Ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception

Abstract

The invention discloses an ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception, particularly relates to the field of ultrasonic imaging, is used for solving the problem of accurate detection and analysis of multi-target labeled microbubbles in a complex blood flow environment, and is characterized in that an ultrasonic probe is triggered by broadband pulse to obtain original data, and information of an included angle between a scanning plane and a blood flow is combined; projection deviation is corrected through a micro-angle resampling technology, an accurate and consistent micro-bubble time-frequency three-dimensional image tensor is generated, and signal quality is ensured. Then, features are extracted by means of sparse phase decoding and complex independent component decomposition, signal separation and classification are optimized in combination with a pre-trained microbubble fingerprint model and an attention network, and the problems of aliasing and distortion are effectively solved; meanwhile, dynamic tracking and density analysis of the microbubbles are realized through frequency attenuation plane projection and particle filtering, and real-time motion and distribution information is provided; and finally, integrating the density curve and the frequency attenuation slope to generate a targeted microbubble concentration time sequence table, and transmitting the targeted microbubble concentration time sequence table to a clinical end through a data interface.

Description

Technical Field

[0001] The present invention relates to the field of ultrasonic imaging, and more particularly to an ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception. Background Art

[0002] In precision diagnosis and treatment scenarios such as ultrasound angiography, physicians often simultaneously inject multiple targeted ultrasound contrast agent microbubbles carrying biomarkers such as VEGF, integrin αvβ3, and HER2, hoping to reveal differences in vascular-cellular pathways with the help of the unique resonant frequency and attenuation slope of each type of microbubble.

[0003] However, when microbubbles move at high speed with blood flow and the ultrasound scanning plane is at a slight angle to the direction of the blood vessels, the time domain trajectory, frequency domain modulation, and spatial domain aggregation overlap. This acoustic fingerprint aliasing makes it difficult to distinguish different target signals in subsequent images, thus limiting accurate assessment. Current threshold segmentation or fixed-band filtering methods cannot simultaneously process the coupled RF echoes in the time, frequency, and spatial domains, thus distorting the classification, dynamic tracking, and dose quantification of multi-target microbubbles. To address these issues, a technical solution is now provided. Summary of the Invention

[0004] To overcome the above-mentioned shortcomings of the prior art, embodiments of the present invention provide an ultrasound contrast agent microbubble detection and analysis system based on intelligent perception. Through adaptive resampling, sparse decoding and attention separation, it extracts independent acoustic fingerprints from aliased radio frequency signals and outputs unified quantization results across devices, thereby solving the problem of decreased specificity caused by multi-target aliasing of microbubbles, thereby addressing the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: An ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception includes the following steps: a data acquisition module, a signal correction module, a feature extraction module, a label generation module, a trajectory tracking module, and a data transmission module; Data acquisition module: The broadband pulse triggers the ultrasound probe to collect B-mode and I / Q frame sequences carrying multi-target labeled microbubbles and writes them into the scanning plane-blood flow angle and time-space index matrix; Signal correction module: performs micro-angle resampling on the frame sequence according to the included angle, maps it to the inertial coordinate system and generates a microbubble time-frequency three-dimensional image tensor; Feature extraction module: performs sparse phase decoding and complex independent component decomposition on the image tensor, extracts the resonant differential entropy vector and the transient phase confluence degree vector, inputs the pre-trained microbubble fingerprint model, and outputs the microbubble unique identification index and residual microbubble fingerprint; Label generation module: The residual microbubble fingerprint and microbubble unique identification index are input into the attention network, and a multi-target microbubble label map is generated based on the resonance similarity and attenuation slope; Trajectory tracking module: projects the label map onto the frequency attenuation plane, calculates the frequency attenuation slope, calls the particle filter to track the microbubble trajectory and draws the transient microbubble density curve; Data transmission module: Combine the microbubble density curve and frequency attenuation slope to generate a targeted microbubble concentration time series table, which is transmitted to the imaging end and the treatment end through the data interface.

[0006] In a preferred embodiment, the data acquisition module includes the following: The B-mode frame sequence provides anatomical information of the tissue, the I / Q frame sequence records the in-phase and quadrature components of the signal to preserve amplitude and phase information, and synchronously measures the angle between the scanning plane and the blood flow direction. The angle between the scanning plane and the blood flow direction is obtained using ultrasound technology to correct the spatial projection deviation of the signal. A spatiotemporal index is assigned to each frame of data. The spatiotemporal index includes a timestamp and spatial coordinates within the image plane. A spatiotemporal index matrix is constructed to record the angle value between the scanning plane and the blood flow direction at the corresponding position.

[0007] In a preferred embodiment, the signal correction module includes the following: Micro-angle resampling is used to correct the spatial projection deviation of the B-mode frame sequence and the I / Q frame sequence caused by the angle between the scanning plane and the blood flow direction. Micro-angle resampling performs coordinate transformation on each pixel point in the B-mode frame sequence and the I / Q frame sequence based on the angle between the scanning plane and the blood flow direction in the spatiotemporal index matrix, and calculates the corrected pixel value through bilinear interpolation. The corrected B-mode frame sequence and I / Q frame sequence are mapped to an inertial coordinate system parallel to the blood flow direction to generate the corrected B-mode frame sequence and I / Q frame sequence, and the microbubble time-frequency three-dimensional image tensor is extracted from the corrected I / Q frame sequence.

[0008] In a preferred embodiment, the signal correction module further includes the following: The microbubble time-frequency three-dimensional image tensor is generated by performing time-frequency analysis on the I / Q signals at each spatial position. The time-frequency analysis uses short-time Fourier transform and a window function to truncate the I / Q signals to reduce edge effects. The time-frequency analysis results are stacked into the microbubble time-frequency three-dimensional image tensor.

[0009] In a preferred embodiment, the feature extraction module includes the following: Sparse phase decoding and complex independent component decomposition are performed on the microbubble time-frequency three-dimensional image tensor. Sparse phase decoding extracts the phase information of the microbubble time-frequency three-dimensional image tensor and applies a sparse coding algorithm to generate a sparse phase matrix. Complex independent component decomposition decomposes the microbubble time-frequency three-dimensional image tensor into independent component matrices.

[0010] In a preferred embodiment, the feature extraction module further includes the following: The resonant differential entropy vector is extracted from the sparse phase matrix. The resonant differential entropy vector characterizes the resonant characteristics of the microbubble by calculating the entropy value of the phase difference. The transient phase confluence vector is extracted from the independent component matrix. The transient phase confluence vector characterizes the phase stability of the microbubble signal by calculating the degree of aggregation of the independent component phase sequence. The resonant differential entropy vector and the transient phase confluence vector are input into the pre-trained microbubble fingerprint model to output the microbubble uniqueness index and residual microbubble fingerprint.

[0011] In a preferred embodiment, the label generation module includes the following: Normalization processing is performed on the residual microbubble fingerprint and the microbubble unique identification index to generate the normalized residual microbubble fingerprint and the normalized microbubble unique identification index, which are then merged into an input feature matrix. The input feature matrix is processed by a multi-head self-attention network to generate an attention feature matrix. The resonance feature subset and the attenuation feature subset are extracted from the attention feature matrix. The resonance similarity is calculated based on the resonance feature subset, and the attenuation slope is calculated based on the attenuation feature subset. The resonance similarity and the attenuation slope are combined to generate a multi-target microbubble label map through a clustering algorithm.

[0012] In a preferred embodiment, the trajectory tracking module includes the following: The multi-targeted microbubble label map is projected onto the frequency attenuation plane to generate a frequency attenuation distribution map. The frequency attenuation distribution map is determined by extracting the peak frequency of the microbubbles and the attenuation rate at the peak frequency in the multi-targeted microbubble label map. The frequency attenuation slope is calculated. The frequency attenuation slope is determined by the ratio of the change in the attenuation rate to the change in the frequency in the frequency attenuation distribution map. Category-specific frequency attenuation slopes are calculated for different categories of microbubbles. Based on the spatial position information and blood flow velocity in the multi-targeted microbubble label map, particle filtering is used to track the microbubble trajectory to generate the microbubble motion trajectory.

[0013] In a preferred embodiment, the trajectory tracking module further includes the following: The particle filter generates microbubble motion trajectories by initializing particle sets, predicting particle positions, updating particle weights, and resampling. Based on the microbubble motion trajectories, the number of microbubbles distributed in space at each time point is counted to calculate the transient microbubble density. The transient microbubble density is the number of microbubbles per unit space per unit time. The transient microbubble density curve is drawn, which shows the dynamic distribution characteristics of microbubbles with time as the horizontal axis and transient microbubble density as the vertical axis.

[0014] In a preferred embodiment, the data transmission module includes the following: A targeted microbubble concentration time series table is generated based on the transient microbubble density curve and the frequency decay slope, specifically including extracting the transient microbubble density curve and the frequency decay slope for microbubbles classified by category, correcting the transient microbubble density curve using the frequency decay slope to generate a corrected density curve, calculating the corrected density curve by multiplying the value of the transient microbubble density curve at each time point by the correction factor, calculating the targeted microbubble concentration based on the corrected density curve, the targeted microbubble concentration being the product of the value of the corrected density curve at each time point and the concentration conversion coefficient, calculating the concentration value for each category of microbubbles separately and summarizing them to form a targeted microbubble concentration time series table; the targeted microbubble concentration time series table records the concentration value of each category of microbubbles using time as an index, converting the targeted microbubble concentration time series table into a standard data structure, and transmitting it to the imaging end and the treatment end through the data interface.

[0015] The technical effects and advantages of the ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception of the present invention are as follows: The present invention improves the detection and analysis capabilities of multi-targeted labeled microbubbles in ultrasound imaging by constructing a complete process; uses a broadband pulse to trigger the ultrasound probe to obtain raw data, and combines the information of the angle between the scanning plane and the blood flow; corrects the projection deviation through micro-angle resampling technology to generate accurate and consistent microbubble time-frequency three-dimensional image tensors to ensure signal quality; then, with the help of sparse phase decoding and complex independent component decomposition, it extracts features, and combines the pre-trained microbubble fingerprint model with the attention network to optimize signal separation and classification, effectively solving the problems of aliasing and distortion; at the same time, dynamic tracking and density analysis of microbubbles are achieved through frequency decay plane projection and particle filtering, providing real-time motion and distribution information; finally, the density curve and frequency decay slope are integrated to generate a targeted microbubble concentration time series table, which is transmitted to the clinical end through the data interface to support real-time monitoring and treatment decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a flow chart of the ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] Example 1: Figure 1The invention provides an ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception, which includes: a data acquisition module, a signal correction module, a feature extraction module, a label generation module, a trajectory tracking module and a data transmission module.

[0019] Data acquisition module: The broadband pulse triggers the ultrasound probe to collect B-mode and I / Q frame sequences carrying multi-target labeled microbubbles, and writes them into the scanning plane-blood flow angle and time-space index matrix.

[0020] Signal correction module: performs micro-angle resampling on the frame sequence according to the included angle, maps it to the inertial coordinate system and generates a microbubble time-frequency three-dimensional image tensor.

[0021] Feature extraction module: performs sparse phase decoding and complex independent component decomposition on the image tensor, extracts the resonant differential entropy vector and the transient phase confluence degree vector, inputs the pre-trained microbubble fingerprint model, and outputs the microbubble unique identification index and residual microbubble fingerprint.

[0022] Label generation module: The residual microbubble fingerprint and microbubble unique identification index are input into the attention network, and a multi-target microbubble label map is generated based on the resonance similarity and attenuation slope.

[0023] Trajectory tracking module: projects the label map onto the frequency attenuation plane, calculates the frequency attenuation slope, calls the particle filter to track the microbubble trajectory and draws the transient microbubble density curve.

[0024] Data transmission module: Combine the microbubble density curve and frequency attenuation slope to generate a targeted microbubble concentration time series table, which is transmitted to the imaging end and the treatment end through the data interface.

[0025] In the field of ultrasound imaging, ultrasound contrast agent microbubbles are widely used due to their ability to significantly enhance image contrast and reveal hemodynamic characteristics. They are particularly valuable in multi-target molecular imaging and hemodynamic monitoring. By injecting microbubbles with different acoustic properties (each microbubble forms a specific "acoustic fingerprint" due to its unique resonant frequency and acoustic response), high-resolution analysis of tissue microenvironment and blood flow characteristics can be achieved.

[0026] However, in practical applications, the high-speed motion of microbubbles in the bloodstream and the angle between the ultrasound probe scanning plane and the blood flow direction can lead to signal overlap and aliasing in the time, frequency, and spatial domains. This signal coupling seriously interferes with the accuracy of microbubble classification, tracking, and quantitative analysis. Traditional methods such as threshold segmentation or fixed-band filtering have difficulty effectively separating these aliased signals, especially when multiple types of microbubbles coexist, where classification distortion and tracking interruption are particularly significant. Therefore, developing a technology that can intelligently sense and accurately decouple aliased signals has become key to achieving accurate multi-target microbubble detection and analysis.

[0027] As the starting point of the entire solution, the data acquisition module aims to establish a complete data set containing microbubble signals and spatial information through broadband pulse triggering and data acquisition, providing a basis for signal correction and feature extraction in subsequent steps.

[0028] The data acquisition module includes the following: S1-1. Broadband pulse triggering and data acquisition: Driven by broadband pulses, the ultrasound probe emits ultrasonic waves to excite multiple target-labeled microbubbles and collect their acoustic responses. The broadband pulse frequency range is designed to cover the resonant frequencies of multiple microbubbles, ensuring simultaneous excitation of different types of microbubbles. This frequency range is defined by a set minimum and maximum frequency, determined based on the resonant frequency characteristics of the microbubbles, ensuring that the resonant frequency of each microbubble falls within this range.

[0029] After receiving the echo signal scattered by the microbubbles, the ultrasound probe generates two types of data: B-mode frame sequence and I / Q frame sequence.

[0030] The B-mode frame sequence provides anatomical information of the tissue for subsequent spatial positioning and visualization; The I / Q frame sequence contains the in-phase and quadrature components of the signal, records the amplitude and phase information of the signal, and provides raw data for subsequent time-frequency analysis and feature extraction.

[0031] A B-mode frame sequence is a series of two-dimensional grayscale images generated by the reflection intensity of ultrasound waves in tissue. In this mode, the brightness of each pixel in the image is proportional to the intensity of the tissue's reflection of ultrasound waves. That is, areas with stronger reflections appear as brighter pixels on the image, while areas with weaker reflections appear as darker pixels. B-mode frame sequences are primarily used to display the anatomical structure of tissues, providing clear spatial positioning and visualization information. They are one of the most commonly used imaging methods in clinical ultrasound diagnosis. For example, they can help doctors observe the morphology, boundaries, and internal structural details of organs.

[0032] The I / Q frame sequence is another representation of the ultrasound echo signal, recording the signal's in-phase component (in-phase) and quadrature component (quadrature). The in-phase component reflects the signal's real-time amplitude, while the quadrature component is related to the signal's phase information. By combining these two components, the I / Q frame sequence retains the complete information of the ultrasound echo signal and is more suitable for complex signal processing and analysis than the B-mode frame sequence. This sequence plays an important role in time-frequency analysis, phase detection, and feature extraction. For example, in the detection of ultrasound contrast agent microbubbles, the I / Q frame sequence can be used to analyze the acoustic response and motion characteristics of the microbubbles, providing deeper physiological or pathological information.

[0033] The process of broadband pulse triggering and data acquisition is coordinated by the system control unit to ensure the time consistency of transmission and reception.

[0034] The use of broadband pulses enables simultaneous excitation of multiple microbubbles, allowing the acquired data to include information about the acoustic responses of different microbubbles. The combination of B-mode frame sequences and I / Q frame sequences provides spatial structure information while preserving the detailed characteristics of the signal.

[0035] S1-2. Measurement of the angle between the scanning plane and the blood flow direction: The angle between the scanning plane and the blood flow direction is a key parameter affecting the spatial projection of microbubble signals. If it is not corrected, it will lead to deviations in subsequent analysis. Therefore, it is necessary to accurately measure the angle. There are two ways to measure it: First, using Doppler ultrasound technology, the angle is calculated based on the geometric relationship between the blood flow velocity vector and the probe scanning plane. Specifically, the angle value is determined by comparing the direction difference between the blood flow velocity vector and the normal vector of the scanning plane. Second, in scenarios where the probe position is fixed, the operator manually enters the angle value based on the actual scanning situation. Angle measurement is synchronized with the acquisition of the B-mode frame sequence and the I / Q frame sequence, ensuring that each frame of data is associated with a corresponding angle value. The synchronization process is controlled by the system clock to ensure accurate time alignment.

[0036] The measurement and recording of the angle provides the necessary information for subsequent signal correction. By synchronously recording the angle, the signal can be spatially corrected to eliminate signal deviations caused by the angle.

[0037] S1-3. Construction of spatiotemporal index matrix: Assign a spatiotemporal index to each frame of data (including B-mode frame sequence and I / Q frame sequence), including a timestamp and spatial coordinates in the image plane.

[0038] The timestamp represents the moment of frame acquisition, and the spatial coordinate represents the pixel's position within the image plane. Based on this, a spatiotemporal index matrix is constructed, whose elements are the angle values corresponding to the spatiotemporal positions. Under the simplifying assumption that the angle varies only with time, all spatial positions with the same timestamp share the same angle value. To maintain universality, the angle is allowed to vary spatially, in which case the angle value is determined by interpolation of local blood flow directions. The interpolation process uses linear or nonlinear estimation based on the angle data of adjacent positions to reflect spatial heterogeneity.

[0039] The construction of the spatiotemporal index matrix closely associates the angle information with the frame data, facilitating targeted signal processing based on specific time and spatial locations.

[0040] S1-4. Data storage and formatting: B-mode frame sequences, I / Q frame sequences, and spatiotemporal index matrices are stored in a database. This storage utilizes standardized data formats, such as DICOM or HDF5, to ensure data compatibility across different devices and analysis platforms. Each data set is keyed by a timestamp and spatial coordinates, linking the corresponding B-mode pixel values, I / Q components, and angle values. The storage process ensures a one-to-one correspondence between timestamps and spatial coordinates, using a validation mechanism to prevent data loss or misalignment. After data formatting, all information is stored in a structured manner for easy subsequent access and processing.

[0041] The data acquisition module successfully acquires B-mode and I / Q frame sequences of multi-targeted microbubbles using a broadband pulse-triggered ultrasound probe, while simultaneously measuring the angle between the scanning plane and the blood flow direction. By constructing a spatiotemporal index matrix containing both spatiotemporal index and angle information, it provides the necessary data foundation for subsequent micro-angle resampling and coordinate system mapping. The B-mode frame sequence provides spatial anatomical information, the I / Q frame sequence preserves the amplitude and phase characteristics of the microbubble signal, and the spatiotemporal index matrix records the signal's spatial deviation parameters. This comprehensive data acquisition and organization ensures data traceability and consistency.

[0042] The data acquisition module uses a broadband pulse-triggered ultrasound probe to collect B-mode and I / Q frame sequences of microbubbles carrying multi-target markers, records the angle between the scanning plane and the blood flow direction, and constructs a spatiotemporal index matrix. This process provides the raw data foundation for the acoustic response analysis of microbubbles. However, due to the angle between the scanning plane and the blood flow direction, the collected signals have projection deviations in the spatial coordinates, which directly affects the accuracy of subsequent feature extraction and classification. The signal correction module aims to correct the frame sequence collected by the data acquisition module through specific technical means, eliminate spatial deviations, and generate a three-dimensional time-frequency image tensor of microbubbles suitable for subsequent analysis.

[0043] The signal correction module includes the following: S2-1. Implementation of micro-angle resampling: Micro-angle resampling aims to correct for signal space projection deviations caused by the angle between the scanning plane and the blood flow direction. Because of this angle, the pixel positions in the B-mode and I / Q frame sequences cannot directly reflect the true location of microbubbles in the bloodstream. This deviation reduces the accuracy of subsequent analysis. To address this issue, micro-angle resampling uses coordinate transformation to remap the B-mode and I / Q frame sequences from the original scanning plane coordinate system to an inertial coordinate system parallel to the blood flow direction.

[0044] In a specific implementation, the angle value of each frame at a specific time and space position is first extracted from the spatiotemporal index matrix provided by the data acquisition module. Then, for each pixel point in the B-mode frame sequence and the I / Q frame sequence, its new position in the inertial coordinate system is calculated based on its horizontal axis position and vertical axis position in the original scanning plane coordinate system, as well as the corresponding angle value. The calculation process is based on the geometric characteristics of the angle: the original horizontal axis position and vertical axis position are rotated and adjusted by the sine and cosine values of the angle to generate new horizontal axis position and vertical axis position. Since the newly calculated position is usually not an integer grid point, it is necessary to use a bilinear interpolation method to perform a weighted average of the values of the four surrounding neighboring pixel points to determine the pixel value at the new position, thereby maintaining the smoothness and continuity of the data.

[0045] For example, coordinate transformations can be performed as follows: Define the original coordinate system as the scanning plane coordinates , the inertial coordinate system is the coordinate system parallel to the direction of blood flow ,in The axis is consistent with the direction of blood flow.

[0046] For each pixel of each frame ,according to Calculate its position in the inertial coordinate system , the formula is as follows:

[0047] : pixel coordinates in the original frame sequence; : spatiotemporal index matrix Corresponding time and location The angle between : The coordinates in the inertial coordinate system after correction.

[0048] Micro-angle resampling eliminates projection deviations caused by the angle between the scanning plane and the blood flow direction through coordinate transformation, enabling B-mode and I / Q frame sequences to truly reflect the spatial distribution of microbubbles in the bloodstream. This correction process improves the accuracy of subsequent feature extraction and signal analysis, laying the foundation for generating reliable microbubble time-frequency 3D image tensors and ensuring the reliability of the entire processing pipeline.

[0049] S2-2. Mapping to the inertial coordinate system: The process of mapping to the inertial coordinate system is to reorganize the micro-angle resampled B-mode frame sequence and I / Q frame sequence into a new data sequence aligned with the direction of blood flow. After completing the micro-angle resampling, the position of each pixel has been corrected to the coordinates in the inertial coordinate system. At this time, these corrected pixels need to be rearranged in time sequence and spatial position to form a new B-mode frame sequence and I / Q frame sequence. The newly generated B-mode frame sequence provides the corrected anatomical structure information, while the newly generated I / Q frame sequence retains the corrected signal amplitude and phase information, both of which are consistent with the direction of blood flow.

[0050] By mapping to an inertial coordinate system, spatial deviations caused by angles in the original scanning plane coordinate system are eliminated, allowing accurate analysis of microbubble motion and acoustic responses in a coordinate frame consistent with the direction of blood flow. This consistency provides a unified spatial reference for subsequent time-frequency feature extraction, ensuring a clearer and more reliable correspondence between the spatial distribution of microbubble signals and hemodynamic characteristics.

[0051] S2-3. Generate microbubble time-frequency three-dimensional image tensor: The purpose of generating a three-dimensional time-frequency microbubble image tensor is to extract the temporal and frequency characteristics of the microbubble signal from the mapped I / Q frame sequence to support subsequent signal decoupling and classification analysis. The specific processing process involves performing a time-frequency analysis of the temporal variation of the signal at each spatial location in the I / Q frame sequence, generating a three-dimensional data structure reflecting the signal energy distribution. Time-frequency analysis reveals the resonant characteristics and dynamic behavior of microbubbles by segmenting the time series signal and calculating its spectrum.

[0052] In implementation, a time window is first selected for the signal at each spatial location in the I / Q frame sequence. For example, the signal is truncated using a Hanning window function to reduce edge effects. The spectrum of the signal within each time window is then calculated by decomposing the signal into the energy distribution of its frequency components, generating the spectrum data for that time period. This process is repeated for all time windows, resulting in a sequence of time-varying spectra at each spatial location.

[0053] The spectral sequences of all spatial locations are stacked along the time, frequency, and space dimensions to form a three-dimensional tensor, the microbubble time-frequency 3D image tensor. To improve computational efficiency, the spectral data can be downsampled to retain only the information within the key frequency range relevant to microbubble characteristics.

[0054] For example, spectrum calculation can be performed in the following way: Input data: Corrected I / Q frame sequence .

[0055] Time-Frequency Analysis: For each spatial position The I / Q signal over time Perform a short-time Fourier transform (STFT) of a variation of to extract the frequency response of the microbubbles:

[0056] Corrected I / Q signal; : Window function (such as Hanning window), used to limit the time window; : Time axis, that is, time dimension; : Frequency axis, that is, frequency dimension; : Time-frequency spectrum of the corresponding position.

[0057] The microbubble time-frequency 3D image tensor comprehensively describes the characteristics of the microbubble signal in time, frequency, and space, enabling effective differentiation of the acoustic behavior of different microbubble types and supporting in-depth analysis of microbubble dynamics. Downsampling reduces the data volume while preserving key features, lowering the complexity of subsequent calculations and improving processing efficiency, providing high-quality input data for analysis in the feature extraction module.

[0058] The signal correction module corrects the spatial deviations between the B-mode and I / Q frame sequences through micro-angle resampling and maps them to an inertial coordinate system parallel to the blood flow direction, generating a corrected frame sequence. Furthermore, time-frequency analysis extracts the time-frequency characteristics of microbubbles from the I / Q frame sequence, forming a three-dimensional time-frequency image tensor of microbubbles. This process provides accurate spatially corrected data and comprehensive time-frequency features for the subsequent feature extraction module's sparse phase decoding and complex independent component decomposition, ensuring the accuracy and reliability of the entire analysis process.

[0059] The signal correction module performs micro-angle resampling on the frame sequence based on the angle, maps it to the inertial coordinate system, and generates a microbubble time-frequency three-dimensional image tensor, which comprehensively characterizes the characteristics of the microbubble signal in time, frequency, and space. However, due to the high-speed movement of multi-targeted labeled microbubbles in the blood flow and the diversity of their acoustic responses, there is aliasing of different microbubble signals in the time domain, frequency domain, and spatial domain in the microbubble time-frequency three-dimensional image tensor, which significantly increases the difficulty of signal separation and classification. The feature extraction module aims to decode and decompose the signal through advanced signal processing technology, extract key feature vectors, and use pre-trained models to achieve preliminary classification and identification of microbubbles, providing accurate input data for the label generation module.

[0060] The feature extraction module includes the following: S3-1. Sparse phase decoding: The goal of sparse phase decoding is to separate the phase characteristics of microbubble signals from the microbubble time-frequency 3D image tensor to address signal aliasing. Microbubble phase information reflects its acoustic response characteristics, and sparse coding techniques can effectively decouple overlapping signals in the phase domain. In specific implementation, phase information is first extracted from the microbubble time-frequency 3D image tensor to generate a phase tensor. This phase tensor is generated by extracting the phase component of each complex signal point in the microbubble time-frequency 3D image tensor. Next, a sparse coding algorithm is applied to decompose the phase tensor to generate a sparse phase matrix. The sparse coding algorithm learns the phase characteristics from historical microbubble data and constructs a phase dictionary matrix containing a set of basis vectors for the phase characteristics. The sparse phase matrix then contains sparse coefficients for the phase characteristics on these basis vectors, indicating that the phase characteristics are represented by the weighted linear combinations of the basis vectors. This decomposition method sparsifies the phase information, facilitating the separation of the phase characteristics of different microbubbles.

[0061] Sparse phase decoding utilizes sparse coding techniques to decouple signals in the phase domain, effectively mitigating aliasing of microbubble signals in the time-frequency domain. This approach improves the discriminability of phase features and provides accurate input data for subsequent extraction of the resonant differential entropy vector, ensuring the reliability and accuracy of feature extraction. By constructing a phase dictionary matrix and generating a sparse phase matrix, the phase features of the microbubble signal are clearly separated.

[0062] S3-2. Complex independent component decomposition: The purpose of complex independent component decomposition is to separate statistically independent signal components from the microbubble time-frequency three-dimensional image tensor and solve the signal aliasing problem in the complex domain. Microbubble signals are expressed in complex form in the time-frequency domain, and complex independent component decomposition can effectively separate the acoustic responses of different microbubbles. In specific implementations, the microbubble time-frequency three-dimensional image tensor is regarded as a multi-channel complex signal, where each time-frequency-space point is regarded as a sample. The complex independent component decomposition algorithm is applied to decompose the microbubble time-frequency three-dimensional image tensor to generate an independent component matrix. The independent component matrix contains the separated independent signal components, each component corresponding to a statistically independent microbubble signal source. Through this decomposition method, the signals of different microbubbles are separated into different independent components, providing a basis for the subsequent extraction of the transient phase confluence degree vector.

[0063] Complex independent component decomposition effectively resolves the aliasing problem of microbubble signals in the complex domain by separating statistically independent signal components. This method improves signal purity, provides a clear signal source for subsequent phase feature analysis, and ensures accurate and effective feature extraction. The generation of an independent component matrix demonstrates the independence of microbubble signals, providing high-quality input data for subsequent steps.

[0064] S3-3. Feature extraction: The goal of feature extraction is to extract the resonant differential entropy vector and the transient phase confluence vector from the sparse phase matrix and independent component matrix to characterize the resonant characteristics and phase stability of the microbubbles. In practice, feature extraction is divided into two parts, processing the resonant differential entropy vector and the transient phase confluence vector respectively.

[0065] Extraction of resonance differential entropy vector: Phase differences are calculated along the frequency dimension from the sparse phase matrix. The specific process is to calculate the phase change between adjacent frequency points. Next, the phase difference values are statistically analyzed to calculate their entropy. The entropy value is calculated by estimating the probability distribution of the phase differences and measuring the change of the phase characteristics with frequency based on the complexity of the probability distribution. Higher entropy values indicate more complex phase changes, reflecting the resonant characteristics of the microbubbles. For example, it can be obtained using the following method: From the sparse phase matrix Mid-edge frequency dimension Calculate the phase difference:

[0066] in Indicates the phase change between adjacent frequency points.

[0067] Calculate the entropy value after difference to measure the complexity of the phase feature change in frequency:

[0068] : The probability distribution of is estimated by the normalized histogram.

[0069] Output: Resonant differential entropy vector , characterizing the resonance characteristics of microbubbles.

[0070] Extraction of transient phase confluence degree vector: Extract the phase sequence of each independent component from the independent component matrix. The specific process is to obtain the phase value of each independent component over time. Next, calculate the degree of temporal convergence of the phase sequence, which is defined as the transient phase convergence. The transient phase convergence is calculated by measuring the amplitude of the phase value change within the time window. The smaller the amplitude of change, the higher the convergence, indicating the temporal stability of the phase. For example, the following method is used to obtain it: From the independent component matrix Extract the phase sequence of each independent component ,in Indicates the extraction of the phase component of the complex signal, Indicates the index of the independent component.

[0071] Calculate the degree of convergence of the phase sequence over time and define the transient phase convergence degree:

[0072] : No. Independent components in the time dimension phase.

[0073] : Neighboring time points within the time window.

[0074] :Small positive number (such as ), to prevent the denominator from being zero.

[0075] Output: Transient phase confluence degree vector , characterizing the spatiotemporal stability of the microbubble signal phase.

[0076] The resonant differential entropy vector characterizes the resonant characteristics of microbubbles by quantifying the complexity of phase variations over frequency, helping to distinguish different types of microbubbles. The transient phase confluence vector reflects the dynamic behavior of the microbubble signal by measuring the temporal stability of the phase, providing a key feature for microbubble classification and identification.

[0077] S3-4. Pre-trained microbubble fingerprint model: The purpose of the pre-trained microbubble fingerprint model is to use deep learning technology to analyze the resonant differential entropy vector and the transient phase confluence vector to generate a microbubble uniqueness index and a residual microbubble fingerprint to achieve preliminary classification and identification of microbubbles. In specific implementation, the pre-trained microbubble fingerprint model is a deep convolutional neural network whose input is the resonant differential entropy vector and the transient phase confluence vector, and whose output is the microbubble uniqueness index and the residual microbubble fingerprint. The model is pre-trained with historical microbubble data to learn the mapping relationship between feature vectors and microbubble types. During the inference phase, the extracted resonant differential entropy vector and the transient phase confluence vector are input into the pre-trained microbubble fingerprint model to generate the microbubble uniqueness index and the residual microbubble fingerprint.

[0078] The microbubble uniqueness index (MUI) is an indicator used to quantify the uniqueness of microbubble signals, primarily used in ultrasound imaging of multi-targeted microbubbles. This index evaluates the degree to which a microbubble signal can be distinguished from other microbubble signals in a specific dataset by analyzing the acoustic properties of microbubbles, such as their resonance characteristics and phase characteristics. A high MUI indicates that the acoustic characteristics of the microbubble are highly unique, facilitating accurate identification and classification in complex signal environments. Conversely, a low MUI indicates that the characteristics of the microbubble are similar to those of other microbubbles, increasing the difficulty of identification and classification. This indicator provides an important quantitative basis for subsequent signal processing and label generation.

[0079] Residual microbubble fingerprints refer to microbubble feature vectors that are not fully identified or classified after the pre-trained microbubble fingerprint model performs preliminary classification of the microbubble signal. During the classification process, some microbubble features may not be accurately processed due to signal aliasing, noise interference, or model limitations. These unclassified features are retained as residual microbubble fingerprints. Residual microbubble fingerprints contain underutilized information from the original signal and can serve as input data for subsequent analysis, such as advanced modules like attention networks, further improving the accuracy of multi-target microbubble label map generation and overall classification accuracy.

[0080] The pre-trained microbubble fingerprint model leverages deep learning technology to efficiently analyze feature vectors, enabling preliminary microbubble classification and identification. The generation of a unique microbubble identification index and residual microbubble fingerprint ensures accurate generation of multi-target microbubble signature maps. This improves the automation and accuracy of microbubble classification and reduces the need for manual intervention.

[0081] The feature extraction module extracts the resonant differential entropy vector and transient phase confluence vector from the microbubble time-frequency 3D image tensor through sparse phase decoding and complex independent component decomposition. It then uses a pre-trained microbubble fingerprint model to generate a microbubble uniqueness index and residual microbubble fingerprint. These outputs effectively separate the characteristics of the microbubble signal, providing precise input to the attention network of the label generation module, supporting further classification of multi-target microbubbles and label map generation.

[0082] The feature extraction module performs sparse phase decoding and complex independent component decomposition on the image tensor, extracting the resonant differential entropy vector and the transient phase confluence degree vector. It then generates a unique microbubble fingerprint and a residual microbubble fingerprint using a pre-trained microbubble fingerprint model. However, due to the complexity of multi-targeted labeled microbubbles and the influence of signal aliasing, the output of the feature extraction module still requires further processing to improve classification accuracy. Based on this, the label generation module analyzes the residual microbubble fingerprint and the unique microbubble fingerprint using an attention network, and generates a multi-targeted microbubble label map based on the resonance similarity and attenuation slope, providing accurate classification results for subsequent steps.

[0083] The label generation module includes the following: S4-1. Input data preparation: The residual microbubble fingerprint and microbubble uniqueness index generated by the feature extraction module are pre-normalized to ensure a consistent numerical scale for subsequent attention network processing. After normalization, the normalized residual microbubble fingerprint and microbubble uniqueness index are combined into an input feature matrix. The number of rows in the input feature matrix corresponds to the number of microbubbles, and the number of columns corresponds to the sum of the feature dimensions of the residual microbubble fingerprint and microbubble uniqueness index.

[0084] S4-2. Attention network processing: The attention network utilizes a multi-head self-attention mechanism to improve feature representation by adaptively learning different representations of the input feature matrix. In practice, the attention network consists of multiple attention heads, each of which independently learns a different feature representation of the input feature matrix. First, a self-attention operation is performed on the input feature matrix to generate a query matrix, a key matrix, and a value matrix. This is achieved by multiplying the input feature matrix by a pre-trained linear transformation matrix, respectively. Next, the attention output is computed by multiplying the query matrix by the transpose of the key matrix to generate a dot product matrix. Each element of the dot product matrix is then divided by the square root of the key matrix dimension to scale the values. The scaled dot product matrix is then normalized using a softmax function to generate an attention weight matrix. Finally, the attention weight matrix is multiplied by the value matrix to generate the single-head attention output. The outputs of multiple attention heads are concatenated and processed through a linear transformation layer to generate the final attention feature matrix.

[0085] The multi-head self-attention mechanism, through parallel learning across multiple attention heads, captures global dependencies within the input feature matrix, thereby improving feature representation. Adaptively learning different feature representations enables the attention network to effectively process complex microbubble signal characteristics, enhancing classification accuracy and robustness. Furthermore, by focusing on key features, the attention mechanism improves the network's ability to identify microbubble types in mixed signals.

[0086] S4-3. Resonance similarity and attenuation slope extraction: The purpose of extracting resonance similarity and attenuation slope is to obtain key information related to microbubble resonance characteristics and signal attenuation from the attention feature matrix, supporting subsequent classification and label generation. In practice, a subset of features related to the resonance characteristics, referred to as the resonance feature subset, is first extracted from the attention feature matrix. This extraction method uses predefined feature indices or network layer outputs to obtain feature components directly related to the resonance characteristics. Next, resonance similarity is calculated by calculating the square of the Euclidean distance between the resonance feature vectors of each pair of microbubbles. This squared Euclidean distance is then converted into a similarity value using a Gaussian kernel function, whose parameters control the decay rate of the similarity. Similarly, a subset of features related to signal attenuation is extracted from the attention feature matrix, referred to as the attenuation feature subset. The attenuation slope is calculated by calculating the temporal change of the attenuation feature subset and then dividing this change by the corresponding time interval, which is provided by the spatiotemporal index matrix generated by the data acquisition module.

[0087] Extracting resonance similarity and decay slope provides quantitative metrics of microbubble acoustic properties, helping to distinguish different types of microbubbles. Resonance similarity uses a Gaussian kernel function to convert distances into similarity values, ensuring smooth and robust calculations and avoiding misclassifications caused by noise. The decay slope reflects the temporal evolution of the microbubble signal, providing an important basis for dynamic characteristic analysis.

[0088] S4-4. Generation of multi-targeted microbubble labeling maps: The purpose of generating a multi-targeted microbubble label map is to classify microbubbles based on resonance similarity and attenuation slope, and to generate a spatial distribution map. In the specific implementation, the microbubbles are first classified by a clustering algorithm based on the resonance similarity and attenuation slope. The clustering algorithm classifies microbubbles with similar resonance characteristics and attenuation slopes into the same category and assigns a classification label to each microbubble. Then, based on the classification label and the spatial position information of the microbubble, a multi-targeted microbubble label map is generated. The multi-targeted microbubble label map marks the specific position of each microbubble in space and the category to which it belongs. Finally, the multi-targeted microbubble label map is output as the input data for the frequency attenuation plane projection in the trajectory tracking module.

[0089] A multi-targeted microbubble signature map is generated using a clustering algorithm, enabling automated microbubble classification and spatial localization. Clustering algorithms can process high-dimensional feature data, ensuring accurate and consistent classification results. The resulting multi-targeted microbubble signature map provides a precise classification basis for subsequent steps, supporting applications such as frequency attenuation plane projection and microbubble trajectory tracking.

[0090] The label generation module uses an attention network to deeply process the residual microbubble fingerprint and microbubble uniqueness index generated by the feature extraction module. It then combines resonance similarity and attenuation slope to generate a multi-target microbubble label map. This multi-target microbubble label map provides accurate microbubble classification and spatial positioning information for the trajectory tracking module's frequency attenuation plane projection, supporting subsequent microbubble trajectory tracking and density curve plotting.

[0091] The label generation module inputs the residual microbubble fingerprint and microbubble uniqueness index into the attention network to generate a multi-target microbubble label map. However, to achieve dynamic microbubble tracking and density analysis, the trajectory tracking module needs to further process the residual microbubble fingerprint. This process projects the fingerprint onto a frequency attenuation plane, calculates the frequency attenuation slope, and uses a particle filter to track the microbubble trajectory. Finally, a transient microbubble density curve is plotted to provide the data transmission module with information on the dynamic distribution and concentration changes of microbubbles.

[0092] The trajectory tracking module includes the following: S5-1. Multi-targeted microbubble labeling map projected onto the frequency attenuation plane: The purpose of projecting the multi-targeted microbubble signature map onto the frequency attenuation plane is to map the time-frequency characteristics of the microbubbles into the space of frequency and attenuation rate, so as to analyze the acoustic properties of the microbubbles. The frequency attenuation plane is a two-dimensional coordinate system with the horizontal axis representing frequency and the vertical axis representing attenuation rate. In a specific implementation, the coordinate axes of the frequency attenuation plane are first defined, with the horizontal axis representing frequency and the vertical axis representing attenuation rate. Next, for each microbubble in the multi-targeted microbubble signature map, the corresponding time-frequency characteristics are extracted from the microbubble time-frequency three-dimensional image tensor generated by the signal correction module based on its category and spatial position. After extraction, the frequency attenuation feature point of each microbubble is calculated. The frequency attenuation feature point consists of the peak frequency and the attenuation rate at the peak frequency. The peak frequency is determined by finding the maximum energy point on the time-frequency spectrum. The specific method is to compare the energy values of each frequency point in the time-frequency spectrum and select the frequency point with the maximum energy value as the peak frequency. The attenuation rate is calculated by analyzing the energy attenuation trend of the time-frequency spectrum over time. The specific method is to take the time series at the peak frequency, calculate the amplitude of the energy reduction over time, and then divide it by the time interval to obtain the attenuation rate. Finally, the frequency attenuation characteristic points of all microbubbles are marked on the frequency attenuation plane to generate a frequency attenuation distribution map.

[0093] The processing method of projecting the multi-targeted microbubble label map onto the frequency attenuation plane can intuitively display the distribution characteristics of microbubbles in frequency and attenuation characteristics, which helps to identify the acoustic behavior of different types of microbubbles.

[0094] S5-2. Calculate the frequency attenuation slope: The purpose of calculating the frequency attenuation slope is to quantify the distribution trend of microbubbles on the frequency attenuation plane and reflect the relationship between the frequency and attenuation rate of the microbubble signal. The frequency attenuation slope is defined as the ratio of the change in attenuation rate to the change in frequency in the frequency attenuation distribution graph. In the specific implementation, the maximum and minimum values of the attenuation rate, as well as the maximum and minimum values of the frequency in the frequency attenuation distribution graph, are first determined. The specific method is to traverse all the frequency attenuation feature points in the frequency attenuation distribution graph and find the maximum and minimum values of the attenuation rate and frequency respectively. Next, the change in attenuation rate is calculated by subtracting the minimum value from the maximum value of the attenuation rate; the change in frequency is calculated by subtracting the minimum value from the maximum value of the frequency. Then, the change in attenuation rate is divided by the change in frequency to obtain the frequency attenuation slope. In addition, in order to more carefully analyze the characteristics of different categories of microbubbles, the above calculation process is performed on different categories of microbubbles in the multi-target microbubble label graph to obtain category-specific frequency attenuation slopes.

[0095] The calculated frequency attenuation slope provides a quantitative index of the acoustic characteristics of microbubbles and helps to distinguish different types of microbubbles.

[0096] S5-3. Particle filter tracking microbubble trajectory: The purpose of particle filtering to track microbubble trajectories is to achieve dynamic tracking of microbubbles in the bloodstream and generate microbubble motion trajectories. Particle filtering is a tracking method based on probability estimation and is suitable for analyzing complex motion states. In specific implementation: First, a set of particles is initialized for each microbubble in the multi-targeted microbubble tag map. Each particle represents a possible position of the microbubble at the current time point. The initial position is determined based on the spatial information in the multi-targeted microbubble tag map.

[0097] Next, based on the blood flow velocity and the motion characteristics of the microbubbles, the position of the particles at the next time point is predicted. The specific method is to calculate the displacement of each particle within the time interval according to the direction and magnitude of the blood flow velocity and update its position.

[0098] Then, the spatial position information of the microbubbles in the multi-targeted microbubble label map is used to update the weight of each particle. The weight reflects the proximity of the particle to the actual observed position. The calculation method is to compare the distance between the predicted position of the particle and the observed position. The closer the distance, the higher the weight.

[0099] Afterwards, through the resampling process, particles with high weights are retained, particles with low weights are eliminated, and a new set of particles is regenerated according to the weights, so that the particle distribution is closer to the actual position.

[0100] Through the above processing of multiple frames, the motion trajectory of the microbubble is generated, and the position sequence of each microbubble changing with time is recorded.

[0101] The particle filter-based microbubble trajectory tracking method simulates the movement and observation of microbubbles, enabling dynamic tracking of microbubbles in complex blood flow environments. This method predicts and updates particle positions, handles nonlinear motion and noise interference, and provides more accurate trajectory estimation. The generated microbubble trajectory provides critical data for subsequent density analysis and concentration time series generation, ensuring dynamic and real-time analysis.

[0102] S5-4. Draw the transient microbubble density curve: The purpose of drawing the transient microbubble density curve is to show the distribution changes of microbubbles in space and time, and to provide an intuitive basis for analyzing the dynamic behavior of microbubbles. The transient microbubble density curve reflects the change of the number of microbubbles per unit space per unit time over time. In the specific implementation, First, based on the microbubble motion trajectory, the number of microbubbles distributed in space at each time point is counted. The specific method is to divide the space into multiple unit areas and count the number of microbubbles in each unit area.

[0103] Next, the transient microbubble density was calculated, which is defined as the number of microbubbles per unit space per unit time. The calculation method is to divide the number of microbubbles per unit area by the spatial volume of the area and the time interval to obtain the density value.

[0104] Finally, a transient microbubble density curve is drawn with time as the horizontal axis and transient microbubble density as the vertical axis. The specific method is to connect the density values of each time point into a curve in chronological order.

[0105] The transient microbubble density curve process visualizes the dynamic distribution of microbubbles, helping researchers quickly understand their movement and aggregation within the bloodstream. It also reveals the aggregation trends of microbubbles in specific areas, providing important insights for clinical diagnosis and treatment decisions, and enhancing the practicality and scientific validity of the analysis results.

[0106] The trajectory tracking module projects the multi-targeted microbubble label map generated by the label generation module onto the frequency attenuation plane to generate a frequency attenuation distribution map, calculate the frequency attenuation slope, and use particle filtering to generate microbubble motion trajectories, ultimately plotting a transient microbubble density curve. These outputs provide the data transmission module with dynamic microbubble distribution characteristics and density change data, enabling the generation of a targeted microbubble concentration time series.

[0107] The trajectory tracking module projects the multi-targeted microbubble label map onto the frequency attenuation plane, calculates the frequency attenuation slope, and generates microbubble motion trajectories through particle filtering, plotting the transient microbubble density curve. However, to achieve quantitative analysis of targeted microbubble concentration and support real-time clinical decision-making, the data transmission module uses the transient microbubble density curve and frequency attenuation slope output by the trajectory tracking module to generate a targeted microbubble concentration time series table and transmit it to the imaging and treatment terminals via a data interface.

[0108] The data transmission module includes the following: S6-1. Combining transient microbubble density curve and frequency decay slope: The purpose of combining the transient microbubble density curve with the frequency decay slope is to correct the transient microbubble density curve using the frequency decay slope, generating a corrected density curve to improve the accuracy of subsequent concentration calculations. In practice, the transient microbubble density curve and corresponding frequency decay slope are first extracted for each category of microbubbles classified by category in the trajectory tracking module. Next, the frequency decay slope is used to correct the transient microbubble density curve to generate a corrected density curve. This correction is accomplished using a correction factor, calculated by comparing the frequency decay slope of each category with a reference frequency decay slope to determine the ratio between the two. This ratio is then added to a reference value of one to obtain the correction factor. The reference frequency decay slope is determined by preset microbubble acoustic properties or historical data and is used to standardize the attenuation differences between different microbubble categories. The corrected density curve is generated by multiplying the value of the transient microbubble density curve at each time point by the corresponding correction factor to obtain the corrected density curve value at each time point.

[0109] The transient microbubble density curve is adjusted by the correction factor to compensate for the density deviation caused by differences in acoustic attenuation characteristics and improve the accuracy of the density data.

[0110] S6-2. Generate a time series table of targeted microbubble concentrations: The purpose of generating a targeted microbubble concentration time series table is to convert the corrected density curve into microbubble concentration values, generating time-varying concentration data that provides a quantitative basis for real-time clinical monitoring of microbubble distribution. In practice, the targeted microbubble concentration is first calculated based on the corrected density curve. The calculation method is to multiply the value of the corrected density curve at each time point by a concentration conversion coefficient to obtain the concentration value at each time point. The concentration conversion coefficient is determined by comprehensively considering the physical properties of the microbubbles, ultrasound equipment parameters, and the influence of the blood flow environment, and is calculated using experimental data or theoretical models. Specifically, the concentration conversion coefficient is related to parameters such as microbubble size, shell material, ultrasound frequency, ultrasound power, blood flow velocity, and blood viscosity. These parameters are determined to fixed values through preliminary experiments or simulation analysis. Next, the concentration value is calculated for each microbubble type, and the results are summarized to form a targeted microbubble concentration time series table. The specific format of the targeted microbubble concentration time series table is a two-dimensional table, with rows representing different time points and columns representing microbubble types and corresponding concentration values. The concentration changes of each microbubble type are recorded using time as the index.

[0111] S6-3. Data interface transmission: The purpose of data interface transmission is to transmit the generated targeted microbubble concentration time series table to the imaging end and the treatment end to ensure the real-time and integrity of the data to support clinical applications.

[0112] First, the targeted microbubble concentration time series table is converted into a standard data structure, such as a comma-separated value format or a key-value pair format, to meet the parsing requirements of the imaging end and the treatment end. The conversion method is: each time point in the targeted microbubble concentration time series table and its corresponding microbubble category and concentration value are organized into a record unit containing a timestamp, category identifier and concentration value. Then, the converted data is transmitted to the imaging end and the treatment end through a network interface or a file sharing protocol. The transmission process adopts a data check and retransmission mechanism to ensure data integrity. The specific method is: a check code is added to each data packet, and the check code is generated by a specific calculation method of the data content; after the receiving end receives the data, the check code is recalculated and compared with the check code provided by the sending end. If the two are inconsistent, the sending end is requested to retransmit the data packet until the check is consistent.

[0113] The data transmission module, based on the transient microbubble density curve and frequency decay slope generated by the trajectory tracking module, completes the generation and transmission of a targeted microbubble concentration time series table through three sub-steps: combining the transient microbubble density curve and frequency decay slope, generating a targeted microbubble concentration time series table, and transmitting the data through the data interface. This targeted microbubble concentration time series table provides critical quantitative data for real-time clinical monitoring of microbubble distribution and optimizing treatment plans.

[0114] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0115] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, and can also be run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0116] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0117] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0118] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. Ultrasound contrast agent microbubble detection and analysis system based on intelligent perception, characterized by: The method comprises the following steps: a data acquisition module, a signal correction module, a feature extraction module, a label generation module, a trajectory tracking module and a data transmission module; Data acquisition module: The broadband pulse triggers the ultrasound probe to collect B-mode and I / Q frame sequences carrying multi-targeted labeled microbubbles and writes them into the scanning plane-blood flow angle and time-space index matrix; Signal correction module: performs micro-angle resampling on the frame sequence according to the included angle, maps it to the inertial coordinate system and generates a microbubble time-frequency three-dimensional image tensor; Feature extraction module: performs sparse phase decoding and complex independent component decomposition on the image tensor, extracts the resonant differential entropy vector and the transient phase confluence degree vector, inputs the pre-trained microbubble fingerprint model, and outputs the microbubble unique identification index and residual microbubble fingerprint; Label generation module: The residual microbubble fingerprint and microbubble unique identification index are input into the attention network, and a multi-target microbubble label map is generated based on the resonance similarity and attenuation slope; Trajectory tracking module: projects the label map onto the frequency attenuation plane, calculates the frequency attenuation slope, calls the particle filter to track the microbubble trajectory and draws the transient microbubble density curve; Data transmission module: Combine the microbubble density curve and frequency attenuation slope to generate a targeted microbubble concentration time series table, which is transmitted to the imaging end and the treatment end through the data interface.

2. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 1, characterized in that: The data acquisition module includes the following: The B-mode frame sequence provides anatomical information of the tissue, the I / Q frame sequence records the in-phase and quadrature components of the signal to preserve amplitude and phase information, and synchronously measures the angle between the scanning plane and the blood flow direction. The angle between the scanning plane and the blood flow direction is obtained using ultrasound technology to correct the spatial projection deviation of the signal. A spatiotemporal index is assigned to each frame of data. The spatiotemporal index includes a timestamp and spatial coordinates within the image plane. A spatiotemporal index matrix is constructed to record the angle value between the scanning plane and the blood flow direction at the corresponding position.

3. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 2, characterized in that: The signal correction module includes the following: Micro-angle resampling is used to correct the spatial projection deviation of the B-mode frame sequence and the I / Q frame sequence caused by the angle between the scanning plane and the blood flow direction. Micro-angle resampling performs coordinate transformation on each pixel point in the B-mode frame sequence and the I / Q frame sequence based on the angle between the scanning plane and the blood flow direction in the spatiotemporal index matrix, and calculates the corrected pixel value through bilinear interpolation. The corrected B-mode frame sequence and I / Q frame sequence are mapped to an inertial coordinate system parallel to the blood flow direction to generate the corrected B-mode frame sequence and I / Q frame sequence, and the microbubble time-frequency three-dimensional image tensor is extracted from the corrected I / Q frame sequence.

4. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 3, characterized in that: The signal correction module also includes the following: The microbubble time-frequency three-dimensional image tensor is generated by performing time-frequency analysis on the I / Q signals at each spatial position. The time-frequency analysis uses short-time Fourier transform and a window function to truncate the I / Q signals to reduce edge effects. The time-frequency analysis results are stacked into the microbubble time-frequency three-dimensional image tensor.

5. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 4, characterized in that: The feature extraction module includes the following: Sparse phase decoding and complex independent component decomposition are performed on the microbubble time-frequency three-dimensional image tensor. Sparse phase decoding extracts the phase information of the microbubble time-frequency three-dimensional image tensor and applies a sparse coding algorithm to generate a sparse phase matrix. Complex independent component decomposition decomposes the microbubble time-frequency three-dimensional image tensor into independent component matrices.

6. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 5, characterized in that: The feature extraction module also includes the following: The resonant differential entropy vector is extracted from the sparse phase matrix. The resonant differential entropy vector characterizes the resonant characteristics of the microbubble by calculating the entropy value of the phase difference. The transient phase confluence vector is extracted from the independent component matrix. The transient phase confluence vector characterizes the phase stability of the microbubble signal by calculating the degree of aggregation of the independent component phase sequence. The resonant differential entropy vector and the transient phase confluence vector are input into the pre-trained microbubble fingerprint model to output the microbubble uniqueness index and residual microbubble fingerprint.

7. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 6, characterized in that: The label generation module includes the following: Normalization processing is performed on the residual microbubble fingerprint and the microbubble unique identification index to generate the normalized residual microbubble fingerprint and the normalized microbubble unique identification index, which are then merged into an input feature matrix. The input feature matrix is processed by a multi-head self-attention network to generate an attention feature matrix. The resonance feature subset and the attenuation feature subset are extracted from the attention feature matrix. The resonance similarity is calculated based on the resonance feature subset, and the attenuation slope is calculated based on the attenuation feature subset. The resonance similarity and the attenuation slope are combined to generate a multi-target microbubble label map through a clustering algorithm.

8. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 7, characterized in that: Trajectory tracking module Includes the following: The multi-targeted microbubble label map is projected onto the frequency attenuation plane to generate a frequency attenuation distribution map. The frequency attenuation distribution map is determined by extracting the peak frequency of the microbubbles and the attenuation rate at the peak frequency in the multi-targeted microbubble label map. The frequency attenuation slope is calculated. The frequency attenuation slope is determined by the ratio of the change in the attenuation rate to the change in the frequency in the frequency attenuation distribution map. Category-specific frequency attenuation slopes are calculated for different categories of microbubbles. Based on the spatial position information and blood flow velocity in the multi-targeted microbubble label map, particle filtering is used to track the microbubble trajectory to generate the microbubble motion trajectory.

9. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 8, characterized in that: The trajectory tracking module also includes the following: The particle filter generates microbubble motion trajectories by initializing particle sets, predicting particle positions, updating particle weights, and resampling. Based on the microbubble motion trajectories, the number of microbubbles distributed in space at each time point is counted to calculate the transient microbubble density. The transient microbubble density is the number of microbubbles per unit space per unit time. The transient microbubble density curve is drawn, which shows the dynamic distribution characteristics of microbubbles with time as the horizontal axis and transient microbubble density as the vertical axis.

10. The ultrasonic contrast agent microbubble detection and analysis system based on intelligent perception according to claim 9, characterized in that: The data transmission module includes the following: A targeted microbubble concentration time series table is generated based on the transient microbubble density curve and the frequency decay slope, specifically including extracting the transient microbubble density curve and the frequency decay slope for microbubbles classified by category, correcting the transient microbubble density curve using the frequency decay slope to generate a corrected density curve, calculating the corrected density curve by multiplying the value of the transient microbubble density curve at each time point by the correction factor, calculating the targeted microbubble concentration based on the corrected density curve, the targeted microbubble concentration being the product of the value of the corrected density curve at each time point and the concentration conversion coefficient, calculating the concentration value for each category of microbubbles separately and summarizing them to form a targeted microbubble concentration time series table; the targeted microbubble concentration time series table records the concentration value of each category of microbubbles using time as an index, converting the targeted microbubble concentration time series table into a standard data structure, and transmitting it to the imaging end and the treatment end through the data interface.

Images