Multi-parameter thrombelastogram instrument for animal blood coagulation function evaluation

Abstract

The invention discloses a multi-parameter thrombelastogram instrument for animal blood coagulation function evaluation. The multi-parameter thrombelastogram instrument comprises a magnetic suspension driving and sensing module, a multi-dimensional signal processing module, a species characteristic database module, a dynamic calibration and quality control module, an artificial intelligence decision module and a multi-mode interaction output module, the magnetic suspension driving and sensing module realizes high-precision blood coagulation monitoring through electromagnetic driving and nanoscale displacement detection; the multi-dimensional signal processing module adopts a composite algorithm to extract key parameters such as an R value and an MA value; the species characteristic database module is used for hierarchically storing multi-species blood coagulation data and supporting dynamic correction; the dynamic calibration and quality control module ensures the stability of the system through a three-level joint control mechanism; the artificial intelligence decision module provides blood coagulation risk assessment and medication suggestions based on the hybrid neural network; and the multi-modal interaction output module dynamically presents a detection result through a visual engine. The system has the characteristics of high sensitivity, interference resistance, cross-species adaptability and low output delay, and is suitable for blood coagulation function evaluation of various animals.

Description

Technical Field

[0001] This invention relates to the fields of intelligent medical care and data processing, specifically to a multi-parameter thromboelastography instrument for assessing coagulation function in animals. Background Technology

[0002] Coagulation function assessment is an important test in clinical medicine and veterinary medicine, used to determine the balance between the blood coagulation and fibrinolytic systems. It is crucial for surgical risk assessment, diagnosis of hemorrhagic diseases, and monitoring of anticoagulation therapy. Traditional coagulation function tests, such as prothrombin time (PT), activated partial thromboplastin time (APTT), and platelet count, only provide single static parameters of the coagulation process and cannot comprehensively reflect the dynamic changes in coagulation and fibrinolysis. Thromboelastography (TEG) technology can monitor the entire coagulation process in real time, including coagulation initiation, clot formation, stability, and fibrinolytic status, and therefore has been widely used in human medicine. However, in the veterinary field, due to the special characteristics of animal blood samples (such as differences in coagulation mechanisms among different species, sample size limitations, and testing environment requirements), existing TEG equipment still has significant limitations in terms of sensitivity, anti-interference ability, and adaptability.

[0003] Currently, thromboelastography (TEG) instruments on the market are mainly based on mechanical or optical detection principles. Mechanical TEG measures torque changes through the physical coupling of a metal probe and the blood clot, but it is susceptible to mechanical vibration and temperature fluctuations, leading to decreased detection accuracy. Optical TEG relies on optical sensors to monitor blood clot formation, but it is sensitive to sample transparency and ambient light, and it is difficult to achieve nanometer-level displacement detection. In addition, the coagulation characteristics of animal blood samples differ significantly from those of humans. For example, canines typically have shorter clotting times, while platelet function in cats may be significantly affected by disease. Existing equipment lacks intelligent analysis algorithms and dynamic calibration mechanisms for different species, resulting in insufficient reliability of the test results.

[0004] In terms of signal processing, traditional TEG instruments employ relatively simple data processing methods, typically using peak detection or mean filtering with fixed thresholds. This makes them ill-suited to handling noise interference (such as motion artifacts, electromagnetic interference, or temperature drift) in complex blood sample signals. Furthermore, the extraction of coagulation parameters (such as R-value and MA value) often relies on manual interpretation, resulting in low automation and a lack of intelligent retesting mechanisms for abnormal signals, thus impacting detection efficiency. In clinical applications, veterinarians typically need to combine multiple test data points and experience for comprehensive judgment; however, existing equipment offers limited data visualization and lacks multi-dimensional interactive analysis capabilities, hindering rapid decision-making. Summary of the Invention

[0005] To address the aforementioned problems, this invention aims to provide a multi-parameter thromboelastography instrument for assessing animal coagulation function. The instrument includes a magnetic levitation drive and sensing module, a multi-dimensional signal processing module, a species characteristic database module, a dynamic calibration and quality control module, an artificial intelligence decision-making module, and a multi-modal interactive output module. The magnetic levitation drive and sensing module achieves nanometer-level displacement monitoring and high-precision suspension control of the probe through an electromagnetic drive unit, a displacement detection unit, and a temperature compensation unit. The blood sample container is made of medical-grade polymer material and undergoes hydrophobic treatment. The multi-dimensional signal processing module, through a composite algorithm architecture and adaptive parameter adjustment technology, achieves high-frequency acquisition and digital filtering of coagulation signals. Wavelength and feature extraction ensure the detection accuracy of R-value and MA-value parameters. The species feature database module adopts a hierarchical storage architecture to manage coagulation parameters of multiple species, supporting dynamic correction and expert rule integration. The dynamic calibration and quality control module ensures system stability through a three-level joint control mechanism, including real-time calibration, periodic detection, and anomaly self-repair. The artificial intelligence decision module analyzes coagulation status based on a hybrid neural network architecture, providing risk assessment and medication suggestions. The multimodal interactive output module dynamically presents the test results through an intelligent visualization engine and supports interactive operation. This system has high sensitivity, anti-interference, and cross-species adaptability, meeting the clinical needs of animal coagulation function assessment.

[0006] Furthermore, the magnetic levitation drive and sensing module includes an electromagnetic drive unit, a displacement detection unit, and a temperature compensation unit. The electromagnetic drive unit consists of a ring-shaped electromagnetic coil array, which generates a levitation magnetic field through current control to achieve stable probe levitation. The current control uses a digital-to-analog converter to control the levitation accuracy. The displacement detection unit integrates a differential Hall sensor and optical interferometry technology to monitor the nanometer-level displacement changes of the probe inside the blood sample container in real time. The optical interferometry technology is based on the principle of white light interference and is equipped with a broadband light source and a spectrometer. The temperature compensation unit adopts a closed-loop feedback mechanism to eliminate the influence of ambient temperature fluctuations on detection accuracy, supports programmable sampling rate adjustment to meet different detection needs, and its temperature adaptation range covers conventional laboratory environments. The blood sample container is made of medical-grade polymer material, and its inner surface is treated with plasma treatment and a hydrophobic coating to reduce blood adhesion. The magnetic levitation drive and sensing module supports multiple preset detection modes and can be upgraded online through a standard interface.

[0007] Furthermore, the multi-dimensional signal processing module has signal acquisition, digital filtering, and feature extraction functions, as detailed below:

[0008] (1) Signal acquisition function

[0009] The signal is collected from the electromagnetic drive unit of the magnetic levitation drive and sensing module.

[0010] (2) Digital filtering function

[0011] A multi-stage cascaded combination of FIR and IIR filters is used to eliminate high-frequency noise through a low-pass filter, remove power frequency interference through an adaptive filter, and finally perform signal noise reduction using wavelet transform.

[0012] (3) Feature extraction function

[0013] By employing a composite algorithm architecture and an adaptive window peak detection algorithm to process the coagulation time R value and the blood clot strength MA value respectively, adjustable parameters are set for different animal blood sample characteristics through signal processing. The processing results can be transmitted, displayed and stored in real time, and an interface for exporting raw data is provided for subsequent analysis.

[0014] Furthermore, the multi-dimensional signal processing module employs a composite algorithm architecture to process coagulation signals, first constructing a constraint variational function:

[0015]

[0016]

[0017] in, For the first Time-domain representation of each intrinsic mode component For the first The center frequency of each modal component The Dirac function is used to construct analytic signals. The imaginary unit, for time, To obtain the partial derivative with respect to t, a quadratic penalty factor is introduced. and Lagrange multipliers Construct the augmented Lagrangian function:

[0018]

[0019] in, For inner product operations, the alternating direction multiplier method is used for iterative solution. After extracting the main modal components, the Hilbert transform is used to calculate the instantaneous frequency. When the instantaneous frequency change rate... First time exceeding the threshold This is determined as the starting point of the R value;

[0020] The blood clot strength MA value is calculated using an adaptive window peak detection algorithm, defining a sliding window energy function. :

[0021]

[0022] in, For window size, Let be the signal amplitude within the i-th window size. Based on signal gradient Dynamically adjusted, the final MA value is from Confirmed, among which For morphological correction coefficients, Let be the sliding window energy function for the i-th window size.

[0023] Furthermore, the species characteristic database module adopts a hierarchical storage architecture to manage the coagulation parameter characteristics of multiple species. The core database includes a three-layer structure: a basic parameter layer, a dynamic correction layer, and an expert rule layer. The basic parameter layer stores the reference range of standard coagulation parameters. The dynamic correction layer records instrument detection data in real time and updates the reference range using a sliding window. The expert rule layer integrates veterinary clinical experience and knowledge, including parameter corrections under special pathological conditions.

[0024] Furthermore, the dynamic calibration and quality control module adopts a three-level joint control mechanism to ensure the stability of the detection system. The three-level joint control mechanism consists of real-time online calibration, periodic quality control detection, and anomaly self-repair. The real-time online calibration achieves continuous monitoring through built-in standard references, and the calibration process performs fully automatic closed-loop adjustment. The periodic quality control detection is automatically executed according to a preset schedule, and the quality control rules adopt the Westgard multi-rule interpretation system.

[0025] Furthermore, the artificial intelligence decision-making module adopts a hybrid neural network architecture to realize multi-dimensional coagulation status analysis, including a feature extraction network, a temporal analysis network, and a decision fusion network;

[0026] The feature extraction network adopts a convolutional neural network structure. The input layer receives the original coagulation parameters, the first hidden layer uses parallel convolutional kernels with variable width to extract multi-scale features, the second hidden layer introduces an attention mechanism to generate feature weights, and the output layer generates feature vectors through feature compression.

[0027] The temporal analysis network is constructed based on bidirectional gated recurrent units. The input layer receives a sequence of feature vectors, the first hidden layer uses a time attention mechanism to enhance the features of key time points, the second hidden layer captures long-term dependencies through a state memory unit, and the output layer generates temporal evolution features.

[0028] The decision fusion network innovatively adopts a hierarchical voting mechanism. The first-level network analyzes coagulation factor activity. The input layer receives R-value features, the hidden layer is mapped to the data space through a fully connected network, and the output layer gives the probability of coagulation factor abnormality. The second-level network evaluates platelet function. The input layer integrates MA values, the hidden layer uses a radial basis function network, and the output layer generates a platelet activity score.

[0029] The outputs of the three sub-networks are integrated through a weighted voting module, with the weights dynamically adjusted according to the animal species, to ultimately generate a coagulation risk assessment index (0-10 points) and medication recommendations.

[0030] Furthermore, the feature extraction network is constructed as a multi-scale convolutional attention model, and the convolution operation is defined as... ,in This represents a convolution operation, where k is the kernel type index of the current operation, and K is the total number of multi-scale convolution kernel types. For activation function, For convolutional layer bias terms, It is a three-dimensional convolution kernel weight matrix, including three scales: 3×3, 5×5, and 7×7. , , For parallel multi-scale convolution kernels, an improved dual-channel mechanism is used for attention weight calculation. ,in For convolutional features, For time-series difference features, This is the attention weight matrix. A learnable query vector;

[0031] The temporal analysis network employs gated spatiotemporal memory units, and the cell state update equation is as follows: The input gate Forgotten Gate Output gate Introducing spatiotemporal correlation factors ,in, Let be the cell state vector. For the Gate of Oblivion For input gate, For output gate, , For the gated weight matrix, This is the bias of the input gate. For the offset of the forget gate, For the output gate bias, For the bias of spatiotemporal correlation factors, For logical functions, It is a learnable spatiotemporal correlation matrix;

[0032] The decision fusion network constructs a hierarchical hybrid expert system. The first-level coagulation factor analysis subnetwork adopts a fully connected layer with residual connections. ,in, , , For the weights of the fully connected layer, , For residual connection bias, x is the input feature vector. The second-level platelet function assessment subnetwork uses the radial basis function kernel. ,in, The radial base core width parameter, Cluster center point pass Algorithm initialization;

[0033] Multi-network collaboration adopts a dynamic weighted fusion mechanism ,in ,in, The weight score of the k-th subnetwork. Let the weight score be the weight score of the j-th subnetwork. Let be the expert-specific transformation matrix of the k-th subnetwork. To share feature vectors, The bias of the k-th subnetwork is given by ReLU, and the activation function is ReLU. An improved hybrid loss function is used during training. Classification loss Ranking loss Comparison of losses ,in, c is the number of categories, and c is the index of the number of categories. for The loss weight hyperparameter, for The loss weight hyperparameter, for The loss weight hyperparameter, For real labels, To predict probabilities, For boundary margin, For cosine similarity, For positive sample anchor point features, For negative sample features, To query sample features, For temperature coefficient, To enhance sample features, For positive sample features, The negative sample feature is represented by 2N, where 2N is the batch size. The similarity function is used to balance various losses through an adaptive weight adjustment strategy.

[0034] Furthermore, the multimodal interactive output module uses an intelligent visualization engine to realize the multi-dimensional presentation of the detection results. The core display interface integrates dynamic TEG curves, two-dimensional parameter radar charts, and risk level dashboards. The dynamic TEG curve drawing adopts anti-aliasing rendering technology, supports gesture-controlled zooming and panning operations, and provides touch query function in the key parameter area. Long-pressing any point can display the detailed parameter values ​​at that moment.

[0035] The beneficial effects of this invention are as follows: In the field of animal coagulation function testing, traditional methods have many limitations. This invention effectively solves these problems through technological innovation. The closed-loop feedback mechanism of the temperature compensation unit further eliminates the influence of environmental fluctuations, enabling the system to obtain reliable test data under normal laboratory conditions. At the signal processing level, traditional methods rely on fixed thresholds or simple filtering algorithms, which are difficult to cope with the complex noise interference of animal blood samples. This invention develops an adaptive algorithm system based on constrained variational mode decomposition. By constructing a special variational function and introducing Lagrange multipliers, the algorithm can accurately decompose the intrinsic mode components in the coagulation signal. It innovatively combines Hilbert transform with dynamic threshold detection, significantly improving the accuracy of R-value determination. For the calculation of MA value, a strategy combining sliding window energy analysis and parametric modeling is adopted. By adjusting the window size and morphological correction coefficient in real time, stable test results are ensured in blood samples from different species. This keeps the coefficient of variation of key coagulation parameters below 3%, which is superior to traditional methods. The hybrid neural network architecture constructed in this invention integrates the advantages of convolutional neural networks, gated recurrent units, and expert systems. The multi-scale convolutional attention mechanism can extract more discriminative features from the original signal, while the introduction of spatiotemporal memory units effectively captures the dynamic evolution of the coagulation process. The hierarchical voting mechanism, through independent analysis of coagulation factor activity and platelet function followed by dynamic weighted fusion, makes the final coagulation risk assessment more scientific and reliable. This architecture not only improves diagnostic accuracy but also provides personalized medication recommendations based on specific animal species. This invention sets a new technical standard for animal coagulation function assessment and has broad application prospects in veterinary clinical practice and scientific research. Attached Figure Description

[0036] The invention will be further illustrated with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For those skilled in the art, other drawings can be obtained based on the following drawings without any creative effort.

[0037] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Please see Figure 1 The present invention will be further described in conjunction with the following examples.

[0040] See Figure 1 This invention aims to provide a multi-parameter thromboelastography instrument for assessing coagulation function in animals. It includes a magnetic levitation drive and sensing module, a multi-dimensional signal processing module, a species characteristic database module, a dynamic calibration and quality control module, an artificial intelligence decision-making module, and a multi-modal interactive output module. The magnetic levitation drive and sensing module achieves nanometer-level displacement monitoring and high-precision levitation control of the probe through an electromagnetic drive unit, a displacement detection unit, and a temperature compensation unit. The blood sample container is made of medical-grade polymer material and has undergone hydrophobic treatment. The multi-dimensional signal processing module achieves high-frequency acquisition, digital filtering, and special processing of coagulation signals through a composite algorithm architecture and adaptive parameter adjustment technology. The system extracts key features to ensure the accuracy of R-value and MA-value detection. The species characteristic database module adopts a hierarchical storage architecture to manage coagulation parameters of multiple species, supporting dynamic correction and expert rule integration. The dynamic calibration and quality control module ensures system stability through a three-level joint control mechanism, including real-time calibration, periodic detection, and anomaly self-repair. The artificial intelligence decision-making module analyzes coagulation status based on a hybrid neural network architecture, providing risk assessment and medication recommendations. The multimodal interactive output module dynamically presents test results through an intelligent visualization engine, supporting interactive operation. This system has high sensitivity, anti-interference, and cross-species adaptability, meeting the clinical needs of animal coagulation function assessment.

[0041] Specifically, the magnetic levitation drive and sensing module includes an electromagnetic drive unit, a displacement detection unit, and a temperature compensation unit. The electromagnetic drive unit consists of a ring-shaped electromagnetic coil array, which generates a levitation magnetic field through current control to achieve stable probe levitation. The current control uses a digital-to-analog converter to control the levitation accuracy. The displacement detection unit integrates a differential Hall sensor and optical interferometry technology to monitor the nanometer-level displacement changes of the probe inside the blood sample container in real time. The optical interferometry technology is based on the principle of white light interference and is equipped with a broadband light source and a spectrometer. The temperature compensation unit adopts a closed-loop feedback mechanism to eliminate the influence of ambient temperature fluctuations on detection accuracy, supports programmable sampling rate adjustment to meet different detection needs, and its temperature adaptation range covers conventional laboratory environments. The blood sample container is made of medical-grade polymer material, and its inner surface is treated with plasma treatment and a hydrophobic coating to reduce blood adhesion. The magnetic levitation drive and sensing module supports multiple preset detection modes and can be upgraded online through a standard interface.

[0042] Specifically, the multi-dimensional signal processing module has signal acquisition, digital filtering, and feature extraction functions, as detailed below:

[0043] (1) Signal acquisition function

[0044] The signal frequency collected from the electromagnetic drive unit of the magnetic levitation drive and sensing module is not less than 100kHz to ensure signal integrity.

[0045] (2) Digital filtering function

[0046] A multi-stage cascaded combination of FIR and IIR filters is used to eliminate high-frequency noise through a low-pass filter, remove power frequency interference through an adaptive filter, and finally perform signal noise reduction using wavelet transform.

[0047] (3) Feature extraction function

[0048] By employing a composite algorithm architecture and an adaptive window peak detection algorithm to process the coagulation time R value and the blood clot strength MA value respectively, adjustable parameters are set for different animal blood sample characteristics through signal processing. The processing results can be transmitted, displayed and stored in real time, and a raw data export interface is provided for subsequent analysis.

[0049] The multi-dimensional signal processing module can periodically calibrate the signal processing channel and automatically trigger a retest mechanism for abnormal signals. The signal processing process adopts double buffering technology to achieve seamless real-time processing. The response speed is improved by FPGA hardware acceleration. The display interface synchronously presents the original waveform and parameters. The display interface supports waveform magnification, measurement and annotation functions. The storage unit adopts a circular buffer structure to save the most recent detection data and supports both USB and wireless data transmission methods for storage. All signal processing parameters can be set and adjusted via touch screen or external keyboard. For special blood sample characteristics, the filtering parameters and feature extraction thresholds can be manually adjusted. The system automatically records the signal quality indicators and processing logs for each detection to ensure the traceability of the detection results.

[0050] Specifically, the multi-dimensional signal processing module uses a composite algorithm architecture to process coagulation signals, first constructing a constraint variational function:

[0051]

[0052]

[0053] in, For the first Time-domain representation of each intrinsic mode component For the first The center frequency of each modal component The Dirac function is used to construct analytic signals. The imaginary unit, for time, To obtain the partial derivative with respect to t, a quadratic penalty factor is introduced. and Lagrange multipliers Construct the augmented Lagrangian function:

[0054]

[0055] in, For inner product operations, the alternating direction multiplier method is used for iterative solution:

[0056]

[0057]

[0058] in, Let be the frequency domain representation of the k-th modal component in the (n+1)th iteration. For the frequency domain sum of other modal components, The Fourier transform of the original coagulation signal. For the frequency domain representation of Lagrange multipliers, For the first The center frequency of each modal component in the (n+1)th iteration For the center frequency parameter, when it satisfies Stop iteration when, where The threshold for terminating the iteration. For the first The modal component in the th ... The frequency domain representation of the next iteration, after extracting the main modal components, uses Hilbert transform to calculate the instantaneous frequency, when the instantaneous frequency change rate... First time exceeding the threshold This is determined as the starting point of the R value;

[0059] The blood clot strength MA value is calculated using an adaptive window peak detection algorithm, defining a sliding window energy function. :

[0060]

[0061] Where n, N, and i are the window sizes. Let be the signal amplitude within the i-th window size. Based on signal gradient Dynamically adjusted, the final MA value is from Confirmed, among which For morphological correction coefficients, Establish a parameterized model for the sliding window energy function under the i-th window size. Where A is the amplitude parameter and B is the coagulation rate constant, used to determine the slope of the curve. Let C be the time offset, and C be the baseline compensation term. Construct the residual function. Jacobian matrix iteration step size , The damping coefficient is used; the signal processing unit has a built-in self-calibration function, which periodically verifies the algorithm accuracy using standard test signals. When parameter drift exceeds a preset tolerance, the calibration process is automatically triggered; algorithm parameter databases are set for different species, including the detection sensitivity of the R-value for dogs. , Sensitivity of R-value detection in cats , Users can fine-tune the parameters according to the actual testing results.

[0062] Specifically, the species characteristic database module adopts a hierarchical storage architecture to manage coagulation parameter characteristics of multiple species. The core database contains a three-layer structure: a basic parameter layer, a dynamic correction layer, and an expert rule layer. The basic parameter layer stores the reference range of standard coagulation parameters. The dynamic correction layer records instrument detection data in real time and updates the reference range using a sliding window. The window size is defined as the most recent 100 valid tests. Falling into the current reference range At that time, through the recursive formula , Gradual parameter updates ensure continuous database optimization. Let be the mean of the normal distribution function. Let be the standard deviation of the normal distribution function. For the recursive mean, The standard deviation is calculated as a recursive standard deviation; the expert rule layer integrates veterinary clinical experience and knowledge, including parameter corrections for specific pathological conditions.

[0063] Specifically, the dynamic calibration and quality control module employs a three-level control mechanism to ensure the stability of the detection system. This mechanism consists of real-time online calibration, periodic quality control testing, and anomaly self-repair. Real-time online calibration utilizes a built-in standard reference for continuous monitoring, and the calibration process is performed with fully automatic closed-loop adjustment. Periodic quality control testing is executed automatically according to a preset schedule: daily startup performs rapid quality control (<3 minutes), weekly performs full quality control (15 minutes), and monthly performs extended quality control (30 minutes). The quality control rules adopt the Westgard multi-rule interpretation system, including 12s warning rules, 13s out-of-control rules, 22s system error rules, and R4s random error rules. The anomaly self-repair strategy includes a three-level response: sensor restart (≤3 attempts), signal path self-check (100% coverage), and backup channel switching (<200ms switching time). If software repair fails, hardware maintenance is prompted.

[0064] Specifically, the artificial intelligence decision-making module adopts a hybrid neural network architecture to realize multi-dimensional coagulation status analysis, including a feature extraction network, a temporal analysis network, and a decision fusion network;

[0065] The feature extraction network adopts a convolutional neural network structure. The input layer receives the original coagulation parameters, the first hidden layer uses parallel convolutional kernels with variable width to extract multi-scale features, the second hidden layer introduces an attention mechanism to generate feature weights, and the output layer generates feature vectors through feature compression.

[0066] The temporal analysis network is constructed based on bidirectional gated recurrent units. The input layer receives a sequence of feature vectors, the first hidden layer uses a time attention mechanism to enhance the features of key time points, the second hidden layer captures long-term dependencies through a state memory unit, and the output layer generates temporal evolution features.

[0067] The decision fusion network innovatively adopts a hierarchical voting mechanism. The first-level network analyzes coagulation factor activity. The input layer receives R-value features, the hidden layer is mapped to the data space through a fully connected network, and the output layer gives the probability of coagulation factor abnormality. The second-level network evaluates platelet function. The input layer integrates MA values, the hidden layer uses a radial basis function network, and the output layer generates a platelet activity score.

[0068] The outputs of the three sub-networks are integrated through a weighted voting module, with the weights dynamically adjusted according to the animal species, to ultimately generate a coagulation risk assessment index (0-10 points) and medication recommendations.

[0069] Specifically, the feature extraction network is constructed as a multi-scale convolutional attention model, and the convolution operation is defined as... ,in This represents a convolution operation, where k is the kernel type index of the current operation, and K is the total number of multi-scale convolution kernel types. For activation function, For convolutional layer bias terms, It is a three-dimensional convolution kernel weight matrix, including three scales: 3×3, 5×5, and 7×7. , , For parallel multi-scale convolution kernels, an improved dual-channel mechanism is used for attention weight calculation. ,in For convolutional features, For time-series difference features, This is the attention weight matrix. A learnable query vector;

[0070] The temporal analysis network employs gated spatiotemporal memory units, and the cell state update equation is as follows: The input gate Forgotten Gate Output gate Introducing spatiotemporal correlation factors ,in, Let be the cell state vector. For the Gate of Oblivion For input gate, For output gate, , For the gated weight matrix, This is the bias of the input gate. For the offset of the forget gate, For the output gate bias, For the bias of spatiotemporal correlation factors, For logical functions, It is a learnable spatiotemporal correlation matrix;

[0071] The decision fusion network constructs a hierarchical hybrid expert system. The first-level coagulation factor analysis subnetwork adopts a fully connected layer with residual connections. ,in, , , For the weights of the fully connected layer, , For residual connection bias, x is the input feature vector. The second-level platelet function assessment subnetwork uses the radial basis function kernel. ,in, The radial base core width parameter, Cluster center point pass Algorithm initialization;

[0072] Multi-network collaboration adopts a dynamic weighted fusion mechanism ,in ,in, The weight score of the k-th subnetwork. Let the weight score be the weight score of the j-th subnetwork. Let be the expert-specific transformation matrix of the k-th subnetwork. To share feature vectors, The bias of the k-th subnetwork is given by ReLU, and the activation function is ReLU. An improved hybrid loss function is used during training. Classification loss Ranking loss Comparison of losses ,in, c is the number of categories, and c is the index of the number of categories. for The loss weight hyperparameter, for The loss weight hyperparameter, for The loss weight hyperparameter, For real labels, To predict probabilities, For boundary margin, For cosine similarity, For positive sample anchor point features, For negative sample features, To query sample features, For temperature coefficient, To enhance sample features, For positive sample features, The negative sample feature is represented by 2N, where 2N is the batch size. The similarity function is used to balance various losses through an adaptive weight adjustment strategy.

[0073] Specifically, the multimodal interactive output module uses an intelligent visualization engine to present the test results in multiple dimensions. The core display interface integrates a dynamic TEG curve, a two-dimensional parameter radar chart, and a risk level dashboard. The dynamic TEG curve is drawn using anti-aliased rendering technology, supports gesture-controlled zooming and panning operations, and provides touch query functionality for key parameter areas. Long-pressing any point displays detailed parameter values ​​at that moment. The coordinate axes of the two-dimensional radar chart represent coagulation initiation time (X-axis) and clot strength (Y-axis), respectively. The normal reference range is displayed in a semi-transparent color gamut, and the current test result is presented as a highlighted line. It supports 360-degree rotation for viewing, and the data points are displayed floatingly with specific values ​​and percentage deviations from the reference values. The risk level dashboard uses a ring-shaped gradient color design, transitioning continuously from green (low risk) to red (high risk), while simultaneously displaying digital readings (0-10 points) and the corresponding clinical recommendation levels (Level I-V).

[0074] The beneficial effects of this embodiment are as follows: In the field of animal coagulation function testing, traditional methods have many limitations. This invention effectively solves these problems through technological innovation. The closed-loop feedback mechanism of the temperature compensation unit further eliminates the influence of environmental fluctuations, enabling the system to obtain reliable test data under normal laboratory conditions. At the signal processing level, traditional methods rely on fixed thresholds or simple filtering algorithms, which are difficult to cope with the complex noise interference of animal blood samples. This invention develops an adaptive algorithm system based on constrained variational mode decomposition. By constructing a special variational function and introducing Lagrange multipliers, the algorithm can accurately decompose the intrinsic mode components in the coagulation signal. It innovatively combines Hilbert transform with dynamic threshold detection, significantly improving the accuracy of R-value determination. For the calculation of MA value, a strategy combining sliding window energy analysis and parametric modeling is adopted. By adjusting the window size and morphological correction coefficient in real time, stable test results are ensured in blood samples from different species. This keeps the coefficient of variation of key coagulation parameters below 3%, which is superior to traditional methods. The hybrid neural network architecture constructed in this invention integrates the advantages of convolutional neural networks, gated recurrent units, and expert systems. The multi-scale convolutional attention mechanism can extract more discriminative features from the original signal, while the introduction of spatiotemporal memory units effectively captures the dynamic evolution of the coagulation process. The hierarchical voting mechanism, through independent analysis of coagulation factor activity and platelet function followed by dynamic weighted fusion, makes the final coagulation risk assessment more scientific and reliable. This architecture not only improves diagnostic accuracy but also provides personalized medication recommendations based on specific animal species. This invention sets a new technical standard for animal coagulation function assessment and has broad application prospects in veterinary clinical practice and scientific research.

[0075] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make substitutions for some of the technical features. Any modifications, substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-parameter thromboelastography instrument for assessing coagulation function in animals, characterized in that, The system comprises a magnetic levitation drive and sensing module, a multi-dimensional signal processing module, a species characteristic database module, a dynamic calibration and quality control module, an artificial intelligence decision-making module, and a multi-modal interactive output module. The magnetic levitation drive and sensing module achieves nanometer-level displacement monitoring and high-precision levitation control of the probe through an electromagnetic drive unit, a displacement detection unit, and a temperature compensation unit. The blood sample container is made of medical-grade polymer material and has undergone hydrophobic treatment. The multi-dimensional signal processing module, through a composite algorithm architecture and adaptive parameter adjustment technology, achieves high-frequency acquisition, digital filtering, and feature extraction of coagulation signals, ensuring the detection accuracy of R-value and MA-value parameters. The species characteristic database module adopts a hierarchical storage architecture to manage coagulation parameters of multiple species, supporting dynamic correction and expert rule integration. The dynamic calibration and quality control module ensures system stability through a three-level joint control mechanism, including real-time calibration, periodic detection, and anomaly self-repair. The artificial intelligence decision-making module analyzes coagulation status based on a hybrid neural network architecture, providing risk assessment and medication recommendations. The multi-modal interactive output module dynamically presents the test results through an intelligent visualization engine.

2. The multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 1, characterized in that, The magnetic levitation drive and sensing module includes an electromagnetic drive unit, a displacement detection unit, and a temperature compensation unit. The electromagnetic drive unit consists of a ring-shaped electromagnetic coil array, which generates a levitation magnetic field through current control to achieve stable probe levitation. The current control uses a digital-to-analog converter to control the levitation accuracy. The displacement detection unit integrates a differential Hall sensor and optical interferometry technology to monitor the nanometer-level displacement changes of the probe inside the blood sample container in real time. The optical interferometry technology is based on the principle of white light interference and is equipped with a broadband light source and a spectrum analyzer. The temperature compensation unit adopts a closed-loop feedback mechanism to eliminate the influence of ambient temperature fluctuations on detection accuracy and supports programmable sampling rate adjustment. The blood sample container is made of medical-grade polymer material, and its inner surface is treated with plasma treatment and a hydrophobic coating. The magnetic levitation drive and sensing module supports multiple preset detection modes and can be upgraded online through a standard interface.

3. The multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 1, characterized in that, The multi-dimensional signal processing module has signal acquisition, digital filtering, and feature extraction functions, as detailed below: (1) Signal acquisition function The signal is collected from the electromagnetic drive unit of the magnetic levitation drive and sensing module. (2) Digital filtering function A multi-stage cascaded combination of FIR and IIR filters is used to eliminate high-frequency noise through a low-pass filter, remove power frequency interference through an adaptive filter, and finally perform signal noise reduction using wavelet transform. (3) Feature extraction function By employing a composite algorithm architecture and an adaptive window peak detection algorithm to process the coagulation time R value and the blood clot strength MA value respectively, adjustable parameters are set for different animal blood sample characteristics through signal processing. The processing results can be transmitted, displayed and stored in real time, and an interface for exporting raw data is provided for subsequent analysis.

4. The multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 3, characterized in that, The multi-dimensional signal processing module uses a composite algorithm architecture to process coagulation signals, first constructing a constraint variational function: in, For the first Time-domain representation of each intrinsic mode component For the first The center frequency of each modal component The Dirac function is used to construct analytic signals. The imaginary unit, for time, To obtain the partial derivative with respect to t, a quadratic penalty factor is introduced. and Lagrange multipliers Construct the augmented Lagrangian function: in, For inner product operations, the alternating direction multiplier method is used for iterative solution. After extracting the main modal components, the Hilbert transform is used to calculate the instantaneous frequency. When the instantaneous frequency change rate... First time exceeding the threshold This is determined as the starting point of the R value; The blood clot strength MA value is calculated using an adaptive window peak detection algorithm, defining a sliding window energy function. : in, For window size, Let be the signal amplitude within the i-th window size. Based on signal gradient Dynamically adjusted, the final MA value is from Confirmed, among which For morphological correction coefficients, Let be the sliding window energy function for the i-th window size.

5. The multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 1, characterized in that, The species characteristic database module adopts a hierarchical storage architecture to manage the coagulation parameter characteristics of multiple species. The core database contains a three-layer structure: a basic parameter layer, a dynamic correction layer, and an expert rule layer. The basic parameter layer stores the reference range of standard coagulation parameters, and the dynamic correction layer records the instrument detection data in real time and updates the reference range using a sliding window. The expert rules layer integrates veterinary clinical experience and knowledge, including parameter corrections for specific pathological conditions.

6. The multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 1, characterized in that, The dynamic calibration and quality control module adopts a three-level joint control mechanism to ensure the stability of the detection system. The three-level joint control mechanism consists of real-time online calibration, periodic quality control detection, and anomaly self-repair. The real-time online calibration achieves continuous monitoring through built-in standard references, and the calibration process performs fully automatic closed-loop adjustment. The periodic quality control detection is automatically executed according to a preset schedule, and the quality control rules adopt the Westgard multi-rule interpretation system.

7. The multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 1, characterized in that, The artificial intelligence decision-making module adopts a hybrid neural network architecture to realize multi-dimensional coagulation status analysis, including a feature extraction network, a temporal analysis network, and a decision fusion network. The feature extraction network adopts a convolutional neural network structure. The input layer receives the original coagulation parameters, the first hidden layer uses parallel convolutional kernels with variable width to extract multi-scale features, the second hidden layer introduces an attention mechanism to generate feature weights, and the output layer generates feature vectors through feature compression. The temporal analysis network is constructed based on bidirectional gated recurrent units. The input layer receives a sequence of feature vectors, the first hidden layer uses a time attention mechanism to enhance the features of key time points, the second hidden layer captures long-term dependencies through a state memory unit, and the output layer generates temporal evolution features. The decision fusion network adopts a hierarchical voting mechanism. The first-level network analyzes coagulation factor activity. The input layer receives R-value features, the hidden layer is mapped to the data space through a fully connected network, and the output layer gives the probability of coagulation factor abnormality. The second-level network evaluates platelet function. The input layer integrates MA values, the hidden layer adopts a radial basis function network, and the output layer generates a platelet activity score. The outputs of the three sub-networks are integrated through a weighted voting module, with the weights dynamically adjusted according to the animal species, ultimately generating a coagulation risk assessment index and medication recommendations.

8. A multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 7, characterized in that, The feature extraction network is constructed as a multi-scale convolutional attention model, and the convolution operation is defined as follows: ,in This represents a convolution operation, where k is the kernel type index of the current operation, and K is the total number of multi-scale convolution kernel types. For activation function, For convolutional layer bias terms, It is a three-dimensional convolution kernel weight matrix, including three scales: 3×3, 5×5, and 7×7. , , For parallel multi-scale convolution kernels, an improved dual-channel mechanism is used for attention weight calculation. ,in For convolutional features, For time-series difference features, This is the attention weight matrix. A learnable query vector; The temporal analysis network employs gated spatiotemporal memory units, and the cell state update equation is as follows: The input gate Forgotten Gate Output gate Introducing spatiotemporal correlation factors ,in, Let be the cell state vector. For the Gate of Oblivion For input gate, For output gate, , For the gated weight matrix, This is the bias of the input gate. For the offset of the forget gate, For the output gate bias, For the bias of spatiotemporal correlation factors, For logical functions, It is a learnable spatiotemporal correlation matrix; The decision fusion network constructs a hierarchical hybrid expert system. The first-level coagulation factor analysis subnetwork adopts a fully connected layer with residual connections. ,in, , , For the weights of the fully connected layer, , For residual connection bias, x is the input feature vector. The second-level platelet function assessment subnetwork uses the radial basis function kernel. ,in, The radial base core width parameter, Cluster center point pass Algorithm initialization; Multi-network collaboration adopts a dynamic weighted fusion mechanism ,in ,in, The weight score of the k-th subnetwork. Let the weight score be the weight score of the j-th subnetwork. Let be the expert-specific transformation matrix of the k-th subnetwork. To share feature vectors, The bias of the k-th subnetwork is given by ReLU, and the activation function is ReLU. An improved hybrid loss function is used during training. Classification loss Ranking loss Comparison of losses ,in, c is the number of categories, and c is the index of the number of categories. for The loss weight hyperparameter, for The loss weight hyperparameter, for The loss weight hyperparameter, For real labels, To predict probabilities, For boundary margin, For cosine similarity, For positive sample anchor point features, For negative sample features, To query sample features, For temperature coefficient, To enhance sample features, For positive sample features, The negative sample feature is represented by 2N, where 2N is the batch size. The similarity function is used to balance various losses through an adaptive weight adjustment strategy.

9. A multi-parameter thromboelastography instrument for assessing animal coagulation function according to claim 1, characterized in that, The multimodal interactive output module uses an intelligent visualization engine to present the detection results in multiple dimensions. The core display interface integrates dynamic TEG curves, two-dimensional parameter radar charts, and risk level dashboards. The dynamic TEG curves are drawn using anti-aliased rendering technology and support gesture-controlled zooming and panning operations. The key parameter area provides a touch query function, and long-pressing any point can display detailed parameter values ​​at that moment.

Images