Microbubble dynamic tracking method and device for diving medicine

Abstract

The application discloses a micro-bubble dynamic tracking method and device for diving medicine, relates to the field of dynamic detection technology, and introduces a variable frame frequency detection mode, controls an ultrasonic imaging device, executes ultrasonic detection based on a first continuous frame, determines a micro-bubble image, classifies a background domain and a bubble domain according to a classifier, executes background domain blank setting and micro-bubble group coding, calibrates each micro-bubble by introducing a center window and a backup window, executes inter-frame dynamic graph network analysis and trend prediction in combination with a generator, generates a step frame, locates a detection node, controls the ultrasonic imaging device to execute ultrasonic detection based on a second continuous frame, executes periodic detection and dynamic analysis under the variable frame frequency detection mode, and integrates a tracking chain, so as to solve the technical problem of insufficient comprehensive, dynamic and automatic monitoring and distinguishing capability of micro-bubbles in the prior art. The application realizes automatic, comprehensive and accurate and effective tracking management in a dynamic operation scene.

Description

Technical Field

[0001] This invention relates to the field of dynamic detection technology, and more specifically to a microbubble dynamic tracking method and device for use in diving medicine. Background Technology

[0002] In diving medicine research and applications, microbubbles are easily generated in the human body due to changes in environmental pressure during diving. If these microbubbles are not identified and treated in time, they may cause serious physiological reactions such as decompression sickness. Traditional monitoring methods mostly rely on static ultrasound imaging technology, using doctors to visually observe or perform simple image processing to identify the location and distribution of microbubbles.

[0003] There are two limitations: First, the resolution and dynamic tracking capabilities are insufficient. The movement of microbubbles in blood flow or tissues is highly time-varying, and traditional devices cannot continuously track their trajectory and evolution process, making it difficult to support accurate diagnosis and early warning in complex physiological states. Second, existing image analysis methods lack deep structure modeling capabilities and can only perform boundary recognition or region segmentation based on static frames, failing to fully explore the evolutionary paths of microbubble movement trends and interactive behaviors.

[0004] In summary, there is an urgent need for a structured tracking mechanism for microbubble monitoring to address the shortcomings of existing technologies in providing comprehensive, dynamic, and automated monitoring and discrimination capabilities for microbubbles, particularly in the analysis of multi-scale, irregular motion characteristics of microbubbles. Summary of the Invention

[0005] This application provides a method and apparatus for dynamic tracking of microbubbles in diving medicine, which addresses the technical problem of insufficient comprehensive, dynamic, and automated monitoring and discrimination capabilities for microbubbles in existing technologies.

[0006] In view of the above problems, this application provides a microbubble dynamic tracking method and device for diving medicine.

[0007] In a first aspect, this application provides a microbubble dynamic tracking method for diving medicine. The method includes: introducing a variable frame rate detection mode, controlling an ultrasound imaging device, performing ultrasound detection based on a first consecutive frame, and determining microbubble images; centrally embedding the ultrasound imaging device through a cascaded classifier and generator, classifying the microbubble images into background and bubble domains according to the classifier, performing background domain blanking and microbubble group encoding, calibrating each microbubble by introducing a central window and alternative windows, and performing inter-frame dynamic graph network analysis and trend prediction in conjunction with the generator to generate step frames, wherein the step frames are identified by a sliding time window; locating detection nodes according to the step frames, controlling the ultrasound imaging device, performing ultrasound detection based on a second consecutive frame, performing periodic detection and dynamic analysis based on the variable frame rate detection mode, and integrating and determining a tracking chain, wherein the tracking chain includes a trajectory dimension and a phase appearance dimension.

[0008] Secondly, this application provides a microbubble dynamic tracking device for diving medicine. The device includes: a first detection module, used to introduce a variable frame rate detection mode, control an ultrasound imaging device, perform ultrasound detection based on a first continuous frame, and determine microbubble images; a step analysis module, used to centrally embed the ultrasound imaging device through a cascaded classifier and generator, classify the microbubble images into background and bubble domains according to the classifier, perform background domain blanking and microbubble group encoding, calibrate each microbubble by introducing a central window and alternative windows, and, in conjunction with the generator, perform inter-frame dynamic graph network analysis and trend prediction to generate step frames, wherein the step frames are identified by a sliding time window; and a second detection module, used to locate detection nodes according to the step frames, control the ultrasound imaging device, perform ultrasound detection based on a second continuous frame, perform periodic detection and dynamic analysis based on the variable frame rate detection mode, and integrate and determine a tracking chain, wherein the tracking chain includes a trajectory dimension and a phase appearance dimension.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] The microbubble dynamic tracking method for diving medicine provided in this application introduces a variable frame rate detection mode, controls an ultrasound imaging device, performs ultrasound detection based on a first consecutive frame, and determines microbubble images. Through a cascaded classifier and generator, the ultrasound imaging device is centrally embedded and deployed. According to the classifier, the microbubble images are classified into background and bubble domains, background domain blanking and microbubble group encoding are performed, and each microbubble is calibrated by introducing a central window and alternative windows. Combined with the generator, inter-frame dynamic graph network analysis is performed and trend prediction is conducted to generate step frames, wherein each step frame is marked with a sliding time window. Based on the step frames, detection nodes are located, and the ultrasound imaging device is controlled to perform ultrasound detection based on a second consecutive frame. Periodic detection and dynamic analysis based on the variable frame rate detection mode are performed, and a tracking chain is integrated and determined. The tracking chain includes trajectory and phase appearance dimensions, addressing the technical problem of insufficient comprehensive, dynamic, and automated monitoring and discrimination capabilities for microbubbles in existing technologies. This achieves automated, comprehensive, accurate, and effective tracking management in dynamic operating scenarios. Attached Figure Description

[0011] Figure 1 This application provides a schematic flowchart of a microbubble dynamic tracking method for diving medicine;

[0012] Figure 2 This application provides a schematic diagram of the acquisition process for the first category of images in a microbubble dynamic tracking method for diving medicine;

[0013] Figure 3 This application provides a schematic diagram of a microbubble dynamic tracking device for diving medicine.

[0014] Explanation of reference numerals in the attached diagram: First detection module 11, step analysis module 12, second detection module 13. Detailed Implementation

[0015] This application provides a method and apparatus for dynamic tracking of microbubbles in diving medicine, which addresses the technical problem of insufficient comprehensive, dynamic, and automated monitoring and discrimination capabilities for microbubbles in existing technologies.

[0016] Example 1: As Figure 1 As shown, this application provides a microbubble dynamic tracking method for diving medicine, the method comprising:

[0017] S1: Introduce a variable frame rate detection mode to control the ultrasound imaging equipment, perform ultrasound detection based on the first consecutive frame, and determine the microbubble image.

[0018] In this embodiment, a variable frame rate detection mode is introduced to control the imaging rhythm of the ultrasound imaging equipment, thereby achieving dynamic response ultrasound detection.

[0019] The variable frame rate detection mode refers to the non-constant adjustment of the frame rate acquired during ultrasound imaging, allowing switching between high and low frame rates to adapt to the microbubble display characteristics under different motion states. For example, two consecutive frame rates are used to analyze the dynamic motion trend between frames to predict the next effective frame rate, such as when active bubble movement or abrupt changes occur. Preferably, during periods of stable motion or no significant change, the frame rate is reduced to lower the frame rate to reduce the pressure on redundant data processing, thereby achieving resource conservation and dynamic optimized data acquisition while ensuring image integrity.

[0020] Specifically, the first consecutive frame refers to a set of ultrasound image frames acquired continuously at fixed time intervals during the initial stage, such as two consecutive frames. Each image frame is acquired using a high-frequency probe, which has good boundary resolution and tissue penetration depth, making it suitable for preliminary detection of the dynamic behavior of microbubbles in blood or tissues.

[0021] Then, using the first consecutive frames, the ultrasound imaging device is controlled to sample the target, and two frames of images are determined as the microbubble images.

[0022] S2: The ultrasound imaging equipment is centrally embedded and deployed through a cascaded classifier and generator. According to the classifier, the microbubble image is classified into background domain and bubble domain. Background domain blanking and microbubble group coding are performed. Each microbubble is calibrated by introducing a central window and alternative windows. Combined with the generator, inter-frame dynamic graph network analysis is performed and trend prediction is performed to generate step frames. The step frames are marked with a sliding time window.

[0023] In one feasible embodiment, the classifier and the generator employ a sample-driven training approach. Specifically, the underlying architecture of the classifier is determined, namely, a two-stage concatenated classification-window calibration. Labeled samples are then acquired, namely, original samples - first-stage classification samples - second-stage classification samples - calibration output samples. Supervised training is performed on the underlying architecture until convergence, resulting in the trained classifier. The underlying architecture of the generator is determined, namely, inter-frame feature network construction under mapping correction - trend prediction - threshold localization. Sample data with the underlying architecture as the processing logic is then acquired, and supervised training is performed until convergence, resulting in the trained generator. The generator is then post-connected to the classifier and embedded in the control center of the ultrasound imaging equipment.

[0024] In this embodiment, the classifier is used to identify and classify the microbubble portion and the background portion of the image. The generator's function is to predict the next step frame, i.e., the next effective acquisition frame node, based on the inter-frame dynamic motion characteristics of the microbubble image. The generator is then connected to the classifier for embedded deployment in the control center of the ultrasound imaging equipment. This allows the ultrasound imaging equipment to integrate preliminary intelligent data processing capabilities, enabling structured processing of image data without relying on an external high-performance computing platform. This deployment method improves the detection response speed and data closed-loop processing capabilities.

[0025] In practice, the classifier performs regional classification processing on the acquired microbubble images, distinguishing between background and bubble regions. In one feasible embodiment, the classifier is based on a lightweight convolutional neural network, which sequentially identifies static background regions (i.e., non-target regions) and dynamic regions containing microbubble signals (i.e., target regions) in the image through a hierarchical structure. The background region typically exhibits uniform tissue echoes and continuous structure, while the bubble region is characterized by drastic changes in echo intensity and prominent edges. After classification, a blanking operation is performed on the background region, that is, its pixel values ​​are reset to a set neutral value (such as grayscale 0), so that subsequent steps focus on image analysis of the target region, thereby improving the accuracy of microbubble encoding and recognition.

[0026] Subsequently, a group coding operation is performed on each microbubble in the bubble domain, that is, by identifying multiple non-overlapping or slightly contacting microbubble targets and assigning a unique code identifier to each microbubble target. Furthermore, a central window and alternative window mechanism is introduced to calibrate each microbubble.

[0027] The central window is a small rectangular or elliptical window that directly covers the central region of the microbubble and is used to extract stable core features; while the alternative windows are at least two windows that surround the central window and cover the edge of the microbubble and are used to extract local perturbation or boundary features.

[0028] In one feasible embodiment, the size of the central window can be preset to 10×10 pixels, and the alternative windows are multiple 5×5 pixel regions distributed around the edge of the microbubble. This multi-window mechanism allows for spatial calibration of the microbubble, providing multi-dimensional references for subsequent correlation analysis.

[0029] Furthermore, the generator extracts the mapping relationship of microbubble features between adjacent frames based on continuous classification images and calibration window information in the time series, and constructs the structural relationship between nodes and edges of a dynamic graph network. Inter-frame dynamic graph network analysis refers to constructing a time evolution graph model using microbubbles as nodes and the interaction features between adjacent frame frequencies as connection edge attributes to characterize the changing trend of the bubble group in the time dimension. Based on this dynamic graph network, a trend prediction mechanism is further introduced, that is, predicting the state evolution trajectory of microbubbles in the graph network and determining whether there are trajectory abrupt changes or state drifts in subsequent frames.

[0030] Furthermore, when the network state evolution reaches a preset threshold for a step variable, a step frame is generated, which is an image frame identified as having an important state change within the current time sliding window, used to identify the stage changes in microbubble behavior. Preferably, the step frame can be labeled with a sliding time window, i.e., the time interval between the first consecutive frame and the step frame triggered by the next sampling, to improve data utilization efficiency and response accuracy of microbubble evolution monitoring.

[0031] Furthermore, such as Figure 2 As shown, the microbubble image is classified into background and bubble regions, background blanking and microbubble group encoding are performed, and individual microbubbles are calibrated by introducing a central window and alternative windows. Step S2 of this application includes:

[0032] The microbubble image is transmitted back, wherein the microbubble image contains at least two frame rate maps; the classifier is triggered to perform domain recognition and classification on the first image in the microbubble image, and to label the background domain and the bubble domain, wherein the first image is any one of the microbubble images; the background domain in the first image is blanked out to determine the first blanked image; the first blanked image is encoded and labeled to determine the first classified image.

[0033] In this embodiment, the microbubble images acquired by the ultrasound imaging equipment are transmitted back to the central control unit of the equipment. The microbubble images include at least two frame-frequency maps, namely, detection images based on consecutive first and second frames, respectively. These are used for motion feature analysis between frame frequencies. The purpose of introducing consecutive frame-frequency images is to enhance the coverage of microbubble motion features.

[0034] Furthermore, the classifier is triggered to perform domain recognition and classification processing on the first image in the microbubble image. The first image is any frame in the microbubble image; that is, both frame rates in the microbubble image need to be processed as follows.

[0035] Among them, domain recognition and classification refers to using a classifier to divide the first image into two types of regions: background region and bubble region. The background region usually refers to the region that is not occupied by microbubbles and has a continuous and uniform echo structure, while the bubble region refers to the target region in the image that has enhanced echo, clear edges or has high brightness features, which is the location where microbubbles may exist.

[0036] After completing the region classification, a blanking operation is performed on the background region in the first image. This involves setting the grayscale value or image feature of all pixels in the background region to zero or specifying an invalid value (such as 255) to achieve an image enhancement effect that highlights the target region and minimizes background interference. This process can effectively reduce the false recognition rate in subsequent encoding processes and improve the accuracy of microbubble calibration to determine the first corrected image, which retains all the information of the bubble region in the original image, while the background region is marked with a uniform color or transparency.

[0037] Furthermore, the first classified image undergoes encoding processing and window calibration to generate a first classified image. Encoding processing includes microbubble cluster identification and unique identifier assignment; that is, using methods such as contour detection and connected component analysis to cluster and identify continuous bright regions in the image, and assigning a unique number or color code to each individual microbubble. Next, window calibration is performed, which involves setting the position and size of a central window and candidate windows on each microbubble target. The central window is aligned with the bubble's centroid region to extract core features; candidate windows are distributed around the central window to capture boundary or perturbation information. These windows are embedded in the classified image as traceable structured labels for subsequent inter-frame matching and dynamic graph modeling processes.

[0038] In summary, transforming raw ultrasound images into classified images with semantic structure and target labeling capabilities lays the foundation for dynamic tracking, phase prediction, and behavior modeling of microbubbles.

[0039] Furthermore, encoding processing and window calibration are performed on the first placement image to determine the first classification image. Step S2 of this application includes:

[0040] The first classified image is traversed, microbubble clusters are identified, coding rules are introduced, and each microbubble is coded and marked to determine the coding group; the window size is determined, and a central window and candidate windows are set, wherein the candidate windows include at least two, the central window is located inside the microbubble, and the candidate windows are located at the boundary of the microbubble; according to the central window and the candidate windows, the microbubble cluster is identified window by window, and the mapping of the coding group and the writing of the central window are performed to determine the first classified image.

[0041] In this embodiment, the non-blank areas (i.e., bubble regions that were not previously blanked) in the image are identified by scanning pixel by pixel or region by region, thereby determining the microbubble groups present in the first placement image.

[0042] Furthermore, after identifying all individual microbubbles, a coding rule is introduced to encode and label each microbubble individually. For example, the coding rule can be designed based on spatial order, morphological features, or center coordinate index, such as numbering the bubbles ID1, ID2, ID3, etc., according to a top-down, left-to-right scanning order. Each identified microbubble target is assigned a unique identifier and stored in a structured coding group list, forming a complete coding group. This coding group will be used for tracking and matching individual microbubbles and updating their states in subsequent image sequences, forming the core foundation for dynamic graph analysis and center window mapping.

[0043] Subsequently, the window size determination and window setting operations are performed. The window size is adjusted according to the preset image resolution and microbubble scale. For example, for microbubble targets with a diameter of 15-50 pixels at a conventional resolution, the central window can be set to 8×8 pixels, and the alternative windows to 5×5 pixels, ensuring that their positions remain consistent relative to the bubble center. In terms of structural settings, the central window must be strictly located within the internal region of the microbubble image target, usually aligned with its center of gravity or the point of maximum brightness; while the alternative windows are symmetrically distributed around the central window or distributed in the bubble edge region to capture boundary information or evolutionary features, and include at least two alternative windows distributed in different directions to enhance the diversity and robustness of spatial features.

[0044] The microbubble cluster is further subjected to window-by-window labeling. Specifically, for each microbubble target, a window is marked within the microbubble target. Simultaneously, a mapping operation between the coding cluster and the window label is performed, that is, the unique coding information of each microbubble is mapped one-to-one with the central window and each alternative window, ensuring that matching channels and time series mapping paths can be quickly established in subsequent inter-frame image analysis.

[0045] Finally, the code corresponding to each microbubble is written into the central window corresponding to that microbubble, completing the generation of the first classification image. This not only preserves the original spatial distribution and brightness information of the microbubbles, but also adds multi-dimensional structural annotations such as the central window, alternative windows, and coding information, so that each microbubble target has a clear identification anchor point and feature index, which facilitates the accurate execution of subsequent dynamic graph network construction and trend evolution analysis.

[0046] In summary, the transformation of microbubble images from unstructured data to structured, trackable targets has been achieved, providing a key foundation for constructing evolutionary relationships across multiple frames.

[0047] Furthermore, step S2 of this application includes: the classification of background domain and bubble domain involves two-stage processing; a first-stage classification process is performed by identifying the boundary between the motion space and non-motion space of the bubble domain to determine a first-stage background domain; a second-stage classification process is performed by identifying the boundary between the bubble domain and non-bubble domain in the motion space to determine a second-stage background domain; and a distinguishable blanking adjustment is performed between the first-stage background domain and the second-stage background domain.

[0048] In this embodiment, to further improve the region segmentation accuracy and background suppression capability of microbubble targets, a two-stage classification process is performed on the microbubble images. First, based on the boundary recognition of the motion space and non-motion space of the bubble domain, a first-stage classification process is performed to initially determine the first-stage background domain. The motion space of the bubble domain refers to image regions with significant displacement, velocity changes, or trajectory activity characteristics in consecutive frames, such as blood vessel regions; while the non-motion space refers to static regions with stable pixel intensity, small temporal changes, or no signs of displacement.

[0049] The first-order background domain often contains non-target interference information such as tissue structure and static liquid interface. Its image features change little, but it usually accounts for a large proportion in the image. It needs to be pre-sorted to reduce its impact on the accuracy of microbubble detection.

[0050] Subsequently, a more fine-grained second-order classification process is applied to the identified motion space regions to further distinguish the boundaries between bubble and non-bubble regions. This involves stripping away non-bubble components, such as motion artifacts, noise reflections, or high-grayscale background stray signals, from the motion space and identifying the regions where microbubbles actually exist. For example, by identifying local grayscale extrema, the closed edges or high-density centers of microbubbles are extracted, defining the region as the bubble region. Image patches adjacent to its boundary that are moving but do not contain bubble structures are classified as second-order background regions.

[0051] Furthermore, after identifying the first-order and second-order background regions, a differentiated blanking process is performed. For example, the first-order background region is typically zeroed out or uniformly suppressed in grayscale to completely eliminate static interference information. The second-order background region, located in motion space and easily confused with the real bubble region, can be occluded using a gradual masking or low-weight masking method to retain some structural information for subsequent verification and analysis. For instance, the system can set the first-order background region pixels to grayscale 0 and the second-order background region pixels to grayscale 50 (in an 8-bit image), thus visually and numerically distinguishing the occupancy levels.

[0052] In summary, this significantly improves the anti-interference capability of the microbubble recognition preprocessing stage, providing a cleaner and more contrasting image foundation for subsequent processing, while effectively avoiding the disturbance of non-target areas such as artifacts and tissue reflections to the construction of the dynamic graph network structure.

[0053] Furthermore, performing inter-frame dynamic graph network analysis, step S2 of this application includes:

[0054] According to the generator, the first and second classified images after the microbubble image classification processing are identified, and window mapping is performed to determine the window mapping group, wherein each microbubble has one central window mapping and at least two spare window mappings; for the window mapping group, trend change analysis of trajectory dimension and phase appearance dimension is performed to determine the inter-frame trend change vector; according to the inter-frame trend change vector, graph nodes are determined, wherein the number of graph nodes is consistent with the number of microbubbles.

[0055] In this embodiment, to construct a microbubble dynamic graph network and realize its structured evolution over time, the generator analyzes the classified microbubble image sequence. Specifically, the system identifies two structured images at different time points: a first-classified image and a second-classified image. The former is typically a reference frame within the current time window, while the latter is the next frame immediately following it. Both have completed microbubble identification, encoding, and window calibration, thus possessing a structured description of a central window and alternative windows.

[0056] Furthermore, a window mapping operation is performed to establish a one-to-one correspondence between the microbubble structures in the two frames, constructing a window mapping group. This involves comparing the spatial location, coding number, and morphological features of each microbubble target in the first and second classification images to determine their temporal continuity. Specific mapping methods can employ nearest neighbor matching, center overlap, etc., to ensure that the central window of a microbubble in the first frame corresponds to the closest structure in the second frame.

[0057] Next, based on the window mapping group, trend analysis was conducted in the trajectory dimension and phase appearance dimension. Trajectory dimension analysis quantifies the positional changes of microbubbles across different frames, typically manifested as the displacement vector, velocity changes, or acceleration trends of the central window coordinates. Phase appearance dimension focuses on the evolution of microbubble morphological attributes, such as changes in grayscale distribution, edge sharpness, and area variations, reflecting changes in the physical state of the bubble, such as expansion, contraction, fusion, or disintegration. By calculating indicators such as the brightness gradient difference, area ratio, and shape distortion rate of each window in consecutive frames, a multi-dimensional inter-frame trend vector can be formed, comprehensively characterizing the spatial and state-related trends of the microbubbles.

[0058] Furthermore, based on the aforementioned inter-frame trend vectors, the basic building blocks of the dynamic graph network—graph nodes—are determined. In this embodiment, each microbubble is considered an independent graph node, and its existence and changes in the time frame sequence are described by its associated trend vector. The uniqueness of a node is determined by its initial encoding and its mapping relationship in consecutive frames, ensuring its continuity and traceability throughout its lifecycle. Since the number of individual microbubbles tracked by the system at each moment is fixed, the final number of graph nodes in the graph network is consistent with the number of microbubbles.

[0059] In the specific implementation process, for existing microbubble disintegration, new additions, etc., bubble formatting and calibration processing is carried out based on the same method as above.

[0060] Furthermore, after determining the graph nodes, step S2 of this application includes:

[0061] For the first classification image, a first neighborhood interaction is determined, wherein the first neighborhood interaction includes the interaction between each microbubble and its neighboring microbubbles; for the second classification image, a second neighborhood interaction is determined; the first neighborhood interaction and the second neighborhood interaction are mapped to determine the interaction vector; the interaction vector is used to determine the edges, and the graph nodes are connected to construct the dynamic graph network.

[0062] In this embodiment, after the graph nodes are set, the neighborhood behavior between each microbubble node is further analyzed to determine the edge attributes between nodes and establish connection relationships that describe their dynamic evolution.

[0063] First, for each microbubble that has been identified and labeled in the first classification image, its first neighborhood interaction is determined, that is, the physical or geometric relationship between any microbubble and its spatially neighboring microbubbles at a specific distance threshold or structural similarity.

[0064] In a practical implementation, a neighborhood radius threshold (such as 20 pixels) can be set to calculate the Euclidean distance between each microbubble and other surrounding microbubbles, identify the interaction state of individual microbubbles within this range, and use it as an adjacency relationship.

[0065] Further interaction features are introduced for measurement, mainly including indicators such as spatial proximity, angle of motion direction, relative velocity change rate, and morphological similarity (such as grayscale histogram cosine distance), which are used to quantitatively describe the possible physical contact, trajectory convergence, or potential fusion trend between microbubbles in the current frame. That is, the adjacency relationship is quantified as the first neighborhood interaction, which includes the specific interaction behavior between each microbubble and its neighboring microbubbles.

[0066] Similarly, the same operation is performed on the second-classified image as the second neighborhood interaction.

[0067] Furthermore, the interactions between the first and second neighborhoods are compared to extract their changing characteristics over time, thereby generating an interaction vector. For example, a multidimensional description can be formed by calculating feature differences (such as changes in direction, approach velocity, and boundary overlap ratio) to reflect the dynamic evolution trend of microbubbles at the local group behavior level. For instance, if two bubbles have a spacing of d1 in the first frame and a spacing of d2 in the second frame, and their orientation angles converge, their interaction vector can be displayed as an attractive state, indicating a potential risk of fusion.

[0068] Furthermore, based on the interaction vector, the connecting edges between graph nodes are determined, and each edge is assigned a weight attribute to characterize the interaction strength or dynamic relationship type between nodes (such as attraction, repulsion, translational consistency, etc.).

[0069] In a preferred embodiment, whether to establish an edge between nodes is usually determined by a threshold mechanism. For example, a connection is established only when the interaction vector exceeds a set threshold in a specific dimension (e.g., the directional angle is <15° and the distance reduction is >20%); otherwise, the nodes are considered unrelated and remain disconnected.

[0070] Through the aforementioned node and edge connection operations, the topology of the dynamic graph network is constructed. Each node represents an individual microbubble, and each edge represents a dynamic interaction path between it and its neighboring bubbles. This dynamic graph network not only preserves the position and attribute change information of microbubbles in the time series but also embeds the linkage behavior patterns among bubble groups, serving as the core foundation for subsequent state prediction, trajectory clustering, phase analysis, and fusion evolution.

[0071] Furthermore, trend prediction is performed to generate step frames. Step S2 of this application includes:

[0072] A step variable threshold is set, which includes a trajectory variable threshold and a phase appearance variable threshold. For the dynamic graph network, the network state is represented by vector values ​​and the evolution direction is represented by vector direction. Dynamic evolution prediction is performed. When any one of the step variable thresholds is satisfied, the step frame is determined.

[0073] In this embodiment, in order to achieve automatic identification of important state change nodes, i.e. step frames, in the microbubble dynamic graph network, a set of step variable thresholds for judging evolutionary abrupt changes is first set. These thresholds are quantitative standards used to measure whether the dynamic graph network undergoes structural changes during time evolution, and include two key dimensions: trajectory variable threshold and phase apparent variable threshold.

[0074] The trajectory variable threshold refers to a standard used to assess the degree of spatial motion change of microbubbles in consecutive frames. For example, when the displacement of the center window position of a microbubble node exceeds a set threshold (e.g., >15 pixels), or the velocity vector direction deviates from the previous state by more than a certain angle (e.g., >30°), its trajectory is considered to have changed drastically, meeting the condition for a trajectory step. The trajectory change threshold can be set through training with historical samples or clinical experience, balancing sensitivity and specificity to avoid misjudging minor perturbations.

[0075] The phase appearance variable threshold is used to determine abrupt changes in microbubble morphology or image characterization features, such as a sudden increase in microbubble area, drastic fluctuations in the mean grayscale distribution, or a sudden decrease in edge sharpness. For example, if the change exceeds a set appearance threshold (e.g., image feature change > 50), it is considered a phase abrupt change, and the condition for a step change is met.

[0076] Furthermore, based on a dynamic graph network, the overall state evolution trend of microbubbles is predicted and analyzed. In one feasible embodiment of this application, using vector values ​​as network states means representing the state information of each graph node in the current frame as a multi-dimensional vector, whose components include spatial position, velocity, morphological features, etc.; while using vector direction as evolution direction means that the system uses the trend of vector change as the evolution path of the bubble state over time to predict its potential development direction. That is, using the dynamic trend direction of the first frame and the second frame as the prediction benchmark, the trend prediction of microbubble behavior in the next few frames can be achieved.

[0077] Furthermore, the evolution vectors of each graph node's state are continuously monitored. Once the evolution value of a node exceeds the trajectory variable threshold or phase appearance variable threshold in any dimension, a step recognition mechanism is triggered. Meeting any threshold condition is considered an evolutionary mutation, and the current frame is thus determined to be a step frame. This step frame serves as a representative frame for potential abnormal changes, critical behavioral transitions, or high-risk states, and becomes the first frame for the next effective tracking acquisition.

[0078] In summary, this significantly improves the response capability to key nodes in microbubble evolution.

[0079] In step S2 of this application, the continuous frame contains at least two consecutive frame frequencies, and the variable frame frequency detection mode is first continuous frame - sliding time window - second continuous frame, wherein the step frame refers to the first and second frame frequencies in the second continuous frame.

[0080] In this embodiment, the continuous frame contains at least two consecutive frame rates. That is, during ultrasound image acquisition, the frame rate is not always maintained, but the sampling frequency is dynamically switched within a specific time series to form image segments with frame rate variation characteristics. Specifically, firstly, two consecutive frame rates are used as the first continuous frame for detection and dynamic analysis to achieve precise capture of the target state change process.

[0081] Furthermore, after performing the above analysis, the sliding time window of the step frame is determined, that is, the time span between the first consecutive frame and the step frame. The step frame is used as the first frame of the second consecutive frame, and then detection and analysis based on the second consecutive frame are performed. The above steps are repeated until the tracking ends.

[0082] The first consecutive frame is a sequence of images from the initial detection phase, used to establish the initial image features and motion baseline of the microbubbles. The sliding time window is a transitional time region connecting the two consecutive frames, used to handle state changes, perform dynamic graph analysis, and construct inter-frame mapping structures. The second consecutive frame is a new round of image sequence acquisition in the later period. Its frame rate is adaptively adjusted based on the state trends identified within the sliding window (such as trend vector anomalies or edge weight mutations), thereby ensuring that the sampling density of key state stages is maximized.

[0083] In summary, this approach achieves flexible response and sensitive capture of the dynamic evolution process of microbubbles.

[0084] S3: Based on the step frame positioning detection node, control the ultrasonic imaging device to perform ultrasonic detection based on the second continuous frame, perform periodic detection and dynamic analysis based on the variable frame frequency detection mode, and integrate and determine the tracking chain, wherein the tracking chain includes trajectory dimension and phase appearance dimension.

[0085] In this embodiment, after generating the step frame and identifying its corresponding sliding time window, the step frame is used as the first frame of the second continuous frame, and then the second continuous frame based on the continuous frame frequency is determined as the detection node.

[0086] Furthermore, the ultrasound imaging equipment is controlled to perform ultrasound detection of the second consecutive frames in a variable frame rate detection mode. Optionally, the second consecutive frames are consistent with the first consecutive frames in terms of sampling method, but the time starting point and frame rate adjustment strategy are adapted according to the window information of the step frame to capture the high-speed evolution behavior of microbubbles during the trajectory change phase, providing sufficient data support for dynamic tracking.

[0087] After the second consecutive frame is acquired, the above image processing steps and the analysis of the next step frame are repeated. During the detection cycle, the above steps are looped, performing periodic detection and dynamic graph analysis operations based on the variable frame rate detection mode. This is used to capture bubble behavior characteristics, while dynamic analysis refers to determining and predicting the evolution path of microbubbles in the continuous time domain by combining node evolution states, edge weight changes, and phase characteristics within the dynamic graph network structure.

[0088] Further integration of graph network sequences from different time periods serves as a tracking chain. This abstract structure, spanning the microbubble tracking cycle, comprises two core dimensions: the trajectory dimension and the phase appearance dimension. The trajectory dimension indicates the microbubble's positional migration path, manifested as a displacement sequence formed by continuous central windows. The phase appearance dimension characterizes the temporal changes in the microbubble's apparent attributes, such as morphology, grayscale distribution, and boundary clarity, to infer key state transitions such as microbubble expansion, rupture, or fusion.

[0089] By constructing this tracking chain, not only can the entire evolution process of a single microbubble be recorded, but also cross-analysis of the correlation behavior between different bubbles can be achieved.

[0090] Furthermore, to integrate and determine the tracking chain, step S3 of this application includes:

[0091] For periodic detection, time series integration is performed on the inter-frame dynamic network to determine the tracking sequence; the tracking sequence is traversed, and sequence decoupling is performed in terms of trajectory dimension and phase appearance dimension to determine the periodic trajectory and periodic phase appearance sequence, which serve as the tracking chain.

[0092] In this embodiment, a longitudinally integrated analysis is first performed on the inter-frame dynamic graph network. Through time-series integration, the dynamic graph nodes and edge information at multiple time points are combined into a continuous graph sequence structure, thereby constructing a complete tracking sequence. That is, it consists of the inter-frame dynamic network from the first consecutive frame frequency to the inter-frame dynamic network of the last consecutive frame.

[0093] Furthermore, using trajectory dimension and phase appearance dimension as core indicators, sequence decoupling operations are performed. Trajectory dimension decoupling refers to independently extracting the spatial position of each microbubble node in each frame of the image, forming a position vector sequence (such as centroid coordinates or center window displacement sequence) to reflect its movement path, velocity changes, and trajectory smoothness. Phase appearance dimension decoupling involves extracting the image representation features of the bubble in each frame, such as area, edge sharpness, grayscale mean, shape compactness, etc., and constructing its time series evolution curve to reveal the periodic change patterns of the microbubble at the state level, such as expansion-contraction cycles and surface tension fluctuations.

[0094] After the above decoupling, periodic trajectory sequences and periodic phase appearance sequences are obtained respectively. Together, they constitute a full-dimensional, multi-time dynamic characterization of the microbubble, and are ultimately stored and retrieved as a complete tracking chain.

[0095] In summary, a tracking chain mechanism spanning the entire timeline has been established, significantly improving the perception accuracy and recognition capability of microbubble evolution paths, and providing a reliable and quantifiable method for multi-temporal dynamic monitoring.

[0096] The microbubble dynamic tracking method for diving medicine provided in this application has the following technical effects:

[0097] 1. A variable frame rate detection mode is introduced, consisting of a first consecutive frame, a sliding time window, and a second consecutive frame. Compared to traditional fixed frame rate detection, this mode allows for flexible adjustment of the detection frequency. It accurately captures the dynamic changes of microbubbles while reducing data redundancy, thus improving detection efficiency and resource utilization.

[0098] 2. A cascaded classifier and generator are used to centrally embed the ultrasound imaging equipment. The classifier performs second-order fine-grained classification of the background and bubble domains, performs blanking and microbubble coding and calibration, and effectively removes background interference. The generator constructs a dynamic graph network to analyze inter-frame changes. The two work together, which significantly improves the accuracy and completeness of microbubble feature extraction compared to a single processing method, and accurately predicts and locates key frame frequencies.

[0099] 3. Based on a dynamic graph network, trend prediction is performed to generate step frames. Trajectory variable thresholds and phase appearance variable thresholds are used as trigger conditions to predict the microbubble's changing trend in advance. Combining periodic detection and dynamic analysis, a tracking chain encompassing trajectory and phase appearance dimensions is formed, enabling continuous and accurate tracking of the microbubble's motion trajectory and state changes.

[0100] Example 2: Based on the same inventive concept as the microbubble dynamic tracking method for diving medicine in the foregoing examples, such as... Figure 3 As shown, this application provides a microbubble dynamic tracking device for diving medicine, the device comprising:

[0101] The first detection module 11 is used to introduce a variable frame rate detection mode, control the ultrasound imaging equipment, perform ultrasound detection based on the first continuous frame, and determine the microbubble image.

[0102] The step analysis module 12 is used to centrally embed and deploy the ultrasound imaging equipment through a cascaded classifier and generator. According to the classifier, the microbubble image is classified into background domain and bubble domain. Background domain blanking and microbubble group coding are performed. Each microbubble is calibrated by introducing a central window and alternative windows. Combined with the generator, inter-frame dynamic graph network analysis is performed and trend prediction is performed to generate a step frame. The step frame is marked with a sliding time window.

[0103] The second detection module 13 is used to locate the detection node according to the step frame, control the ultrasound imaging equipment, perform ultrasound detection based on the second consecutive frame, perform periodic detection and dynamic analysis based on the variable frame rate detection mode, and integrate and determine the tracking chain, wherein the tracking chain includes trajectory dimension and phase appearance dimension.

[0104] Furthermore, the first detection module 11 is used to perform the following steps: transmit the microbubble image back, wherein the microbubble image contains at least two frame rate maps; trigger the classifier to perform domain recognition and classification on the first image in the microbubble image, and label the background domain and bubble domain, wherein the first image is any one of the microbubble images; perform blanking on the background domain in the first image to determine the first corrected image; perform encoding processing and window labeling on the first corrected image to determine the first classified image.

[0105] Furthermore, the step analysis module 12 is used to perform the following steps: traversing the first classified image, identifying microbubble clusters, introducing coding rules, coding and marking each microbubble, and determining the coding cluster; determining the window size, setting a central window and candidate windows, wherein the candidate windows include at least two, the central window is located inside the microbubble, and the candidate windows are located at the microbubble boundary; according to the central window and the candidate windows, identifying each window of the microbubble cluster, and performing the mapping of the coding cluster and writing of the central window to determine the first classified image.

[0106] Furthermore, the step analysis module 12 is used to perform the following steps: the classification of the background domain and the bubble domain includes two-level processing; the first-level classification processing is performed by identifying the boundary between the motion space and the non-motion space of the bubble domain to determine the first-level background domain; the second-level classification processing is performed by identifying the boundary between the bubble domain and the non-bubble domain in the motion space to determine the second-level background domain; and the distinguishable blanking of the first-level background domain and the second-level background domain is performed.

[0107] Furthermore, the step analysis module 12 is used to perform the following steps: according to the generator, identify the first classification image and the second classification image after the microbubble image classification processing, and perform window mapping to determine the window mapping group, wherein each microbubble has one central window mapping and at least two spare window mappings; for the window mapping group, perform trend change analysis of trajectory dimension and phase appearance dimension to determine the inter-frame trend change vector; according to the inter-frame trend change vector, determine the graph nodes, wherein the number of graph nodes is consistent with the number of microbubbles.

[0108] Furthermore, the step analysis module 12 is used to perform the following steps: for the first classification image, determine the first neighborhood interaction, wherein the first neighborhood interaction includes the interaction between each microbubble and the neighboring microbubbles; for the second classification image, determine the second neighborhood interaction; map the first neighborhood interaction and the second neighborhood interaction to determine the interaction vector; determine the edges with the interaction vector, connect the graph nodes, and construct the dynamic graph network.

[0109] Furthermore, the step analysis module 12 is used to perform the following steps: setting step variable thresholds, wherein the step variable thresholds include trajectory variable thresholds and phase apparent variable thresholds; for the dynamic graph network, using vector values ​​as network states and vector directions as evolution directions, performing dynamic evolution prediction; and determining the step frame when any one of the step variable thresholds is satisfied.

[0110] The continuous frame contains at least two consecutive frame frequencies, and the variable frame frequency detection mode is a first continuous frame - sliding time window - second continuous frame, wherein the step frame refers to the first and second frame frequencies in the second continuous frame.

[0111] Furthermore, the second detection module 13 is used to perform the following steps: for periodic detection, perform time series integration on the inter-frame dynamic network to determine the tracking sequence; traverse the tracking sequence, and perform sequence decoupling in terms of trajectory dimension and phase appearance dimension to determine the periodic trajectory and periodic phase appearance sequence as the tracking chain.

[0112] Through the foregoing detailed description of the microbubble dynamic tracking method for diving medicine, those skilled in the art can clearly understand the microbubble dynamic tracking method and apparatus for diving medicine in this embodiment. As for the apparatus disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section description.

[0113] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. 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 this application. Therefore, this application 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 microbubble dynamic tracking method for diving medicine, characterized in that, The method includes: A variable frame rate detection mode is introduced to control the ultrasound imaging equipment and perform ultrasound detection based on the first consecutive frame to determine the microbubble image; A cascaded classifier and generator are used to centrally embed and deploy ultrasound imaging equipment. According to the classifier, the microbubble image is classified into background domain and bubble domain. Background domain blanking and microbubble group coding are performed. Each microbubble is calibrated by introducing a central window and alternative windows. Combined with the generator, inter-frame dynamic graph network analysis is performed and trend prediction is performed to generate step frames. The step frames are marked with a sliding time window. Background domain blanking means setting the gray value or image feature of all pixels in the background domain to zero or specifying an invalid value. Based on the step frame positioning detection node, the ultrasonic imaging device is controlled to perform ultrasonic detection based on the second consecutive frame, perform periodic detection and dynamic analysis based on the variable frame frequency detection mode, and integrate and determine the tracking chain, wherein the tracking chain includes trajectory dimension and phase appearance dimension.

2. The microbubble dynamic tracking method for diving medicine as described in claim 1, characterized in that, The microbubble image is classified into background and bubble regions. Background blanking and microbubble group encoding are performed. Individual microbubbles are calibrated by introducing a central window and alternative windows, including: The microbubble image is transmitted back, wherein the microbubble image contains at least two frame rate maps; The classifier is triggered to perform domain recognition and classification on the first image in the microbubble image, and to label the background domain and the bubble domain, wherein the first image is any one of the microbubble images; The background region in the first image is blanked out to determine the first blanked image; The first categorized image is encoded and windowed to determine the first classification image.

3. The microbubble dynamic tracking method for diving medicine as described in claim 2, characterized in that, Encoding and windowing are performed on the first positioned image to determine the first category image, including: Traverse the first placement image, identify microbubble clusters, introduce coding rules, encode and mark each microbubble individually, and determine the coding clusters; Determine the window size, set the center window and alternative windows, wherein the alternative windows include at least two, the center window is located inside the microbubble, and the alternative windows are located at the boundary of the microbubble; Based on the central window and the alternative windows, the microbubble clusters are identified window by window, and the mapping of the coding clusters and the writing of the central window are performed to determine the first classification image.

4. The microbubble dynamic tracking method for diving medicine as described in claim 2, characterized in that, The background and bubble domain classification involves two levels of processing; By identifying the boundary between the motion space and non-motion space of the bubble domain, first-order classification is performed to determine the first-order background domain. By identifying the boundaries between bubble and non-bubble domains within the motion space, a second-order classification process is performed to determine the second-order background domain. Perform a distinguishable blanking adjustment between the first-order background domain and the second-order background domain.

5. The microbubble dynamic tracking method for diving medicine as described in claim 1, characterized in that, Perform inter-frame dynamic graph network analysis, including: According to the generator, the first classification image and the second classification image after the microbubble image classification processing are identified, and window mapping is performed to determine the window mapping group, wherein each microbubble has a central window mapping and at least two spare window mappings; For the window mapping group, trend change analysis of trajectory dimension and phase appearance dimension is performed to determine the inter-frame trend change vector; Graph nodes are determined based on the inter-frame trend vector, wherein the number of graph nodes is consistent with the number of microbubbles.

6. The microbubble dynamic tracking method for diving medicine as described in claim 5, characterized in that, After determining the graph nodes, the following is included: For the first classification image, a first neighborhood interaction is determined, wherein the first neighborhood interaction includes the interaction between each microbubble and its neighboring microbubbles; For the second-classified images, determine the second-neighborhood interactions; Map the first neighborhood interaction with the second neighborhood interaction to determine the interaction vector; Edges are determined using the interaction vectors, and the graph nodes are connected to construct the dynamic graph network.

7. The microbubble dynamic tracking method for diving medicine as described in claim 6, characterized in that, Perform trend prediction and generate step frames, including: Set a threshold for step variables, wherein the threshold for step variables includes a threshold for trajectory variables and a threshold for phase appearance variables; For the dynamic graph network, the vector values ​​are used as the network state and the vector direction is used as the evolution direction to perform dynamic evolution prediction. When any one of the step variable thresholds is satisfied, the step frame is determined.

8. The microbubble dynamic tracking method for diving medicine as described in claim 1, characterized in that, The continuous frames contain at least two consecutive frame frequencies, and the variable frame frequency detection mode is a first continuous frame - sliding time window - second continuous frame, wherein the step frame refers to the first frame in the second continuous frame.

9. The microbubble dynamic tracking method for diving medicine as described in claim 1, characterized in that, Integrating and defining the tracking chain includes: For periodic detection, time-series integration is performed on the inter-frame dynamic network to determine the tracking sequence; Traverse the tracking sequence, perform sequence decoupling using the trajectory dimension and phase appearance dimension, and determine the periodic trajectory and periodic phase appearance sequence as the tracking chain.

10. A microbubble dynamic tracking device for diving medicine, characterized in that, The apparatus for performing the microbubble dynamic tracking method for diving medicine according to any one of claims 1-9, the apparatus comprising: The first detection module is used to introduce a variable frame rate detection mode, control the ultrasound imaging equipment, perform ultrasound detection based on the first consecutive frames, and determine the microbubble image. The step analysis module is used to centrally embed and deploy ultrasound imaging equipment through a cascaded classifier and generator. According to the classifier, the microbubble image is classified into background domain and bubble domain. Background domain blanking and microbubble group encoding are performed. Each microbubble is calibrated by introducing a central window and alternative windows. Combined with the generator, inter-frame dynamic graph network analysis is performed and trend prediction is performed to generate step frames. The step frames are marked with a sliding time window. Background domain blanking means setting the gray value or image feature of all pixels in the background domain to zero or specifying an invalid value. The second detection module is used to locate the detection node according to the step frame, control the ultrasound imaging equipment, perform ultrasound detection based on the second continuous frame, perform periodic detection and dynamic analysis based on the variable frame frequency detection mode, and integrate and determine the tracking chain, wherein the tracking chain includes trajectory dimension and phase appearance dimension.

Images