A carbon fiber guide wire positioning method and device for interventional surgery

Abstract

The present invention provides a method and device for positioning a carbon fiber guide wire in an interventional operation, relating to the technical field of medical image processing. The method includes: obtaining multi-source data in real time during an interventional operation; performing time synchronization processing on the multi-source data to obtain synchronized image data and synchronized electromagnetic positioning data; extracting features from the synchronized image data to obtain first position features; updating the spatial position of the carbon fiber guide wire through the synchronized electromagnetic positioning data to obtain second position features; performing multi-source fusion through the first position features and the second position features to obtain final position features; constructing a carbon fiber guide wire positioning network through the final position features, and positioning the carbon fiber guide wire in the interventional operation through the carbon fiber guide wire positioning network to obtain a carbon fiber guide wire positioning result. The present invention solves the problems of unclear imaging, low positioning accuracy, and insufficient real-time performance existing in the existing carbon fiber guide wire positioning methods.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and more particularly, to a method and device for positioning a carbon fiber guide wire in an interventional operation. Background Art

[0002] The positioning of a carbon fiber guide wire plays a crucial role in interventional operations. It has high flexibility, small size, and excellent mechanical properties. However, due to its unique properties, positioning errors are prone to occur in electromagnetic positioning.

[0003] Due to the size of the carbon fiber guide wire, it is difficult to position it using an electromagnetic sensor. Although electromagnetic markers can use nano-level adhesives or fix magnetic markers inside the carbon fiber guide wire through an embedded design. However, due to the small diameter of the carbon nano guide wire, the signals generated by the magnetic markers are weak, which makes precise positioning even more difficult. At the same time, external electromagnetic fields and other medical devices may interfere with the signals, affecting the positioning accuracy.

[0004] On the other hand, carbon fiber guide wires are usually made of carbon fiber or carbon nano materials, and these materials themselves have low radioactivity contrast under CT or X-ray. The X-ray absorption coefficient of carbon itself is much lower than that of metal materials. Therefore, in conventional X-ray or CT scans, carbon nano guide wires may be difficult to clearly display, resulting in unclear imaging and affecting the positioning accuracy. And the carbon nano guide wire may bend or deform during the operation. When using X-ray or CT for positioning, the three-dimensional shape of the guide wire (especially the bent part) may be difficult to accurately display. Especially when passing through curved blood vessels or tortuous channels, the shape of the guide wire makes its positioning more difficult. Therefore, there is an urgent need for a high-precision and real-time carbon fiber guide wire positioning method. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for positioning a carbon fiber guide wire in an interventional operation to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0006] In a first aspect, the present application provides a method for positioning a carbon fiber guide wire in an interventional operation, including:

[0007] Obtaining multi-source data in real time during an interventional operation, where the multi-source data includes image data and electromagnetic positioning data;

[0008] Performing time synchronization processing on the image data and the electromagnetic positioning data to obtain synchronized multi-source data, where the synchronized multi-source data includes synchronized image data and synchronized electromagnetic positioning data;

[0009] Performing feature extraction on the synchronized image data to obtain a first position feature;

[0010] Update the spatial position of the carbon fiber guide wire through the synchronized electromagnetic positioning data to obtain a second position feature;

[0011] Perform multi-source fusion through the first position feature and the second position feature to obtain a final position feature;

[0012] Construct a carbon fiber guide wire positioning network through the final position feature, and position the carbon fiber guide wire in the interventional surgery through the carbon fiber guide wire positioning network to obtain a carbon fiber guide wire positioning result.

[0013] In a second aspect, the present application also provides a carbon fiber guide wire positioning device for interventional surgery, including:

[0014] An acquisition unit for real-time acquisition of multi-source data in interventional surgery, the multi-source data including image data and electromagnetic positioning data;

[0015] A synchronization unit for performing time synchronization processing on the image data and the electromagnetic positioning data to obtain synchronized multi-source data, the synchronized multi-source data including synchronized image data and synchronized electromagnetic positioning data;

[0016] A first feature extraction unit for extracting features from the synchronized image data to obtain a first position feature;

[0017] A second feature extraction unit for updating the spatial position of the carbon fiber guide wire through the synchronized electromagnetic positioning data to obtain a second position feature;

[0018] A fusion unit for performing multi-source fusion through the first position feature and the second position feature to obtain a final position feature;

[0019] A positioning unit for constructing a carbon fiber guide wire positioning network through the final position feature, and positioning the carbon fiber guide wire in the interventional surgery through the carbon fiber guide wire positioning network to obtain a carbon fiber guide wire positioning result.

[0020] The beneficial effects of the present invention are as follows: Through multi-source data fusion, the present invention adopts time synchronization processing of image data and electromagnetic positioning data to ensure the temporal consistency of different data sources, reducing the positioning error. Secondly, by combining multi-resolution dense convolution sub-modules, the network can effectively capture the detailed information of the guide wire at different scales, improving the expressive ability of the image data, suppressing background noise, enhancing the signal quality of the image features, and thus improving the positioning accuracy. The electromagnetic positioning data makes up for the deficiency of electromagnetic navigation in positioning in deep or occluded areas through the carbon fiber guide wire positioning network. In addition, by fusing the position features of multiple sources, the network can learn richer and more accurate spatial information, improving the positioning accuracy and real-time performance of the carbon fiber guide wire in complex anatomical environments.

[0021] Other features and advantages of the present invention will be described in the following specification, and some of them will become apparent from the specification or can be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 It is a schematic flow chart of the carbon fiber guide wire positioning method for interventional surgery described in the embodiments of the present invention;

[0024] Figure 2 It is a schematic structural diagram of the carbon fiber guide wire positioning network described in the embodiments of the present invention;

[0025] Figure 3 It is a schematic structural diagram of the first feature extraction branch described in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive distinction and cannot be understood as indicating or implying relative importance.

[0028] Embodiment 1:

[0029] This embodiment provides a method for positioning a carbon fiber guide wire in an interventional operation.

[0030] See Figure 1 , which shows that this method includes step S100, step S200, step S300, step S400, step S500, and step S600.

[0031] Step S100: Obtain multi-source data in real time during an interventional operation, where the multi-source data includes image data and electromagnetic positioning data;

[0032] In this embodiment, the image data can be real-time X-ray or CT images, and the electromagnetic positioning data is obtained through magnetic markers on the carbon nanotube guide wire and externally paired electromagnetic sensor devices. Specifically, by combining X-ray or CT images with electromagnetic positioning data, the deficiencies of a single data source are made up for.

[0033] Step S200: Perform time synchronization processing on the image data and the electromagnetic positioning data to obtain synchronized multi-source data, where the synchronized multi-source data includes synchronized image data and synchronized electromagnetic positioning data;

[0034] In this embodiment, the image data and the electromagnetic positioning data come from different devices, and the sampling frequencies and time points may be different. If time synchronization processing is not performed, there will be problems such as data mismatch, real-time positioning lag or inaccuracy. At the same time, multi-source data needs to be used as input for feature fusion simultaneously. If the time steps of different modality data are not synchronized, it will be difficult for the network to capture effective spatio-temporal correlation characteristics.

[0035] The step S200 includes:

[0036] Step S201: Obtain the timestamps of the image data and the electromagnetic positioning data respectively;

[0037] Step S202: Perform nearest neighbor matching on the image data and the electromagnetic positioning data based on the timestamps to obtain preliminary alignment data;

[0038] Step S203: Perform time compensation on the preliminary alignment data through interpolation to obtain synchronized multi-source data.

[0039] In this embodiment, through preliminary alignment by nearest neighbor matching, the closest value between the time points of the electromagnetic data and the image data can be quickly found, and the data error can be controlled within the time resolution range, avoiding significant misalignment problems caused by too large differences in time points. The interpolation method can perform refined adjustment on the preliminarily aligned data, further reducing the residual error at the time point and ensuring high data synchronization. It is compatible with different sampling frequencies, supplements the low-frequency data, and seamlessly integrates with the high-frequency data, which helps to capture effective features during feature extraction and fusion, thereby improving the positioning performance.

[0040] Step S300: Extract features from the synchronized image data to obtain first position features;

[0041] In this embodiment, a carbon fiber guidewire positioning network is constructed as Figure 2 shown. The carbon fiber guidewire positioning network includes two feature extraction branches, namely the first feature extraction branch and the second feature extraction branch. Among them, the first feature extraction branch is used to extract the first position features of the synchronized image data, and the second feature extraction branch is used to extract the second position features of the synchronized electromagnetic positioning data. ⊕ represents an addition operation.

[0042] The step S300 includes:

[0043] Step S301: Construct a feature extraction module, a feature fusion module and a timing module based on the synchronized image data;

[0044] As Figure 3 shown, it is the structure diagram of the first feature extraction branch, where UP represents upsampling and ⊕ represents an addition operation.

[0045] Step S302: Extract features from the synchronized image data through the feature extraction module to obtain preliminary position features;

[0046] The step S302 includes:

[0047] Step A100: Extract features from the synchronized image data through the convolutional sub-module to obtain the first extracted feature;

[0048] In this embodiment, due to the high transparency of the carbon fiber guide wire, some auxiliary techniques will be adopted during the interventional operation to enhance the visibility of the guide wire. For example, fluorescence or other special markers are added or trace radioactive imaging materials are embedded to improve the visibility in low-contrast images, which appears as a linear structure in the image. Therefore, this local characteristic can be effectively captured through the convolution operation. The convolutional sub-module can clearly extract the guide wire features, including low-level information such as the edges and brightness gradients of the carbon fiber guide wire. At the same time, it can also reduce background interference and initially filter the interference of background structures (such as bones and soft tissues).

[0049] Step A200: Sequentially extract features from the first extracted feature through multiple dense convolutional sub-modules to obtain multiple dense features with different resolutions. Among them, every two adjacent dense convolutional sub-modules are connected by a sampling sub-module, and the sampling sub-module includes a batch normalization, a convolutional layer, and an average pooling layer connected in sequence;

[0050] In this step, the number of fusion convolutional layers included in each dense convolutional sub-module is different. In this embodiment, for lightweight design, the number of fusion convolutional layers in multiple dense convolutional sub-modules is set to 4, 8, 12, and 8. At the same time, the sampling sub-module is used to reduce the number of channels of the features and further downsample the feature map.

[0051] Step A300: Use multiple dense features with different resolutions together as the preliminary position features output by the feature extraction module.

[0052] In step S302, the steps for the dense convolutional sub-module to extract features are:

[0053] Step B100: Obtain the input features of the dense convolutional sub-module;

[0054] Step B200: Perform multi-scale perception on the input features of the dense convolutional sub-module through multiple sequentially connected fusion convolutional layers to obtain the first convolutional feature. The fusion convolutional layer includes a dimensionality reduction convolutional block and a fusion convolutional block connected in sequence;

[0055] Step B300: Enhance the global perception ability of the first convolutional feature through a shared perceptron to obtain the output features of the dense convolutional sub-module.

[0056] In this embodiment, since the features of the carbon fiber guide wire are small in the image data, and the importance of different scale information varies, and the overall position information needs to be considered during positioning, while simple local features are likely to cause positioning errors, a dense convolution sub-module is constructed for feature extraction.

[0057] Among them, the input of each fusion convolution layer in the dense convolution sub-module comes from the outputs of all previous layers, and the output of each layer is also passed to all subsequent layers, enhancing feature propagation and reuse, reducing feature redundancy, and at the same time being able to extract information from all previous layers, improving feature diversity. Since each layer only needs to learn a small number of new features, because most features can be reused from the previous layers, the number of parameters is reduced, and the output of each layer contains the feature information of all previous layers, forming multi-level feature fusion.

[0058] At the same time, a dimensionality reduction convolution block, which is a 1x1 convolution, is set for each fusion convolution layer to reduce the number of channels of the input feature map and reduce the computational amount.

[0059] The expression of the fusion convolution layer is:

[0060] F i =SE(α1·f dilated (X di )+α2·f depthwise (X di )+α3·f adaptive (X di ))

[0061] X di =Conv(f li ([F1,F2,…,F i-1 ))

[0062] In the formula, F n represents the output feature of the nth fusion convolution layer, n = 1, 2, …, i, n represents the number of fusion convolution layers, i represents the index parameter of the fusion convolution layer, SE(·) represents the channel attention mechanism, α1, α2, and α3 all represent weight factors, f dilated (·) represents dilated convolution, f depthwise (·) represents depthwise separable convolution, f adaptive (·) represents adaptive convolution, X di represents the input feature of the ith fusion convolution layer, f li (·) represents the transformation function of the ith fusion convolution layer, Conv(·) represents the dimensionality reduction convolution operation.

[0063] In this embodiment, the fusion convolution block includes dilated convolution, depthwise separable convolution, and adaptive convolution. These three convolution methods can effectively capture multi-scale features, and their respective characteristics can complement each other, thereby enhancing the network's perception ability. Especially in processing tasks such as carbon fiber guide wire positioning with multi-scale features, it can significantly improve the positioning performance.

[0064] Step S303: Feature-fuse the preliminary position features through the feature fusion module to obtain fused position features;

[0065] The said step S303 includes:

[0066] Step C100: Upsample and feature-fuse multiple dense features in sequence according to the resolution size to obtain preliminary fused features containing low-level details and high-level semantic information;

[0067] In this embodiment, high-resolution features contain more spatial details, and low-resolution features have stronger global semantic information. The upsampling operation can restore low-resolution features to a higher resolution and align them with high-resolution features, thus achieving better fusion. Multi-level feature fusion helps to capture both the detailed information (such as edge information) and context information (such as surrounding tissue structures) of the carbon fiber guide wire.

[0068] In guide wire positioning, high-resolution features help to identify the boundaries and shapes of carbon fiber guide wires, while low-resolution features provide overall spatial positions and semantic clues. Through fusion, the position of the carbon fiber guide wire in a complex background can be accurately located.

[0069] Step C200: Suppress the background noise of the preliminary fused features through the self-attention mechanism to obtain weight-fused features;

[0070] In this embodiment, carbon fiber guide wires are usually small and are easily interfered by the surrounding complex background (such as blood vessels, soft tissues, etc.). The self-attention mechanism can adaptively assign higher weights to the regions where carbon fiber guide wires are located. By focusing on the relevant features of carbon fiber guide wires, weakening the useless interference in the complex background, and effectively distinguishing carbon fiber guide wires from the background, such as the contrast agent flowing in blood vessels and the reflection of surgical instruments, the accuracy of carbon fiber guide wire positioning can be improved.

[0071] Step C300: Non-linearly map the weight-fused features through a shared perceptron to obtain fused position features.

[0072] In this embodiment, in the carbon fiber guide wire positioning task, complex structures may lead to non-linear relationships between features, and the shared perceptron can learn these complex relationships, enhance the expression ability of features, reduce the computational cost through parameter sharing, and retain the global perception ability at the same time.

[0073] Step S304: Learn the context information of the fused position feature through the timing module to obtain a first position feature, and the timing module adopts a gated recurrent unit.

[0074] In this embodiment, during the interventional surgery, the position and movement of the carbon fiber guide wire are continuously changing, and the timing information is crucial for the positioning of the carbon fiber guide wire. The timing module can effectively learn the dynamic information of the position feature changing with time, and can predict the current position of the carbon fiber guide wire according to the historical positioning information, improving the accuracy and stability of the positioning, especially during the rapid operation process.

[0075] Step S400: Update the spatial position of the carbon fiber guide wire through the synchronized electromagnetic positioning data to obtain a second position feature;

[0076] In this embodiment, the electromagnetic positioning can provide the real-time position of the carbon fiber guide wire in the three-dimensional space, but due to the influence of factors such as electromagnetic interference, the position may drift. Therefore, by updating the spatial position, the noise is filtered out, and a more accurate spatial position feature is obtained, ensuring the accuracy and stability of the position update of the carbon fiber guide wire.

[0077] Specifically, the second position feature is obtained through Figure 2 the second feature extraction branch, the spatial position is updated through the update module, and the spatial position is encoded through the spatial position encoding module, and finally the second position feature is obtained through the shared perceptron.

[0078] The step S400 includes:

[0079] Step S401: Perform position update on the synchronized electromagnetic positioning data through Kalman filtering to obtain the updated spatial position of the carbon fiber guide wire;

[0080] In this embodiment, during the interventional surgery, the position of the carbon fiber guide wire needs to be updated with high frequency and low latency. Kalman filtering can filter out the noise in the synchronized electromagnetic positioning data, ensuring the accuracy and stability of the position update of the carbon fiber guide wire.

[0081] Step S402: Perform spatial position encoding on the updated spatial position of the carbon fiber guide wire to obtain a position encoding feature;

[0082] In this embodiment, after obtaining the updated spatial position, this information must be converted into a feature representation suitable for further processing. The spatial position encoding can convert the position data into a feature suitable for neural network processing, providing more effective information for subsequent network processing.

[0083] Meanwhile, the spatial position encoding can fuse the position information of the guide wire with other features, enhancing the network's understanding of spatial information.

[0084] Step S403: Perform feature mapping on the position encoding feature through a shared perceptron to obtain a second position feature.

[0085] In this embodiment, the shared perceptron processes the input features by sharing parameters, enhancing the perception ability of the entire spatial environment and avoiding local overfitting.

[0086] Step S500: Perform multi-source fusion through the first position feature and the second position feature to obtain a final position feature;

[0087] In this embodiment, in the carbon fiber guide wire positioning task during interventional surgery, the data sources come from different modalities, such as image data and electromagnetic positioning data. To ensure that the information of each data source can be fully utilized, these multi-source data are effectively fused to improve the accuracy and stability of the final positioning result.

[0088] The step S501 includes:

[0089] Step S502: Perform normalization processing on the first position feature and the second position feature to obtain a first normalized feature and a second normalized feature;

[0090] In this embodiment, the first position feature and the second position feature come from different modalities, and their value ranges and scales are different. Normalization processing can unify these features to the same scale, thereby reducing the bias caused by scale differences. And after normalization, the features can be smoother in the subsequent fusion and learning processes, avoiding a certain feature having too much influence on network training.

[0091] Step S503: Perform feature splicing on the first normalized feature and the second normalized feature to obtain a position vector;

[0092] Step S504: Perform cross-fusion on the position vector through a cross-attention mechanism to obtain a cross position vector;

[0093] In this embodiment, the cross-attention mechanism can adaptively weight different features in the feature space, utilize the mutual relationship between each feature, and select the most important information for fusion. The cross-attention mechanism is more effective than simple weighted average or splicing methods because it can learn the correlation between different features in a specific task and automatically adjust the fusion weights.

[0094] Step S505: Perform feature mapping on the cross position vector through a shared perceptron to obtain a final position feature.

[0095] In this embodiment, the cross position vector is further processed by a shared perceptron, mapped to a higher dimensional space through nonlinear transformation, and potential complex relationships are mined to generate the final high-quality position feature, i.e., the final position feature. At the same time, the generalization ability of the network is enhanced by sharing parameters, and it has good adaptability and stability between different input data.

[0096] Step S600: constructing a carbon fiber guide wire positioning network through the final position feature, and positioning the carbon fiber guide wire in the interventional surgery through the carbon fiber guide wire positioning network to obtain a carbon fiber guide wire positioning result.

[0097] In this embodiment, different branches in the carbon fiber guide wire positioning network are shared for training, that is, different tasks are shared for training, and information is transferred between different tasks, thereby improving the comprehensive ability of the network. At the same time, a joint loss function is set to balance the learning objectives between different tasks.

[0098] In summary, the present invention aims to solve the problems of high flexibility, small size, low contrast and weak electromagnetic signal of carbon fiber guidewires during positioning in interventional surgery by using technologies such as time synchronization, feature extraction, noise suppression, multi-source fusion and timing modeling to solve the problems of unclear imaging, low positioning accuracy and insufficient real-time performance of carbon nano guidewires during positioning in interventional surgery.

[0099] Specifically, the present invention fuses multi-source data, learns the position features of different data sources through the carbon fiber guidewire positioning network, and uses the cross-attention mechanism to optimize the interaction between the two data sources during the feature fusion process, so that the network can learn richer and more accurate spatial information. It improves the positioning accuracy and real-time performance of carbon fiber guidewires in interventional surgery, and can make precise adjustments during dynamic surgery, providing doctors with more reliable positioning support.

[0100] Embodiment 2:

[0101] This embodiment provides a carbon fiber guidewire positioning device for use in interventional surgery, the device comprising:

[0102] An acquisition unit, used for acquiring multi-source data in interventional surgery in real time, wherein the multi-source data includes image data and electromagnetic positioning data;

[0103] A synchronization unit, used for performing time synchronization processing on the image data and the electromagnetic positioning data to obtain synchronized multi-source data, wherein the synchronized multi-source data includes synchronized image data and synchronized electromagnetic positioning data;

[0104] A first feature extraction unit, used for performing feature extraction on the synchronous image data to obtain a first position feature;

[0105] A second feature extraction unit, configured to update the spatial position of the carbon fiber guide wire through the synchronous electromagnetic positioning data to obtain a second position feature;

[0106] A fusion unit, configured to perform multi-source fusion through the first position feature and the second position feature to obtain a final position feature;

[0107] A positioning unit, configured to construct a carbon fiber guide wire positioning network through the final position feature, and position the carbon fiber guide wire in the interventional operation through the carbon fiber guide wire positioning network to obtain a carbon fiber guide wire positioning result.

[0108] It should be noted that regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0109] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0110] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A carbon fiber guide wire positioning device for interventional surgery, characterized in that, Including: An acquisition unit for acquiring multi-source data in an interventional operation in real time, where the multi-source data includes image data and electromagnetic positioning data; A synchronization unit for performing time synchronization processing on the image data and the electromagnetic positioning data to obtain synchronized multi-source data, where the synchronized multi-source data includes synchronized image data and synchronized electromagnetic positioning data; A first feature extraction unit for extracting features from the synchronized image data to obtain a first position feature, including: constructing a feature extraction module, a feature fusion module, and a timing module based on the synchronized image data; Extracting features from the synchronized image data through the feature extraction module to obtain a preliminary position feature; Performing feature fusion on the preliminary position feature through the feature fusion module to obtain a fused position feature; Learning the context information of the fused position feature through the timing module to obtain a first position feature, where the timing module uses a gated recurrent unit; A second feature extraction unit for updating the spatial position of a carbon fiber guide wire through the synchronized electromagnetic positioning data to obtain a second position feature, including: performing position update on the synchronized electromagnetic positioning data through Kalman filtering to obtain the updated spatial position of the carbon fiber guide wire; Performing spatial position encoding on the updated spatial position of the carbon fiber guide wire to obtain a position encoding feature; Performing feature mapping on the position encoding feature through a shared perceptron to obtain a second position feature; A fusion unit for performing multi-source fusion through the first position feature and the second position feature to obtain a final position feature, including: performing normalization processing on the first position feature and the second position feature to obtain a first normalized feature and a second normalized feature; Performing feature splicing on the first normalized feature and the second normalized feature to obtain a position vector; Performing cross-fusion on the position vector through a cross-attention mechanism to obtain a cross position vector; Performing feature mapping on the cross position vector through a shared perceptron to obtain a final position feature; A positioning unit for constructing a carbon fiber guide wire positioning network through the final position feature and positioning the carbon fiber guide wire in the interventional operation through the carbon fiber guide wire positioning network to obtain a carbon fiber guide wire positioning result.

2. The carbon fiber guide wire positioning device for interventional surgery according to claim 1, wherein , The performing time synchronization processing on the image data and the electromagnetic positioning data to obtain synchronized multi-source data includes: Respectively obtaining the timestamps of the image data and the electromagnetic positioning data; Performing nearest neighbor matching on the image data and the electromagnetic positioning data based on the timestamps to obtain preliminary alignment data; Performing time compensation on the preliminary alignment data through interpolation to obtain synchronized multi-source data.

3. The carbon fiber guide wire positioning device for interventional surgery according to claim 1, wherein , The extracting features from the synchronized image data through the feature extraction module to obtain a preliminary position feature includes: Performing feature extraction on the synchronized image data through a convolutional sub-module to obtain a first extraction feature; The first extracted feature is successively subjected to feature extraction through a plurality of dense convolution sub-modules to obtain a plurality of dense features with different resolutions. Among them, every two adjacent dense convolution sub-modules are connected by a sampling sub-module, and the sampling sub-module includes a batch normalization, a convolution layer, and an average pooling layer connected in sequence; The plurality of dense features with different resolutions are jointly used as the preliminary position features output by the feature extraction module.

4. The carbon fiber guide wire positioning device for interventional surgery according to claim 3, wherein , The steps of the dense convolution sub-module for feature extraction are: Obtain the input feature of the dense convolution sub-module; The input feature of the dense convolution sub-module is subjected to multi-scale perception through a plurality of fusion convolution layers connected in sequence to obtain a first convolution feature, and the fusion convolution layer includes a dimensionality reduction convolution block and a fusion convolution block connected in sequence; Enhance the global perception ability of the first convolution feature through a shared perceptron to obtain the output feature of the dense convolution sub-module.

5. The carbon fiber guide wire positioning device for interventional surgery according to claim 4, characterized in that , The expression of the fusion convolution layer is: F i = SE(α1·f dilated (X di ) + α2·f depthwise (X di ) + α3·f adaptive (X di )) X di = Conv(f li ([F1,F2,…,F i-1 )) where, F n represents the output feature of the nth fusion convolutional layer, n = 1, 2, …, i, n represents the number of fusion convolutional layers, i represents the index parameter of the fusion convolutional layer, SE(·) represents the channel attention mechanism, α1, α2, and α3 all represent weight factors, f dilated (·) represents dilated convolution, f depthwise (·) represents depthwise separable convolution, f adaptive (·) represents adaptive convolution, X di represents the input feature of the ith fusion convolutional layer, f li (·) represents the transformation function of the ith fusion convolutional layer, Conv(·) represents the dimensionality reduction convolution operation.

6. The carbon fiber guide wire positioning device for interventional surgery according to claim 3, characterized in that , The process of the feature fusion module fusing the preliminary position features to obtain the fused position features includes: The plurality of dense features are successively upsampled and feature-fused according to the size of the resolution to obtain a preliminary fused feature containing low-level details and high-level semantic information; Suppress the background noise of the preliminary fused feature through a self-attention mechanism to obtain a weighted fused feature; The weighted fused feature is subjected to non-linear mapping processing through a shared perceptron to obtain the fused position feature.

Images