Method for segmenting aorta blood vessel image from X-ray angiography image and related device

Abstract

The invention discloses a method for segmenting an aorta blood vessel image from an X-ray angiography image and a related device, and belongs to the technical field of medical instruments. The method comprises the following steps: extracting blood vessel features from an X-ray angiography image, constructing and optimizing an aorta blood vessel segmentation network according to the extracted blood vessel features, and segmenting an aorta blood vessel image from the X-ray angiography image by using the optimized aorta blood vessel segmentation network. According to the method, learning simulation is carried out on an X-ray angiography image through computer software, calibration and optimization are carried out on the X-ray angiography image in a DSA operation, registration of an actual operation image and a pre-stored image can be achieved, a high-pixel-level accurate segmentation result can be provided, and the segmentation accuracy is improved. In addition, the integrity of the topological structure of the blood vessel can be kept in the process, and more detailed image information is provided for surgeons.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical devices, and in particular relates to a method for segmenting an aortic vascular image from an X-ray angiography image and a related device. Background Art

[0002] Digital subtraction angiography (DSA) is the main auxiliary technology for interventional surgery. By adding contrast agents to blood vessels and performing digital processing, the different densities of tissues and blood vessels are contrasted, thereby excluding tissue components and retaining vascular components, and accurately reflecting the location, morphology, and range of vascular lesions. By analyzing X-ray angiography images, doctors can directly observe the location of the thoracic and abdominal arteries and their branches, whether they are dilated or narrowed, and the location and degree of narrowing, thereby helping doctors make a clear diagnosis. In clinical work, the results of angiography often affect treatment decisions. The relationship between the location of branch blood vessels and the aorta determines the positioning of devices such as stents and balloons. Vascular segmentation is to amplify and refine this relationship to make treatment more accurate.

[0003] Current vascular segmentation methods can be divided into two categories: traditional vascular segmentation methods and deep learning-based vascular segmentation methods. Traditional vascular segmentation methods usually focus on specific image filters, tracking operators, and optimization models. However, these traditional vascular segmentation methods are based on predefined rules, require a lot of expertise and manual model adjustment, and are often affected by low-quality X-ray angiography and motion artifacts, resulting in low segmentation accuracy and inaccurate segmentation boundaries. Deep learning-based vascular segmentation methods can extract features with more semantic information. At present, deep learning-based vascular segmentation methods are divided into three categories according to the purpose of segmentation: semantic segmentation, topology-preserving methods, and hybrid methods. Semantic segmentation methods are committed to achieving high pixel-level accuracy, while topology-preserving methods aim to obtain the complete topological structure of blood vessels, and hybrid methods combine these two strategies. The use of a large amount of radiation and contrast agents in the implementation of DSA is the reason that limits the development of DSA. In the process of real-time imaging, the images of catheters and guidewires cannot be displayed simultaneously with angiography, and the current vascular segmentation technology is not mature enough to take into account both segmentation accuracy and topological structure.

[0004] The Chinese invention (CN105741299B) provides a method for segmenting coronary CT angiography images. The method includes the following steps: first, the CTA image is preprocessed, including median filtering and threshold segmentation, and a binary image is obtained after preprocessing; then, all connected areas in the binary image are marked, and the feature information of each connected area is obtained; then, a feature matching mechanism is used to determine whether the coronary artery is split; the target is tracked using the feature matching mechanism in the initial dozen frames of images, and then the target is tracked using Kalman. By combining the feature matching mechanism with the Kalman filter to track the target, the tracking of the target area under some special forms can be solved, and the tracking accuracy of the target area can be improved. Moreover, the entire tracking process uses the feature matching mechanism to determine whether the coronary artery in the image has split, and accurately determines whether the coronary artery has split and which parts it has split into, which can further ensure the accuracy of tracking. However, this method mainly focuses on the collection and analysis of images, and cannot perform fusion registration of simulated images and real images. Summary of the invention

[0005] In order to overcome the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a method and related device for segmenting an aortic vascular image from an X-ray angiography image, so as to solve the technical problem that the existing method cannot take into account both segmentation accuracy and topological structure.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: The first aspect of the present invention discloses a method for segmenting an aortic vascular image from an X-ray angiography image, extracting vascular features from the X-ray angiography image, constructing and optimizing an aortic vascular segmentation network based on the extracted vascular features, and segmenting the aortic vascular image from the X-ray angiography image using the optimized aortic vascular segmentation network.

[0007] Preferably, before using the blood vessel features, the blood vessel features are processed by horizontal flipping, vertical flipping, random rotation, random adjustment of color indexes and adjustment of pixel size.

[0008] Preferably, the steps of constructing an aortic vessel segmentation network based on the extracted vascular features are: encoding the extracted vascular features to extract feature maps of different scales and rich semantic information; decoding the extracted feature maps of different scales and rich semantic information to achieve feature fusion and feature enhancement to obtain an aortic vessel segmentation network.

[0009] Further preferably, the extracted vascular features are encoded by constructing a feature coding model, wherein the feature coding model includes five layers, each layer contains a residual unit, the trunk of the residual unit is composed of two convolutional layers, and the side branch is composed of one convolutional layer; the second to fifth layers of the feature coding model also include a downsampling layer of a convolutional layer.

[0010] Preferably, the extracted feature maps of different scales and rich semantic information are decoded by constructing a feature enhancement decoding model, wherein the feature enhancement decoding model includes a shallow semantic segmentation decoding unit, a deep semantic segmentation decoding unit, and an attention feature enhancement unit.

[0011] Further preferably, the attention feature enhancement unit includes a channel dimension enhancement subunit, a spatial dimension enhancement subunit and a feature enhancement fusion subunit.

[0012] Preferably, feature maps of different scales and rich semantic information are input into the feature enhancement decoding model for upsampling, and then sent to the shallow semantic segmentation decoding unit and the deep semantic segmentation decoding unit to generate corresponding feature maps, and then feature fusion is used to form a multi-channel feature map containing shallow and deep semantic information. Thereafter, the multi-channel feature map containing shallow and deep semantic information and the output of the shallow semantic segmentation decoding unit are used together as the input of the attention feature enhancement unit to obtain the channel attention weight and the attention weight in the spatial dimension, and then the fusion subunit of the enhanced feature is combined with the obtained channel attention weight and the attention weight in the spatial dimension to enhance and normalize the feature map to obtain the aortic vascular image segmentation result.

[0013] Preferably, the constructed aortic vessel segmentation network is optimized using a binary cross entropy loss function and a Dice loss function.

[0014] A second aspect of the present invention discloses a system for segmenting an aortic vessel image in an X-ray angiography image, comprising: A vascular feature extraction module extracts vascular features from X-ray angiography images; Modeling module, which constructs and optimizes the aortic vessel segmentation network based on the extracted vascular features; The segmentation module uses the optimized aortic vessel segmentation network to segment the aortic vessel image from the X-ray angiography image.

[0015] The third aspect of the present invention discloses a computer device, including a display device, a data processing terminal, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for segmenting an aortic vascular image from an X-ray angiography image when executing the computer program.

[0016] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for segmenting an aortic vascular image from an X-ray angiography image.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for segmenting an aortic vascular image from an X-ray angiography image. By extracting vascular features from the X-ray angiography image, an aortic vascular segmentation network is constructed and optimized. The optimized aortic vascular segmentation network can be used to segment the aortic vascular image from the X-ray angiography image. The method uses computer software to learn and simulate X-ray angiography images, calibrate and optimize X-ray angiography data in DSA surgery, and can achieve registration of actual operation images and pre-stored images. It can not only provide accurate segmentation results at a high pixel level, but also maintain the integrity of the vascular topology structure in the process, providing more detailed image information for surgeons. By using simulated images to assist real-time operations in interventional surgery, surgical complications are reduced, the efficiency of actual surgery is greatly increased, the operation time is reduced, and the amount of radiation and contrast agents can be further reduced.

[0018] Furthermore, by performing change processing on the vascular features extracted from the X-ray angiography data by horizontally flipping, vertically flipping, randomly rotating, randomly adjusting the color index and adjusting the pixel size, the input X-ray angiography data samples can have more changes and increase the diversity of the samples.

[0019] Furthermore, in the encoding stage, by constructing a feature encoding model, vascular features are extracted while reducing the dimension; the feature encoding model uses a residual structure to deepen the depth of the vascular segmentation network, so that the vascular segmentation network can learn richer semantic information.

[0020] Furthermore, in the decoding stage, feature fusion and feature enhancement are performed by constructing a feature enhancement decoding model, in which channel and spatial attention mechanisms are used to learn feature weights to enhance the expressiveness of feature maps.

[0021] Furthermore, in the model training stage, the binary cross entropy loss function and the Dice loss function are introduced to optimize the parameters of the vascular segmentation network, and the parameters of the model are adjusted to minimize the loss, so that the model can better fit the data, thereby improving the accuracy and precision of the vascular segmentation network in segmenting vascular images, and guiding the vascular segmentation network to pay more attention to high-pixel-level accuracy and the integrity of the vascular topological structure, which helps to improve the accuracy and robustness of vascular segmentation.

[0022] The present invention provides a computer device that digitally segments X-ray angiography images through external images and guides the segmentation process through an attention mechanism, aiming to achieve higher pixel-level segmentation accuracy while maintaining the integrity of the vascular topology structure. At the same time, it exhibits good robustness when processing images of different contrasts, so that it can achieve good segmentation effects in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flow chart of a method for segmenting an aortic vessel image in an X-ray angiography image; Figure 2 A schematic diagram of a method for segmenting an arterial blood vessel image in an X-ray angiography image; Figure 3 Schematic diagram of the feature encoding model; Figure 4 Schematic diagram for feature enhancement; Figure 5 It is a visualization segmentation result diagram of the present invention; wherein A is the vascular feature of the thoracic aorta angiography under X-ray, B is the extracted thoracic aorta vascular feature, C is the target fusion area after segmentation, and D is the enhanced fusion result; Figure 6 The figure is a schematic diagram of an electronic device to which the method for segmenting an aortic blood vessel image in an X-ray angiography image of the present invention is applied. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] The present invention provides a method for segmenting an aortic vessel image from an X-ray angiography image. Figure 1 and Figure 2 As shown, the following steps are included: S1. Data preparation and preprocessing process The data set is divided into training images and test images, and the training images and the test images are changed to obtain the changed data set; wherein the changing method is: horizontally flipping, vertically flipping, randomly rotating, randomly adjusting color indicators and adjusting pixel size of the training images and the test images in sequence.

[0027] The thoracic aortography images annotated by experienced radiologists were used as the true labels.

[0028] S2. Constructing the aortic vessel segmentation network S2.1. Vascular feature extraction S2.1.1. Constructing feature encoding model like Figure 3 As shown, the feature encoding model includes five layers, each layer contains a residual unit, the trunk of the residual unit is composed of two convolutional layers, and the side branch is composed of one convolutional layer. The second to fifth layers of the feature encoding model also include a downsampling layer of the convolutional layer.

[0029] S2.1.2. Constructing data shape The changed dataset obtained in S1 is converted into a [B, C, H, W] shape, where B is the batch size, C is the channel dimension, H is the height dimension, and W is the width dimension. The converted shape is input into the feature encoding model constructed in S2.1.1, and through five encoding layers, five feature maps of different scales and rich semantic information are extracted.

[0030] S2.2. Vascular feature enhancement S2.2.1. Constructing a feature-enhanced decoding model The feature enhancement decoding model includes a shallow semantic segmentation decoding unit, a deep semantic segmentation decoding unit and an attention feature enhancement unit; the attention feature enhancement unit includes a channel dimension enhancement subunit, a spatial dimension enhancement subunit and an enhanced feature fusion subunit; wherein the shallow semantic segmentation decoding unit is connected to the feature encoding model through a jump connection.

[0031] S2.2.2 Feature Fusion S2.2.2.1. Input the five feature maps of different scales and rich semantic information extracted by S2.1.2 into the feature enhancement decoding model constructed in S2.1.1 for upsampling, so that the size and number of channels of the input feature map are doubled.

[0032] The doubled feature maps of S2.2.2.2 and S2.2.2.1 are sent to the shallow semantic segmentation decoding unit and the deep semantic segmentation decoding unit to generate low-level semantic feature maps and high-level semantic feature maps respectively. The output of the shallow semantic segmentation decoding unit is combined with the output of the deep semantic segmentation decoding unit, and feature fusion is performed through concatenation and convolution operations to form a multi-channel feature map containing shallow and deep semantic information.

[0033] S2.2.3. Feature Enhancement S2.2.3.1, if Figure 4 As shown, first, the multi-channel feature map containing shallow and deep semantic information formed by S2.2.2.2 and the output of the shallow semantic segmentation decoding unit are used as the input of the attention feature enhancement unit, and global pooling and average pooling operations are performed through the channel dimension enhancement sub-unit to obtain feature representations of different scales. Secondly, the pooled features are processed by Conv2d, and then nonlinear transformation is performed through the rectified linear unit (ReLU) activation function. Again, the channel weights learned by the average pooling and maximum pooling operations are added to form the total weight of the channel dimension. Finally, the output is normalized using the Sigmoid function to obtain the channel attention weight. ;

[0034] Among them, in is the input feature map, is the channel attention weight processed by the Sigmoid function.

[0035] S2.2.3.2. Input the features processed by the channel dimension enhancement subunit in S2.2.3.1 into the spatial dimension enhancement subunit, and undergo global pooling and average pooling operations in the channel dimension. Then, concatenate the results after global pooling and average pooling according to the channel dimension. After concatenation, perform convolution on the concatenated feature map. Finally, normalize the output after convolution through the Sigmoid function to obtain the attention weight in the spatial dimension. ;

[0036] Where in is the input feature map, is the spatial attention weight processed by the Sigmoid function.

[0037] S2.2.3.3. First, the channel attention weights obtained from S2.2.3.1 are obtained by using the fusion subunit of the enhanced features in the attention feature enhancement unit. And the attention weights on the spatial dimension obtained in S2.2.3.2 It is used to enhance the feature map output by the shallow semantic segmentation decoding unit to obtain the enhanced feature map. Secondly, the enhanced feature map is passed through a convolution layer to obtain the processed feature map. Finally, the processed feature map is added to the feature map output by the shallow semantic segmentation decoding unit, and then the added feature map is normalized by the Sigmoid function to obtain the aortic vessel segmentation result.

[0038] S3. Optimizing the aortic vessel segmentation network S3.1. Optimization of segmentation accuracy The training images in S1 are changed, and then the changed data set is divided into multiple batches and sent to the aortic vessel segmentation network model for training to obtain the training segmentation results of the vessels. The binary cross entropy loss function is used to judge the difference between the probability distribution of the training segmentation results of the vessels and the probability distribution of the true labels to measure the accuracy of the classification. The binary cross entropy loss function is calculated as follows: ] in, is the binary cross entropy loss function, N is the total number of pixels, is a binary label, Take 0 or 1, For output belonging to The probability of the label.

[0039] S3.2 Optimization of segmentation accuracy The Dice loss function is used to calculate the ratio between the intersection of the training segmentation results of the blood vessels and the true labels and their union, measuring the similarity between the training segmentation results of the blood vessels and the true labels. The Dice loss function is calculated as follows:

[0040] in, is the Dice loss function, is the intersection of P and G, is the number of its elements, is the segmented predict_mask(label), is the ground truth (label) of the blood vessel.

[0041] S3.3. Use the binary cross entropy loss function of S3.1 and the Dice loss function of S3.2 to construct a joint loss function , the parameters in the aortic vessel segmentation network model are adjusted by the joint loss function, and the optimal aortic vessel segmentation network model is trained.

[0042]

[0043] in, is the joint loss function, and are the penalty parameter and loss of the binary cross entropy loss function, respectively. and are the penalty parameter and loss of the Dice loss function respectively.

[0044] The present invention is further described in detail below in conjunction with the accompanying drawings: Example 1 S1. Data preparation and preprocessing process like Figure 5 As shown in A, 126 independent thoracic aorta DSA X-ray angiography images were obtained as a data set. The resolution of each image is 512×512 pixels and is single-channel. The data set is divided into two groups: 84 training images and 42 test images. The two groups of images are then flipped horizontally to introduce left-right symmetric changes; the two groups of images are then flipped vertically to increase top-bottom symmetric changes; the two groups of images are then randomly rotated to increase the diversity of rotation; the color indicators of the two groups of images are randomly adjusted to introduce color changes; and the training images are randomly cropped to a size of 256×256 pixels to increase the diversity of samples.

[0045] In addition, the X-ray angiography images of the thoracic aorta DSA annotated by experienced radiologists are used as the true labels to provide an accurate reference benchmark for the calculation of the loss function.

[0046] S2. Constructing the aortic vessel segmentation network S2.1. Vascular feature extraction S2.1.1. Constructing feature encoding model The feature encoding model includes five layers, each layer contains a residual unit, the trunk of the residual unit consists of two convolutional layers, and the side branch consists of one convolutional layer. The second to fifth layers of the feature encoding model also include a downsampling layer of the convolutional layer.

[0047] S2.1.2. Constructing data shape The changed dataset obtained by S1 is converted into a shape of [B, 1, 256, 256]. Then the image of [B, 1, 256, 256] is input into the feature encoding model of S2.1. Through five encoding layers, five feature maps of different scales and rich semantic information are extracted. The feature channel dimensions of each layer output are [64, 128, 256, 512, 512], and the output feature map shapes of each layer are [B, 64, 256, 256], [B, 128, 128, 128], [B, 256, 64, 64], [B, 512, 32, 32], [B, 512, 16, 16], and the extracted vascular features are shown in the figure. Figure 5 As shown in B.

[0048] S2.2. Vascular feature enhancement S2.2.1. Constructing a feature-enhanced decoding model The feature enhancement decoding model includes a shallow semantic segmentation decoding unit, a deep semantic segmentation decoding unit, and an attention feature enhancement unit; the attention feature enhancement unit includes a channel dimension enhancement subunit, a spatial dimension enhancement subunit, and an enhanced feature fusion subunit. The shallow semantic segmentation decoding unit is connected to the feature encoding model through a skip connection to ensure that the network does not lose low-level semantic features during the learning process.

[0049] S2.2.2 Feature Fusion S2.2.2.1. Input the five feature maps of different scales and rich semantic information extracted by S2.1.2 into the feature enhancement decoding model constructed by S2.1.1, and use the two-dimensional deconvolution (ConvTranspose2d; Conv2d) operation to upsample, so that the size and number of channels of the input feature map are doubled.

[0050] The doubled feature maps of S2.2.2.2 and S2.2.2.1 are sent to the shallow semantic segmentation decoding unit and the deep semantic segmentation decoding unit to generate low-level semantic feature maps and high-level semantic feature maps respectively. Then, the output of the shallow semantic segmentation decoding unit is combined with the output of the deep semantic segmentation decoding unit, and feature fusion is performed through concatenation and convolution operations to form an 8-channel feature map containing shallow and deep semantic information ( Figure 5 Middle C).

[0051] S2.2.3. Feature Enhancement S2.2.3.1. The 8-channel feature map containing shallow and deep semantic information formed by S2.2.2.2 and the output of the shallow semantic segmentation decoding unit are used as the input of the attention feature enhancement unit. Global pooling and average pooling operations are performed through the channel dimension enhancement sub-unit to obtain feature representations of different scales. Secondly, the pooled features are processed by Conv2d, and then nonlinear transformation is performed through the ReLU activation function. The channel weights learned by the average pooling and maximum pooling operations are added again to form the total weight of the channel dimension. Finally, the output is normalized using the Sigmoid function to obtain the channel attention weight. ;

[0052] Among them, in is the input feature map, is the channel attention weight processed by the Sigmoid function.

[0053] S2.2.3.2. Input the features processed by the channel dimension enhancement subunit in S2.2.3.1 into the spatial dimension enhancement subunit, and undergo global pooling and average pooling operations on the channel dimension to reduce the number of channels, so that the network can focus more on learning the features in the spatial dimension. Then, the results after global pooling and average pooling are spliced ​​according to the channel dimension. After splicing, the spliced ​​feature map is convolved. Finally, the output after convolution is normalized by the Sigmoid function to obtain the attention weight in the spatial dimension. ;

[0054] Where in is the input feature map, is the spatial attention weight processed by the Sigmoid function.

[0055] S2.2.3.3. First, the fusion subunit of the enhanced feature in the attention feature enhancement unit is used to enhance the channel attention weight obtained from S2.2.3.1 and the attention weight on the spatial dimension obtained from S2.2.3.2 to enhance the feature map output by the shallow semantic segmentation decoding unit to obtain the enhanced feature map. Secondly, the enhanced feature map is passed through a convolution layer to obtain the processed feature map. Finally, the processed feature map is added to the feature map output by the shallow semantic segmentation decoding unit, and then the extracted discriminative features that fuse the shallow and deep semantic information are normalized by the Sigmoid function to obtain the segmentation result of the blood vessel. Subsequently, the threshold processing method is used to generate the blood vessel segmentation map. The visualization result of the segmentation is shown as follows. Figure 5 As shown in D.

[0056] S3. Optimizing the aortic vessel segmentation network S3.1. Optimization of segmentation accuracy The 84 training image datasets after the changes in S1 are divided into multiple batches and sent to the aortic vessel segmentation network model constructed by S2 for training, and the training segmentation results of the vessels are obtained. The binary cross entropy loss function is used to judge the difference between the probability distribution of the training segmentation results of the vessels and the probability distribution of the true labels obtained by S1 to measure the accuracy of the classification. The binary cross entropy loss function is calculated as follows:

[0057] in, is the binary cross entropy loss function, is a binary label, Take 0 or 1, For output belonging to The probability of the label.

[0058] S3.2 Optimization of segmentation accuracy The Dice loss function is used to calculate the ratio between the intersection of the training segmentation results of the blood vessels and the true labels obtained by S1 and their union to measure the similarity between the predicted results and the true labels. The Dice loss function is calculated as follows:

[0059] in, is the Dice loss function, is the intersection of P and G, is the number of its elements, is the segmented predict_mask(label), is the ground truth (label) of the blood vessel.

[0060] S3.3, use the binary cross entropy loss function of S3.1 and the Dice loss function of S3.2 to adjust the network parameters as follows:

[0061] in, is the joint loss function, and are the penalty parameter and loss of the binary cross entropy loss function, respectively. and are the penalty parameter and loss of the Dice loss function respectively.

[0062] S4. Validity Verification In order to verify the effectiveness of this method, this experiment uses Precision, Recall and F1-score as evaluation indicators. Precision measures the proportion of blood vessel pixels (positive class) predicted by the model to be blood vessel pixels that actually belong to blood vessels; it reflects the accuracy of the model in predicting blood vessels. If the precision is high, it means that the number of non-vascular pixels incorrectly marked as blood vessel pixels is small. Recall measures the proportion of all actual blood vessel pixels correctly predicted by the model to be blood vessel pixels; it reflects the completeness of the model in identifying blood vessel structures; if the recall rate is high, the model can find most of the blood vessel pixels. F1-score is the harmonic mean of Precision and Recall, which is used to consider these two indicators at the same time; in the blood vessel segmentation task, F1-score can provide an indicator for comprehensively evaluating model performance. It can balance Precision and Recall, so that the model is both accurate and complete when predicting blood vessels; the higher the F1-score, the better the balance between Precision and Recall is achieved.

[0063] In the blood vessel segmentation task, if the part of the input image containing the real label information in S1 is correctly identified as the corresponding real label information, it is recorded as (TP); if the part of the input image containing background information is incorrectly identified as the real label information, it is recorded as (FP); if the part of the input image containing background information is correctly identified as the corresponding background information, it is recorded as (TN); if the part of the input image containing real label information is incorrectly identified as background information, it is recorded as (FN). The calculation formulas of the indicators Precision, Recall, and F1-score are as follows:

[0064]

[0065]

[0066] The performance of the method for segmenting aortic vessels in X-ray angiography of Example 1 was tested on a dataset of thoracic aortic angiography images, and the obtained segmentation accuracy of the indicators of F1-score, Precision and Recall are shown in Table 1: Table 1 Performance test results

[0067] As can be seen from Table 1, although Precision is lower than Recall, both are relatively high, and the F1-score is also high. This shows that the vessel segmentation model not only ensures the accuracy of the blood vessels in identifying blood vessels, but also ensures that most of the blood vessels can be found, and performs well in the vessel segmentation task.

[0068] Example 2 Embodiment 2 provided by the present invention is an embodiment of a system for segmenting an aortic vascular image in an X-ray angiography image provided by the present invention. The embodiment of the system includes: a vascular feature extraction module, a modeling module and a segmentation module.

[0069] A vascular feature extraction module extracts vascular features from X-ray angiography images; Modeling module, which constructs and optimizes the aortic vessel segmentation network based on the extracted vascular features; The segmentation module uses the optimized aortic vessel segmentation network to segment the aortic vessel image from the X-ray angiography image.

[0070] It can be understood that the system for segmenting an aortic vessel image in an X-ray angiography image provided by the present invention corresponds to the method for segmenting an aortic vessel image in an X-ray angiography image provided by the aforementioned embodiments. The relevant technical features of the system for segmenting an aortic vessel image in an X-ray angiography image can refer to the relevant technical features of the method for segmenting an aortic vessel image in an X-ray angiography image, which specifically includes the following steps: Extracting vascular features from X-ray angiography images; Construct and optimize the aortic vessel segmentation network based on the extracted vascular features; The optimized aortic vessel segmentation network is used to segment the aortic vessel image from the X-ray angiography image.

[0071] Example 3 This embodiment provides an electronic device, such as Figure 6 As shown, it includes a display device, a data processing terminal, a memory, a processor, and a computer program stored in the memory and can be run on the processor. The display device provides visual presentation for the final fusion image; the data processing terminal imports external DSA image data through a data line, uses an internal software system to simulate, and then outputs the synthesized image to a real device. The processor executes the steps of the method for segmenting the aortic vascular image in the X-ray angiography image.

[0072] Extracting vascular features from X-ray angiography images; Construct and optimize the aortic vessel segmentation network based on the extracted vascular features; The optimized aortic vessel segmentation network is used to segment the aortic vessel image from the X-ray angiography image.

[0073] This electronic device can segment the thoracic and abdominal arteries while analyzing the images by externalizing the DSA images. On the device display, vascular imaging that originally required the injection of contrast agents to obtain can be auxiliary superimposed on the existing real-time images, reducing the use of contrast agents and radiation during surgery, shortening the operation time, improving surgical efficiency, and playing a positive role in reducing intraoperative and postoperative complications for patients.

[0074] Example 4 This embodiment provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method for segmenting an aortic vessel image in an X-ray angiography image are implemented.

[0075] Extracting vascular features from X-ray angiography images; Construct and optimize the aortic vessel segmentation network based on the extracted vascular features; The optimized aortic vessel segmentation network is used to segment the aortic vessel image from the X-ray angiography image.

[0076] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may 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.

[0077] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0078] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for segmenting an aortic vessel image from an X-ray angiography image, characterized in that: The vascular features are extracted from the X-ray angiography image, an aortic vessel segmentation network is constructed and optimized according to the extracted vascular features, and the aortic vessel image is segmented from the X-ray angiography image using the optimized aortic vessel segmentation network.

2. The method for segmenting an aortic vessel image from an X-ray angiography image according to claim 1, characterized in that: Before using the vascular features, the vascular features are processed by horizontal flipping, vertical flipping, random rotation, random adjustment of color indexes and pixel size adjustment.

3. A method for segmenting an aortic vessel image from an X-ray angiography image according to claim 1 or 2, characterized in that: The steps of constructing an aortic vessel segmentation network based on the extracted vascular features are: encoding the extracted vascular features to extract feature maps of different scales and rich semantic information; decoding the extracted feature maps of different scales and rich semantic information to achieve feature fusion and feature enhancement to obtain an aortic vessel segmentation network.

4. The method for segmenting an aortic vessel image from an X-ray angiography image according to claim 3, characterized in that: The extracted vascular features are encoded by constructing a feature encoding model, wherein the feature encoding model includes five layers, each layer contains a residual unit, the trunk of the residual unit is composed of two convolutional layers, and the side branch is composed of one convolutional layer; the second to fifth layers of the feature encoding model also include a downsampling layer of a convolutional layer.

5. The method for segmenting an aortic vessel image from an X-ray angiography image according to claim 3, characterized in that: The extracted feature maps of different scales and rich semantic information are decoded by constructing a feature enhancement decoding model, wherein the feature enhancement decoding model includes a shallow semantic segmentation decoding unit, a deep semantic segmentation decoding unit, and an attention feature enhancement unit.

6. The method for segmenting an aortic vessel image from an X-ray angiography image according to claim 5, characterized in that: The attention feature enhancement unit includes a channel dimension enhancement subunit, a spatial dimension enhancement subunit and an enhanced feature fusion subunit.

7. The method for segmenting an aortic vessel image from an X-ray angiography image according to claim 1, characterized in that: The constructed aortic vessel segmentation network was optimized using the binary cross entropy loss function and the Dice loss function.

8. A system for segmenting an aortic vessel image in an X-ray angiography image, characterized in that: include: A vascular feature extraction module extracts vascular features from X-ray angiography images; Modeling module, which constructs and optimizes the aortic vessel segmentation network based on the extracted vascular features; The segmentation module uses the optimized aortic vessel segmentation network to segment the aortic vessel image from the X-ray angiography image.

9. A computer device comprising a display device, a data processing terminal, a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the processor implements the steps of a method for segmenting an aortic vessel image from an X-ray angiography image as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of a method for segmenting an aortic vessel image from an X-ray angiography image as described in any one of claims 1 to 7 are implemented.

Images