Intelligent monitoring analysis method for cardiology department pressing hemostasis device

Abstract

The invention discloses an intelligent monitoring and analyzing method for a cardiology department pressing hemostasis device. The method comprises the steps that the pressing hemostasis device collects blood vessel surface pressure, blood flow velocity and blood vessel wall vibration signals of a pressing area; after signal interference is removed through a blood flow signal noise reduction enhancement algorithm, a blood vessel pressure stress feedback prediction model is called, and real-time pressure stress in the blood vessel is calculated in combination with pressure and the processed blood flow signals; performing time-frequency domain feature extraction and fusion on the blood flow, the vibration signal and the pressure stress value by adopting a biological signal time-frequency fusion algorithm to generate a multi-dimensional fusion feature vector; inputting the vector into an AI auxiliary hemostasis decision support platform, and analyzing and outputting a pressing pressure adjustment suggestion, a duration evaluation result and a hemostasis effect index through a deep learning model; the device execution module receives the suggestion to adjust the pressing pressure, closed-loop monitoring is formed, intelligent monitoring and accurate control over pressing hemostasis are achieved, and operation safety and effectiveness are improved.

Description

Technical Field

[0001] This invention relates to the field of hemostasis monitoring technology in cardiology, and more particularly to an intelligent monitoring and analysis method for cardiology pressure hemostasis devices. Background Technology

[0002] In the diagnosis and treatment of cardiovascular diseases, cardiology compression hemostasis is a crucial procedure for preventing bleeding complications after interventional treatment. Precise control of compression pressure, duration, and hemostatic effect directly impacts postoperative recovery. Traditional compression hemostasis relies heavily on the experience and judgment of medical staff, requiring real-time monitoring of parameters such as vascular status and blood flow in the compression area. However, manual monitoring suffers from response lag and incomplete parameter collection, making it difficult to dynamically adapt to individual differences in vascular elasticity and coagulation function among patients. With the development of intelligent medical technology, the clinical demand for automated and precise monitoring of compression hemostasis devices is increasingly urgent. There is a pressing need to achieve intelligent monitoring and decision support for the compression hemostasis process through multi-signal collaborative acquisition, algorithmic model analysis, and closed-loop control, thereby improving the safety and effectiveness of hemostasis operations.

[0003] Existing monitoring technologies for compression hemostasis in cardiology have two significant drawbacks. Firstly, in signal processing, most technologies can only perform simple filtering on single-type signals, failing to effectively fuse multi-dimensional signals such as blood flow, vascular vibration, and pressure. Furthermore, they struggle to remove the influence of complex interference factors on signal quality, resulting in insufficient accuracy of signals used for subsequent analysis and an inability to truly reflect critical states such as intravascular pressure stress. Secondly, at the decision support level, existing technologies lack intelligent analysis models based on multi-signal fusion characteristics. They cannot dynamically output compression pressure adjustment suggestions and hemostasis effect judgments based on real-time monitoring data, still relying on human experience for decision-making. This leads to untimely and inaccurate adjustments to compression parameters, easily resulting in situations where excessive compression pressure causes vascular damage or insufficient pressure leads to hemostasis failure. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides an intelligent monitoring and analysis method for a cardiology compression hemostasis device.

[0005] The technical solution adopted in this invention is an intelligent monitoring and analysis method for a cardiology compression hemostasis device, comprising the following steps: S1, acquiring pressure signals on the surface of blood vessels in the compression area through the pressure sensing module of the cardiology compression hemostasis device, and simultaneously acquiring blood flow velocity signals and blood vessel wall vibration signals corresponding to that area through a biosignal acquisition module; S2, transmitting the pressure signals, blood flow velocity signals, and blood vessel wall vibration signals acquired in S1 to a signal processing module, and processing the blood flow velocity signals using a blood flow signal noise reduction and enhancement algorithm to remove power frequency interference, electromyographic interference, and random noise from the signals; S3, calling a preset vascular pressure stress feedback prediction model, using the blood flow velocity signals processed in S2 and the pressure signals acquired in S1 as model inputs, and calculating the blood vessel pressure stress during the compression process. S4. Using a biosignal time-frequency fusion algorithm, the blood flow velocity signal processed in S2, the blood vessel wall vibration signal, and the internal compressive stress value obtained in S3 are subjected to time-frequency domain feature extraction and fusion to generate a multi-dimensional fusion feature vector; S5. The multi-dimensional fusion feature vector generated in S4 is input into the AI-assisted hemostasis decision support platform. The platform analyzes the fusion feature vector through a trained deep learning model and outputs pressure adjustment suggestions, pressure duration evaluation results, and hemostasis effect judgment indicators; S6. The execution module of the cardiology compression hemostasis device receives the pressure adjustment suggestions output in S5, adjusts the pressure output value of the compression component in real time, and the monitoring module continuously collects the adjusted data and repeats S2-S5 to form a closed-loop monitoring and analysis process.

[0006] Furthermore, the expression for the vascular compressive stress feedback prediction model is as follows: ,in Let be the compressive stress value in the j-th direction at the i-th monitoring point inside the blood vessel at time t. , , These are the model weight coefficients. The pressure signal value on the surface of the blood vessel in the pressed area at time t. The processed blood flow velocity signal value at time t. , These are the model fitting coefficients. For integration variables, Let be the second-order Laplace operator for the blood vessel surface pressure signal at time t. It is an integral variable The blood flow velocity signal value after processing at any given time. It is an integral variable The pressure signal value on the surface of the blood vessels in the pressing area at all times.

[0007] Furthermore, the expression for the biosignal time-frequency fusion algorithm is as follows: ,in, Frequency at time t The time-frequency characteristic value of multi-signal fusion at point n, where n is the number of biological signal types involved in the fusion. The fusion weights for the k-th type of biological signal are... For the kth type of biological signal, This is a blood flow velocity signal. This is a vibration signal from the blood vessel wall. It is a short-time Fourier transform operator. For wavelet transform operators, For time-frequency domain coupling coefficients, This represents the real-time compressive stress value inside the blood vessel. For continuous wavelet transform operators, This is the Hilbert transform operator.

[0008] Furthermore, the expression for the blood flow signal noise reduction and enhancement algorithm is as follows: ,in, This is the enhanced blood flow velocity signal after noise reduction. The original blood flow velocity signal is represented by M, which is the number of intrinsic mode functions obtained from empirical mode decomposition. The filter coefficients are for the m-th intrinsic mode function. Let m be the m-th eigenmode function. These are the frequency domain filter strength coefficients. For signal frequency, This is the power frequency interference frequency. For rectangular window functions, , These are the Fast Fourier Transform and Inverse Fourier Transform operators, respectively.

[0009] Furthermore, the output layer expression of the deep learning model in the AI-assisted hemostasis decision support platform is as follows: ,in The model output vector includes the values ​​corresponding to the suggested compression pressure, the evaluation results of the compression duration, and the indicators for judging hemostasis effectiveness. , , These are the weight matrices from the model input layer to hidden layer 1, from hidden layer 1 to hidden layer 2, and from hidden layer 2 to the output layer, respectively. , , These are the bias vectors for the corresponding layers. To fuse feature vectors from multiple dimensions, The activation function for the rectified linear unit is... This is the softmax activation function.

[0010] Furthermore, the pressure adjustment control expression of the execution module of the cardiology compression hemostasis device is as follows: ,in Let t be the pressing force value output by the module at time t. Let t be the current pressing pressure value of the module at time t. The target compression pressure value output by the AI-assisted hemostasis decision support platform. , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. For integration variables, It is an integral variable At this moment, the AI-assisted hemostasis decision support platform outputs the target compression pressure value. It is an integral variable At any given moment, the current pressing pressure value of the execution module.

[0011] Further, step S3 includes the following sub-steps: S31, retrieving the parameter configuration file of the pre-trained vascular compressive stress feedback prediction model from the storage module. This file includes the initial values ​​of the model weight coefficients, fitting coefficients, and update rules; S32, performing time axis alignment processing on the blood flow velocity signal processed in S2 to ensure that it corresponds one-to-one with the sampling points of the pressure signal collected in S1 in the time dimension; S33, inputting the aligned blood flow velocity signal value and pressure signal value into the vascular compressive stress feedback prediction model time by time, and calculating the initial compressive stress value of different monitoring points in different directions inside the blood vessel at each time according to the model expression; S34, correcting the initial compressive stress value calculated in S33 according to the preset range of the elastic coefficient of the blood vessel wall, eliminating abnormal values ​​that exceed the physiologically reasonable range, and obtaining the final real-time compressive stress value inside the blood vessel.

[0012] Further, step S4 includes the following sub-steps: S41, performing short-time Fourier transform on the blood flow velocity signal and the blood vessel wall vibration signal processed in S2 respectively to obtain the frequency domain feature matrices of the two signals under different time windows; simultaneously, performing continuous wavelet transform on the intravascular compressive stress value obtained in S3 to obtain the time-frequency feature matrix of the compressive stress value; S42, using principal component analysis to perform dimensionality reduction processing on the time-frequency feature matrices respectively, retaining the principal component feature vectors whose contribution rate exceeds a preset threshold in each matrix; S43, concatenating the three principal component feature vectors after dimensionality reduction in chronological order to form a preliminary fusion feature matrix, and then assigning different weights to different types of feature vectors through an attention mechanism; S44, normalizing the weighted preliminary fusion feature matrix to obtain a multi-dimensional fusion feature vector with unified dimensions, which includes the time-frequency domain information of the three types of signals: blood flow, vibration, and compressive stress.

[0013] Further, S5 includes the following sub-steps: S51, inputting the multi-dimensional fused feature vector generated in S4 into the feature preprocessing module of the AI-assisted hemostasis decision support platform. This module performs outlier detection and correction on the feature vector to ensure the validity of the input data; S52, calling the deep learning model trained internally by the platform. This model adopts a structure combining convolutional neural networks and recurrent neural networks to perform spatial feature extraction and time series analysis on the preprocessed feature vector; S53, the model output layer obtains the probability distribution corresponding to the pressure adjustment suggestion, pressure duration evaluation result, and hemostasis effect judgment index through the softmax activation function; S54, determining the final output result from the probability distribution according to the preset probability threshold, and converting it into a control signal format recognizable by the cardiology pressure hemostasis device.

[0014] A smart monitoring and analysis method for a cardiology compression hemostasis device is disclosed. This method is implemented through different units, including: a vascular pressure and biosignal synchronous acquisition unit, which consists of a pressure sensor array, a blood flow velocity sensor, a vascular wall vibration sensor, and a signal synchronization triggering circuit. This unit synchronously acquires vascular surface pressure signals, blood flow velocity signals, and vascular wall vibration signals in the compression area and transmits the acquired signals to a signal preprocessing unit; a blood flow signal noise reduction and enhancement processing unit, connected to the vascular pressure and biosignal synchronous acquisition unit, which incorporates a blood flow signal noise reduction and enhancement algorithm module. This module receives the blood flow velocity signal transmitted from the acquisition unit, removes interference noise, and transmits the processed signal to a vascular pressure stress calculation unit and a biosignal time-frequency fusion unit, respectively; and a vascular pressure stress calculation unit, connected to both the blood flow signal noise reduction and enhancement processing unit and the vascular pressure and biosignal synchronous acquisition unit, which is loaded with a vascular pressure stress feedback prediction model. This unit receives the processed blood flow velocity signal and the original vascular surface pressure signal, calculates the real-time internal vascular pressure stress value, and transmits it to the biosignal time-frequency fusion unit. The system comprises several units: a time-frequency fusion unit and a biosignal time-frequency fusion unit. The former is connected to the blood flow signal denoising and enhancement processing unit and the vascular pressure stress calculation unit, respectively. It employs a biosignal time-frequency fusion algorithm to fuse the processed blood flow velocity signal, vascular wall vibration signal, and vascular internal pressure stress value in the time and frequency domains, generating a multi-dimensional fused feature vector which is then transmitted to the AI-assisted hemostasis decision unit. The latter, connected to the biosignal time-frequency fusion unit, is equipped with an AI-assisted hemostasis decision support platform and a deep learning model. It analyzes the multi-dimensional fused feature vector, outputs pressure adjustment suggestions, pressure duration evaluation results, and hemostasis effect judgment indicators, and transmits the results to the pressure execution control unit. The former, connected to the AI-assisted hemostasis decision unit, consists of a pressure execution mechanism, a pressure adjustment module, and real-time monitoring sensors. It receives control signals from the AI-assisted hemostasis decision unit, adjusts the pressure value, and simultaneously collects the adjusted signal through real-time monitoring sensors and feeds it back to the vascular pressure and biosignal synchronous acquisition unit, forming a closed-loop monitoring and control process.

[0015] Beneficial Effects: This invention proposes an intelligent monitoring and analysis method for cardiology compression hemostasis devices. By simultaneously collecting multiple signals such as vascular surface pressure, blood flow velocity, and vascular wall vibration in the compression area, and combining this with a blood flow signal noise reduction and enhancement algorithm to remove complex interference, it solves the problems of single signal processing and insufficient accuracy in existing technologies, ensuring that the signals used for analysis can truly reflect the internal state of the blood vessels. A vascular pressure stress feedback prediction model is used to calculate real-time internal pressure stress in the blood vessels, and a biosignal time-frequency fusion algorithm is used to achieve multi-dimensional signal feature fusion, providing a comprehensive and accurate feature foundation for subsequent analysis. Utilizing a deep learning model of an AI-assisted hemostasis decision support platform, the method dynamically outputs compression pressure adjustment suggestions, duration evaluation results, and hemostasis effect indicators based on the fused features, replacing manual experience-based decision-making and solving the problems of lack of intelligent decision-making, lagging parameter adjustment, and low accuracy in existing technologies. Simultaneously, the execution module receives suggestions to adjust the compression pressure in real time and continuously collects data for repeated analysis to form a closed-loop process, achieving fully automated and precise monitoring and control of compression hemostasis. This significantly improves the safety and effectiveness of hemostasis operations, adapts to individual differences among patients, and reduces the risk of vascular damage or hemostasis failure. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a diagram showing the unit composition for implementing the method of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, an intelligent monitoring and analysis method for a cardiology compression hemostasis device includes the following steps: S1. The pressure sensor module of the cardiology compression hemostasis device collects the pressure signal of the blood vessel surface in the compression area, and at the same time, the biosignal acquisition module obtains the blood flow velocity signal and blood vessel wall vibration signal in the corresponding area. Specifically, step S1 is the signal acquisition foundation of the entire intelligent monitoring and analysis process. Multi-parameter synchronous acquisition is completed through a dedicated sensing module mounted on the cardiology compression hemostasis device. The pressure sensing module employs a thin-film pressure sensor array with an accuracy of ±0.1 kPa. This array includes 16 evenly distributed sensing units, covering a compression area diameter of 5-8 cm, and acquires pressure signals on the blood vessel surface in real time during compression. The sampling frequency is set to 100 Hz to ensure the capture of subtle fluctuations in pressure. Simultaneously, the biosignal acquisition module includes two key sensors: an ultrasonic blood flow velocity sensor, which transmits and receives signals at a 2 MHz ultrasonic frequency with a sampling frequency of 50 Hz to acquire blood flow velocity signals within the blood vessels in the compression area, with a measurement range of 0-100 cm / s, accurately reflecting hemodynamic changes; and a miniature vibration sensor with a sensitivity of 0.01 g and a sampling frequency of 200 Hz, acquiring vibration signals generated by blood flow impact and compression on the blood vessel wall, with a frequency response range of 1-500 Hz, capable of capturing vibration characteristics related to changes in blood vessel wall elasticity. All sensors achieve synchronized signal acquisition through the device's built-in synchronous trigger circuit, with time error controlled within ±1ms. This ensures that the acquired pressure, blood flow velocity, and blood vessel wall vibration signals correspond one-to-one in the time dimension, providing accurate and complete raw data support for subsequent signal processing and analysis, and avoiding analysis errors caused by time asynchrony.

[0019] S2. The pressure signal, blood flow velocity signal, and blood vessel wall vibration signal collected in S1 are transmitted to the signal processing module. The blood flow velocity signal is processed by the blood flow signal noise reduction and enhancement algorithm to remove power frequency interference, electromyographic interference and random noise from the signal. Specifically, step S2 is the preprocessing stage of the original acquired signals. A blood flow signal denoising and enhancement algorithm is used to improve the quality of the blood flow velocity signal, laying the foundation for subsequent model calculations. In this step, the signal processing module first receives the three types of original signals synchronously transmitted in step S1. First, denoising processing is performed on the blood flow velocity signal. Because the original blood flow signal is easily affected by 50Hz power frequency interference, electromyographic interference (frequency range 20-500Hz) generated by medical personnel's operations, and electronic noise from equipment (randomly distributed across the entire frequency band), the algorithm achieves interference removal through multi-stage filtering: The first stage uses adaptive notch filtering, with the center frequency locked at 50Hz and the notch bandwidth set to 2Hz, accurately filtering out power frequency interference; the second stage uses wavelet threshold filtering, selecting the db6 wavelet basis, setting the decomposition level to 5 levels, and applying soft thresholding to the high-frequency coefficients after decomposition. The threshold is dynamically adjusted according to the signal-to-noise ratio, effectively suppressing electromyographic interference; the third stage uses moving average filtering, with the window size set to 5 sampling points, smoothing random noise. After the above processing, the signal-to-noise ratio of the blood flow velocity signal is improved to over 30dB, the signal distortion is controlled within 5%, and the key characteristics of blood flow velocity changes are preserved. The processed blood flow velocity signal is temporarily stored in the cache unit of the signal processing module (capacity 16GB, read / write speed 100MB / s), and the original pressure signal and blood vessel wall vibration signal are simultaneously transmitted to the cache unit, awaiting use in subsequent steps.

[0020] S3. Call the preset vascular pressure stress feedback prediction model, take the blood flow velocity signal processed by S2 and the pressure signal collected by S1 as the model input, and calculate the real-time pressure stress value inside the blood vessel during the pressing process. Specifically, step S3 is the step of monitoring the internal state of the blood vessel. By calling the preset vascular pressure stress feedback prediction model, the blood flow velocity signal processed in step S2 is combined with the pressure signal collected in step S1 to calculate the real-time pressure stress value inside the blood vessel during the compression process. First, the model calling module reads the configuration file of the vascular pressure stress feedback prediction model from the device's storage unit (using SSD, capacity 512GB). This file includes the model's structural parameters, weight coefficients, and calculation rules. The weight coefficients are obtained by training with 500 sets of clinical sample data, including compression hemostasis monitoring data of patients of different ages (20-70 years old) and weights (40-100kg), to ensure that the model is adapted to different individual differences. During model calculation, the input pressure and blood flow velocity signals are first aligned and verified on the time axis to ensure a one-to-one correspondence between the pressure and blood flow velocity values ​​at each sampling moment, with alignment accuracy controlled within ±0.5ms. Then, a calculation window of 100ms is used, and the average value of the pressure signal within the window is taken as the baseline pressure value for that window. Simultaneously, the rate of change of the blood flow velocity signal within the window is calculated to reflect the dynamic characteristics of hemodynamics. The baseline pressure value and the rate of change of blood flow velocity are then input into the model. Through multivariate nonlinear calculation, the compressive stress values ​​at different monitoring points (a total of 8, evenly distributed across the vessel cross-section) within the window are obtained. The compressive stress calculation range covers 0-50kPa, with calculation accuracy controlled within ±0.2kPa. After each window's calculation is completed, the model output unit transmits the compressive stress value to the data interaction module in real time and stores it in the historical database (supporting real-time data backup with a storage period of 3 months) for subsequent signal fusion and decision analysis. This step achieves accurate estimation from vessel surface pressure to internal compressive stress, providing crucial data support for assessing vascular injury risk.

[0021] S4. Using a biological signal time-frequency fusion algorithm, the blood flow velocity signal processed in S2, the blood vessel wall vibration signal, and the internal pressure stress value of the blood vessel obtained in S3 are extracted and fused in the time-frequency domain to generate a multi-dimensional fusion feature vector. Specifically, step S4 is the step of realizing multi-signal collaborative analysis. Through the biological signal time-frequency fusion algorithm, the blood flow velocity signal processed in step S2, the blood vessel wall vibration signal and the internal pressure stress value of the blood vessel obtained in step S3 are extracted and fused in the time-frequency domain to generate a multi-dimensional fused feature vector. This step first extracts time-frequency domain features from three types of signals: For blood flow velocity signals, short-time Fourier transform is used for time-frequency analysis with a time window length of 200ms and a window overlap rate of 50%. Three types of frequency domain features are extracted in the 0-10Hz frequency band: power spectral density, centroid frequency, and bandwidth. Simultaneously, three types of time domain features are calculated in the time domain: mean, variance, and peak value. For vascular wall vibration signals, continuous wavelet transform is used with a Morlet wavelet basis and a scale range of 1-32. Three types of time-frequency features are extracted in the 5-50Hz frequency band: wavelet energy entropy, wavelet variance, and wavelet correlation. For intravascular compressive stress values, four types of time domain features are calculated: rate of change, peak value, trough value, and duration. Simultaneously, Hilbert-Huang transform is used to perform empirical mode decomposition on the compressive stress signal, obtaining three intrinsic mode functions. The energy proportion of each function is extracted as a frequency domain feature. The feature fusion stage then begins. First, all extracted features are standardized, mapping feature values ​​to the 0-1 range to eliminate dimensional differences. Next, an attention mechanism is used to assign weights to different features. The power spectral density of blood flow velocity signals, wavelet energy entropy of vascular vibration signals, and rate of change of compressive stress values ​​are weighted at 0.2 (highest weight), while the weights of other features are set to 0.05-0.15 based on clinical relevance. Finally, the weighted features are concatenated in a preset order to form a 28-dimensional multi-dimensional fusion feature vector. The precision of each feature in the vector is retained to 4 decimal places, and the storage format is floating-point. This vector fully preserves the key time-frequency domain information of the three types of signals, providing comprehensive feature input for subsequent AI decision-making.

[0022] S5. Input the multi-dimensional fusion feature vector generated in S4 into the AI-assisted hemostasis decision support platform. The platform analyzes the fusion feature vector through the trained deep learning model and outputs suggestions for adjusting the pressure, evaluation results of the pressure duration, and indicators for judging the hemostasis effect. Specifically, step S5 is the step of realizing intelligent decision support. The multi-dimensional fusion feature vector generated in step S4 is input into the AI-assisted hemostasis decision support platform. Through the analysis of the trained deep learning model, the platform outputs suggestions for adjusting the pressure, evaluation results of the pressure duration, and indicators for judging the hemostasis effect. The platform first verifies the validity of the input fused feature vectors, eliminating abnormal vectors by using a preset threshold range (the upper and lower limits of each feature value are set according to the clinical normal range). If three consecutive vectors are abnormal, an alarm mechanism is triggered. After the verification is passed, the feature vectors are input into a deep learning model, which adopts a hybrid structure of "convolutional neural network + long short-term memory network". The convolutional layer has three layers (with kernel sizes of 3×1, 5×1, and 7×1, and numbers of 32, 64, and 128 respectively) to extract spatial correlation features from the feature vectors. The pooling layer uses max pooling (with a pooling kernel size of 2×1) to reduce the number of parameters. The long short-term memory network layer has two layers (with 128 and 64 hidden units respectively) to capture the time-series variation patterns of the feature vectors. The fully connected layer has two layers (with 64 and 32 neurons respectively), and the output layer has three neurons, corresponding to the three types of decision results. The model was trained using 1000 sets of valid clinical data (including successful and abnormal hemostasis cases), with 500 training iterations. The cross-entropy loss function was used, and the model achieved an accuracy of over 92% on the test set. During decision output, the compression pressure adjustment suggestion was output in the form of "current pressure ± adjustment value" (adjustment value range 0-2 kPa, step size 0.1 kPa), the compression duration assessment result was output in the form of "suggested remaining duration" (range 0-30 min, accuracy 1 min), and the hemostasis effect judgment index was output in the form of "hemostasis level" (divided into 1-5 levels, level 1 for no hemostasis, level 5 for complete hemostasis). All output results were displayed in real-time on the device's touchscreen (resolution 1920×1080) and simultaneously transmitted to the execution module.

[0023] S6, the execution module of the cardiology compression hemostasis device receives the compression pressure adjustment suggestion output by S5, adjusts the pressure output value of the compression component in real time, and at the same time the monitoring module continuously collects the adjusted data and repeats S2-S5 to form a closed-loop monitoring and analysis process.

[0024] Specifically, step S6 is the stage for achieving closed-loop monitoring and control. The execution module of the cardiology compression hemostasis device receives the compression pressure adjustment suggestion output in step S5, adjusts the pressure output value of the compression component in real time, and simultaneously the monitoring module continuously collects the adjusted data and repeats steps S2-S5, forming a continuous closed-loop monitoring and analysis process. The execution module includes a pressure drive unit and a compression component. The pressure drive unit adopts a ball screw structure driven by a servo motor, with a transmission accuracy of 0.01mm, a pressure adjustment range of 0-30kPa, and an adjustment response time of ≤100ms. The compression component uses a silicone compression head (diameter 3-5cm, hardness Shore A50±5) to ensure compression comfort and sealing. After receiving the pressure adjustment suggestion, the execution module first compares the difference between the current actual pressing pressure (provided in real-time feedback from the pressure sensing module, with a sampling frequency of 100Hz) and the target pressure. If the difference is greater than 0.5kPa, a fast adjustment mode is activated, and the servo motor drives the pressing head to move at its maximum speed (1000r / min) to quickly reduce the pressure difference. If the difference is less than or equal to 0.5kPa, the fine adjustment mode is switched, and the servo motor speed is reduced to 100r / min to ensure a pressure adjustment accuracy of ±0.1kPa. Simultaneously with pressure adjustment, the monitoring module continuously collects the adjusted blood vessel surface pressure, blood flow velocity, and blood vessel wall vibration signals at the same frequency as in step S1. The newly collected data is transmitted to the signal processing module in real-time, repeating the noise reduction processing in step S2, the compressive stress calculation in step S3, the feature fusion in step S4, and the decision analysis in step S5, forming a closed-loop cycle every 100ms. If the hemostasis effect assessment index remains stable at level 4-5 for 5 consecutive cycles, and the recommended adjustment value for the compression pressure is ≤0.2kPa, the system will automatically determine that the hemostasis effect has met the standard and enter the pressure maintenance mode (maintain the current pressure and reduce the sampling frequency to 20Hz). If the hemostasis effect index decreases (<3 levels), the closed-loop cycle frequency will be increased to 50ms / cycle to ensure timely adjustment of the compression parameters and avoid hemostasis failure or vascular damage.

[0025] Preferably, the expression for the vascular compressive stress feedback prediction model is: ,in Let be the compressive stress value in the j-th direction at the i-th monitoring point inside the blood vessel at time t. , , These are the model weight coefficients. The pressure signal value on the surface of the blood vessel in the pressed area at time t. The processed blood flow velocity signal value at time t. , These are the model fitting coefficients. For integration variables, Let be the second-order Laplace operator for the blood vessel surface pressure signal at time t. It is an integral variable The blood flow velocity signal value after processing at any given time. It is an integral variable The pressure signal value on the surface of the blood vessels in the pressing area at all times.

[0026] Specifically, the vascular compressive stress feedback prediction model is used to accurately calculate the real-time compressive stress value inside the blood vessel during compression, providing core data for assessing the risk of vascular injury. When implementing the model, the specific values ​​of each parameter are first determined: among the weighting coefficients, the first coefficient is set to 0.6, the second to 0.3, and the third to 0.1. These coefficients are optimized through training with 500 clinical samples (including patients aged 20-70 years and weighing 40-100 kg) to ensure they are adapted to the differences in vascular elasticity among different individuals; among the fitting coefficients, the first coefficient is set to 0.02, and the second to 0.005, used to correct the nonlinear relationship between blood flow velocity and compressive stress. The calculation process uses a 100ms time window. First, it acquires the average value of the vascular surface pressure signal within the pressure area (sampling frequency 100Hz, average of 10 sampling points) and the rate of change of the processed blood flow velocity signal (sampling frequency 50Hz, average of the differences between adjacent sampling points). Then, it uses the model's exponential, integral, and second-order Laplace operator terms in synergistic calculation: the exponential term reflects the dynamic influence of blood flow velocity on compressive stress; the integral term accumulates the interaction between pressure and blood flow velocity over a period of time; and the second-order Laplace operator term corrects for spatial distribution differences in the pressure signal. The model's calculation accuracy is controlled within ±0.2kPa, and the output compressive stress value range covers 0-50kPa. After each calculation, the results are transmitted to the data interaction module and simultaneously stored in the historical database (storage period 3 months). This model achieves accurate conversion from vascular surface pressure to internal compressive stress, avoiding misjudgments of vascular damage caused by relying solely on surface pressure.

[0027] Preferably, the expression for the biosignal time-frequency fusion algorithm is: ,in, Frequency at time t The time-frequency characteristic value of multi-signal fusion at point n, where n is the number of biological signal types involved in the fusion. The fusion weights for the k-th type of biological signal are... For the kth type of biological signal, This is a blood flow velocity signal. This is a vibration signal from the blood vessel wall. It is a short-time Fourier transform operator. For wavelet transform operators, For time-frequency domain coupling coefficients, This represents the real-time compressive stress value inside the blood vessel. For continuous wavelet transform operators, This is the Hilbert transform operator.

[0028] Specifically, the biosignal time-frequency fusion algorithm integrates the time-frequency domain features of three types of signals: blood flow velocity, vessel wall vibration, and intravascular compressive stress, generating a comprehensive fused feature vector. During implementation, the parameter values ​​are first determined: in the fusion weights, the weight corresponding to the blood flow velocity signal is set to 0.4, and the weight corresponding to the vessel wall vibration signal is set to 0.35. The weights of the two signal types are set according to their contribution to the hemostasis effect assessment; the time-frequency coupling coefficient is set to 0.25 to enhance the time-frequency correlation between the compressive stress signal and the other two types of signals. In the signal processing stage, short-time Fourier transform was used for blood flow velocity signals (with a signal-to-noise ratio of over 30dB after processing), with a time window of 200ms and an overlap rate of 50%, to extract features in the 0-10Hz frequency band. Wavelet transform was used for blood vessel wall vibration signals (sensitivity 0.01g, sampling frequency 200Hz), selecting a specific wavelet basis and a scale range of 1-32 to extract features in the 5-50Hz frequency band. A combination of continuous wavelet transform and Hilbert transform was used for the intravascular compressive stress values ​​(calculation accuracy ±0.2kPa) to capture their instantaneous changes. The fusion process consisted of three steps: First, the time-frequency features of each type of signal were calculated separately to ensure complete feature extraction; second, the compressive stress signal features were associated with the features of the other two types of signals through coupling coefficients to enhance cross-signal dimension information interaction; third, all features were weighted and summed according to the fusion weights to generate a multi-dimensional fusion feature vector with 28 dimensions, and each feature precision retained to 4 decimal places. This algorithm effectively avoids the one-sidedness of single signal features, providing comprehensive and accurate feature input for subsequent AI decision-making and improving decision accuracy.

[0029] Preferably, the expression for the blood flow signal noise reduction and enhancement algorithm is: ,in, This is the enhanced blood flow velocity signal after noise reduction. The original blood flow velocity signal is represented by M, which is the number of intrinsic mode functions obtained from empirical mode decomposition. The filter coefficients are for the m-th intrinsic mode function. Let m be the m-th eigenmode function. These are the frequency domain filter strength coefficients. For signal frequency, This is the power frequency interference frequency. For rectangular window functions, , These are the Fast Fourier Transform and Inverse Fourier Transform operators, respectively.

[0030] Specifically, the blood flow signal denoising and enhancement algorithm is used to remove interference noise from the original blood flow velocity signal, ensuring signal quality. During implementation, parameter values ​​need to be determined based on the interference type: the number of intrinsic mode functions (IMFs) is set to 5. The original blood flow signal (sampling frequency 50Hz, measurement range 0-100cm / s) is decomposed into 5 IMFs of different frequencies through empirical mode decomposition. The first 3 are used to retain effective signal components, and the last 2 are used to filter out noise. In the filtering coefficients, the coefficients corresponding to the first 3 IMFs are set to 0.1-0.2, and the last 2 are set to 0.8-0.9, achieving focused suppression of noise components. The frequency domain filtering strength coefficient is set to 0.3, the power frequency interference frequency is locked at 50Hz, and the rectangular window function bandwidth is set to 2Hz to accurately cover the power frequency interference band. The algorithm is implemented in three stages: The first stage uses Empirical Mode Decomposition (EMD) to decompose the original signal into five intrinsic mode functions (EMFs), weighting each function according to the filtering coefficients to suppress noise-related components. The second stage performs a Fast Fourier Transform (FFT) on the processed signal, filtering out power frequency interference within the 50Hz±1Hz range in the frequency domain using a rectangular window function. The third stage uses an Inverse Fast Fourier Transform (IFFT) to return the signal to the time domain, completing noise reduction and enhancement. The signal-to-noise ratio of the processed blood flow velocity signal is improved to over 30dB, with signal distortion controlled within 5%, effectively preserving key characteristics of blood flow velocity changes (such as peak values, trough values, and rate of change), avoiding deviations in subsequent compressive stress calculations and decision analysis due to noise interference, and providing a high-quality blood flow signal foundation for the entire monitoring process.

[0031] Preferably, the output layer expression of the deep learning model in the AI-assisted hemostasis decision support platform is: ,in The model output vector includes the values ​​corresponding to the suggested compression pressure, the evaluation results of the compression duration, and the indicators for judging hemostasis effectiveness. , , These are the weight matrices from the model input layer to hidden layer 1, from hidden layer 1 to hidden layer 2, and from hidden layer 2 to the output layer, respectively. , , These are the bias vectors for the corresponding layers. To fuse feature vectors from multiple dimensions, The activation function for the rectified linear unit is... This is the softmax activation function.

[0032] Specifically, the implementation of the deep learning model output layer in the AI-assisted hemostasis decision support platform is described. This output layer is used to transform the fused feature vector into specific pressure adjustment suggestions and hemostasis effect indicators. During implementation, the model parameter values ​​were first determined: in the weight matrix, the matrix dimension from the input layer to the first hidden layer is 28×64 (28 is the dimension of the fused feature vector, and 64 is the number of neurons in the first hidden layer), the matrix dimension from the first hidden layer to the second hidden layer is 64×32, and the matrix dimension from the second hidden layer to the output layer is 32×3 (3 corresponds to the three types of output results). The bias vector dimensions are consistent with the number of neurons in the corresponding layer: the first bias vector is 64-dimensional, the second is 32-dimensional, and the third is 3-dimensional. All weights and biases were obtained through training on 1000 sets of clinical data (including successful hemostasis and abnormal cases), with 500 training iterations. The cross-entropy loss function was used, and the model achieved an accuracy of over 92% on the test set. The implementation process consists of three steps: First, the 28-dimensional fused feature vector is input into the model. The basic correlation information of the features is extracted through calculation using the first weight matrix and bias vector, combined with the rectified linear unit activation function. Second, the result from the first step is calculated using the second weight matrix and bias vector, and activated again by the rectified linear unit to deepen the mining of complex relationships among the features. Third, after calculation using the third weight matrix and bias vector, the result is transformed into a probability distribution of three output classes using the softmax activation function, with probability values ​​ranging from 0 to 1. The final output vector includes pressure adjustment suggestions (value range 0-2 kPa, step size 0.1 kPa), pressure duration evaluation results (value range 0-30 min, accuracy 1 min), and hemostasis effect judgment indicators (values ​​corresponding to levels 1-5). The output results are displayed in real-time on the touchscreen (resolution 1920×1080) and transmitted to the execution module, providing accurate basis for adjusting pressure parameters.

[0033] Preferably, the pressure adjustment control expression of the execution module of the cardiology compression hemostasis device is: ,in Let t be the pressing force value output by the module at time t. Let t be the current pressing pressure value of the module at time t. The target compression pressure value output by the AI-assisted hemostasis decision support platform. , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. For integration variables, It is an integral variable At this moment, the AI-assisted hemostasis decision support platform outputs the target compression pressure value. It is an integral variable At any given moment, the current pressing pressure value of the execution module.

[0034] Specifically, the pressure adjustment control module of the cardiology compression hemostasis device adjusts the compression pressure in real time based on AI decision results to ensure hemostasis effectiveness and safety. During implementation, parameter values ​​must balance adjustment speed and accuracy: the proportional coefficient is set to 0.8 to quickly respond to the difference between the current and target pressures; the integral coefficient is set to 0.2 to eliminate long-term pressure deviations; and the derivative coefficient is set to 0.1 to suppress overshoot during pressure adjustment. The execution process operates on a 100ms control cycle. First, it acquires the target compression pressure value (range 0-30kPa, accuracy ±0.1kPa) output by the AI-assisted hemostasis decision support platform and the current compression pressure value of the execution module (real-time feedback from the pressure sensor module, sampling frequency 100Hz), calculating the difference between the two. Then, it collaboratively calculates the adjustment amount through proportional, integral, and derivative terms: the proportional term directly outputs an adjustment signal based on the current difference, achieving rapid response; the integral term accumulates the sum of differences over the past 1 second (summing the differences over 10 control cycles), eliminating persistent deviations; the derivative term calculates the rate of change of the difference (the difference between adjacent control cycles), predicting pressure change trends and avoiding overshoot. The adjusted compression pressure value output ranges from 0-30kPa, with an adjustment response time ≤100ms and a pressure control accuracy of ±0.1kPa. The execution module adopts a ball screw structure driven by a servo motor (transmission accuracy 0.01mm). It drives the pressing head (silicone material, diameter 3-5cm, hardness Shore A50±5) to move according to the calculated pressing pressure value, thereby realizing pressure adjustment. This control method ensures that the pressing pressure can follow the target value change in real time and accurately, avoiding blood vessel damage due to excessive pressure or hemostasis failure due to insufficient pressure.

[0035] Preferably, step S3 includes the following sub-steps: S31, retrieving the parameter configuration file of the pre-trained vascular compressive stress feedback prediction model from the storage module. This file includes the initial values ​​of the model weight coefficients, fitting coefficients, and update rules; S32, performing time axis alignment processing on the blood flow velocity signal processed in S2 to ensure that it corresponds one-to-one with the sampling points of the pressure signal collected in S1 in the time dimension; S33, inputting the aligned blood flow velocity signal value and pressure signal value into the vascular compressive stress feedback prediction model time by time, and calculating the initial compressive stress value of different monitoring points in different directions inside the blood vessel at each time according to the model expression; S34, correcting the initial compressive stress value calculated in S33 according to the preset range of the elastic coefficient of the blood vessel wall, eliminating abnormal values ​​that exceed the physiologically reasonable range, and obtaining the final real-time compressive stress value inside the blood vessel.

[0036] Specifically, step S3, the invocation and calculation of the vascular compressive stress feedback prediction model, includes four sub-steps to ensure the accuracy and reliability of the compressive stress value. In stage S31, the model configuration file is retrieved from the device's storage unit (using a 512GB SSD with a read / write speed ≥200MB / s). This file includes weight coefficients trained on 500 clinical samples (covering patients aged 20-70 years and weighing 40-100kg), initial values ​​of the fitting coefficients, and rules for updating the coefficients every 100ms, ensuring the model adapts to different individual vascular characteristics. In stage S32, the blood flow velocity signal (sampling frequency 50Hz) processed in step S2 is time-aligned with the pressure signal (sampling frequency 100Hz) acquired in step S1. Through the device's built-in synchronous calibration module, the blood flow velocity signal is upscaled to a 100Hz sampling frequency using an interpolation algorithm, ensuring a one-to-one correspondence between the two types of signals at each sampling moment (10ms time interval), with the alignment error controlled within a specified range. Within ±0.5ms; In stage S33, with a calculation window of 100ms, the aligned signal values ​​are input into the model time by time. The average value of 10 pressure signal sampling points within the window is taken as the baseline pressure. The rate of change of 5 blood flow velocity sampling points is calculated and substituted into the model to obtain the initial compressive stress values ​​of 8 vascular monitoring points (uniformly distributed in the cross-section of the blood vessel). The calculation accuracy is temporarily controlled within ±0.3kPa; In stage S34, based on the preset range of the elastic coefficient of the blood vessel wall (the elastic coefficient of the normal adult artery is 1.5-3.0MPa), the initial values ​​are corrected and abnormal values ​​exceeding the physiologically reasonable range of 0-50kPa are removed. Finally, the real-time compressive stress value inside the blood vessel with an accuracy of ±0.2kPa is output, providing accurate data on the internal state of the blood vessel for subsequent signal fusion and avoiding analysis deviations caused by signal misalignment or abnormal values.

[0037] Preferably, step S4 includes the following sub-steps: S41, performing short-time Fourier transform on the blood flow velocity signal and the blood vessel wall vibration signal processed in S2 respectively to obtain the frequency domain feature matrices of the two signals under different time windows, and simultaneously performing continuous wavelet transform on the intravascular compressive stress value obtained in S3 to obtain the time-frequency feature matrix of the compressive stress value; S42, using principal component analysis to perform dimensionality reduction processing on the time-frequency feature matrices respectively, retaining the principal component feature vectors whose contribution rate exceeds a preset threshold in each matrix; S43, concatenating the three principal component feature vectors after dimensionality reduction in chronological order to form a preliminary fusion feature matrix, and then assigning different weights to different types of feature vectors through an attention mechanism; S44, normalizing the weighted preliminary fusion feature matrix to obtain a multi-dimensional fusion feature vector with unified dimensions, which includes the time-frequency domain information of the three types of signals: blood flow, vibration, and compressive stress.

[0038] Specifically, in step S4, the biosignal time-frequency fusion algorithm effectively integrates multiple signal features through four sub-steps. In stage S41, time-frequency features are extracted from three types of signals: the processed blood flow velocity signal (signal-to-noise ratio ≥30dB) is subjected to short-time Fourier transform with a time window of 200ms and an overlap rate of 50%, extracting three types of frequency domain features (power spectral density, centroid frequency, etc.) and three types of time domain features (mean, variance, etc.) in the 0-10Hz frequency band; the vascular wall vibration signal (sensitivity 0.01g, sampling frequency 200Hz) is subjected to continuous wavelet transform, selecting a Morlet wavelet basis and a scale range of 1-32, extracting three types of time-frequency features (wavelet energy entropy, wavelet variance, etc.) in the 5-50Hz frequency band; the intravascular compressive stress value (accuracy ±0.2kPa) is calculated for four types of features in the time domain (rate of change, peak value, etc.), and three intrinsic mode functions are obtained through empirical mode decomposition. The energy proportion of each function is extracted as a frequency domain feature, with each type of signal feature extraction taking ≤50ms; in stage S42, the following steps are taken... Principal component analysis (PCA) was used to reduce the dimensionality of the feature matrices of the three types of signals (6×100 for blood flow signals, 3×200 for vibration signals, and 7×100 for compressive stress signals), retaining principal components with a contribution rate ≥85%. After dimensionality reduction, the feature vectors of blood flow signals were 6-dimensional, vibration signals were 3-dimensional, and compressive stress signals were 4-dimensional. In stage S43, the dimensionality-reduced feature vectors were concatenated in chronological order to form a preliminary fusion matrix of 13×100-dimensional. An attention mechanism was used to assign the highest weight of 0.2 to the power spectral density of blood flow signals, the wavelet energy entropy of vibration signals, and the rate of change of compressive stress, while the weights of the remaining features were set to 0.05-0.15 based on clinical relevance. In stage S44, the weighted matrix was standardized in the 0-1 range to generate a 28-dimensional multi-dimensional fusion feature vector (each feature retains 4 decimal places). This vector fully retains the key information of the three types of signals, providing comprehensive feature input for AI decision-making and improving decision accuracy.

[0039] Preferably, step S5 includes the following sub-steps: S51, inputting the multi-dimensional fused feature vector generated in S4 into the feature preprocessing module of the AI-assisted hemostasis decision support platform. This module performs outlier detection and correction on the feature vector to ensure the validity of the input data; S52, calling the deep learning model trained internally by the platform. This model adopts a structure combining convolutional neural networks and recurrent neural networks to perform spatial feature extraction and time series analysis on the preprocessed feature vector; S53, the model output layer obtains the probability distribution corresponding to the pressure adjustment suggestion, pressure duration evaluation result, and hemostasis effect judgment index through the softmax activation function; S54, determining the final output result from the probability distribution according to the preset probability threshold, and converting it into a control signal format recognizable by the cardiology pressure hemostasis device.

[0040] Specifically, step S5, the analysis process of the AI-assisted hemostasis decision support platform, achieves accurate decision output through four sub-steps. In stage S51, the 28-dimensional fused feature vector generated in step S4 is input into the platform's feature preprocessing module. The module has a built-in outlier detection algorithm that screens for abnormal vectors based on preset threshold ranges for each feature (e.g., blood flow power spectral density 0-50dB, compressive stress change rate 0-2kPa / ms). If three consecutive vectors are abnormal, an audible and visual alarm is triggered (alarm sound pressure level ≥60dB, alarm light flashing frequency 2Hz). At the same time, the normal feature vector from the previous cycle is automatically used to replace the abnormal vectors, ensuring the validity of the input data. In stage S52, the deep learning model trained within the platform is invoked. The model adopts a "3-layer convolutional neural network + 2-layer long short-term memory network" structure. The convolutional layer kernel sizes are 3×1, 5×1, and 7×1, with 32, 64, and 128 kernels respectively. The pooling layer uses max pooling (kernel size 2×1). The long short-term memory network has 128 and 64 hidden units. The model is tested with 1000 sets of clinical data (including 300 cases of abnormal hemostasis). Training was performed with 500 iterations, achieving a test set accuracy of ≥92%. Spatial feature extraction and time series analysis were conducted on the preprocessed feature vectors, with a processing time of ≤100ms. In stage S53, the model output layer used a softmax activation function to convert the analysis results into a probability distribution of three types of outputs (the probability values ​​of the pressure adjustment suggestion, pressure duration assessment, and hemostasis effect judgment are all in the range of 0-1). The probability threshold was set to 0.7, and a result was considered valid when the probability of a certain type of output was ≥0.7. In stage S54, the results corresponding to the valid probabilities were converted into a control signal format recognizable by the device (using the RS485 communication protocol with a transmission rate of 115200bps). The accuracy of the pressure adjustment suggestion was ±0.1kPa, the accuracy of the pressure duration assessment was ±1min, and the hemostasis effect judgment was divided into 1-5 levels. The signal transmission delay was ≤20ms to ensure that the execution module could receive the adjustment instructions in a timely manner and achieve accurate dynamic optimization of the pressure parameters.

[0041] The vascular compressive stress feedback prediction model in this invention is a core model for calculating real-time internal compressive stress from monitoring signals on the vascular surface, and can achieve accurate quantification of the internal state of the vascular vessel. The implementation process includes: First, retrieving the configuration file trained with 500 clinical samples (covering patients aged 20-70 years and weighing 40-100 kg) from the device's storage unit, obtaining the weighting coefficients, fitting coefficients, and rules for updating the coefficients every 100 ms; then, aligning the processed blood flow velocity signal (sampling frequency 50 Hz) with the original pressure signal (sampling frequency 100 Hz) on the time axis, and using an interpolation algorithm to increase the blood flow signal sampling frequency to 100 Hz to ensure that the sampling time corresponds one-to-one, with an alignment error within ±0.5 ms; then, using 100 ms as the calculation window, taking the average of 10 pressure sampling points within the window as the baseline pressure, calculating the change rate of 5 blood flow sampling points, and substituting them into the model to obtain the initial compressive stress values ​​of 8 vascular monitoring points, with an initial calculation accuracy of ±0.3 kPa; finally, correcting the initial values ​​based on the normal adult arterial elasticity coefficient (1.5-3.0 MPa), removing abnormal values ​​outside the physiological range of 0-50 kPa, and finally outputting a compressive stress value with an accuracy of ±0.2 kPa. This model transforms the internal compressive stress of blood vessels, which cannot be directly measured, into calculable quantitative data, providing accurate information on the internal state of blood vessels for subsequent signal fusion. It avoids misjudgment of vascular damage caused by relying solely on surface pressure, provides key data support for the safety of pressure-based hemostasis, and adapts to the differences in individual vascular characteristics of different patients.

[0042] The biosignal time-frequency fusion algorithm in this invention integrates the time-frequency domain features of three types of signals: blood flow velocity, vessel wall vibration, and intravascular compressive stress. It generates a multi-dimensional feature vector that comprehensively reflects the hemostasis state. The implementation process includes: First, extracting features from the three types of signals: blood flow velocity signal (signal-to-noise ratio ≥30dB) is extracted using short-time Fourier transform (time window 200ms, overlap rate 50%) to obtain 6 types of time-frequency features in the 0-10Hz frequency band; vessel wall vibration signal (sensitivity 0.01g, sampling frequency 200Hz) is extracted using continuous wavelet transform (Morrlet wavelet basis, scale 1-32) to obtain 3 types of time-frequency features in the 5-50Hz frequency band; compressive stress value (accuracy ±0.2kPa) is used to calculate 7 types of time-frequency features. The extraction time for each type of signal feature is ≤50ms. Second, the feature vectors of the three types of signals are processed... Principal component analysis (PCA) was performed on the arrays (blood flow 6×100 dimensionality, vibration 3×200 dimensionality, and compressive stress 7×100 dimensionality) for dimensionality reduction, retaining principal components with a contribution rate ≥85%, resulting in 6-dimensional, 3-dimensional, and 4-dimensional feature vectors, respectively. In the third step, the dimensionality-reduced vectors were concatenated in chronological order to form a preliminary 13×100-dimensional fusion matrix. An attention mechanism was used to assign a maximum weight of 0.2 to key features (blood flow power spectral density, vibration wavelet energy entropy, and compressive stress change rate), with the remaining features weighted between 0.05 and 0.15. In the fourth step, the weighted matrix was standardized within the 0-1 range to generate a 28-dimensional fusion feature vector (retaining 4 decimal places). This algorithm overcomes the limitations of single-signal features, integrating key information from multiple dimensions to provide comprehensive and accurate feature input for AI-assisted decision-making, improving decision accuracy and ensuring the reliability of subsequent hemostasis effect assessment and parameter adjustment.

[0043] The blood flow signal denoising and enhancement algorithm in this invention is a preprocessing algorithm that removes interference noise from the original blood flow velocity signal and improves signal quality, laying the foundation for subsequent model calculation and analysis. Its implementation process focuses on targeted treatment of three types of interference: First, the algorithm parameters are determined, setting the number of intrinsic mode functions (IMFs) in the empirical mode decomposition to 5. The first three retain valid signals, and the last two filter out noise, with corresponding filter coefficients of 0.1-0.2 and 0.8-0.9, respectively. The frequency domain filter strength coefficient is 0.3, the power frequency interference frequency is locked at 50Hz, and the rectangular window function bandwidth is 2Hz. Next, the original blood flow signal (sampling frequency 50Hz, measurement range 0-100cm / s) is decomposed into 5 IMFs through empirical mode decomposition, and noise components are suppressed by weighting according to the filter coefficients. Then, a fast Fourier transform is performed on the processed signal, and power frequency interference of 50Hz±1Hz is filtered out in the frequency domain using a rectangular window function. Finally, the signal is converted back to the time domain through an inverse fast Fourier transform, completing the denoising and enhancement. The processed blood flow signal signal-to-noise ratio is improved to over 30dB, with signal distortion ≤5%, while fully preserving key features such as peak values, trough values, and rates of change of blood flow velocity. This algorithm eliminates power frequency interference, electromyographic interference (20-500Hz), and random noise in the original blood flow signal, ensuring signal authenticity and reliability. It avoids noise-induced deviations in compressive stress calculation, feature fusion errors, and decision-making mistakes, providing a high-quality blood flow signal foundation for the entire intelligent monitoring process and ensuring the accuracy of subsequent analysis results.

[0044] The AI-assisted hemostasis decision support platform of this invention is an intelligent decision-making system based on multi-dimensional fusion feature vectors to output pressure adjustment suggestions and hemostasis effect indicators, achieving precise and automated decision-making for pressure-based hemostasis. Its implementation process includes: First, feature preprocessing: 28-dimensional fusion feature vectors are input into the preprocessing module. Abnormal vectors are screened through preset threshold ranges (e.g., blood flow power spectral density 0-50dB, compressive stress change rate 0-2kPa / ms). Three consecutive abnormal vectors trigger an audible and visual alarm (sound pressure level ≥60dB, light flashing frequency 2Hz), and the normal vector from the previous cycle is used to replace it, ensuring the validity of the input data. Second, a deep learning model is invoked. The model adopts a "3-layer convolutional neural network + 2-layer long short-term memory network" structure, with convolutional kernel sizes of 3×1 / 5×1 / 7×1 and a number of 32 / 64 / 128, pooling kernels of 2×1, and long short-term memory network hidden units of 128 / 64. The model has been tested on 1000 sets of clinical data (including 3... The platform was trained with 00 abnormal cases, iterated 500 times, and achieved a test set accuracy of ≥92%. Spatial feature extraction and time series analysis of the feature vectors were performed, taking ≤100ms. The third step generated probability distributions. The model output layer used a softmax activation function to convert the analysis results into probability distributions (0-1 intervals) for compression pressure adjustment suggestions, compression duration assessments, and hemostasis effect judgments. Probabilities ≥0.7 were considered valid results. The fourth step converted control signals, converting valid results into a device-recognizable format according to the RS485 protocol (transmission rate 115200bps), outputting pressure adjustment suggestions (accuracy ±0.1kPa), duration assessments (accuracy ±1min), and hemostasis levels (1-5), with a transmission delay ≤20ms. This platform replaces human experience-based decision-making, dynamically outputting precise compression adjustment commands, achieving real-time optimization of compression parameters, avoiding the lag and subjectivity of human judgment, reducing the risk of vascular injury or hemostasis failure, and improving the safety and effectiveness of compression hemostasis operations in cardiology.

[0045] like Figure 2As shown, an intelligent monitoring and analysis method for a cardiology compression hemostasis device is implemented through different units, including: a vascular pressure and biosignal synchronous acquisition unit, which consists of a pressure sensor array, a blood flow velocity sensor, a vascular wall vibration sensor, and a signal synchronization triggering circuit, used to synchronously acquire vascular surface pressure signals, blood flow velocity signals, and vascular wall vibration signals in the compression area, and transmit the acquired signals to a signal preprocessing unit; a blood flow signal noise reduction and enhancement processing unit, which is connected to the vascular pressure and biosignal synchronous acquisition unit, and has a built-in blood flow signal noise reduction and enhancement algorithm processing module, used to receive the blood flow velocity signal transmitted by the acquisition unit, remove the interference noise, and transmit the processed signal to a vascular pressure stress calculation unit and a biosignal time-frequency fusion unit respectively; and a vascular pressure stress calculation unit, which is connected to both the blood flow signal noise reduction and enhancement processing unit and the vascular pressure and biosignal synchronous acquisition unit, loaded with a vascular pressure stress feedback prediction model, used to receive the processed blood flow velocity signal and the original vascular surface pressure signal, calculate the real-time pressure stress value inside the vascular body, and transmit it to the biosignal unit. The system comprises the following components: a time-frequency fusion unit and a biosignal time-frequency fusion unit. The former is connected to the blood flow signal denoising and enhancement processing unit and the vascular pressure stress calculation unit. Using a biosignal time-frequency fusion algorithm, it fuses the processed blood flow velocity signal, vascular wall vibration signal, and vascular internal pressure stress value in the time and frequency domains, generating a multi-dimensional fused feature vector which is then transmitted to the AI-assisted hemostasis decision unit. The latter, connected to the biosignal time-frequency fusion unit, is equipped with an AI-assisted hemostasis decision support platform and a deep learning model. It analyzes the multi-dimensional fused feature vector, outputs pressure adjustment suggestions, pressure duration evaluation results, and hemostasis effect judgment indicators, and transmits the results to the pressure execution control unit. The former, connected to the AI-assisted hemostasis decision unit, consists of a pressure execution mechanism, a pressure adjustment module, and a real-time monitoring sensor. It receives control signals from the AI-assisted hemostasis decision unit, adjusts the pressure value, and simultaneously collects the adjusted signal through the real-time monitoring sensor and feeds it back to the vascular pressure and biosignal synchronous acquisition unit, forming a closed-loop monitoring and control process.

[0046] An intelligent monitoring and analysis method for cardiology compression hemostasis devices enables the coordinated acquisition and precise processing of multiple signals. It simultaneously acquires signals of vascular surface pressure, blood flow velocity, and vascular wall vibration in the compression area, and then uses a dedicated algorithm to remove various interferences from the signals, ensuring their authenticity and reliability. Deep analysis is achieved using professional models and algorithms. A vascular pressure stress feedback prediction model accurately calculates real-time internal pressure stress in the blood vessel, and a biosignal time-frequency fusion algorithm integrates multi-dimensional signal features, providing comprehensive and accurate data support for subsequent decision-making. An intelligent decision-making and closed-loop control system is constructed, utilizing an AI-assisted hemostasis decision support platform to dynamically output compression adjustment suggestions. Combined with the execution module, compression parameters are adjusted in real time and continuously monitored, achieving fully automated and precise operation, adapting to individual differences among patients.

[0047] This method addresses the shortcomings of existing technologies, such as single signal processing and insufficient accuracy. It simultaneously acquires multiple types of signals and employs a dedicated noise reduction and enhancement algorithm to remove interference and fuse multi-dimensional signal features, ensuring that the signals accurately reflect the internal state of blood vessels and providing precise data for analysis. Furthermore, it addresses the lack of intelligent decision-making and the lag in parameter adjustment in existing technologies, by leveraging an AI-assisted hemostasis decision support platform. Based on the fused features, it dynamically outputs pressure adjustment suggestions, replacing manual experience-based decision-making. Simultaneously, it adjusts pressure parameters in real time through a closed-loop process, avoiding untimely or inaccurate adjustments caused by manual judgment, reducing the risk of vascular damage or hemostasis failure, and improving the safety and effectiveness of hemostasis operations.

[0048] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart monitoring and analysis method for a cardiology compression hemostasis device, characterized in that, Includes the following steps: S1. The pressure sensor module of the cardiology compression hemostasis device collects the pressure signal on the surface of the blood vessel in the compression area, and simultaneously acquires the corresponding blood flow velocity signal and blood vessel wall vibration signal through the biosignal acquisition module. S2. The pressure signal, blood flow velocity signal, and blood vessel wall vibration signal collected in S1 are transmitted to the signal processing module. A blood flow signal noise reduction and enhancement algorithm is used to process the blood flow velocity signal, removing power frequency interference, electromyographic interference, and random noise. S3. A preset blood vessel pressure stress feedback prediction model is invoked, using the blood flow velocity signal processed in S2 and the pressure signal collected in S1 as model inputs to calculate the real-time pressure stress value inside the blood vessel during compression. S4. Biosignal time-frequency fusion is employed. The algorithm extracts and fuses time-frequency domain features from the blood flow velocity signal processed in S2, the blood vessel wall vibration signal, and the intravascular compressive stress value obtained in S3, generating a multi-dimensional fused feature vector. In S5, the multi-dimensional fused feature vector generated in S4 is input into the AI-assisted hemostasis decision support platform. This platform analyzes the fused feature vector through a trained deep learning model and outputs pressure adjustment suggestions, pressure duration evaluation results, and hemostasis effect judgment indicators. In S6, the execution module of the cardiology pressure hemostasis device receives the pressure adjustment suggestions output in S5, adjusts the pressure output value of the pressing component in real time, and the monitoring module continuously collects the adjusted data and repeats S2-S5 to form a closed-loop monitoring and analysis process.

2. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, The expression for the vascular compressive stress feedback prediction model is as follows: ,in Let be the compressive stress value in the j-th direction at the i-th monitoring point inside the blood vessel at time t. , , These are the model weight coefficients. The pressure signal value on the surface of the blood vessel in the pressed area at time t. The processed blood flow velocity signal value at time t. , These are the model fitting coefficients. For integration variables, Let be the second-order Laplace operator for the blood vessel surface pressure signal at time t. It is an integral variable The blood flow velocity signal value after processing at any time. It is an integral variable The pressure signal value on the surface of the blood vessels in the pressing area at all times.

3. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, The expression for the biosignal time-frequency fusion algorithm is as follows: ,in, Frequency at time t The time-frequency characteristic value of multi-signal fusion at point n, where n is the number of biological signal types involved in the fusion. The fusion weights for the k-th type of biological signal are... For the kth type of biological signal, This is a blood flow velocity signal. This is a vibration signal from the blood vessel wall. It is a short-time Fourier transform operator. For wavelet transform operators, For time-frequency domain coupling coefficients, This represents the real-time compressive stress value inside the blood vessel. For continuous wavelet transform operators, For Hilbert transformation operators.

4. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, The expression for the blood flow signal noise reduction and enhancement algorithm is: ,in, This is the enhanced blood flow velocity signal after noise reduction. The original blood flow velocity signal is represented by M, which is the number of intrinsic mode functions obtained from empirical mode decomposition. The filter coefficients are for the m-th intrinsic mode function. Let m be the m-th eigenmode function. These are the frequency domain filter strength coefficients. For signal frequency, This is the power frequency interference frequency. For rectangular window functions, , These are the Fast Fourier Transform and Inverse Fourier Transform operators, respectively.

5. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, The output layer expression of the deep learning model in the AI-assisted hemostasis decision support platform is: ,in The model output vector includes the values ​​corresponding to the suggested compression pressure, the evaluation results of the compression duration, and the indicators for judging hemostasis effectiveness. , , These are the weight matrices from the model input layer to hidden layer 1, from hidden layer 1 to hidden layer 2, and from hidden layer 2 to the output layer, respectively. , , These are the bias vectors for the corresponding layers. To fuse feature vectors from multiple dimensions, The activation function for the rectified linear unit is... This is the softmax activation function.

6. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, The pressure adjustment control expression for the execution module of the cardiology compression hemostasis device is: ,in Let t be the pressing force value output by the module at time t. Let t be the current pressing pressure value of the module at time t. The target compression pressure value output by the AI-assisted hemostasis decision support platform. , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. For integration variables, It is an integral variable At this moment, the AI-assisted hemostasis decision support platform outputs the target compression pressure value. It is an integral variable At any given moment, the current pressing pressure value of the execution module.

7. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, S3 includes the following steps: S31, retrieve the parameter configuration file of the pre-trained vascular compressive stress feedback prediction model from the storage module. This file includes the initial values ​​of the model weight coefficients, fitting coefficients, and update rules; S32, perform time axis alignment processing on the blood flow velocity signal processed in S2 to ensure that it corresponds one-to-one with the sampling points of the pressure signal collected in S1 in the time dimension; S33, input the aligned blood flow velocity signal value and pressure signal value into the vascular compressive stress feedback prediction model time by time, and calculate the initial compressive stress value of different monitoring points in different directions inside the blood vessel at each time according to the model expression; S34, correct the initial compressive stress value calculated in S33 according to the preset range of the elastic coefficient of the blood vessel wall, remove abnormal values ​​that exceed the physiologically reasonable range, and obtain the final real-time compressive stress value inside the blood vessel.

8. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, S4 includes the following sub-steps: S41, Perform short-time Fourier transform on the blood flow velocity signal and the blood vessel wall vibration signal after processing in S2 respectively to obtain the frequency domain feature matrices of the two signals under different time windows, and at the same time, perform continuous wavelet transform on the internal compressive stress value of the blood vessel obtained in S3 to obtain the time-frequency feature matrix of the compressive stress value. S42. Principal component analysis is used to reduce the dimensionality of the time-frequency feature matrices, retaining the principal component feature vectors whose contribution rate exceeds a preset threshold in each matrix; S43. The three types of principal component feature vectors after dimensionality reduction are concatenated in chronological order to form a preliminary fusion feature matrix, and then different weights are assigned to different types of feature vectors through an attention mechanism; S44. The weighted preliminary fusion feature matrix is ​​normalized to obtain a multi-dimensional fusion feature vector with unified dimensions, which includes the time-frequency domain information of the three types of signals: blood flow, vibration, and compressive stress.

9. The intelligent monitoring and analysis method for a cardiology compression hemostasis device according to claim 1, characterized in that, S5 includes the following steps: S51, inputting the multi-dimensional fused feature vector generated in S4 into the feature preprocessing module of the AI-assisted hemostasis decision support platform. This module performs outlier detection and correction on the feature vector to ensure the validity of the input data; S52, calling the deep learning model trained internally by the platform. This model adopts a structure combining convolutional neural networks and recurrent neural networks to perform spatial feature extraction and time series analysis on the preprocessed feature vector; S53, the model output layer obtains the probability distribution corresponding to the pressure adjustment suggestion, pressure duration evaluation result, and hemostasis effect judgment index through the softmax activation function; S54, according to the preset probability threshold, determining the final output result from the probability distribution and converting it into a control signal format recognizable by the cardiology pressure hemostasis device.

10. A smart monitoring and analysis method for a cardiology compression hemostasis device according to any one of claims 1-9, characterized in that, This method is implemented through different units, including: a vascular pressure and biosignal synchronous acquisition unit, which consists of a pressure sensor array, a blood flow velocity sensor, a vascular wall vibration sensor, and a signal synchronization triggering circuit, used to synchronously acquire vascular surface pressure signals, blood flow velocity signals, and vascular wall vibration signals in the pressure area, and transmit the acquired signals to a signal preprocessing unit; a blood flow signal noise reduction and enhancement processing unit, which is connected to the vascular pressure and biosignal synchronous acquisition unit, and has a built-in blood flow signal noise reduction and enhancement algorithm processing module, used to receive the blood flow velocity signal transmitted by the acquisition unit, remove the interference noise, and then transmit the processed signal to the vascular pressure stress calculation unit and the biosignal time-frequency fusion unit respectively; a vascular pressure stress calculation unit, which is connected to both the blood flow signal noise reduction and enhancement processing unit and the vascular pressure and biosignal synchronous acquisition unit, and is loaded with a vascular pressure stress feedback prediction model, used to receive the processed blood flow velocity signal and the original vascular surface pressure signal, calculate the real-time pressure stress value inside the vascular body, and transmit it to the biosignal time-frequency fusion unit; and a biosignal time-frequency fusion unit. The fusion unit, connected to both the blood flow signal denoising and enhancement processing unit and the vascular pressure stress calculation unit, employs a biosignal time-frequency fusion algorithm to fuse the processed blood flow velocity signal, vascular wall vibration signal, and vascular internal pressure stress value in the time and frequency domains. This generates a multi-dimensional fused feature vector, which is then transmitted to the AI-assisted hemostasis decision unit. The AI-assisted hemostasis decision unit, connected to the biosignal time-frequency fusion unit, is equipped with an AI-assisted hemostasis decision support platform and a deep learning model. It analyzes the multi-dimensional fused feature vector, outputs pressure adjustment suggestions, pressure duration assessment results, and hemostasis effect judgment indicators, and transmits the results to the pressure execution control unit. The pressure execution and closed-loop monitoring unit, connected to the AI-assisted hemostasis decision unit, consists of a pressure execution mechanism, a pressure adjustment module, and a real-time monitoring sensor. It receives control signals from the AI-assisted hemostasis decision unit, adjusts the pressure value, and simultaneously collects the adjusted signal through the real-time monitoring sensor and feeds it back to the vascular pressure and biosignal synchronous acquisition unit, forming a closed-loop monitoring and control process.

Images