Lightweight wavelet convolution guidewire segmentation network model construction method and dual guidewire generation method

Abstract

The present invention proposes a method for constructing a lightweight wavelet convolution guidewire segmentation network model and a dual guidewire generation method. The construction method includes: obtaining an input image, using a Stem module to perform feature extraction on the input image to obtain underlying features, iteratively processing the underlying features through five wavelet encoders to obtain features output by the five wavelet encoders respectively, and iteratively inputting the features output by the wavelet encoders into a decoder for decoding processing to obtain a segmentation prediction map. The wavelet transform attention convolution proposed in the present invention is specifically used to capture the complex edges and texture subtle differences of the guidewire. A channel attention mechanism is introduced after the wavelet decomposition process to enhance feature representation. The combination of channel attention can selectively emphasize key frequency components, significantly improve the network's ability to capture complex texture patterns, and improve the accuracy of guidewire segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and in particular to a method for constructing a lightweight wavelet convolution guidewire segmentation network model and a method for generating dual guidewires. Background Art

[0002] Cardiovascular and cerebrovascular diseases have a high morbidity, morbidity, and mortality rates. Given current technological advances, vascular intervention is considered the most effective treatment. To avoid damaging the vessel wall, the movement of the guidewire within the blood vessel must be very precise. Therefore, accurate guidewire segmentation technology is crucial to the success of vascular intervention. However, guidewire segmentation poses significant challenges. In X-ray sequences, slender guidewires often exhibit class imbalance and low signal-to-noise ratio, resulting in potential artifacts caused by the patient's heartbeat and respiration. In addition, the guidewire may be obscured by blood vessels and ribs, and the similarity of the structure makes the guidewire easily misclassified.

[0003] Currently, guidewire segmentation methods are primarily categorized into traditional methods and deep learning-based methods. Traditional methods, similar to edge detection techniques, focus on extracting the curved contour of the guidewire trajectory. For example, B-spline curve fitting techniques are commonly used to model guidewires, while fluorescent agents can be used to locate endpoints for tracking. Deep learning methods primarily fall into two categories: convolutional neural network (CNN)-based models and Transformer-based models. Convolutional neural networks offer the unique advantage of inductive biases in medical image segmentation, particularly with limited datasets. Thanks to the self-attention mechanism, Transformers can effectively capture global dependencies, but this also comes with the trade-off of a large number of parameters. To further address the computational overhead associated with real-time guidewire segmentation, the development of lightweight architectures is imperative. While there are currently no dedicated lightweight networks specifically designed for guidewire segmentation, several general-purpose lightweight medical image segmentation networks, such as UNeXt and CMUNeXT, have been proposed. While this approach reduces parameters and improves efficiency by simplifying the model, it compromises the preservation of critical texture details. Summary of the Invention

[0004] In view of the above situation, the main purpose of the present invention is to propose a method for constructing a lightweight wavelet convolution guidewire segmentation network model and a dual guidewire generation method to solve the above technical problems.

[0005] The present invention proposes a method for constructing a lightweight wavelet convolution guidewire segmentation network model, which comprises the following steps:

[0006] Step 1: Construct a Stem module based on two-dimensional convolution, construct a wavelet encoder based on wavelet transform attention convolution, point-by-point convolution and two-dimensional convolution, and construct a decoder based on grouped convolution and point-by-point convolution. The Stem module, wavelet encoder and decoder constitute a guidewire segmentation network model;

[0007] Step 2: Get the input image, use the Stem module to extract the features of the input image, obtain the underlying features, and iteratively process the underlying features through five wavelet encoders in turn to obtain the features output by the five wavelet encoders respectively;

[0008] Step 3: Input the features output by the wavelet encoder into the decoder in an iterative manner for decoding to obtain a segmentation prediction map;

[0009] Step 4: Based on the segmentation prediction graph, a binary cross entropy loss function and a Dice loss function are constructed respectively. The guidewire segmentation network model is optimized using the binary cross entropy loss function and the Dice loss function to obtain the final lightweight wavelet convolution guidewire segmentation network model.

[0010] The present invention also proposes a dual guidewire generation method, wherein the method applies the lightweight wavelet convolution guidewire segmentation network model as described above, and the method comprises the following steps:

[0011] S801, obtaining two single guide wire images and their corresponding true labels, and generating a binary mask based on the true label corresponding to the second single guide wire image;

[0012] S802: Based on the binary mask, the two single-wire images and their corresponding true labels are pixel-replaced to obtain a pseudo double-wire image and its corresponding true label;

[0013] S803: Input the pseudo dual-guidewire image into the final lightweight wavelet convolution guidewire segmentation network model for processing to obtain a segmentation prediction image.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] 1. We propose a lightweight guidewire segmentation network model to adapt to the medical scenario of real-time guidewire segmentation. Using a fully convolutional architecture, we design a novel, lightweight network model specifically for real-time guidewire segmentation tasks, achieving an optimal trade-off between segmentation performance and computational cost. Among various guidewire segmentation network models, this model exhibits the smallest number of parameters and computational complexity, has a fast training method, and can also produce the best segmentation results.

[0016] 2. The wavelet transform attention convolution proposed in this paper is specifically designed to capture the complex edges and texture nuances of the guidewire. A channel attention mechanism is introduced after the wavelet decomposition process to enhance feature representation. The combination of channel attention can selectively emphasize key frequency components, significantly improving the network's ability to capture complex texture patterns and enhancing the accuracy of guidewire segmentation.

[0017] 3. The proposed dual-wire generation algorithm effectively addresses the problem of scarce dual-wire data. In previous guidewire segmentation training, insufficient guidewire samples, particularly dual-wire samples, hampered model training. This paper proposes a guidewire data generation method that converts single-wire images into dual-wire images. This method aims to overcome the issue of limited dual-wire data and establish a more robust and versatile solution for guidewire segmentation.

[0018] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flowchart of the steps of the method for constructing a lightweight wavelet convolution guidewire segmentation network model proposed in the present invention;

[0020] Figure 2 This is a model framework diagram of the lightweight wavelet convolution guidewire segmentation network model proposed in the present invention;

[0021] Figure 3 This is an architecture diagram of the wavelet transform attention convolution in a specific embodiment of the present invention;

[0022] Figure 4 Flowchart of the processing of wavelet transform attention convolution in a specific embodiment of the present invention;

[0023] Figure 5 This is an algorithm framework diagram of the dual guidewire generation method proposed in the present invention;

[0024] Figure 6 This is a comparison chart of the dual guidewire data generated by the dual guidewire generation method proposed in the present invention and the actual dual guidewire data. DETAILED DESCRIPTION

[0025] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0026] These and other aspects of the embodiments of the present invention will become clear with reference to the following description and accompanying drawings. In these descriptions and accompanying drawings, some specific implementations of the embodiments of the present invention are specifically disclosed to illustrate some ways of implementing the principles of the embodiments of the present invention, but it should be understood that the scope of the embodiments of the present invention is not limited thereto.

[0027] See also Figure 1 This embodiment provides a method for constructing a lightweight wavelet convolution guidewire segmentation network model, which includes the following steps:

[0028] Step 1. Construct the Stem module based on two-dimensional convolution, construct a wavelet encoder based on wavelet transform attention convolution, point-by-point convolution and two-dimensional convolution, and construct a decoder based on grouped convolution and point-by-point convolution. The Stem module, wavelet encoder and decoder constitute the guidewire segmentation network model.

[0029] Step 2: Get the input image, use the Stem module to extract the features of the input image, obtain the underlying features, and iteratively process the underlying features through five wavelet encoders in turn to obtain the features output by the five wavelet encoders respectively.

[0030] See also Figure 2 In step 2, the input image is obtained, and the Stem module is used to extract the features of the input image to obtain the underlying features. The underlying features are processed in an iterative manner through five wavelet encoders to obtain the features output by the five wavelet encoders respectively. The specific sub-steps are as follows:

[0031] Get the input image, and process it through two-dimensional convolution, ReLU activation function and batch normalization in sequence to obtain the underlying features. The following relationship exists in the corresponding process:

[0032] ;

[0033] in, Represents the underlying features, It means that after batch normalization, Indicates that it has been processed by the ReLU activation function. Indicates that after a two-dimensional convolution operation, represents the input image;

[0034] S201, the underlying features are sequentially processed by wavelet transform attention convolution, GELU activation function and batch normalization, and residual addition is performed with the underlying features to obtain the intermediate features after wavelet transform attention convolution and residual addition in the first cycle. The corresponding process has the following relationship:

[0035] ;

[0036] in, represents the intermediate features after wavelet transform attention convolution and residual addition in the first cycle, It means that after batch normalization, Indicates that it has been processed by the GELU activation function. represents the attention convolution operation after wavelet transform, represents the addition of residuals;

[0037] S202, the intermediate features that have undergone wavelet transform attention convolution and residual addition in the first cycle are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the intermediate features that have undergone the first point-by-point convolution in the first cycle. The following relationship exists in the corresponding process:

[0038] ;

[0039] in, Represents the intermediate features after the first point-by-point convolution in the first cycle, Indicates point-by-point convolution operation;

[0040] S203, the intermediate features that have undergone the first point-by-point convolution in the first cycle are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the features output by the first cycle. The following relationship exists in the corresponding process:

[0041] ;

[0042] in, Indicates the characteristics of the output of the first cycle;

[0043] S204, repeat steps S201 to S203 in an iterative manner using the features output from the first cycle as input After the second time, The characteristics of the output of the sub-cycle correspond to the following relationship in the process:

[0044] ;

[0045] in, Indicates the The intermediate features of the wavelet transform attention convolution and residual addition in the second cycle, represents the index of the loop count, and ; Indicates the total number of cycles, Indicates the The characteristics of the output of the sub-cycle are: Indicates the The intermediate features after the first point-by-point convolution in the second cycle, Indicates the Characteristics of the secondary cycle output;

[0046] S205, the The features output by the second cycle are processed in sequence by two-dimensional convolution, GELU activation function, batch normalization and maximum pooling to obtain the features output by the first wavelet encoder. The following relationship exists in the corresponding process:

[0047] ;

[0048] in, represents the characteristics of the first wavelet encoder output, Indicates that the maximum pooling operation has been performed.

[0049] After iteratively taking the features output by the first wavelet encoder as input and repeating steps S201 to S205 four times, the features output by the second wavelet encoder, the features output by the third wavelet encoder, the features output by the fourth wavelet encoder and the features output by the fifth wavelet encoder are obtained respectively, and the features output by the fifth wavelet encoder are used as the bottleneck features.

[0050] For further information, see Figure 3 and Figure 4 , the wavelet transform attention convolution operation includes the following sub-steps:

[0051] The input features are sequentially subjected to two-dimensional Haar wavelet transform and two-dimensional convolution operations to obtain low-frequency components and high-frequency components. The following relationship exists in the corresponding process:

[0052] ;

[0053] in, represents a low-pass filter, Both represent high-pass filters. express The low-frequency component of represents the index of the wavelet level, express The high-frequency components in the horizontal direction, express The high-frequency components in the vertical direction, express High-frequency components in the diagonal direction;

[0054] The SE module is constructed based on the channel attention mechanism, and the three high-frequency components generated by the wavelet transform are processed by the SE module to obtain three enhanced high-frequency components. The following relationship exists in the corresponding process:

[0055] ;

[0056] in, Both represent the enhanced high-frequency components. Indicates that it has been processed by the SE module;

[0057] Perform a two-dimensional convolution operation on the low-frequency component and the enhanced high-frequency component to obtain the low-frequency component and high-frequency component after the two-dimensional convolution operation. The following relationship exists in the corresponding process:

[0058] ;

[0059] in, Represents the low-frequency component after the two-dimensional convolution operation, Both represent the high-frequency components after two-dimensional convolution operation. represents the weight matrix;

[0060] The low-frequency component and high-frequency component after the two-dimensional convolution operation are transposed and convolved to obtain the wavelet level The output characteristics of the inverse wavelet transform correspond to the following relationship:

[0061] ;

[0062] in, Indicates the wavelet level is The output characteristics of the inverse wavelet transform are Indicates that the convolution operation has been performed;

[0063] The input features are subjected to a two-dimensional convolution operation and the residual is added to the output features of the inverse wavelet transform with a wavelet level of 1 to obtain the features of the wavelet transform attention convolution output. The corresponding process has the following relationship:

[0064] ;

[0065] in, represents the features of the wavelet transform attention convolution output, represents the input features, Represents the output features of the inverse wavelet transform with wavelet level 1.

[0066] It should be noted that the input image includes a single-wire image and a pseudo-double-wire image generated by the double-wire generation method; Figure 3 middle, Represents element-wise multiplication; Figure 4(a) in the figure compares the LH features processed and not processed by the SE module: the left figure shows the LH features without the SE module, and the right figure shows the features after the SE module is weighted. It can be seen that the application of the SE module significantly enhances the details of the high-frequency components, thereby more effectively highlighting the key features while suppressing irrelevant information. In addition, Figure 4 (b) in the figure compares the results of using ordinary convolution (5×5) and using wavelet transform attention convolution: the left figure shows the result of ordinary convolution, and the right figure shows the result of wavelet transform attention convolution. It can be seen that wavelet transform attention convolution can better emphasize certain horizontal or vertical texture details in the image, while ordinary convolution tends to maintain global smoothness at the expense of accurate depiction of local details.

[0067] Step 3: Input the features output by the wavelet encoder into the decoder in an iterative manner for decoding to obtain a segmentation prediction map.

[0068] In step 3, the features output by the wavelet encoder are input into the decoder in an iterative manner for decoding to obtain a segmentation prediction map, which specifically includes the following sub-steps:

[0069] S301. The bottleneck features are sequentially processed by upsampling, two-dimensional convolution, ReLU activation function and batch normalization to obtain the pre-processed bottleneck features. The following relationship exists in the corresponding process:

[0070] ;

[0071] in, represents the bottleneck feature after preprocessing, represents the bottleneck feature, Indicates that it has undergone upsampling operation;

[0072] S302: After concatenating the preprocessed bottleneck features with the output features of the corresponding encoder, the features are sequentially processed by group convolution, GELU activation function and batch normalization to obtain the fusion features of the first encoder. The following relationship exists in the corresponding process:

[0073] ;

[0074] in, represents the fusion feature of the first encoder, Indicates that after group convolution operation, Indicates that after feature splicing operation, Represents the characteristics of the 4th wavelet encoder output;

[0075] S303: The fusion features of the first encoder are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the intermediate features of the first encoder. The following relationship exists in the corresponding process:

[0076] ;

[0077] in, Represents the intermediate features of the first encoder;

[0078] S304: The intermediate features of the first encoder are sequentially processed by point-by-point convolution, GELU activation function and batch normalization to obtain the features output by the first decoder. The following relationship exists in the corresponding process:

[0079] ;

[0080] in, Represents the features of the first decoder output;

[0081] In an iterative form, the features output by the first decoder are used as input to repeat steps S301 to S304 three times, and the features output by the second decoder, the features output by the third decoder, and the features output by the fourth decoder are obtained respectively. The features output by the fourth encoder are used as the segmentation prediction map. The following relationship exists in the corresponding process:

[0082] ;

[0083] in, After preprocessing, The features output by the decoder are Indicates the The features output by the decoder are Indicates that after upsampling operation, represents the index of the encoder, and ; Indicates the The fusion features of the encoders, Indicates the The characteristics of the wavelet encoder output, represents the output features of the previous decoder, Indicates the The intermediate features of the encoder, Indicates the The features output by the decoder.

[0084] Step 4: Based on the segmentation prediction graph, a binary cross entropy loss function and a Dice loss function are constructed respectively. The guidewire segmentation network model is optimized using the binary cross entropy loss function and the Dice loss function to obtain the final lightweight wavelet convolution guidewire segmentation network model.

[0085] In step 4, the binary cross entropy loss function and the Dice loss function are constructed based on the segmentation prediction graph. The guidewire segmentation network model is optimized using the binary cross entropy loss function and the Dice loss function to obtain the final lightweight wavelet convolution guidewire segmentation network model. The expression of the binary cross entropy loss function is:

[0086] ;

[0087] in, represents the binary cross entropy loss, Indicates the total number of pixels. Represents pixels The label value of Represents pixels The predicted value of

[0088] The expression of Dice loss function is:

[0089] ;

[0090] in, Denotes Dice loss.

[0091] Furthermore, the total training loss function is:

[0092] ;

[0093] in, represents the total training loss.

[0094] Please refer to Figure 5 and Figure 6 This embodiment further provides a dual guidewire generation method, wherein the method applies the lightweight wavelet convolution guidewire segmentation network model as described above, and the method comprises the following steps:

[0095] S801: Obtain two single guide wire images and their corresponding true labels, and generate a binary mask based on the true label corresponding to the second single guide wire image.

[0096] In S801, two single-wire images and their corresponding true labels are obtained, and a binary mask is generated based on the true label corresponding to the second single-wire image. The following relationship exists in the corresponding process:

[0097] ;

[0098] in, Indicates the binary mask at the pixel point The value of represents pixel coordinates, Indicates that the true label corresponding to the second single wire image is at pixel point The value of .

[0099] S802: Based on the binary mask, the two single-wire images and their corresponding true labels are pixel-replaced to obtain a pseudo double-wire image and its corresponding true label.

[0100] In S802, based on the binary mask, the two single-wire images and their corresponding true labels are pixel-replaced to obtain a pseudo double-wire image and its corresponding true label. The following relationship exists in the corresponding process:

[0101] ;

[0102] in, Indicates the pseudo double-wire image at pixel point The value of Indicates the true label corresponding to the pseudo double-wire image at pixel point The value of Indicates the first single wire image at pixel point The value of Indicates the second single wire image at pixel point The value of Indicates the true label corresponding to the first single wire image at pixel point The value of

[0103] S803: Input the pseudo dual-guidewire image into the final lightweight wavelet convolution guidewire segmentation network model for processing to obtain a segmentation prediction image.

[0104] It should be noted that in Figure 6 In the figure, the three groups of figures on the left show some examples of pseudo-double-wire data generated, and two groups of real double-wire samples are also shown on the right. The double-wire generation method proposed in this paper not only expands the range of single-wire images to double-wire images, but also implements various forms of data augmentation. This method supports the simultaneous input of multiple guidewire information to generate composite images containing three or more guidewires, significantly expanding the coverage of training data, providing more diverse training samples for the guidewire segmentation network model, and significantly enhancing the generalization ability.

[0105] It should be understood that, although the various steps in the flow chart of each embodiment of the present invention are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0106] It should be understood that various components of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0107] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

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

Claims

1. A method for constructing a lightweight wavelet convolution guidewire segmentation network model, characterized in that: The construction method comprises the following steps: Step 1: Construct a Stem module based on two-dimensional convolution, construct a wavelet encoder based on wavelet transform attention convolution, point-by-point convolution and two-dimensional convolution, and construct a decoder based on grouped convolution and point-by-point convolution. The Stem module, wavelet encoder and decoder constitute a guidewire segmentation network model; Step 2: Get the input image, use the Stem module to extract the features of the input image, obtain the underlying features, and iteratively process the underlying features through five wavelet encoders in turn to obtain the features output by the five wavelet encoders respectively; Step 3: Input the features output by the wavelet encoder into the decoder in an iterative manner for decoding to obtain a segmentation prediction map; Step 4: Based on the segmentation prediction graph, a binary cross entropy loss function and a Dice loss function are constructed respectively. The guidewire segmentation network model is optimized using the binary cross entropy loss function and the Dice loss function to obtain the final lightweight wavelet convolution guidewire segmentation network model; Among them, in the said step 2, an input image is obtained, and the Stem module is used to extract features of the input image to obtain underlying features. The underlying features are sequentially processed by five wavelet encoders in an iterative form to obtain features output by the five wavelet encoders respectively, which specifically includes the following sub-steps: Get the input image, and process it through two-dimensional convolution, ReLU activation function and batch normalization in order to obtain the underlying features; S201, the underlying features are sequentially processed by wavelet transform attention convolution, GELU activation function and batch normalization, and residual addition is performed with the underlying features to obtain the intermediate features that have undergone wavelet transform attention convolution and residual addition in the first cycle; S202, the intermediate features that have undergone wavelet transform attention convolution and residual addition in the first cycle are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the intermediate features that have undergone the first point-by-point convolution in the first cycle; S203, the intermediate features that have undergone the first point-by-point convolution in the first cycle are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the features output by the first cycle; S204, in an iterative form, taking the features outputted in the first cycle as input, and repeating the steps S201 to S203 for a certain number of times, to obtain the first Characteristics of the secondary cycle output; S205, the The features output by the second cycle are processed in sequence by two-dimensional convolution, GELU activation function, batch normalization and maximum pooling to obtain the features output by the first wavelet encoder; After iteratively taking the features output by the first wavelet encoder as input and repeating steps S201 to S205 a certain number of times, the features output by the second wavelet encoder, the features output by the third wavelet encoder, the features output by the fourth wavelet encoder, and the features output by the fifth wavelet encoder are obtained respectively, and the features output by the fifth wavelet encoder are used as the bottleneck features; The input images include single-wire images and pseudo-double-wire images generated by a double-wire generation method.

2. The method for constructing a lightweight wavelet convolution guidewire segmentation network model according to claim 1, characterized in that: In the steps of obtaining the input image, passing the input image through two-dimensional convolution, ReLU activation function and batch normalization in sequence to obtain the underlying features, the following relationship exists: ； in, Represents the underlying features, It means that after batch normalization, Indicates that it has been processed by the ReLU activation function. Indicates that after a two-dimensional convolution operation, represents the input image; In the step of sequentially subjecting the underlying features to wavelet transform attention convolution, GELU activation function and batch normalization, and performing residual addition with the underlying features to obtain the intermediate features that have undergone wavelet transform attention convolution and residual addition in the first cycle, the following relationship exists: ； in, represents the intermediate features after wavelet transform attention convolution and residual addition in the first cycle, It means that after batch normalization, Indicates that it has been processed by the GELU activation function. represents the attention convolution operation after wavelet transform, represents the addition of residuals; In the step of processing the intermediate features of the wavelet transform attention convolution and residual addition in the first cycle through point-by-point convolution, GELU activation function and batch normalization in sequence to obtain the intermediate features of the first point-by-point convolution in the first cycle, the following relationship exists: ； in, Represents the intermediate features after the first point-by-point convolution in the first cycle, Indicates point-by-point convolution operation; In the step of processing the intermediate features of the first point-by-point convolution in the first cycle through point-by-point convolution, GELU activation function and batch normalization in sequence to obtain the features output by the first cycle, the following relationship exists: ； in, Indicates the characteristics of the output of the first cycle; After the features outputted from the first cycle are inputted in an iterative manner and steps S201 to S203 are repeated a certain number of times, the first In the step of the feature output of the secondary cycle, the following relationship exists: ； in, Indicates the The intermediate features of the wavelet transform attention convolution and residual addition in the second cycle, The index indicating the number of loops, Indicates the total number of cycles, Indicates the The characteristics of the secondary cycle output, Indicates the The intermediate features after the first point-by-point convolution in the second cycle, Indicates the Characteristics of the secondary cycle output; In the The features output by the first cycle are processed in sequence through two-dimensional convolution, GELU activation function, batch normalization and maximum pooling to obtain the features output by the first wavelet encoder. The following relationship exists: ； in, represents the characteristics of the first wavelet encoder output, Indicates that the maximum pooling operation has been performed.

3. The method for constructing a lightweight wavelet convolution guidewire segmentation network model according to claim 2, characterized in that: Wavelet transform attention convolution includes the following sub-steps: The input features are sequentially subjected to two-dimensional Haar wavelet transform and two-dimensional convolution operations to obtain low-frequency components and high-frequency components. The following relationship exists in the corresponding process: ； in, represents a low-pass filter, Both represent high-pass filters. express The low-frequency component of represents the index of the wavelet level, express The high-frequency components in the horizontal direction, express The high-frequency components in the vertical direction, express High-frequency components in the diagonal direction; The SE module is constructed based on the channel attention mechanism, and the three high-frequency components generated by the wavelet transform are processed by the SE module to obtain three enhanced high-frequency components. The following relationship exists in the corresponding process: ； in, Both represent the enhanced high-frequency components. Indicates that it has been processed by the SE module; Perform a two-dimensional convolution operation on the low-frequency component and the enhanced high-frequency component to obtain the low-frequency component and high-frequency component after the two-dimensional convolution operation. The following relationship exists in the corresponding process: ； in, Represents the low-frequency component after the two-dimensional convolution operation, Both represent the high-frequency components after two-dimensional convolution operation. represents the weight matrix; The low-frequency component and high-frequency component after the two-dimensional convolution operation are transposed and convolved to obtain the wavelet level The output characteristics of the inverse wavelet transform correspond to the following relationship: ； in, Indicates the wavelet level is The output characteristics of the inverse wavelet transform are Indicates that the convolution operation has been performed; The input features are subjected to a two-dimensional convolution operation and the residual is added to the output features of the inverse wavelet transform with a wavelet level of 1 to obtain the features of the wavelet transform attention convolution output. The corresponding process has the following relationship: ； in, represents the features of the wavelet transform attention convolution output, represents the input features, Represents the output features of the inverse wavelet transform with wavelet level 1.

4. The method for constructing a lightweight wavelet convolution guidewire segmentation network model according to claim 3, characterized in that: In step 3, the features output by the wavelet encoder are input into the decoder in an iterative manner for decoding to obtain a segmentation prediction map, which specifically includes the following sub-steps: S301, the bottleneck features are sequentially subjected to upsampling, two-dimensional convolution, ReLU activation function and batch normalization to obtain preprocessed bottleneck features; S302: Concatenate the preprocessed bottleneck features with the output features of the corresponding encoder, and then perform group convolution, GELU activation function, and batch normalization processing in sequence to obtain the fusion features of the first encoder; S303, the fusion features of the first encoder are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the intermediate features of the first encoder; S304: The intermediate features of the first encoder are sequentially subjected to point-by-point convolution, GELU activation function and batch normalization to obtain the features output by the first decoder; In an iterative form, after repeating steps S301 to S304 a certain number of times with the features output by the first decoder as input, the features output by the second decoder, the features output by the third decoder, and the features output by the fourth decoder are obtained respectively, and the features output by the fourth encoder are used as the segmentation prediction map.

5. The method for constructing a lightweight wavelet convolution guidewire segmentation network model according to claim 4, characterized in that: In the steps of sequentially subjecting the bottleneck features to upsampling, two-dimensional convolution, ReLU activation function, and batch normalization to obtain the preprocessed bottleneck features, the following relationship exists: ； in, represents the bottleneck feature after preprocessing, represents the bottleneck feature, Indicates that it has undergone upsampling operation; After concatenating the preprocessed bottleneck features with the output features of the corresponding encoder, the features are sequentially processed through group convolution, GELU activation function, and batch normalization to obtain the fusion features of the first encoder. The following relationship exists: ； in, represents the fusion feature of the first encoder, Indicates that after group convolution operation, Indicates that after feature splicing operation, Represents the characteristics of the 4th wavelet encoder output; In the step of obtaining the intermediate features of the first encoder by subjecting the fused features of the first encoder to point-by-point convolution, GELU activation function and batch normalization, the following relationship exists: ； in, Represents the intermediate features of the first encoder; In the step of processing the intermediate features of the first encoder through point-by-point convolution, GELU activation function and batch normalization to obtain the features output by the first decoder, the following relationship exists: ； in, Represents the features of the first decoder output; After iteratively repeating steps S301 to S304 a certain number of times using the features output by the first decoder as input, the features output by the second decoder, the features output by the third decoder, and the features output by the fourth decoder are obtained respectively. In the step of using the features output by the fourth encoder as the segmentation prediction map, the following relationship exists: ； in, After preprocessing, The features output by the decoder are Indicates the The features output by the decoder are represents the index of the encoder, Indicates the The fusion features of the encoders, Indicates the The characteristics of the wavelet encoder output, represents the output features of the previous decoder, Indicates the The intermediate features of the encoder, Indicates the The features output by the decoder.

6. The method for constructing a lightweight wavelet convolution guidewire segmentation network model according to claim 5, characterized in that: In step 4, a binary cross entropy loss function and a Dice loss function are constructed based on the segmentation prediction graph, and the guidewire segmentation network model is optimized using the binary cross entropy loss function and the Dice loss function to obtain the final lightweight wavelet convolution guidewire segmentation network model, wherein the expression of the binary cross entropy loss function is: ； in, represents the binary cross entropy loss, Indicates the total number of pixels. Represents pixels The label value of Represents pixels The predicted value of The expression of Dice loss function is: ； in, Denotes Dice loss.

7. A method for generating a double guide wire, characterized in that: The method applies the lightweight wavelet convolution guidewire segmentation network model according to any one of claims 1 to 6, and the method comprises the following steps: S801, obtaining two single guide wire images and their corresponding true labels, and generating a binary mask based on the true label corresponding to the second single guide wire image; S802: Based on the binary mask, the two single-wire images and their corresponding true labels are pixel-replaced to obtain a pseudo double-wire image and its corresponding true label; S803: Input the pseudo dual-guidewire image into the final lightweight wavelet convolution guidewire segmentation network model for processing to obtain a segmentation prediction image.

8. The method for generating a double guide wire according to claim 7, wherein: In the steps of obtaining two single-wire images and their corresponding true labels and generating a binary mask based on the true label corresponding to the second single-wire image, the following relationship exists: ； in, Indicates the binary mask at the pixel point The value of represents pixel coordinates, Indicates that the true label corresponding to the second single wire image is at pixel point value.

9. The method for generating a double guide wire according to claim 8, wherein: In the step of performing pixel replacement on the two single-wire images and their corresponding true labels based on the binary mask to obtain the pseudo double-wire image and its corresponding true label, the following relationship exists: ； in, Indicates the pseudo double-wire image at pixel point The value of Indicates the true label corresponding to the pseudo double-wire image at pixel point The value of Indicates the first single wire image at pixel point The value of Indicates the second single wire image at pixel point The value of Indicates the true label corresponding to the first single wire image at pixel point value.

Images