Guidewire tip tracking method and system

Abstract

The present invention discloses a guidewire tip tracking method and system. The method obtains an initial frame image and obtains a feature cluster using a feature extraction method FE, and then obtains a current frame image in real time, and iteratively processes the feature cluster. In the iterative processing, a polygonal mask is used in combination with a feature extraction method FE to obtain a feature cluster, and then all historical information of the feature cluster of branch i is used to determine a unique guidewire feature. After obtaining the guidewire feature, the guidewire feature is also processed to obtain a continuous point set for representing the guidewire tip trajectory. The guidewire tip tracking method and system of the present invention can improve real-time performance and accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of X-ray fluoroscopy, and in particular to a guide wire tip tracking method and system. Background Art

[0002] With the rapid development of computer-assisted intervention technology, many related robotic intervention systems have been proposed. For example, cardiovascular robotic intervention systems. The key step in cardiovascular robotic intervention systems is to achieve the tracking and segmentation of the guidewire tip. Guidewire tip tracking and segmentation can be used to analyze the collected X-ray images online, thereby providing visual feedback of the guidewire to the robotic intervention system, reducing human intervention and laying the foundation for the development of semi-automatic or fully automatic robotic intervention systems.

[0003] Traditional guidewire tracking and segmentation methods are mainly based on curve fitting. These methods use the natural bending characteristics of the guidewire tip to detect the guidewire tip. However, this method usually does not produce ideal segmentation results in complex backgrounds. At present, with the rapid development of deep learning and machine learning technologies, many deep learning-based methods have been proposed. Deep learning-based methods use learned detectors to extract wires or use recurrent neural networks to solve the problem of guidewire tracking and segmentation. Although some related models have been proposed, the real-time and accuracy of the guidewire tip tracking and segmentation model still has a lot of room for improvement. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides a guidewire tip tracking method that can improve real-time performance and accuracy.

[0005] The invention also provides a guidewire tip tracking system.

[0006] According to the first aspect of the present invention, the guidewire tip tracking method comprises the following steps: obtaining an initial frame image, and generating a feature cluster by using a feature extraction method FE Where i is the sequence number, t is the time; for the feature cluster Performing iterative processing to obtain a unique guidewire feature; processing the guidewire feature to obtain a continuous point set for representing the guidewire tip trajectory; wherein the iterative processing includes the following steps: generating a feature cluster using a feature extraction method P-FE According to the feature cluster of branch i All historical information of the branch i is used to determine the feature cluster of the branch i. Is it the only guidewire feature? The feature extraction method P-FE comprises the following steps: obtaining a feature cluster And for the feature cluster Preprocessing is performed to obtain a polygonal mask; a current frame image is acquired, and a polygonal image is extracted from the current frame image using the polygonal mask; a feature cluster is generated based on the polygonal image and using the feature extraction method FE

[0007] The guidewire tip tracking method according to the embodiment of the present invention has at least the following beneficial effects: the method obtains the initial frame image and obtains the feature cluster by using the feature extraction method FE Then obtain the current frame image in real time and analyze the feature cluster Iterative processing is performed to obtain a unique guidewire feature, which is beneficial to improving real-time performance. In the iterative processing, the polygonal mask is combined with the feature extraction method FE to obtain the feature cluster Reuse the feature cluster of branch i All historical information is used to determine the unique guidewire feature, which is beneficial to improving accuracy; after obtaining the guidewire feature, the guidewire feature is also processed to obtain a continuous point set for representing the guidewire tip trajectory, so as to facilitate real-time tracking of the guidewire tip, which is beneficial to improving real-time performance.

[0008] According to some embodiments of the present invention, the feature extraction method FE comprises the following steps: obtaining an input image and generating a binary segmentation image using a neural network model; and generating feature clusters based on the binary segmentation image using a neighborhood growing clustering method.

[0009] According to some embodiments of the present invention, the neural network model adopts a Unet model.

[0010] According to some embodiments of the present invention, the feature cluster Preprocessing is performed to obtain a polygonal mask, including the following steps: calculating the polygonal mask according to a preset calculation formula; wherein the calculation formula is: represents the polygonal mask, f(·) represents the dilation operation in the image morphological operation method, MFD represents the maximum forward moving distance, MBD represents the maximum backward moving distance, and MAmp represents the maximum amplitude of the guide wire.

[0011] According to some embodiments of the present invention, the feature cluster is generated based on the polygonal image and using the feature extraction method. The method comprises the following steps: generating a plurality of fine-grained clusters according to the polygonal image by using the feature extraction method FE; merging the plurality of fine-grained clusters into the feature cluster by using the fine-grained denoising merging method;

[0012] According to some embodiments of the present invention, the feature cluster according to branch i All historical information of the branch i is used to determine the feature cluster of the branch i. Whether it is the only guidewire feature, comprising the following steps: according to the feature cluster of the branch i All the historical information of the branch i is used to determine the feature cluster of the branch i. Whether the decision condition is met; when the feature cluster of the branch i Satisfy the decision condition and determine the feature cluster of the branch i is the guidewire feature, and retains the feature cluster of the branch i When the feature cluster of the branch i If the decision condition is not met, the feature cluster of branch i is discarded. Determine whether the guidewire feature is unique; when the guidewire feature is unique, end the iterative processing; when the number of the guidewire features is greater than one, continue the iterative processing; when the number of the guidewire features is less than one, restart the guidewire tip tracking method.

[0013] According to some embodiments of the present invention, the feature cluster according to the branch i All the historical information of the branch i is used to determine the feature cluster of the branch i. Whether the decision condition is met, comprising the following steps: using the decision formula, calculating the feature cluster of each time t of the branch i The centroid of the feature cluster of the branch i The mean square error between the centroids of max ; When the maximum mean square error V max is less than a preset threshold, determining the feature cluster of the branch i Satisfy the decision conditions; wherein the decision formula is:

[0014]

[0015] p represents the maximum sequence number, K represents the maximum time, d t,i Represents the feature cluster With the feature cluster The Euclidean distance between a,i Represents the feature cluster The center of mass.

[0016] According to some embodiments of the present invention, the processing of the guidewire features to obtain a continuous point set for representing the guidewire tip trajectory includes the following steps: performing guidewire segmentation on the guidewire features using a feature extraction method P-FE to obtain first data; processing the first data using a skeletonization method to obtain a guidewire tip skeleton; calculating and obtaining a neighborhood matrix of the skeleton points of the guidewire tip skeleton based on a neighborhood relationship; and calculating a continuous point set for representing the guidewire tip trajectory using the neighborhood matrix.

[0017] According to some embodiments of the present invention, the following steps are also included: using topological analysis of the guidewire structure to detect whether the guidewire tip is bent; when the guidewire tip is bent, sending a warning signal to a control system.

[0018] According to the second aspect of the present invention, the guidewire tip tracking system comprises: a feature extraction module FE, which is used to obtain an initial frame image and generate a feature cluster using a feature extraction method FE Where i is the sequence number and t is the time; the feature extraction module P-FE is used to generate feature clusters using the feature extraction method P-FE The guidewire judgment module GJ is used to determine the characteristic cluster of branch i. All historical information of the branch i is used to determine the feature cluster of the branch i. Is it a unique guidewire feature; an iterative module for using the feature extraction module P-FE and the guidewire judgment module to extract the feature cluster Iterative processing is performed to obtain the unique guidewire feature; a tracking module is used to process the guidewire feature to obtain a continuous point set for representing the guidewire tip trajectory.

[0019] The guidewire tip tracking system according to the present invention has at least the following beneficial effects: the feature cluster is obtained by the feature extraction module FE. The iterative module reuses the feature extraction module P-FE to extract the feature clusters Real-time iterative processing is beneficial to improve real-time performance, and the iterative module also uses the guide wire judgment module GJ to identify the feature clusters. The system uses a tracking module to track the guidewire feature in real time, which is conducive to improving real-time performance.

[0020] 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 practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0022] Figure 1 is a flow chart of a guidewire tracking method according to an embodiment of the present invention;

[0023] Figure 2 for Figure 1 A first specific flow chart of the guidewire tracking method shown;

[0024] Figure 3 for Figure 1 A second specific flow chart of the guidewire tracking method shown;

[0025] Figure 4 for Figure 3 A third specific flow chart of the guidewire tracking method shown;

[0026] Figure 5 for Figure 4 A fourth specific flow chart of the guidewire tracking method shown;

[0027] Figure 6 for Figure 3 A fifth specific flow chart of the guidewire tracking method shown;

[0028] Figure 7 for Figure 6 A sixth specific flow chart of the guidewire tracking method shown;

[0029] Figure 8 for Figure 1 A seventh specific flow chart of the guidewire tracking method shown;

[0030] Fig. 9 for Figure 1 A flowchart of the additional steps of the guidewire tracking method shown;

[0031] Fig.10 A schematic structural diagram of a guidewire tracking system according to an embodiment of the present invention;

[0032] Fig.11 for Fig.10 Schematic diagram of the structure of the feature extraction module P-FE of the guidewire tracking system shown. DETAILED DESCRIPTION

[0033] Embodiments of the present invention are described in detail below, examples of which 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 only used to explain the present invention, and cannot be understood as limiting the present invention.

[0034] In the description of the present invention, "several" means one or more, "more" means two or more, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0035] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0036] FE (Feature Extraction module), feature extraction module.

[0037] P-FE (Polygon-mask-based Feature Extraction module), a feature extraction module based on polygon masks.

[0038] GJ (Guide-wire Judgement module), guide wire judgment module.

[0039] MBD (Maximum Move-Backwards Distance), maximum backward moving distance.

[0040] MFD (Maximum Move-Forward Distance), maximum forward moving distance.

[0041] MAmp (Maximum Amplitude of Guide-Wire), maximum guide wire amplitude.

[0042] TPS (Thin-Plate spline), thin plate spline interpolation.

[0043] FCN (Fully Convolutional Neural Network), fully convolutional neural network.

[0044] First aspect

[0045] Reference Figure 1 , a guidewire tracking method, comprising the following steps: step S1000, step S2000 and step S3000.

[0046] Step S1000, obtaining an initial frame image and generating a feature cluster using a feature extraction method FE Where i is the sequence number and t is the time.

[0047] The purpose of step S1000 is to obtain initial data and extract feature clusters similar to the guidewire features from the initial frame image. Then a screening range for subsequent screening of unique guidewire features is determined to facilitate subsequent screening. Among them, i is used to represent the sequence number. For example, if three feature clusters similar to the guidewire features are extracted, they are respectively used and Indicates; t is used to indicate the moment, the moment of the initial frame image is defined as t=a, where the value of a can be set according to actual needs. For example, the image at time 0 is taken as the initial frame image, then the value of a is 0; or, the image at time 1 is taken as the initial frame image, then the value of a is 1.

[0048] Step S2000: feature cluster An iterative process is performed to obtain a unique guidewire feature.

[0049] Step S3000: Process the guidewire features to obtain a continuous point set for representing the guidewire tip trajectory.

[0050] The purpose of step S2000 is to filter out a unique guidewire feature from the multiple feature clusters obtained in step S1000. The purpose of step S3000 is to process the filtered guidewire features. The guidewire features before processing are represented by images, and after processing, a continuous point set that can be used to represent the guidewire tip trajectory is obtained. Through iterative processing, a unique guidewire feature can be judged and filtered in real time, which is conducive to improving real-time performance.

[0051] Among them, refer to Figure 2 , the feature extraction method FE includes step S1100 and step S1200.

[0052] Step S1100, obtaining an input image and generating a binary segmentation image using a neural network model.

[0053] Step S1200, generating feature clusters using a neighborhood growing clustering method based on the binary segmentation image.

[0054] Specifically, the input image is input into the Unet model. In step S1000, the input image is the initial frame image. Then, the Unet model outputs a binary segmentation image. Then, each segmented pixel in the binary segmentation image is transferred to the corresponding cluster through the neighborhood growth clustering method, thereby generating a feature cluster. In step S1000, the feature cluster generated by the feature extraction method FE is

[0055] It should be noted that the Unet model is a neural network model based on the FCN framework, which is more suitable for the segmentation of medical images with fuzzy boundaries and complex gradients, and is conducive to improving accuracy. In addition to the Unet model, convolutional neural network models can also be used.

[0056] In addition, in addition to using the feature extraction method FE to extract feature clusters, the combination of nessel filtering and GVF ridge point extraction method can also be used to achieve the extraction of feature clusters.

[0057] Reference Figure 3 , the iterative process includes step S2100 and step S2200.

[0058] Step S2100, generating feature clusters using feature extraction method P-FE

[0059] Step S2200, based on the feature cluster of branch i All historical information of branch i is used to determine the feature cluster of branch i Is this the only guidewire feature?

[0060] Specifically, for the feature cluster Iterative processing is performed, that is, the feature extraction method P-FE is used to Iterate to get the feature cluster And, after each iteration, the feature cluster of branch i All historical information of branch i is used to determine the feature cluster of branch i Is it a unique guidewire feature, where the historical information refers to the feature cluster of branch i Information before iteration and current information. For example, when a is 0, if after iteration, the feature cluster is obtained According to the feature cluster All historical information of Is it a unique guidewire feature, where historical information refers to the feature cluster Feature Cluster and feature clusters Or, if after iteration, the feature cluster is obtained According to the feature cluster All historical information of Is it a unique guidewire feature, where historical information refers to a feature cluster Feature Cluster Feature Cluster and feature clusters

[0061] By the feature cluster Iterative processing can be performed to update the feature cluster in real time And after each iteration, according to the feature cluster of branch i All historical information on the feature cluster of branch i Making judgments to screen out unique guidewire features is beneficial to improving real-time performance and accuracy.

[0062] Among them, refer to Figure 4 , the feature extraction method P-FE includes the following steps: step S2110, step S2120 and step S2130.

[0063] Step S2110, obtaining feature clusters And for feature clusters Preprocessing is performed to obtain a polygonal mask.

[0064] Specifically, the difference between t and 1 is greater than or equal to a, and the acquired feature cluster Preprocessing refers to using a preset calculation formula to Calculation is performed to obtain a polygonal mask, where the calculation formula is: represents the polygonal mask, f(·) represents the dilation operation in the image morphological operation method, MFD represents the maximum forward moving distance, MBD represents the maximum backward moving distance, and MAmp represents the maximum amplitude of the guide wire.

[0065] Step S2120, obtaining the current frame image, and extracting the polygonal image from the current frame image using the polygonal mask.

[0066] By using the polygonal mask, a polygonal image can be extracted from the current frame image. Compared with the current frame image, the polygonal image eliminates a lot of noise, such as rib features with similar structures to the guidewire features. Among them, by using the polygonal mask, a polygonal area in the current frame image can be confirmed, so that the image in the polygonal area is extracted, and the image outside the polygonal area is eliminated, thereby achieving the purpose of eliminating noise, which is conducive to improving accuracy.

[0067] Step S2130, generating a feature cluster using a feature extraction method FE according to the polygon image

[0068] Specifically, the polygonal image with a lot of noise removed is used to obtain the corresponding feature clusters through the feature extraction method FE. Then the feature cluster More similar to guidewire characteristics, which helps improve accuracy.

[0069] Among them, refer to Figure 5 , step S2130 includes the following steps: step S2131 and step S2132.

[0070] Step S2131, generating multiple fine-grained clusters based on the polygonal image using a feature extraction method FE.

[0071] Step S2132: merging multiple fine-grained clusters into a feature cluster using a fine-grained denoising merging method

[0072] Specifically, multiple fine-grained clusters can be extracted from polygonal images using the feature extraction method FE. Multiple fine-grained clusters have high similarity, so the noise on the fine-grained clusters is eliminated by denoising and merging, so that multiple fine-grained clusters are merged into feature clusters. It is helpful to improve accuracy.

[0073] Reference Figure 6 , step S2200 includes the following steps: step S2210, step S2220, step S2230, step S2240, step S2250, step S2260 and step S2270.

[0074] Step S2210, based on the feature cluster of branch i All historical information of branch i is used to determine the feature cluster of branch i Whether the decision conditions are met.

[0075] Step S2220, when the feature cluster of branch i Satisfy the decision conditions and determine the feature cluster of branch i is the guide wire feature, and retains the feature cluster of branch i

[0076] Step S2230, when the feature cluster of branch i If the decision condition is not met, the feature cluster of branch i is discarded

[0077] Step S2240, determine whether the guidewire feature is unique.

[0078] Step S2250: When the guidewire feature is unique, the iterative process ends.

[0079] Step S2260, when the number of guidewire features is greater than one, continue iterative processing.

[0080] Step S2270, when the number of guidewire features is less than one, restart the guidewire tip tracking method.

[0081] Specifically, the decision condition is used to determine the feature cluster of branch i Is it a guide wire feature? If the feature cluster of branch i If the decision condition is met, the feature cluster of branch i is the guide wire feature, and the feature cluster of branch i Keep, otherwise, the feature cluster of branch i Abandon, that is, use the decision conditions to separately analyze the feature clusters of each branch Make a judgment, so as to classify the feature clusters belonging to the guidewire features For example, the feature cluster There are 2 branches in total. Use the decision conditions to determine the feature clusters of branch 1 respectively. and the characteristic cluster of branch 2 Whether it is a guidewire feature.

[0082] By using decision conditions to determine the feature cluster After determining whether it is a guidewire feature, it is necessary to judge the retained guidewire feature to confirm whether the retained guidewire feature is unique. If the retained guidewire feature is unique, the retained guidewire feature represents the guidewire tip, thereby terminating the iterative processing; if the number of retained guidewire features is greater than one, it means that the guidewire feature that can represent the guidewire tip has not been confirmed, and it is necessary to continue the iterative processing; if the number of retained guidewire features is less than one, it means that there is an error in the process of executing the guidewire tip tracking method, and it is necessary to restart the method, that is, re-execute the method.

[0083] Among them, refer to Figure 7 , step S2210 includes step S2211 and step S2212.

[0084] Step S2211, using the decision formula, calculate the feature cluster of each time t of branch i The centroid and characteristic cluster of branch i The mean square error between the centroids of max .

[0085] Step S2212, when the maximum mean square error V max Less than the preset threshold, determine the feature cluster of branch i Satisfy the decision conditions.

[0086] Specifically, the decision formula is: p represents the maximum sequence number, K represents the maximum time, d t,i Represents a feature cluster With feature cluster The Euclidean distance between a,i Represents a feature cluster In addition, the decision formula can be set according to actual needs. For example, in a branch, there are multiple mean square errors. In this embodiment, the maximum mean square error V among the multiple mean square errors is taken as max Feature cluster used to judge branch i Whether the decision condition is met, and in some embodiments, the sum of the mean square errors of each branch is calculated, and the feature cluster of branch i is judged by comparing the sum of the mean square errors of each branch. Whether the decision conditions are met.

[0087] In addition, the maximum mean square error V is calculated by the decision formula max After that, we need to calculate the maximum mean square error V max Compared with the preset threshold, when the maximum mean square error V max If it is less than the preset threshold, the feature cluster of branch i is determined The decision condition is satisfied. In this embodiment, the preset threshold value can be set within the range of 10 pixels to 20 pixels. For example, the preset threshold value is 10 pixels, 15 pixels or 20 pixels.

[0088] Reference Figure 8 , step S3000 includes the following steps: step S3100, step S3200, step S3300 and step S3400.

[0089] Step S3100, using feature extraction method P-FE to segment the guidewire features to obtain first data.

[0090] Step S3200: Process the first data using a skeletonization method to obtain a guidewire tip skeleton.

[0091] Step S3300, calculating and obtaining the neighborhood matrix of the skeleton points of the guidewire tip skeleton according to the neighborhood relationship.

[0092] Step S3400, using the neighborhood matrix, calculate a continuous point set used to represent the trajectory of the guidewire tip.

[0093] Specifically, after determining the unique guidewire feature, it is necessary to track the guidewire feature in real time. Using the feature extraction method P-FE, the guidewire feature is updated and the guidewire feature is segmented to obtain the first data, for example, the guidewire feature and the current frame image are input into the feature extraction method P-FE to obtain the first data. The first data is then processed by the skeletonization method to obtain the skeleton of the guidewire tip, for example, the morphological skeletonization method is used to process the first data, or the centerline extraction method is used to process the first data. Then, the skeleton points of the guidewire tip skeleton are calculated according to the neighborhood relationship to obtain a neighborhood matrix, wherein the neighborhood relationship can adopt an 8-neighborhood relationship or a 4-neighborhood relationship, etc. Through the neighborhood matrix, the guidewire tip trajectory can be represented by a continuous point set, so as to facilitate real-time tracking of the guidewire tip, which is conducive to improving real-time performance.

[0094] In addition, refer to Fig. 9 The guidewire tip tracking method also includes step S4100 and step S4200.

[0095] Step S4100, using topological analysis of the guidewire structure to detect whether the guidewire tip is bent.

[0096] Step S4200: When the tip of the guide wire bends, a warning signal is sent to the control system.

[0097] Specifically, after the guidewire tip is represented by a continuous point set in step S3000, the guidewire tip can be tracked in real time. At this stage, the topological analysis of the guidewire structure is used to detect whether the guidewire tip is bent. Specifically, the bending energy can be calculated based on whether there is a loop at the guidewire tip in the image and the TPS model, so as to comprehensively judge whether the guidewire tip is bent. When the guidewire tip is bent, the processor sends a warning signal to the control system so as to handle the bending of the guidewire tip in a timely manner.

[0098] Second aspect

[0099] Reference Fig.10 A guidewire tracking system includes a feature extraction module FE, a feature extraction module P-FE, a guidewire judgment module GJ, an iteration module, and a tracking module (not shown in the figure). The feature extraction module FE is used to obtain an initial frame image and generate a feature cluster using a feature extraction method FE. Where i is the sequence number and t is the time; the feature extraction module P-FE is used to generate feature clusters using the feature extraction method P-FE The guidewire judgment module GJ is used to identify the characteristic cluster of branch i. All historical information of branch i is used to determine the feature cluster of branch i Is it a unique guidewire feature? The iteration module is used to use the feature extraction module P-FE and the guidewire judgment module to identify the feature cluster. Iterative processing is performed to obtain a unique guidewire feature; the tracking module is used to process the guidewire feature to obtain a continuous point set for representing the guidewire tip trajectory.

[0100] Reference Fig.10 ,exist Fig.10 In the example, the value of a is 0, then I t＝0 Represents the initial frame image, I t＝1 ,I t＝n ,I t＝n+1 ,I t＝T Respectively represent the current frame image at different times. Fig.10 As an example, the current frame image is input into the feature extraction module FE to obtain the feature cluster Feature Cluster and feature clusters The feature cluster Feature Cluster Feature Cluster and the current frame image It＝1 Input the iterative module, and after being processed by the feature extraction module P-FE and the guidewire judgment module GJ, the feature cluster is obtained. Feature Cluster and feature clusters Then, for the feature cluster Feature Cluster and feature clusters Perform multiple iterations. Fig.10 After multiple iterations, a feature cluster is left. The feature cluster is the only guide wire feature. At this point, the iteration ends, thus converting the feature cluster and the current frame image I t＝T Input the feature extraction module P-FE to obtain the updated feature cluster This allows the tracking module to process the guidewire features to obtain an updated continuous point set for representing the guidewire tip trajectory, which is beneficial to improving real-time performance.

[0101] Reference Fig.11 , the feature cluster and the current frame image I t Input feature extraction module P-FE to get feature clusters Specifically, the feature extraction module P-FE uses the input feature cluster Generate a polygonal mask, and then use the polygonal mask to input the current frame image I t The polygonal image is processed, and then the feature extraction module FE is used to process the polygonal image to obtain multiple fine-grained clusters. Finally, the multiple fine-grained clusters are denoised and merged into feature clusters through the fine-grained denoising and merging method. In order to obtain the feature cluster More accurate, which is conducive to improving accuracy.

[0102] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A guidewire tip tracking method, characterized in that: The following steps are involved: Get the initial frame image and use the feature extraction method FE to generate feature clusters Where i is the sequence number and t is the time; For the feature cluster Performing iterative processing to obtain a unique guidewire feature; Processing the guidewire features to obtain a continuous point set for representing the guidewire tip trajectory; The iterative process comprises the following steps: Generate feature clusters using feature extraction method P-FE According to the feature cluster of branch i All historical information of branch i is used to determine the feature cluster Whether it is the only guidewire feature; The feature extraction method P-FE comprises the following steps: Get feature clusters And for the feature cluster Performing preprocessing to obtain a polygonal mask; Acquire a current frame image, and extract a polygonal image from the current frame image using the polygonal mask; According to the polygonal image, a feature cluster is generated using the feature extraction method FE The feature cluster according to branch i All historical information of branch i is used to determine the feature cluster Whether it is the only guidewire feature, comprising the following steps: According to the feature cluster of the branch i All the historical information of the branch i is used to determine the feature cluster of the branch i. Whether the decision conditions are met; When the feature cluster of the branch i Satisfy the decision condition and determine the feature cluster of the branch i is the guidewire feature, and retains the feature cluster of the branch i When the feature cluster of the branch i If the decision condition is not met, the feature cluster of branch i is discarded. Determining whether the guidewire feature is unique; When the guidewire feature is unique, the iterative process ends; When the number of the guidewire features is greater than one, continuing the iterative process; When the number of guidewire features is less than one, the guidewire tip tracking method is restarted.

2. The guidewire tip tracking method according to claim 1, characterized in that: The feature extraction method FE comprises the following steps: Get the input image and generate a binary segmentation image using the neural network model; According to the binary segmentation image, a feature cluster is generated using a neighborhood growing clustering method.

3. The guidewire tip tracking method according to claim 2, characterized in that: The neural network model adopts the Unet model.

4. The guidewire tip tracking method according to claim 1, characterized in that: The feature cluster FC i t-1 Preprocessing is performed to obtain a polygonal mask, including the following steps: The polygonal mask is obtained by calculation according to a preset calculation formula; Wherein, the calculation formula is: represents the polygonal mask, f(·) represents the dilation operation in the image morphological operation method, MFD represents the maximum forward moving distance, MBD represents the maximum backward moving distance, and MAmp represents the maximum amplitude of the guide wire.

5. The guidewire tip tracking method according to claim 1, characterized in that: The method of extracting the polygonal image and generating a feature cluster using the feature extraction method The following steps are involved: According to the polygonal image, a plurality of fine-grained clusters are generated using the feature extraction method FE; Using the fine-grained denoising merging method, multiple fine-grained clusters are merged into the feature cluster 6. The guidewire tip tracking method according to claim 1, characterized in that: The feature cluster according to the branch i All the historical information of the branch i is used to determine the feature cluster of the branch i. Whether the decision conditions are met, including the following steps: Using the decision formula, calculate the feature cluster at each time t of the branch i The centroid of the feature cluster of the branch i The mean square error between the centroids of max ; When the maximum mean square error V max is less than a preset threshold, determining the feature cluster of the branch i satisfying the decision conditions; The decision formula is: p represents the maximum sequence number, K represents the maximum time, d t,i Represents the feature cluster With the feature cluster The Euclidean distance between a,i Represents the feature cluster The center of mass.

7. The guidewire tip tracking method according to claim 1, characterized in that: The processing of the guidewire features to obtain a continuous point set for representing the guidewire tip trajectory comprises the following steps: Segmenting the guidewire features using a feature extraction method P-FE to obtain first data; Processing the first data using a skeletonization method to obtain a guidewire tip skeleton; Calculate and obtain a neighborhood matrix of the skeleton points of the guidewire tip skeleton according to the neighborhood relationship; Using the neighborhood matrix, a continuous point set for representing the trajectory of the guidewire tip is calculated.

8. The guidewire tip tracking method according to claim 1, characterized in that: The following steps are also included: Using topological analysis of the guidewire structure to detect whether the guidewire tip is bent; When the guidewire tip bends, a warning signal is sent to the control system.

9. A guidewire tracking system, characterized in that: include: The feature extraction module FE is used to obtain the initial frame image and generate feature clusters using the feature extraction method FE Where i is the sequence number and t is the time; Feature extraction module P-FE, used to generate feature clusters using feature extraction method P-FE The guidewire judgment module GJ is used to determine the characteristic cluster of branch i. All historical information of the branch i is used to determine the feature cluster of the branch i. Whether it is the only guidewire feature; Iteration module, used for using the feature extraction module P-FE and the guidewire judgment module to extract the feature cluster Performing iterative processing to obtain unique guidewire features; A tracking module, used for processing the guidewire features to obtain a continuous point set for representing the guidewire tip trajectory; The guidewire judgment module GJ is also used to determine the characteristic cluster of the branch i. All the historical information of the branch i is used to determine the feature cluster of the branch i. Whether the decision conditions are met; When the feature cluster of the branch i Satisfy the decision condition and determine the feature cluster of the branch i is the guidewire feature, and retains the feature cluster of the branch i When the feature cluster of the branch i If the decision condition is not met, the feature cluster of branch i is discarded. Determine whether the guidewire feature is unique; when the guidewire feature is unique, end the iterative processing; when the number of the guidewire features is greater than one, continue the iterative processing; when the number of the guidewire features is less than one, restart the guidewire tip tracking method.

Images