent Information
- Application Number
- CN202410672400.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-28
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2044-05-28
AI Technical Summary
Existing coronary artery segmentation algorithms suffer from information loss, blurred boundaries, and background interference in coronary angiography images, resulting in inaccurate segmentation and failing to meet the needs of clinical diagnosis.
An improved U-Net neural network is used, which combines pyramid feature cascade branches, ConvNeXt modules, attention mechanism modules, and large-size convolutional layers to construct a coronary artery segmentation model. The model parameters are optimized by a hybrid loss function to retain more semantic features and balance the gap between encoding and decoding features, thereby improving the edge segmentation accuracy.
It significantly improves the accuracy and generalization ability of coronary artery segmentation, reduces the rate of missed diagnosis and misdiagnosis, and assists doctors in making accurate diagnoses of coronary artery diseases.
Smart Images

Figure CN118644511B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical artificial intelligence technology, and in particular to a method, device, and medium for coronary artery segmentation based on an improved U-Net neural network. Background Technology
[0002] Coronary artery disease (CAD), the most common type of heart disease, is a leading cause of death worldwide. Therefore, early detection and accurate diagnosis are crucial for effective treatment and reducing mortality. Accurate coronary artery segmentation plays a key role in early diagnosis and treatment. Currently, the standard for diagnosing CAD is coronary angiography, which involves injecting contrast agents into the blood vessels and then using X-ray imaging to determine the structure and shape of the coronary arteries. This allows for rapid diagnosis of the lesion site and assessment of the degree of blockage, enabling effective treatment. However, visual assessment is often unreliable, and manually extracting the vessel contour is time-consuming and labor-intensive. Therefore, utilizing computer-aided diagnostic technology helps doctors improve examination speed and reduce missed and misdiagnosed rates, thus achieving a supportive medical outcome.
[0003] Coronary angiography images suffer from low contrast between vascular structures and background areas. The presence of irrelevant tissues (such as catheters, diaphragms, and the spine) leads to uneven illumination, and the coronary artery topology is complex and variable. These issues make coronary artery segmentation more difficult. Analysis of polyp segmentation and traditional early segmentation algorithms reveals the following challenges in applying deep learning to clinical diagnosis at this stage.
[0004] 1) Traditional CNN segmentation algorithms only use global information from the last encoder block, which leads to the loss of local information in the intermediate layers.
[0005] 2) Existing segmentation networks are not specifically designed to address foreground and background interference in coronary angiography images, making it difficult to adapt to the diversity of interference in coronary vessel shape and size.
[0006] 3) Due to the loss of information in the downsampling operators, their boundary information is not fully preserved, which is not conducive to accurately delineating the boundaries and leads to blurred segmentation edges.
[0007] Therefore, there is an urgent need for an accurate and efficient automatic coronary artery segmentation model to assist doctors in diagnosis, improve the early detection rate of coronary artery disease, and further improve the cure rate of patients. Summary of the Invention
[0008] To achieve the above-mentioned objectives and other advantages of the present invention, a first objective of the present invention is to provide a coronary artery segmentation method based on an improved U-Net neural network, comprising the following steps:
[0009] Acquire coronary angiography images and establish a coronary angiography contour annotation dataset;
[0010] The data in the dataset is preprocessed and the dataset is divided.
[0011] The encoder and decoder are constructed using U-Net as the backbone structure;
[0012] By integrating pyramid feature cascade branches, ConvNeXt modules, attention mechanism modules, and large-size convolutional layers into the network model, a coronary artery segmentation model can be constructed.
[0013] The coronary artery segmentation model is trained in a supervised manner using the training set, and the parameters are optimized using a hybrid loss function. The output is then used as the coronary artery segmentation prediction result.
[0014] Each evaluation metric is calculated using the test set.
[0015] Furthermore, the preprocessing of the data in the dataset and the partitioning of the dataset include the following steps:
[0016] Data augmentation was used to expand the training set, and histogram equalization was used for data preprocessing.
[0017] A comparative experiment using five-fold cross-validation was conducted on the dataset.
[0018] Furthermore, the encoder and the decoder adopt a multi-layer symmetrical structure. The encoder is used for image downsampling to extract feature information, and the decoder is used for image upsampling to restore the original size and segment blood vessel contours.
[0019] Furthermore, the construction of the coronary artery segmentation model by integrating pyramid feature cascade branches, ConvNeXt modules, attention mechanism modules, and large-size convolutional layers into the network model includes the following steps:
[0020] The pyramid feature concatenation branch is placed in the decoding part of the U-Net network. The pyramid feature concatenation branch is used to fuse multi-scale feature maps and retain more semantic features.
[0021] The attention mechanism module is placed at each layer of the pyramid feature cascade branch so that the network learns the correlation between channels, has channel-specific attention, and is used to filter effective information in the feature map;
[0022] The ConvNeXt module is placed at the cross-link of the codec structure to balance the semantic gap between codec features;
[0023] The first convolutional layer of the U-Net encoding part is modified to the large-size convolutional layer to realize the construction of the coronary artery segmentation model.
[0024] Furthermore, the hybrid loss function comprises a binary cross-entropy loss function and a Dice loss function; wherein, the expression of the hybrid loss function is as follows:
[0025] Loss=λ1L BCE (y n ,y n ′)+λ2L DICE (y n ,y n ′),
[0026] The binary cross-entropy loss function L BCE The expression is as follows:
[0027]
[0028] The Dice loss function L Dice The expression is as follows:
[0029]
[0030] Among them, y n Representing the true value, y′ n λ1 and λ2 represent the actual output values of the model, N represents the number of samples, and o represents the smoothing coefficient.
[0031] Furthermore, the evaluation metrics include accuracy, sensitivity, specificity, precision, and AUC value; wherein, the accuracy is expressed as follows:
[0032]
[0033] The sensitivity expression is as follows:
[0034]
[0035] The specificity expression is as follows:
[0036]
[0037] The precision expression is as follows:
[0038]
[0039] Among them, T P T N F P and F N These represent true positive, true negative, false positive, and false negative, respectively.
[0040] The AUC value is the area under the ROC curve.
[0041] Furthermore, the evaluation metrics also include the Dice score and the Jaccard score;
[0042] The Dice scoring expression is as follows:
[0043]
[0044] The Jaccard scoring expression is as follows:
[0045]
[0046] Furthermore, the encoder and the decoder adopt a 4-layer symmetrical structure, and the large-size convolutional layer is a 7x7 convolutional layer.
[0047] A second object of the present invention is to provide an electronic device comprising: a memory having program code stored thereon; and a processor connected to the memory, wherein the above-described method is implemented when the program code is executed by the processor.
[0048] A third objective of the present invention is to provide a computer-readable storage medium having program instructions stored thereon, which, when executed, implement the method described above.
[0049] Compared with the prior art, the beneficial effects of the embodiments of the present invention are:
[0050] This invention provides a coronary artery segmentation method, device, and medium based on an improved U-Net neural network. The first convolutional layer of the U-Net encoding part is modified to a large-size convolutional layer. Larger kernels can cover a wider range of input information, thereby increasing the receptive field of each layer, helping to capture more contextual information, and significantly improving the performance of the neural network. An optimized ConvNeXt module is introduced at the encoder-decoder jump connection. This module reduces the computational complexity of the network while maintaining a performance level similar to Transformer, balancing the semantic gap between encoder and decoder features. Blood vessels of different thicknesses exhibit different activation levels in feature maps at different scales, and noise in coronary angiography images can be effectively removed at deeper layers after multi-layer feature extraction. The pyramid feature cascade branch module fuses multi-scale feature maps, retaining more semantic features, compensating for the loss of global information during downsampling, suppressing the influence of noise, and improving the accuracy of vessel segmentation. The attention mechanism module directly acts on the channel dimension to filter effective information in the feature map, focusing on the target detail location during feature extraction, thereby improving the segmentation of coronary artery edges. The coronary artery segmentation model constructed using these modules has high accuracy and generalization ability, which is of great significance for assisting clinical diagnosis.
[0051] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description
[0052] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0053] Figure 1 This is a flowchart of the coronary artery segmentation method based on the improved U-Net neural network in Example 1;
[0054] Figure 2 This is a flowchart of the data processing and dataset partitioning process in Example 1;
[0055] Figure 3 This is a flowchart of the coronary artery segmentation model construction in Example 1;
[0056] Figure 4 This is a schematic diagram of the coronary artery segmentation model in Example 1;
[0057] Figure 5 This is a schematic diagram of the pyramid feature cascaded branch structure of Example 1;
[0058] Figure 6 This is a schematic diagram of the optimized ConvNeXt module structure in Example 1;
[0059] Figure 7 This is a schematic diagram of the attention mechanism module structure in Example 1;
[0060] Figure 8 This is a schematic diagram comparing the coronary artery segmentation results of Example 1;
[0061] Figure 9 This is a comparison chart of the segmentation results of the present invention in Example 1 with the gold standard;
[0062] Figure 10 This is a rendering of the segmentation result of the present invention in Example 1;
[0063] Figure 11 This is a schematic diagram of the electronic device in Example 2;
[0064] Figure 12 This is a schematic diagram of the storage medium in Example 3. Detailed Implementation
[0065] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.
[0066] In the following description, the use of suffixes such as "part" or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "part" or "unit" may be used interchangeably.
[0067] Coronary angiography is the gold standard for diagnosing coronary artery disease. By injecting contrast agent into the coronary arteries and examining its flow within the vascular structure through two-dimensional projection, it helps clinicians locate potential stenosis, visually measure its severity, and determine appropriate interventional treatment. However, visual assessment of stenosis is often unreliable. To more objectively assess luminal diameter, quantitative coronary angiography has been introduced. The vessel segmentation step in quantitative coronary angiography is a prerequisite for calculating the percentage of arterial stenosis.
[0068] This invention combines coronary angiography images and deep learning algorithms to assist in the development of accurate and efficient automatic coronary artery segmentation models, laying the foundation for subsequent quantitative diagnosis, helping doctors improve examination speed and reduce the rate of missed diagnoses and misdiagnoses, thus achieving the effect of assisting medical treatment.
[0069] Example 1
[0070] A coronary artery segmentation method based on an improved U-Net neural network can accurately and efficiently segment coronary angiography images with uneven illumination, low contrast between vascular structures and background regions, and complex coronary artery topology. Figure 1 As shown, the method includes the following steps:
[0071] S1. Acquire coronary angiography images and create a coronary angiography contour annotation dataset; for example, create a coronary angiography contour annotation dataset based on manual annotations by cardiology experts. It should be noted that a coronary angiography contour annotation dataset can also be created automatically through a terminal.
[0072] S2. Preprocess the data in the dataset and divide the dataset; specifically, such as... Figure 2 As shown, it includes the following steps:
[0073] S21. Data augmentation is used to expand the training set, including random horizontal and vertical flipping, rotation and scaling, and histogram equalization is used for data preprocessing.
[0074] S22. A comparative experiment with five-fold cross-validation was conducted on the established coronary angiography contour annotation dataset. For example, 80% of this data was used for model training, 10% for parameter optimization and validation, and 10% for result testing to ensure the final results were convincing.
[0075] S3. The encoder and decoder are constructed using U-Net as the backbone structure; specifically, the encoder and decoder adopt a multi-layer symmetrical structure. Preferably, the encoder and decoder adopt a 4-layer symmetrical structure, with the encoder used for image downsampling to extract feature information and the decoder used for image upsampling to restore the original size and segment blood vessel contours.
[0076] S4. The pyramid feature cascade branch (PFC), the optimized ConvNeXt module, the attention mechanism (SE) module, and large-size convolutional layers are integrated into the network model to construct the coronary artery segmentation model; specifically, such as Figure 3 As shown, it includes the following steps:
[0077] S41. Place the pyramid feature cascade branch in the decoding part of the U-Net network. The pyramid feature cascade branch is used to fuse multi-scale feature maps and retain more semantic features; the structure of the pyramid feature cascade branch is as follows: Figure 5 As shown.
[0078] S42. Place the attention mechanism module in each layer of the pyramid feature cascade branch to enable the network to learn the correlation between channels, providing channel-specific attention for filtering effective information in the feature maps; the structure of the attention mechanism module is as follows: Figure 7 As shown.
[0079] The ConvNeXt module is a convolutional block optimized for the characteristics of the Transformer architecture, aiming to fully utilize the advantages of convolutional networks while adhering to the design principles of Transformer.
[0080] S43. Place the ConvNeXt module at the cross-link of the encoding / decoding structure to balance the semantic gap between encoding / decoding features; the structure of the optimized ConvNeXt module is as follows: Figure 6 As shown.
[0081] S44. Modify the first convolutional layer of the U-Net encoding part to a large-size convolutional layer to construct the coronary artery segmentation model. Preferably, the large-size convolutional layer is a 7x7 convolutional layer.
[0082] Figure 4 This invention presents the network model for a coronary artery segmentation method based on an improved U-Net neural network. The design employs a four-layer symmetric encoder-decoder structure. To address potential semantic gaps between encoder and decoder levels in U-Net, the proposed model introduces a ConvNeXt module across connections to balance these gaps. Furthermore, the first convolutional layer of the U-Net encoding part is modified to a 7x7 convolutional layer to expand the receptive field of each layer. Finally, based on the Feature Pyramid Network (FPN), this invention introduces another branch, Pyramid Feature Concatenation (PFC), at the decoder to fuse features at various scales. An attention mechanism, Squeeze-and-Excitation (SE), is added to each layer of the PFC to filter effective information from the feature maps.
[0083] S5. Supervised training of the coronary artery segmentation model is performed using the training set, and the parameters are optimized using a hybrid loss function. The output is used as the coronary artery segmentation prediction result.
[0084] In this embodiment, the hybrid loss function comprises a binary cross-entropy loss function and a Dice loss function; the expression for the hybrid loss function is as follows:
[0085] Loss=λ1L BCE (y n ,y n ′)+λ2L DICE (y n ,y n ′),
[0086] Binary cross-entropy loss function L BCE The expression is as follows:
[0087]
[0088] Dice loss function L Dice The expression is as follows:
[0089]
[0090] Among them, y n Representing the true value, y′ n λ1 and λ2 represent the actual output values of the model, N represents the number of samples, and o represents the smoothing coefficient.
[0091] The coronary artery segmentation is completed through the above steps.
[0092] S6. Calculate the evaluation metrics using the test set. Specifically, compare the segmentation results on the test set with the results of manual segmentation by doctors to verify the model's effectiveness. The evaluation metrics used are accuracy (Acc), sensitivity (Sen), specificity (Spe), precision (Pre), and AUC (area under curve) to objectively evaluate the segmentation performance of coronary arteries. The accuracy expression is as follows:
[0093]
[0094] The sensitivity expression is as follows:
[0095]
[0096] The specific expression is as follows:
[0097]
[0098] The precision expression is as follows:
[0099]
[0100] Among them, T P T N F P and F N These represent true positive, true negative, false positive, and false negative, respectively.
[0101] The receiver operating characteristic (ROC) curve is a commonly used evaluation metric for medical image segmentation tasks. The curve is plotted with the false positive rate (FPR) on the horizontal axis and the true positive rate (TPR) on the vertical axis. The closer the area under the ROC curve (AUC) is to 1, the better the segmentation performance of the model.
[0102] Since coronary artery segmentation tasks consist of mostly easily segmentable pixels, the improvements in the aforementioned metrics are not significant. To more intuitively reflect the performance differences between different methods, the Dice score and Jaccard score are added to measure segmentation performance. The Dice score expression is as follows:
[0103]
[0104] The Jaccard scoring formula is as follows:
[0105]
[0106] The effectiveness and accuracy of the model are demonstrated by examining the predicted segmentation results using the above indicators.
[0107] Figure 8 This is an example of the coronary artery segmentation method based on the improved U-Net neural network proposed in this invention, and the coronary artery segmentation results obtained by the comparison algorithm. Figure 8 The data includes segmentation results from five groups using six different models: the original image, the gold standard, ResUNeXt, U-Net, TransUnet, MultiResUnet, and the segmentation results from this invention. As can be seen from the results, the segmentation results from this invention are more accurate for coronary arteries, with complete vessel segmentation, good connectivity, clear and distinct edges, and better segmentation ability for small terminal vessels.
[0108] Figure 9 This image shows a comparison between the segmentation results of the coronary artery segmentation method based on the improved U-Net neural network proposed in this invention and the gold standard annotation by doctors. The segmentation shape is highly similar to that of the manual annotation by professional doctors.
[0109] Figure 10 This is a rendered image showing the segmentation result of the coronary artery segmentation method based on the improved U-Net neural network proposed in this invention. The segmentation result of this invention is accurate, clear, and has complete edges.
[0110] This invention provides a coronary artery segmentation method based on an improved U-Net neural network. Evaluation metrics include Dice score, Jaccard score, accuracy (Acc), sensitivity (Sen), specificity (Spe), precision (Pre), and AUC value. The model's predicted segmentation results were tested. Results show that Dice = 0.8811, Jaccard = 0.7875, Accuracy = 0.9785, Sensitivity = 0.8589, Specificity = 0.9906, Precision = 0.9019, and AUC = 0.9903, indicating a strong correlation between the predicted segmentation results and the manual standard segmentation results. In other words, there is no significant difference between the predicted segmentation results of this invention and the gold standard segmentation.
[0111] Example 2
[0112] An electronic device, such as Figure 11 As shown, it includes: a memory storing program code; and a processor connected to the memory, which, when the program code is executed by the processor, implements a coronary artery segmentation method based on an improved U-Net neural network. For a detailed description of the method, please refer to the corresponding description in the above method embodiments, which will not be repeated here.
[0113] Example 3
[0114] A computer-readable storage medium, such as Figure 12 As shown, it stores program instructions, which, when executed, implement a coronary artery segmentation method based on an improved U-Net neural network. For a detailed description of the method, please refer to the corresponding description in the above method embodiments; it will not be repeated here.
[0115] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.
[0116] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.
[0117] The apparatus, electronic device, and non-volatile computer storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, electronic device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, electronic device, and non-volatile computer storage medium will not be repeated here.
[0118] Those skilled in the art will also know that, besides implementing the controller in the form of purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller take the form of logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices included within it for implementing various functions can also be considered structures within that hardware component. Alternatively, the devices for implementing various functions can be considered as both software units implementing the method and structures within a hardware component.
[0119] The systems, apparatuses, or units described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions. For ease of description, the above apparatuses are described separately as various units based on their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.
[0120] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0121] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0122] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0123] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0124] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0125] This specification may be described in the general context of computer-executable instructions, such as program units, that are executed by a computer. Generally, program units include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification may also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program units may reside in local and remote computer storage media, including storage devices.
[0126] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0127] The above description is merely an embodiment of this specification and is not intended to limit the scope of one or more embodiments of this specification. Various modifications and variations can be made to one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of one or more embodiments of this specification.
Claims
1. A coronary artery segmentation method based on an improved U-Net neural network, characterized in that, Includes the following steps: Acquire coronary angiography images and establish a coronary angiography contour annotation dataset; The data in the dataset is preprocessed and the dataset is divided. The encoder and decoder are constructed using U-Net as the backbone structure; By integrating pyramid feature cascade branches, ConvNeXt modules, attention mechanism modules, and large-size convolutional layers into the network model, a coronary artery segmentation model can be constructed. The coronary artery segmentation model is trained in a supervised manner using the training set, and the parameters are optimized using a hybrid loss function. The output is then used as the coronary artery segmentation prediction result. Each evaluation metric is calculated using the test set; The hybrid loss function comprises a binary cross-entropy loss function and a Dice loss function; the expression for the hybrid loss function is as follows: , The binary cross-entropy loss function The expression is as follows: , The Dice loss function The expression is as follows: , in, Represents the true value. This represents the actual output value of the model. This is the proportionality coefficient. N Indicates the number of samples. o This represents the smoothing coefficient.
2. The coronary artery segmentation method based on the improved U-Net neural network as described in claim 1, characterized in that: The preprocessing and partitioning of the dataset includes the following steps: Data augmentation was used to expand the training set, and histogram equalization was used for data preprocessing. A comparative experiment using five-fold cross-validation was conducted on the dataset.
3. The coronary artery segmentation method based on the improved U-Net neural network as described in claim 1, characterized in that: The encoder and the decoder adopt a multi-layer symmetrical structure. The encoder is used for image downsampling to extract feature information, and the decoder is used for image upsampling to restore the original size and segment blood vessel contours.
4. The coronary artery segmentation method based on the improved U-Net neural network as described in claim 1 or 3, characterized in that: The method of integrating pyramid feature cascade branches, ConvNeXt modules, attention mechanism modules, and large-size convolutional layers into the network model to construct a coronary artery segmentation model includes the following steps: The pyramid feature concatenation branch is placed in the decoding part of the U-Net network. The pyramid feature concatenation branch is used to fuse multi-scale feature maps and retain more semantic features. The attention mechanism module is placed at each layer of the pyramid feature cascade branch so that the network learns the correlation between channels, has channel-specific attention, and is used to filter effective information in the feature map; The ConvNeXt module is placed at the cross-link of the codec structure to balance the semantic gap between codec features; The first convolutional layer of the U-Net encoding part is modified to the large-size convolutional layer to realize the construction of the coronary artery segmentation model.
5. The coronary artery segmentation method based on the improved U-Net neural network as described in claim 1, characterized in that: The evaluation metrics include accuracy, sensitivity, specificity, precision, and AUC value; wherein, the accuracy is expressed as follows: , The sensitivity expression is as follows: , The specificity expression is as follows: , The precision expression is as follows: , in, , , and These represent true positive, true negative, false positive, and false negative, respectively. The AUC value is the area under the ROC curve.
6. The coronary artery segmentation method based on the improved U-Net neural network as described in claim 5, characterized in that: The evaluation metrics also include the Dice score and the Jaccard score; The Dice scoring expression is as follows: , The Jaccard scoring expression is as follows: 。 7. The coronary artery segmentation method based on the improved U-Net neural network as described in claim 4, characterized in that: The encoder and the decoder adopt a 4-layer symmetrical structure, and the large-size convolutional layer is a 7x7 convolutional layer.
8. An electronic device, characterized in that, include: A memory that stores program code; A processor, which is connected to the memory, and which, when the program code is executed by the processor, implements the method as described in any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that, It stores program instructions that, when executed, implement the method as described in any one of claims 1 to 7.
