A blood vessel positioning method based on multi-parameter fusion and self-learning model

Abstract

The application discloses a blood vessel positioning method based on a multi-parameter fusion and self-learning model, and relates to the technical field of blood vessel positioning. The method comprises the following steps: collecting three-dimensional medical image data, and pre-processing the image; segmenting the aorta and branch blood vessels by using an image segmentation algorithm, determining a region where a target blood vessel is located, continuously collecting infrared blood vessel images in the region where the target blood vessel is located, and naming the images as a first image and a second image respectively; determining the position of the target blood vessel in the second image by using an optical flow tracking method according to the position of the target blood vessel in the first image, and realizing blood vessel positioning. The application fuses multi-scale, multi-direction and multi-source heterogeneous information, and introduces a self-learning mechanism to optimize the segmentation and tracking process, so that real-time positioning of the target blood vessel is realized.

Description

Technical Field

[0001] This invention relates to the field of vascular localization technology, and in particular to a vascular localization method based on multi-parameter fusion and a self-learning model. Background Technology

[0002] In modern medical diagnosis and treatment, vascular localization technology is one of the core supporting technologies for clinical procedures such as minimally invasive surgery, interventional vascular therapy, and puncture biopsy. Accurate and real-time vascular localization provides clinicians with precise anatomical references, effectively reducing surgical risks, decreasing the incidence of complications, and improving treatment outcomes and patient prognosis. With the rapid development of medical imaging technology and artificial intelligence algorithms, vascular localization technology is evolving from traditional manual localization and single-modal image localization to multimodal fusion and intelligent self-learning localization.

[0003] Traditional vascular localization methods have many limitations: localization methods that rely solely on 3D medical images can provide anatomical information about blood vessels, but there is a time delay in image acquisition and processing, which cannot meet the needs of real-time navigation; localization methods based solely on infrared vascular imaging are easily affected by physiological factors such as skin thickness, pigmentation, and subcutaneous fat content, and have limited ability to detect deep blood vessels, making it difficult to guarantee localization accuracy; while traditional segmentation and tracking algorithms lack adaptability and are difficult to cope with complex scenarios such as differences in vascular morphology among different patients and interference from surrounding tissues.

[0004] Therefore, providing a blood vessel localization method based on multi-parameter fusion and self-learning model to overcome the difficulties of existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a blood vessel localization method based on multi-parameter fusion and self-learning model, which integrates multi-scale, multi-directional, and multi-source heterogeneous information, and introduces a self-learning mechanism to optimize the segmentation and tracking process, thereby achieving real-time localization of the target blood vessel.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A blood vessel localization method based on multi-parameter fusion and self-learning model includes the following steps: Acquire 3D medical image data and preprocess the images; Image segmentation algorithms are used to segment the aorta and its branch vessels, and to determine the region where the target vessel is located. Infrared images of blood vessels were continuously acquired in the area where the target blood vessel was located, and named as the first image and the second image respectively. Based on the position of the target blood vessel in the first image, the optical flow tracing method is used to determine the position of the target blood vessel in the second image, thereby achieving blood vessel localization.

[0007] Optionally, image preprocessing involves using mapping localization to extract the main vascular network, and then using multi-source heterogeneous information to enhance the image.

[0008] Optionally, extracting the main vascular network includes: Channel decomposition is performed on the blood vessel image to extract the green channel image; The morphological bottom-hat transformation is used: the green channel image is closed, and then the original green channel image is subtracted to obtain the transformed image; The highest-order plane of the transformed image is extracted using the bit-plane segmentation method to obtain a binary blood vessel image; The main vascular network is obtained by determining the horizontal and vertical coordinates of the center of the main vascular network based on binary vascular images.

[0009] Optionally, image enhancement using multi-source heterogeneous information includes: Multi-scale Hessian matrix filtering was used to perform multi-scale gray-level stretching on the blood vessel image to obtain the multi-scale stretched image. Multi-directional two-dimensional matched filtering is used to perform multi-directional grayscale stretching on blood vessel images to obtain multi-directional stretched images; The enhanced image is obtained by fusing images stretched at multiple scales and in multiple directions.

[0010] Optionally, segmenting the aorta and its branch vessels includes: Initialize the centerline of the blood vessel to be segmented corresponding to the blood vessel to be segmented from the image, and use it as the starting point of the blood vessel; The probability that each neighboring point of the identified blood vessel point belongs to the centerline of the blood vessel to be segmented is calculated based on the feature values ​​of each point on the previous blood vessel centerline of the identified blood vessel point. The neighborhood point with the highest probability is added as a new identified blood vessel point to the center line of the blood vessel to be segmented. The probability of each neighborhood point of the identified blood vessel point belonging to the center line of the blood vessel to be segmented is calculated based on the feature values ​​of each point on the previous blood vessel center line of the identified blood vessel point until the endpoint of the blood vessel to be segmented from the image is added as the neighborhood point with the highest probability to the center line of the blood vessel to be segmented. The segmented blood vessels are formed based on the positions of each point on the center line of the blood vessel to be segmented, which includes the endpoint, and the corresponding blood vessel radius.

[0011] Optionally, determining the target blood vessel location includes: determining the location of the target blood vessel in the second infrared blood vessel image based on the final layer optical flow vector in the pyramid optical flow tracing method and the target solution corresponding to the final layer optical flow vector; the number of layers is determined as follows: Determine the number of layers N in the pyramid optical flow tracing method, where N is a positive integer greater than or equal to 1; Based on the target blood vessel and the second infrared blood vessel image, the first layer optical flow vector of the pyramid in the pyramid optical flow tracing method is determined, and the target solution corresponding to the first layer optical flow vector is obtained by iterative calculation of the first layer optical flow vector. The optical flow vectors of the (i+1)th layer are determined based on the optical flow vector of the i-th layer and the target solution corresponding to the optical flow vector of the i-th layer. The optical flow vector of the (i+1)th layer is then iteratively calculated to obtain the target solution corresponding to the optical flow vector of the (i+1)th layer. The final optical flow vector is the optical flow vector of the N-th layer.

[0012] As can be seen from the above technical solution, compared with the prior art, the present invention provides a vascular localization method based on multi-parameter fusion and self-learning model, which has the following beneficial effects: 1) The present invention integrates information from two modalities, namely three-dimensional medical imaging data and infrared vascular imaging data, effectively improving the contrast between blood vessels and surrounding tissues, suppressing noise interference, and significantly improving the accuracy and robustness of localization; 2) The present invention does not rely on a large amount of labeled data for offline training, and can make real-time adaptive adjustments according to different clinical scenarios and individual patient differences, thus improving versatility and flexibility. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0014] Figure 1 This is a flowchart of a blood vessel localization method based on multi-parameter fusion and self-learning model disclosed in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Reference Figure 1 As shown, this invention discloses a blood vessel localization method based on multi-parameter fusion and a self-learning model, comprising the following steps: Acquire 3D medical image data and preprocess the images; Image segmentation algorithms are used to segment the aorta and its branch vessels, and to determine the region where the target vessel is located. Infrared images of blood vessels were continuously acquired in the area where the target blood vessel was located, and named as the first image and the second image respectively. Based on the position of the target blood vessel in the first image, the optical flow tracing method is used to determine the position of the target blood vessel in the second image, thereby achieving blood vessel localization.

[0017] Furthermore, image preprocessing involves using mapping localization to extract the main vascular network, and then using multi-source heterogeneous information to enhance the image.

[0018] Furthermore, the extraction of the main vascular network includes: Channel decomposition is performed on the blood vessel image to extract the green channel image; The morphological bottom-hat transformation is used: the green channel image is closed, and then the original green channel image is subtracted to obtain the transformed image; The highest-order plane of the transformed image is extracted using the bit-plane segmentation method to obtain a binary blood vessel image; The main vascular network is obtained by determining the horizontal and vertical coordinates of the center of the main vascular network based on binary vascular images.

[0019] Specifically, in the extracted green channel image, the blood vessel area is mostly dark, while the surrounding tissue is relatively bright. However, there is still some noise and uneven grayscale, resulting in blurred blood vessel edges and unclear details. Therefore, the corresponding morphological undercap transformation includes: First, a suitable structuring element is selected to perform morphological closing operation on the green channel image. Then, a dilation operation is performed on the image to fill the tiny holes and gaps in the image. Next, an erosion operation is performed to restore the original size of the target area in the image. Finally, the image processed by the closing operation is subtracted from the original green channel image to obtain the image after the bottom cap transformation. Obtaining binary blood vessel images includes: The grayscale values ​​of the image are converted into 8-bit binary numbers, and then the highest-order bit plane is extracted. After bit plane segmentation, the image is converted into a binary image, in which the blood vessel region is presented as continuous dark pixels, while the surrounding background is bright pixels, and the morphological structure of the blood vessels is clearly represented by binary values.

[0020] Obtaining the main vascular network includes: Connectivity detection is performed on binary blood vessel images, and connected regions with an area smaller than a set threshold are removed, while the main blood vessel connected regions with larger areas and continuous morphology are retained. The centroid coordinates of each main blood vessel connected region are calculated, and the set of centroid coordinates of all main blood vessel connected regions constitutes the central coordinate system of the main blood vessel network. Finally, based on the central coordinate system and the morphological continuity of the blood vessels, the approximate direction and spatial distribution of the main blood vessel network are fitted.

[0021] Furthermore, image enhancement using multi-source heterogeneous information includes: Multi-scale Hessian matrix filtering is used to perform multi-scale gray-scale stretching on vascular images to obtain multi-scale stretched images. Multi-scale Hessian matrix filtering includes constructing Hessian matrices of different scales and performing multi-scale traversal filtering on vascular images, thereby achieving accurate detection and gray-scale stretching of blood vessels of different diameters. Multi-directional two-dimensional matched filtering is used to perform multi-directional gray-level stretching on blood vessel images to obtain multi-directional stretched images. Multi-directional two-dimensional matched filtering includes constructing filter templates in different directions and performing multi-directional traversal filtering on blood vessel images to achieve precise enhancement of blood vessels with different orientations. A weighted average fusion strategy is used to fuse images stretched at multiple scales and stretched in multiple directions to obtain an enhanced image.

[0022] Furthermore, the division of the aorta and its branch vessels includes: Initialize the centerline of the blood vessel to be segmented corresponding to the blood vessel to be segmented from the image, and use it as the starting point of the blood vessel; The probability that each neighboring point of the identified blood vessel belongs to the center line of the blood vessel to be segmented is calculated based on the feature values ​​of each point on the previous blood vessel center line of the identified blood vessel. The feature values ​​include: the gray value of the point, the average gray value and the gray value variance, the gradient values ​​of the point in the x, y and z directions, and the distance of the point to the center line of the identified blood vessel. The neighborhood point with the highest probability is added as a new identified blood vessel point to the center line of the blood vessel to be segmented. The probability P of each neighborhood point of the identified blood vessel point belonging to the center line of the blood vessel to be segmented is calculated based on the feature values ​​of each point on the previous blood vessel center line of the identified blood vessel point until the endpoint of the blood vessel to be segmented from the image is added as the neighborhood point with the highest probability to the center line of the blood vessel to be segmented. The segmented blood vessels are formed based on the positions of each point on the center line of the blood vessel to be segmented, which includes the endpoint, and the corresponding blood vessel radius.

[0023] Specifically, the standardization of blood vessels to be segmented includes: Based on clinical diagnostic and treatment needs, the target blood vessel to be segmented is determined; from the set of central coordinates of the main vascular network, the centroid coordinates (x0, y0, z0) corresponding to the starting position of the target blood vessel are selected; this coordinate point is used as the initial point of the center line of the blood vessel to be segmented, and the segmentation initialization is completed.

[0024] The expression for calculating probability P is: , Where F is the standardized feature vector, w is the weight coefficient of the feature vector, and b is the bias term.

[0025] Furthermore, forming the segmented blood vessels also includes determining the vessel radius, specifically: For each point on the blood vessel centerline, a grayscale profile line is constructed with that point as the center and along a direction perpendicular to the blood vessel centerline. The grayscale profile line is then Gaussian smoothed to suppress noise interference. An adaptive thresholding method is used to find the pixel points corresponding to the blood vessel edges on the grayscale profile line. The distances from the point on the blood vessel centerline to the two edge points are calculated, and the average value is taken as the blood vessel radius corresponding to that point.

[0026] Specifically, by combining the position of each point on the center line of the blood vessel with its corresponding radius, a three-dimensional geometric model of the target blood vessel is constructed, enabling precise segmentation of the aorta and its branch vessels, and providing a precise spatial range reference for subsequent infrared vascular image acquisition of the target blood vessel area.

[0027] Furthermore, determining the location of the target blood vessel includes: determining the location of the target blood vessel in the second infrared blood vessel image based on the final layer optical flow vector in the pyramid optical flow tracing method and the target solution corresponding to the final layer optical flow vector; determining the number of layers as follows: Determine the number of layers N in the pyramid optical flow tracing method, where N is a positive integer greater than or equal to 1; Based on the target blood vessel and the second infrared blood vessel image, the first layer optical flow vector of the pyramid in the pyramid optical flow tracing method is determined, and the target solution corresponding to the first layer optical flow vector is obtained by iterative calculation of the first layer optical flow vector. The optical flow vectors of the (i+1)th layer are determined based on the optical flow vector of the i-th layer and the target solution corresponding to the optical flow vector of the i-th layer. The optical flow vector of the (i+1)th layer is then iteratively calculated to obtain the target solution corresponding to the optical flow vector of the (i+1)th layer. The final optical flow vector is the optical flow vector of the N-th layer.

[0028] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A blood vessel localization method based on multi-parameter fusion and self-learning model, characterized in that, Includes the following steps: Acquire 3D medical image data and preprocess the images; Image segmentation algorithms are used to segment the aorta and its branch vessels, and to determine the region where the target vessel is located. Infrared images of blood vessels were continuously acquired in the area where the target blood vessel was located, and named as the first image and the second image respectively. Based on the position of the target blood vessel in the first image, the optical flow tracing method is used to determine the position of the target blood vessel in the second image, thereby achieving blood vessel localization.

2. The blood vessel localization method based on multi-parameter fusion and self-learning model according to claim 1, characterized in that, Image preprocessing involves using mapping localization to extract the main vascular network, and then using multi-source heterogeneous information to enhance the image.

3. The blood vessel localization method based on multi-parameter fusion and self-learning model according to claim 2, characterized in that, Extraction of the main vascular network includes: Channel decomposition is performed on the blood vessel image to extract the green channel image; The morphological bottom-hat transformation is used: the green channel image is closed, and then the original green channel image is subtracted to obtain the transformed image; The highest-order plane of the transformed image is extracted using the bit-plane segmentation method to obtain a binary blood vessel image; The main vascular network is obtained by determining the horizontal and vertical coordinates of the center of the main vascular network based on binary vascular images.

4. The blood vessel localization method based on multi-parameter fusion and self-learning model according to claim 2, characterized in that, Image enhancement using multi-source heterogeneous information includes: Multi-scale Hessian matrix filtering was used to perform multi-scale gray-level stretching on the blood vessel image to obtain the multi-scale stretched image. Multi-directional two-dimensional matched filtering is used to perform multi-directional grayscale stretching on blood vessel images to obtain multi-directional stretched images; The enhanced image is obtained by fusing images stretched at multiple scales and in multiple directions.

5. The blood vessel localization method based on multi-parameter fusion and self-learning model according to claim 1, characterized in that, The aorta and its branch vessels are divided into: Initialize the centerline of the blood vessel to be segmented corresponding to the blood vessel to be segmented from the image, and use it as the starting point of the blood vessel; The probability that each neighboring point of the identified blood vessel point belongs to the centerline of the blood vessel to be segmented is calculated based on the feature values ​​of each point on the previous blood vessel centerline of the identified blood vessel point. The neighborhood point with the highest probability is added as a new identified blood vessel point to the center line of the blood vessel to be segmented. The probability of each neighborhood point of the identified blood vessel point belonging to the center line of the blood vessel to be segmented is calculated based on the feature values ​​of each point on the previous blood vessel center line of the identified blood vessel point until the endpoint of the blood vessel to be segmented from the image is added as the neighborhood point with the highest probability to the center line of the blood vessel to be segmented. The segmented blood vessels are formed based on the positions of each point on the center line of the blood vessel to be segmented, which includes the endpoint, and the corresponding blood vessel radius.

6. The blood vessel localization method based on multi-parameter fusion and self-learning model according to claim 1, characterized in that, Determining the location of the target blood vessel includes: determining the location of the target blood vessel in the second infrared blood vessel image based on the final layer optical flow vector in the pyramid optical flow tracing method and the target solution corresponding to the final layer optical flow vector; the number of layers is determined as follows: Determine the number of layers N in the pyramid optical flow tracing method, where N is a positive integer greater than or equal to 1; Based on the target blood vessel and the second infrared blood vessel image, the first layer optical flow vector of the pyramid in the pyramid optical flow tracing method is determined, and the target solution corresponding to the first layer optical flow vector is obtained by iterative calculation of the first layer optical flow vector. The optical flow vectors of the (i+1)th layer are determined based on the optical flow vector of the i-th layer and the target solution corresponding to the optical flow vector of the i-th layer. The optical flow vector of the (i+1)th layer is then iteratively calculated to obtain the target solution corresponding to the optical flow vector of the (i+1)th layer. The final optical flow vector is the optical flow vector of the N-th layer.

Images