Intelligent recognition method for deep vein blood vessel position for peripheral catheterization

Abstract

This invention relates to the field of image enhancement technology, specifically to an intelligent method for identifying the location of deep veins during peripheral catheter insertion. The invention first extracts the rebound start frame and then backtracks to extract the resting image. Through inter-frame difference comparison, the initial vein contour region and vein center point of the rebound start frame are extracted. Further analysis of the displacement features of the current frame relative to the adjacent previous frame is performed to construct a tissue relative displacement field. A tissue connectivity weight map is then generated to identify highly connected regions. Based on the spatial statistical characteristics of highly connected regions within the tissue relative displacement field, combined with the tissue relative displacement field, a single-frame net displacement field is obtained. Finally, the cumulative total displacement field is updated based on the single-frame net displacement field of the current frame. The vein center point of the current frame is updated using the cumulative total displacement field, and the initial vein contour region is mapped to the current frame for display. This achieves continuous and stable tracking and visual guidance of deep vein location, effectively solving the problem of tracking loss.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology, and more specifically to a method for intelligent identification of deep vein location for peripheral catheter insertion. Background Technology

[0002] Clinical procedures involving the insertion of a peripherally inserted central venous catheter (PICC) typically rely on real-time guidance using B-mode ultrasound. During puncture localization, physicians often use probe compression to distinguish between veins and arteries.

[0003] However, during the rebound phase of the probe's pressure release, the subcutaneous tissue undergoes rapid, non-rigid deformation, accompanied by the Rouleaux effect caused by the blood flow not yet returning to laminar flow, resulting in a significant reduction in the contrast of the vessel boundaries. Existing vessel recognition algorithms based on strong edge gradients are prone to failure at this time, leading to unstable visualization and localization of deep veins in ultrasound images and affecting the physician's continuous observation of anatomical structures. Summary of the Invention

[0004] To address the technical problem of lost deep vein image tracking due to degradation of vascular anatomy during the ultrasound probe rebound phase, this invention aims to provide an intelligent method for identifying the location of deep veins during peripheral catheter insertion. The specific technical solution adopted is as follows: A historical image buffer is established. Based on the longitudinal displacement features of adjacent frames, the rebound start frame is extracted and the resting image is extracted back. After inter-frame difference comparison, the initial vein contour region and vein center point of the rebound start frame are extracted. Based on optical flow analysis of the displacement features of the current frame relative to the adjacent previous frame, a tissue relative displacement field is constructed. Based on the vein center point determined in the previous frame, the gradient cumulative cost value of all pixels in the image is calculated using the shortest path algorithm, and a tissue connectivity weight map is generated based on the gradient cumulative cost value. Highly connected regions are identified based on the tissue connectivity weight map, and the net displacement field of a single frame is obtained based on the spatial statistical characteristics of the highly connected regions within the tissue relative displacement field and the tissue relative displacement field itself. The cumulative total displacement field is updated based on the single-frame net displacement field of the current frame, the vein center point of the current frame is updated using the cumulative total displacement field, and the initial vein contour region is mapped to the current frame for display.

[0005] Furthermore, the method for obtaining the bounce start frame includes: The vertical displacement of each frame relative to the preset central region of the previous frame is obtained based on the optical flow method. For multiple consecutive frames in the historical image buffer, when the vertical displacement is continuously non-negative and the next vertical displacement is negative, the acquisition frame corresponding to the turning point of the vertical displacement is taken as the rebound start frame.

[0006] Furthermore, the method for obtaining the initial vein contour region includes: Obtain the absolute difference map between the image of the rebound start frame and the resting image, extract candidate connected regions from the absolute difference map, and select the largest candidate connected region that meets the conditions of having an eccentricity greater than a preset eccentricity threshold and a region area within a preset area range, as the initial vein contour region.

[0007] Furthermore, the method for obtaining the relative displacement field of the tissue includes: The motion vectors of pixels within a preset vertical strip region of the current frame are tracked using optical flow. Based on the fluctuation characteristics and overall statistical characteristics of the motion vectors corresponding to the preset vertical strip region, the lateral slip component of the probe is obtained. The motion vectors of each pixel in the current frame image relative to the previous frame are obtained using optical flow. The lateral slip component of the probe is fused to construct the tissue relative displacement field of the current frame.

[0008] Furthermore, the method for obtaining the gradient accumulation cost value includes: The gradient magnitudes of any two adjacent pixels are fused to obtain the gradient cost between the two corresponding pixels; the vein center point of the current frame is used as the source node, and the cumulative gradient cost from the source node to each pixel is obtained using the Dijkstra algorithm.

[0009] Furthermore, the method for obtaining the net displacement field of a single frame includes: The average vector value of the highly connected regions within the relative displacement field of the tissue is extracted as the spatial average vector. Based on the element values ​​of each position in the tissue connectivity weight map, a weighted weight is assigned to the corresponding pixel. The spatial average vector and the vector of each pixel within the relative displacement field of the tissue are weighted and fused to obtain the single-frame net displacement vector of each pixel, thus obtaining the single-frame net displacement field.

[0010] Furthermore, the method for updating the cumulative total displacement field includes: For each pixel coordinate at the start of the rebound, the intermediate mapping position of the pixel coordinate in the current frame is calculated using the cumulative total displacement field of the previous frame; the displacement vector value of the single frame net displacement field of the current frame at the intermediate mapping position is obtained using the bilinear interpolation algorithm; the cumulative total displacement field of the previous frame and the displacement vector value are vector superimposed to obtain the cumulative total displacement field of the current frame.

[0011] Furthermore, the method for obtaining the organizational connectivity weight graph includes: The attenuation radius parameter is obtained based on the equivalent radius of the initial vein contour region; based on the attenuation radius parameter, the gradient accumulation value of each pixel is mapped using a Gaussian attenuation function to construct a tissue connectivity weight map.

[0012] Furthermore, the method for obtaining the highly connected region includes: In the organization connectivity weight graph, a set of pixels with element values ​​greater than a preset weight threshold is selected to construct a highly connected region.

[0013] Furthermore, the method for acquiring the resting image includes: Before the rebound start frame, the grayscale difference factor of each frame image is obtained based on the grayscale difference between each frame image and the adjacent previous frame image; when the grayscale difference factor is less than the preset static difference threshold, the corresponding frame image is marked as a static image.

[0014] The present invention has the following beneficial effects: This invention first extracts the rebound initiation frame and backtracks to extract the resting image. After inter-frame difference comparison, the initial vein contour region and vein center point of the rebound initiation frame are extracted, providing a reliable anatomical benchmark for subsequent position tracking in low-contrast environments. Further analysis of the displacement features of the current frame relative to the adjacent previous frame is performed to construct a tissue relative displacement field, providing a foundation for subsequent tracking. A tissue connectivity weight map is then generated to obtain a weight field reflecting the continuity of the anatomical structure, identifying highly connected regions. Blood vessels and their tightly wrapped tissues are separated as a single moving unit from the complex background. Based on the spatial statistical characteristics of highly connected regions within the tissue relative displacement field, combined with the tissue relative displacement field, a single-frame net displacement field is obtained, effectively offsetting random high-frequency disturbances caused by speckle noise, providing a reliable input for final accurate integration and mapping. Finally, the cumulative total displacement field is updated based on the single-frame net displacement field of the current frame, and the vein center point of the current frame is updated using the cumulative total displacement field. The initial vein contour region is then mapped to the current frame for display, ensuring continuous and stable tracking. This invention overcomes the blurring of vascular boundaries and tissue deformation during the ultrasound probe rebound phase by using spatiotemporal reference anchoring, connectivity weight analysis and optical flow weighted fusion, to achieve continuous and stable tracking and visual guidance of deep vein location, effectively solving the problem of tracking loss. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a method for intelligent identification of deep vein location during peripheral catheter insertion, provided in one embodiment of the present invention; Figure 2 This is a flowchart of a method for obtaining gradient cumulative cost values ​​according to an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for intelligent identification of deep vein location for peripheral catheter insertion based on the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following describes in detail, with reference to the accompanying drawings, a specific scheme of the intelligent identification method for deep vein location for peripheral catheter insertion provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a method for intelligent identification of deep vein location for peripheral catheter insertion according to an embodiment of the present invention, specifically including: Step S1: Establish a historical image buffer. Based on the longitudinal displacement features of adjacent frame images, extract the rebound start frame and backtrack to extract the resting image. After inter-frame difference comparison, extract the initial vein contour region and vein center point of the rebound start frame.

[0021] During ultrasound-guided puncture, the moment the probe is pressed and released is the critical point for establishing the vascular tracking baseline. The tissue movement direction reversal and the morphological characteristics of the vein being compressed and closed caused by this operation provide the algorithm with a unique spatiotemporal origin. Therefore, a historical image buffer is established, and the rebound start frame is extracted based on the longitudinal displacement characteristics of adjacent frames to determine the starting time and baseline state of vascular tracking. At the same time, the resting image is extracted back to obtain the original anatomical background that is not affected by probe pressure. Then, after inter-frame differential comparison, the initial vein contour region and vein center point of the rebound start frame are extracted. Taking advantage of the physical characteristics of deep veins being easily deformed by pressure while the surrounding tissues are relatively static, the initial vascular morphology with a high signal-to-noise ratio is accurately captured before the echo in the blood vessel increases and the boundary becomes blurred, providing a reliable anatomical baseline for subsequent position tracking in low-contrast environments.

[0022] Preferably, in one embodiment of the present invention, the system sets up a first-in-first-out (FIFO) historical image buffer in memory, with the latest frame corresponding to the last frame of the historical image buffer; the length of the buffer is set to... The time interval is 2 seconds in this example. When there are fewer than 10 frames in the historical image buffer, the system does not perform bounce analysis, but only image accumulation.

[0023] To avoid global calculation errors caused by probe edge lifting or tilting, the system extracts the central region of the image. For consecutive image frames, the system uses optical flow to obtain the average motion vector (the average value of the motion vectors of pixels within the region) of each frame relative to the preset central region of the previous frame. Then, it extracts the vertical component to obtain the longitudinal displacement in the vertical direction. ; For multiple consecutive frames of images within the historical image buffer, if the previous frames are... All values ​​are non-negative, indicating that the probe is in the downward pressing phase, and the subcutaneous tissue is compressed. In this image frame, The first time the value turns negative, it indicates that the probe pressure has been released and the subcutaneous tissue has begun to rebound upwards. Therefore, when the longitudinal displacement is continuously non-negative and the next longitudinal displacement is negative, the acquisition frame corresponding to the turning point of the longitudinal displacement to negative is taken as the rebound start frame.

[0024] The vertical direction is defined as the direction that increases in depth along the ultrasound scan line (i.e., the Y-axis direction of the image coordinate system). The positive Y-axis direction is defined as the direction of deep tissue (away from the probe), and the negative direction as the direction of superficial skin (closer to the probe). The preset central region is set to have a width W occupying 50% of the entire image and a height H occupying 80% of the entire image, located at the center of the entire image; before the rebound start frame. The duration is at least 2.

[0025] In other embodiments of the present invention, the implementer may also use conventional block matching algorithms to obtain the average motion vector; using optical flow or block matching algorithms to calculate the average motion vector of an image region is a conventional motion estimation method in the art.

[0026] Before the rebound start frame (temporal domain), the grayscale difference factor of each frame image is obtained based on the grayscale difference between each frame image and the adjacent previous frame image; when the grayscale difference factor is less than the preset static difference threshold, the corresponding frame image is marked as a resting image.

[0027] As an example, the sum of absolute differences (SAD) between each frame and the adjacent previous frame is calculated as the grayscale difference factor for each frame. A preset static difference threshold is set to... When the grayscale difference factor is less than the preset static difference threshold, it indicates that the subcutaneous tissue has not undergone significant deformation and the probe is in a resting state before stable pressure. At this time, the corresponding frame image is marked as a resting image.

[0028] The summation range covers a preset central region of the image to exclude probe edge slippage or noise interference. A preset static difference threshold corresponds to an average grayscale change of no more than 5 per pixel; if no grayscale difference factor meets the preset static difference threshold, the image corresponding to the smallest grayscale difference factor is marked as the resting image.

[0029] The system obtains the absolute difference map between the image of the rebound start frame and the resting image, which is the absolute value map of the difference between the gray values ​​of corresponding pixels in the two frames. In this map, the high gray value area represents the part of the tissue that has undergone significant compression deformation (such as a flattened vein), and the low gray value area represents the part of the tissue that is relatively still or not compressed (such as arteries or surrounding muscles). The system performs binarization on the absolute difference map and performs morphological opening operation to remove noise, and obtains the candidate connected components in the absolute difference map. The largest candidate connected region that meets the conditions of having an eccentricity greater than a preset eccentricity threshold and a region area within a preset area range is selected as the initial vein contour region. Then, the geometric centroid of the initial vein contour region (such as the average coordinates of the pixels within the region) is calculated as the vein center point.

[0030] In this example, the preset eccentricity threshold is 0.8, the preset area range is 10~200 square millimeters, the number of pixels in the candidate connected region is used as the pixel area, and the physical area is converted into pixel area using the horizontal and vertical resolution of the ultrasound device (a known technique).

[0031] It should be noted that the embodiments of the present invention are applicable to conventional B-ultrasound equipment. The system is set to acquire image streams from the ultrasound equipment at a frequency of 20-30 fps; the resolution and size of the images are determined by the equipment parameters in the specific implementation scenario. Here, it is only limited that the acquired images are conventional rectangular shapes, and the analyzed images are grayscale images (non-grayscale images are preprocessed using conventional grayscale conversion methods). The calculation of eccentricity and geometric centroid, as well as optical flow methods, difference maps, morphological operations, etc., are all well-known technologies in the field of digital image processing. In other embodiments of the present invention, the implementer can adjust the preset eccentricity threshold, preset area range, and preset static difference threshold according to the needs of the actual application scenario.

[0032] Step S2: Analyze the displacement features of the current frame relative to the adjacent previous frame using optical flow method to construct a tissue relative displacement field; based on the vein center point determined in the previous frame, calculate the gradient cumulative cost of all pixels in the image using the shortest path algorithm, and generate a tissue connectivity weight map based on the gradient cumulative cost; identify highly connected regions based on the tissue connectivity weight map, and obtain the net displacement field of a single frame based on the spatial statistical characteristics of highly connected regions within the tissue relative displacement field and the tissue relative displacement field.

[0033] During the probe's rebound, the entire image area contains two main types of motion: first, the elastic rebound motion of subcutaneous tissue (including target blood vessels) due to pressure release, which is a valid signal that needs attention; second, the unintentional global rigid slippage or tilt that may occur when the doctor holds the probe, which is a type of interference noise. If the raw optical flow is used directly, the displacement of the blood vessels will be mixed with the displacement of the probe, resulting in severe distortion of tracking and positioning.

[0034] Therefore, motion tracking decoupling is required. Based on optical flow analysis, the displacement characteristics of the current frame relative to the adjacent previous frame are analyzed to construct the tissue relative displacement field. By calculating and subtracting the rigid motion component of the probe, the displacement field generated purely by tissue deformation can be separated, providing a basis for subsequent tracking.

[0035] It should be noted that, except for the backtracking extraction of the resting image which is analyzed before the rebound start frame, all other analysis processes begin after the rebound start frame. For example, if the index of the rebound start frame is k=0, then the analysis starts from the next frame k=1 in the positive temporal direction.

[0036] Preferably, in one embodiment of the present invention, considering that when a doctor releases pressure by holding the probe, there may be unintentional lateral slippage, and this global rigid motion will contaminate the local displacement signal reflecting tissue deformation, by analyzing the consistency (fluctuation characteristics) of the motion vectors of feature points within the preset vertical strip at the edge of the image, it can be determined whether the motion source is overall probe movement or local tissue deformation. If the motion is consistent, this average vector is extracted as the slip component.

[0037] Based on this, the motion vectors of pixels within a preset vertical strip region of the current frame are tracked using optical flow. The transverse slip component of the probe is obtained based on the fluctuation characteristics and overall statistical characteristics of the motion vectors corresponding to the preset vertical strip region. The motion vectors of each pixel in the current frame relative to the previous frame are obtained using optical flow. The transverse slip component of the probe is fused to construct the tissue relative displacement field of the current frame.

[0038] As an example, the width of the preset vertical strip region is fixed, but its effective range is not simply defined from the physical edge of the image. The system starts from the left and right edges of the image and performs validity checks column by column towards the image center. For each candidate column, the system calculates the maximum grayscale value of all pixels within that column. If this maximum grayscale value is less than a preset invalid region threshold (for example, a threshold of 5 in the grayscale range of 0-255), then the column is determined to be outside the effective acoustic field of view of the image, belonging to a completely black area with no signal or a noise area, and is skipped.

[0039] The system continuously moves towards the image center until it finds a column where the maximum grayscale value exceeds the invalid region threshold. This column is then used as the starting boundary of a preset vertical strip region, which is then extended towards the image center by a preset fixed width. This determines the final valid vertical strip region used for optical flow tracing. This mechanism ensures that the reference region used for slip calculations necessarily contains image content with actual anatomical texture, avoiding slip vector calculation errors caused by tracing textureless pure black areas.

[0040] In this example, there is one preset vertical strip region on each side of the current frame image, with a fixed width of 5% of the image width (rounded down if not an integer). The two preset vertical strip regions together constitute the background reference region, and the motion vector of the points in the background reference region is tracked using the Lucas-Kanade sparse optical flow algorithm.

[0041] Considering the significant non-uniform deformation (such as muscle contraction or large-scale tissue rebound) within the background reference area, the background reference is unreliable. Therefore, we first analyze the fluctuation characteristics of the motion vectors corresponding to the preset vertical strip regions. Specifically, we extract the variance of the motion vectors corresponding to the pixels within the background reference area. ,like If the value is greater than or equal to a preset rigid threshold (e.g., 1 pixel squared), the probe will be forced to slide laterally. If the vector is zero, skip the slip compensation step to avoid incorrectly deducting valid tissue motion.

[0042] like If the value is less than the preset rigidity threshold, it indicates that all points within the background reference area have the same direction, magnitude, and height of motion, which conforms to the characteristics of rigid body translation of the probe. At this point, the average value of the motion vectors of all points within the background reference area is calculated, and its horizontal component (i.e., displacement along the X-axis of the image) is extracted as the lateral slip component of the probe. The mean value represents the overall statistical characteristics of the data.

[0043] Then, the Farneback dense optical flow algorithm is used to obtain the motion vector of each pixel in the current frame relative to the previous frame. ,Will minus The difference is used as the relative displacement vector of each pixel in the current frame. Rigid background motion is eliminated, and all relative displacement vectors are collected to construct the organization relative displacement field of the current frame.

[0044] It should be noted that, considering that if the two preset vertical strip regions obtained in the end overlap or cannot be found, extreme situations may occur (such as the probe not contacting the skin, resulting in extremely weak image signal). In such cases, the current frame is determined to be an invalid frame, and the probe lateral sliding component of the current frame is forcibly set to zero vector. At the same time, a warning log is generated to prompt the operator to check the image quality, or the current frame is directly discarded.

[0045] The Farneback dense optical flow algorithm and the Lucas-Kanade sparse optical flow algorithm are well-known techniques for tracking adjacent frames and extracting motion vectors, and will not be described in detail here. In other embodiments of the present invention, the implementer may set other preset rigid thresholds and preset vertical strip regions.

[0046] At the moment of rebound, the echo inside the blood vessel is enhanced due to the red blood cell stacking effect, significantly reducing the acoustic difference (i.e., image gradient) between the vessel wall and surrounding tissue, causing traditional edge detectors to fail. However, the blood vessel and its tightly wrapped connective tissue sheath are anatomically a continuous entity. The shortest path algorithm treats the image as a topological topographic map. Based on the vein center point determined in the previous frame, it calculates the cumulative gradient cost of all pixels in the image using the shortest path algorithm, and generates a tissue connectivity weight map based on the cumulative gradient cost. This constructs a weight field reflecting the continuity of the anatomical structure, providing a foundation for subsequent motion information purification.

[0047] Although the original "tissue relative displacement field" is clean, it is still affected by ultrasound speckle noise. Even in a uniform vascular region, there will be chaotic motion vectors. Direct analysis will introduce huge errors. Physically, the deep vein and its sheath can be approximated as a local rigid body in short-term rebound. The internal motion should be highly consistent, which is a highly connected region. Therefore, identifying highly connected regions based on the tissue connectivity weight map can separate the blood vessel and its tightly wrapped tissue as a whole motion unit from the complex background, laying the foundation for extracting consistent motion trends. Then, the motion vectors within the highly connected regions should fluctuate around a common true value. Based on the spatial statistical characteristics of the highly connected regions within the tissue relative displacement field, combined with the tissue relative displacement field, the net displacement field of a single frame can be obtained, which can effectively cancel the random high-frequency disturbances caused by speckle noise, providing a reliable input for the final accurate integration and mapping.

[0048] Preferably, in one embodiment of the present invention, please refer to Figure 2 The flowchart illustrates a method for obtaining gradient accumulation cost values ​​according to an embodiment of the present invention, specifically including: Step S201: Fuse the gradient magnitudes of any two adjacent pixels to obtain the gradient cost between the two corresponding pixels.

[0049] Calculate the gradient magnitude map of the current frame image. Construct a grid diagram that corresponds one-to-one with the image pixels; Because the gradient at the blood vessel boundary is weak or even discontinuous during the rebound process, the gradient value of a single pixel cannot reliably define the blood vessel region. However, the blood vessel and its surrounding tissue are anatomically connected, which means that the gradient magnitude of adjacent pixels changes gently within the tissue and increases significantly when crossing the anatomical boundary. Therefore, the gradient magnitudes of any two adjacent pixels are fused to obtain the gradient cost between the corresponding two pixels. As an example, the sum of the gradient magnitudes of two pixels is multiplied by a preset gradient penalty coefficient. The product of the two, plus the preset benchmark value. The sum of these values ​​is used as the gradient cost between the corresponding two pixels. In this example, , dimensionless, is designed to amplify the blocking effect of the boundary gradient to ensure that the connected region can accurately fit the anatomical boundary of the blood vessel. In this example, we take 2; The unit is grayscale / pixel, which aims to reduce the contribution of geometric distance to connectivity calculation, so that the algorithm can tolerate longer vascular connectivity regions. In this example, it is set to 0.05.

[0050] It should be noted that the analysis process is consistent for each pair of adjacent pixels, and the analysis process is consistent for each frame of the image after the bounce start frame. Only one example is described here.

[0051] Step S202: Using the vein center point of the current frame as the source node, the cumulative gradient cost from the source node to each pixel is obtained using the Dijkstra algorithm.

[0052] Here, the gradient cost between any two adjacent pixels represents the local texture resistance that needs to be overcome to move from one pixel to its neighboring pixels during the path search process. Dijkstra's algorithm calculates the shortest path from the source node to each pixel using a greedy strategy. The "shortest" means that the sum of the gradient costs between all adjacent pixel pairs on the path is minimized. The total cost of this path is the cumulative gradient cost of the corresponding pixel.

[0053] It should be noted that the embodiments of the present invention use an 8-neighborhood connectivity method. Dijkstra's algorithm is a well-known method in graph theory for solving the single-source shortest path problem. In other embodiments of the present invention, implementers may use other shortest path algorithms, and need to adjust the neighborhood connectivity method (such as 4-neighborhood / 8-neighborhood) and the above two parameters accordingly. and The range of values ​​for is determined to suit the numerical stability requirements of different algorithms, and will not be elaborated further.

[0054] Further construct the organizational connectivity weight graph: In practice, due to differences in resolution, imaging depth, and anatomical variations at the puncture site of different ultrasound devices, the scale of the acquired images and the apparent size of the blood vessels exhibit significant individuality. This variability makes it impossible to use a fixed absolute value to uniformly define the tissue extent consistent with vascular movement. Therefore, the attenuation radius parameter is obtained based on the equivalent radius of the initial vein contour region. As an example, the equivalent radius of the initial vein contour region (i.e., the radius of a circle with the same area) is compared with a preset empirical coefficient. The product of these two factors, used as the attenuation radius parameter, is discussed in this example. The attenuation radius parameter setting ensures that the range of subsequently generated connectivity weights can cover the deep vein lumen and its surrounding tightly connected connective tissue sheath, allowing the algorithm to tolerate certain changes in vascular morphology.

[0055] Then, based on the decay radius parameter, the gradient accumulation cost of each pixel is mapped using the Gaussian decay function to construct an organizational connectivity weight graph.

[0056] As an example, the Gaussian decay function is Where z is the independent variable, corresponding to the cumulative gradient cost of each pixel in the current frame. The corresponding decay radius parameter; after applying a Gaussian decay function, the gradient accumulation cost is mapped to a range of 0 to 1, yielding the organization connectivity weights for each pixel. Near the center of a blood vessel and in the closely surrounding connected tissue region, the smaller the gradient accumulation cost, the closer the weight is to 1. As the path crosses anatomical boundaries, the gradient accumulation cost exceeds the denominator. .

[0057] Each pixel By assigning corresponding values ​​according to spatial distribution locations, an organizational connectivity weight graph is constructed.

[0058] Next, in the tissue connectivity weight map, a set of pixels with element values ​​greater than a preset weight threshold is selected to construct highly connected regions. These highly connected regions represent the most coherent and reliable parts of vascular motion. Separating these regions from the surrounding noisy background regions lays the foundation for accurately calculating the average vector representing the overall vascular motion.

[0059] As an example, with a preset weight threshold of 0.8, a set of pixels with element values ​​greater than 0.8 is selected. To construct a continuous, connected region from these potentially discrete or isolated pixels, the system performs a morphological closing operation (dilation followed by erosion) to fill small holes within the region and connect neighboring pixels, thereby forming one or more connected regions. Subsequently, the system selects the connected region with the largest area and defines it as the highly connected region of the current frame.

[0060] It should be noted that when the total number of pixels in a set whose element value is greater than the preset weight threshold is less than the preset effective number threshold (e.g., 10 pixels), it indicates that the vascular features in the current frame are extremely weak or the weight map calculation has failed, making it impossible to extract effective highly connected regions. In this case, the system will activate a degradation processing strategy: abandon the motion correction based on connectivity for this frame and instead directly use the unweighted original tissue relative displacement field for subsequent calculations, or reuse the spatial average vector successfully calculated in the previous frame (which can be multiplied by an attenuation coefficient such as 0.8) to ensure the continuous and stable operation of the algorithm under extreme conditions.

[0061] It should be noted that, in other embodiments of the present invention, the implementer may adjust the control according to specific application scenarios. The Gaussian decay function can be replaced with other functions that have similar monotonically decreasing and smooth characteristics (such as Cauchy function, exponential decay function, etc.) by setting the preset weight threshold.

[0062] Finally, the net displacement field of a single frame is obtained: First, the average vector value of the highly connected regions within the tissue's relative displacement field is extracted as the spatial average vector. This yields the spatial average vector representing the overall movement trend of blood vessels. To ensure a smooth transition of the motion field in the image space and avoid tearing artifacts at the blood vessel boundaries, the system uses the element values ​​of each position in the tissue connectivity weight map to assign weights to the corresponding pixels, and then weights and fuses the spatial average vector and the vector of each pixel in the tissue relative displacement field to obtain the single-frame net displacement vector of each pixel, thus obtaining the single-frame net displacement field.

[0063] As an example, the net displacement vector of a single frame The calculation formulas include: ; Let x be the organizational connectivity weight of the x-th pixel; This represents the spatial average vector corresponding to a highly connected region. Let be the relative displacement vector in the relative displacement field of the x-th pixel.

[0064] Among them, each pixel Values ​​are assigned according to spatial distribution locations to construct a single-frame net displacement field; within blood vessels and connected tissues... When the value is close to 1, the system forces the displacement of pixels in that region to equal the spatial average vector, effectively suppressing speckle noise; in the background region, Approaching zero preserves natural deformation; at the boundaries, a smooth transition is achieved. The final obtained single-frame net displacement field maintains the integrity of the blood vessel morphology while eliminating internal noise, ensuring the stability and accuracy of the displacement integral results.

[0065] Step S3: Update the cumulative total displacement field based on the single-frame net displacement field of the current frame, update the vein center point of the current frame using the cumulative total displacement field, and map the initial vein contour region to the current frame for display.

[0066] To accurately map the clear vascular contours at the reference time onto the blurred image at the current time, the system must maintain a cumulative total displacement field describing the total displacement from the rebound initiation frame to the current frame. Therefore, the cumulative total displacement field is updated based on the single-frame net displacement field of the current frame. To ensure that the shortest path algorithm can continue to spread from the correct intravascular origin in the next frame, the gradient accumulation cost is obtained, and the vein center point of the current frame is updated using the cumulative total displacement field. Finally, the initial vein contour region is mapped onto the current frame for display. By stably projecting vascular projections based on clear reference contours into low-contrast images with tissue rebound and blurred vascular boundaries, continuous and stable tracking is achieved during the temporary degradation of vascular anatomical features, providing reliable reference information for relevant personnel.

[0067] Preferably, in one embodiment of the present invention, since the analysis starts from the rebound start frame (k=0) and the initial vein contour region and vein center point are extracted, the initial cumulative total displacement field is all zero, the initialization of the cumulative total displacement field is completed, and the comparison analysis is performed with the previous frame starting from frame k=1. Because the tissue is constantly in motion, simple fixed-point accumulation will lead to errors. Therefore, the system uses a semi-Lagrangian integration strategy for updates, specifically: For each pixel coordinate at the start of the rebound, the intermediate mapping position of that pixel coordinate in the current frame is calculated using the cumulative total displacement field of the previous frame: ; In the formula, This represents the coordinates of the x-th pixel in the initial frame of the bounce. This represents the cumulative total displacement vector of the x-th pixel (at position) in the cumulative total displacement field corresponding to the (k-1)-th frame. This represents the current position of the x-th pixel in the (k-1)-th frame image coordinate system, corresponding to the backward tracking starting point in the semi-Lagrange integral. It is used to obtain the instantaneous displacement increment of this point through bilinear interpolation in the single-frame net displacement field of the current frame. It is an "intermediate" state variable in the semi-Lagrange integral step, and therefore corresponds to the intermediate mapping position of the current frame.

[0068] Then perform boundary checks. Check if it is within the image's physical coordinate range. If it is outside the range, set the incremental displacement at that point to (0,0) to prevent index out-of-bounds errors. If the image is located within its physical coordinate range, then the net displacement field of a single frame is read using a bilinear interpolation algorithm. Displacement vector value at .

[0069] Finally, the cumulative total displacement field of the previous frame is vector-superimposed with the displacement vector value to obtain the cumulative total displacement field of the current frame. ,Right now: .

[0070] Updated It fully records the total displacement path of each pixel from the rebound start frame to the current moment.

[0071] To further update the vein center point, the system reads the coordinates of the vein center point from the rebound start frame. ,exist Query the displacement vector at this point The coordinates of the center point of the vein in the current frame. for: .

[0072] To prevent the center point from drifting outside the blood vessel due to abnormal optical flow calculations (which would trigger a logic avalanche in the next frame), the system performs out-of-bounds and validity checks here: [determining...] Is it located within the image boundary and not deviated from the reference position? If the distance is too far (e.g., the displacement does not exceed 20mm), or if it exceeds the image boundary or deviates too far from the reference position, the verification is deemed to have failed.

[0073] If the verification fails, At that time, linear extrapolation prediction is enabled. ;when Then let .

[0074] The system obtains the edge coordinate set of the initial vein contour region, uses the cumulative total displacement field of the current frame to perform coordinate transformation on each point, obtains the vein projection contour of the current frame, and displays it on the screen with a highlighted color.

[0075] It should be noted that the dynamic tracking process described in steps S2 to S3 of the present invention is strictly limited to the dynamic stage of tissue rebound triggered by the release of probe pressure.

[0076] Specifically, after the system detects the rebound start time (k=0) in step S1, it starts a preset rebound monitoring window (e.g., 2 seconds). During this window, the system continues to execute the closed-loop tracking process from S2 to S3.

[0077] The system will automatically terminate the update and integration of the displacement field once any of the following conditions are met: 1. Timeout condition: The cumulative time exceeds the preset maximum window period.

[0078] 2. Stability condition: The system calculates the movement speed of the vein center point in two adjacent frames in real time. When the speed is lower than the preset stability threshold (e.g., 0.5 mm / s) for multiple consecutive frames (e.g., 5 frames), it indicates that the active rebound movement of the subcutaneous tissue has basically ended and the position tends to stabilize.

[0079] In summary, to address the technical problem of deep vein image tracking loss due to vascular anatomy degradation during the ultrasound probe rebound phase, this invention provides an intelligent deep vein location identification method for peripheral catheter insertion. This invention first extracts the rebound start frame and retrospectively extracts the resting image. After inter-frame difference comparison, the initial vein contour region and vein center point of the rebound start frame are extracted. Further analysis of the displacement features of the current frame relative to the adjacent previous frame is performed to construct a tissue relative displacement field. A tissue connectivity weight map is then generated to identify highly connected regions. Based on the spatial statistical characteristics of highly connected regions within the tissue relative displacement field, combined with the tissue relative displacement field, a single-frame net displacement field is obtained. Finally, the cumulative total displacement field is updated based on the single-frame net displacement field of the current frame, and the vein center point of the current frame is updated using the cumulative total displacement field. The initial vein contour region is then mapped to the current frame for display. This invention, through spatiotemporal reference anchoring, connectivity weight analysis, and optical flow weighted fusion, overcomes vascular boundary blurring and tissue deformation during the ultrasound probe rebound phase, achieving continuous and stable tracking and visual guidance of deep vein location, effectively solving the tracking loss problem.

[0080] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0081] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for intelligent identification of deep vein location for peripheral catheter insertion, characterized in that, The method includes: A historical image buffer is established. Based on the longitudinal displacement features of adjacent frames, the rebound start frame is extracted and the resting image is extracted back. After inter-frame difference comparison, the initial vein contour region and vein center point of the rebound start frame are extracted. Based on optical flow analysis of the displacement features of the current frame relative to the adjacent previous frame, a tissue relative displacement field is constructed. Based on the vein center point determined in the previous frame, the gradient cumulative cost value of all pixels in the image is calculated using the shortest path algorithm, and a tissue connectivity weight map is generated based on the gradient cumulative cost value. Highly connected regions are identified based on the tissue connectivity weight map, and the net displacement field of a single frame is obtained based on the spatial statistical characteristics of the highly connected regions within the tissue relative displacement field and the tissue relative displacement field itself. The cumulative total displacement field is updated based on the single-frame net displacement field of the current frame, the vein center point of the current frame is updated using the cumulative total displacement field, and the initial vein contour region is mapped to the current frame for display.

2. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the rebound start frame includes: The vertical displacement of each frame relative to the preset central region of the previous frame is obtained based on the optical flow method. For multiple consecutive frames in the historical image buffer, when the vertical displacement is continuously non-negative and the next vertical displacement is negative, the acquisition frame corresponding to the turning point of the vertical displacement is taken as the rebound start frame.

3. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the initial vein contour region includes: Obtain the absolute difference map between the image of the rebound start frame and the resting image, extract candidate connected regions from the absolute difference map, and select the largest candidate connected region that meets the conditions of having an eccentricity greater than a preset eccentricity threshold and a region area within a preset area range, as the initial vein contour region.

4. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the relative displacement field of the tissue includes: The motion vectors of pixels within a preset vertical strip region of the current frame are tracked using optical flow. Based on the fluctuation characteristics and overall statistical characteristics of the motion vectors corresponding to the preset vertical strip region, the lateral slip component of the probe is obtained. The motion vectors of each pixel in the current frame image relative to the previous frame are obtained using optical flow. The lateral slip component of the probe is fused to construct the tissue relative displacement field of the current frame.

5. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the gradient accumulation cost value includes: The gradient magnitudes of any two adjacent pixels are fused to obtain the gradient cost between the two corresponding pixels; the vein center point of the current frame is used as the source node, and the cumulative gradient cost from the source node to each pixel is obtained using the Dijkstra algorithm.

6. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the net displacement field of a single frame includes: The average vector value of the highly connected regions within the relative displacement field of the tissue is extracted as the spatial average vector. Based on the element values ​​of each position in the tissue connectivity weight map, a weighted weight is assigned to the corresponding pixel. The spatial average vector and the vector of each pixel within the relative displacement field of the tissue are weighted and fused to obtain the single-frame net displacement vector of each pixel, thus obtaining the single-frame net displacement field.

7. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for updating the cumulative total displacement field includes: For each pixel coordinate at the start of the rebound, the intermediate mapping position of the pixel coordinate in the current frame is calculated using the cumulative total displacement field of the previous frame; the displacement vector value of the single frame net displacement field of the current frame at the intermediate mapping position is obtained using the bilinear interpolation algorithm; the cumulative total displacement field of the previous frame and the displacement vector value are vector superimposed to obtain the cumulative total displacement field of the current frame.

8. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the organizational connectivity weight graph includes: The attenuation radius parameter is obtained based on the equivalent radius of the initial vein contour region; based on the attenuation radius parameter, the gradient accumulation value of each pixel is mapped using a Gaussian attenuation function to construct a tissue connectivity weight map.

9. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the highly connected region includes: In the organization connectivity weight graph, a set of pixels with element values ​​greater than a preset weight threshold is selected to construct a highly connected region.

10. The intelligent identification method for deep vein location during peripheral catheter insertion according to claim 1, characterized in that, The method for obtaining the resting image includes: Before the rebound start frame, the grayscale difference factor of each frame image is obtained based on the grayscale difference between each frame image and the adjacent previous frame image; when the grayscale difference factor is less than the preset static difference threshold, the corresponding frame image is marked as a static image.

Images