A digital analysis processing control system for cardiac surgical medical information

Abstract

This application relates to the field of medical data management technology and discloses a digital analysis and processing control system for cardiac surgery medical information. The system includes modules for sensing, synchronization, normalization, covariance, imbalance determination, gain mapping, and safety output. It acquires mechanical and physiological parameters and aligns them along the time axis; extracts temperature parameters to generate a diagonal compensation matrix; calculates a thermodynamically normalized physiological vector stripped of physical thermal drift and concatenates them to generate a comprehensive state vector; calculates the covariance matrix of this vector set; extracts the principal eigenvalues ​​and principal eigenvectors; triggers an imbalance determination when the deviation rate exceeds a threshold; addresses the Jacobian matrix to map the principal eigenvectors to actuator pointing vectors; extracts error input terms to calculate control commands; and issues commands for execution after physical threshold interception. This invention eliminates thermal drift interference during temperature changes, achieves single-variable decoupled intervention, and avoids cross-coupling oscillations caused by simultaneous adjustment of multiple mechanisms.

Description

Technical Field

[0001] This invention relates to the field of medical data management technology, specifically to a digital analysis, processing, and control system for cardiac surgical medical information. Background Technology

[0002] In cardiac cardiopulmonary bypass surgery, the system needs to process mechanical execution parameters and human physiological feedback parameters from heterogeneous medical devices in real time. During the temperature-changing phase of cardiopulmonary bypass, changes in blood temperature directly alter rheological properties such as blood viscosity, leading to physical baseline drift in physiological feedback data such as vascular resistance. Existing data processing methods typically directly collect and use raw data containing this thermal drift, failing to isolate non-pathological interference caused by temperature changes. This masks the body's true physiological state, resulting in inaccurate subsequent data analysis and calculations.

[0003] Furthermore, there is a multi-dimensional nonlinear coupling relationship between the physical intervention actions of mechanical devices such as extracorporeal circulation machines and the patient's physiological response. Conventional monitoring and control methods often rely on setting absolute thresholds for single parameters for independent judgment, making it difficult to quantitatively extract the state correlation features between mechanical input and physiological output from multi-dimensional data. This judgment method cannot timely and accurately identify implicit multi-dimensional coupling imbalances in the system, and due to the lack of individual baseline comparison and anti-shake logic, it is prone to misjudgment during normal physiological fluctuations, resulting in inaccurate control triggering mechanisms.

[0004] When performing closed-loop regulation of the above parameters, existing automatic control logic typically issues intervention commands to multiple actuators simultaneously. Because physiological and mechanical parameters are coupled, simultaneous action of multiple actuators not only makes it difficult to decouple the control variables but also causes cross-coupling oscillations in the system, reducing the stability of the underlying mechanical regulation. Furthermore, if multiple intervention commands are superimposed, there is a risk of mechanical components exceeding their safe operating limits. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a digital analysis and processing control system for cardiac surgical medical information. This system solves the problems of inaccurate analysis due to the failure to separate physiological data from physical thermal drift, difficulty in accurately identifying implicit multidimensional coupling imbalances in the system, and the tendency to induce cross-coupling oscillations during multivariate interventions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a digital analysis, processing, and control system for cardiac surgical medical information, comprising: The sensing module is used to acquire mechanical and physiological parameters transmitted by heterogeneous medical devices; The synchronization module is used to align the mechanical execution parameters and physiological parameters on the time axis and output the aligned data sequence. The normalization module is used to extract the aligned temperature parameters, calculate the thermodynamically normalized physiological vector, and concatenate them to generate a comprehensive state vector. The covariance module is used to calculate and update the covariance matrix in real time for the comprehensive state vector set; The imbalance determination module is used to perform eigenvalue decomposition on the covariance matrix to extract the principal eigenvalues ​​and principal eigenvectors, and output a trigger signal and latch the principal eigenvectors when the deviation rate of the principal eigenvalues ​​is greater than the determination threshold. The gain mapping module is used to map the latched main feature vector to generate the actuator pointing vector and extract the error input term after receiving the trigger signal. The safety output module is used to calculate the total amount of original control commands based on the error input items, perform physical threshold interception, and generate the final control command value to be sent to the lower-level actuator of the extracorporeal circulation machine.

[0007] Preferably, the sensing module reads the blood flow of the main pump of the extracorporeal circulation machine and the water temperature of the variable temperature water tank through the serial communication bus to construct a mechanical state column vector as mechanical execution parameters; at the same time, it parses the mean arterial pressure, central venous pressure and patient's nasopharyngeal temperature parameters output by the multi-parameter monitor through the Ethernet interface to construct a physiological feedback column vector as physiological parameters.

[0008] Preferably, the synchronization module has a physically isolated circular buffer within the system, and uses a hardware timer to assign a microsecond-level timestamp to each data packet entering the circular buffer.

[0009] Preferably, the synchronization module establishes a unified system clock domain and uses a linear interpolation algorithm to perform interpolation calculations on heterogeneous data with timestamps to estimate the data values ​​at the standard system sampling time to eliminate the transmission delay differences caused by different communication buses; then, it uses a Butterworth low-pass filter with a preset cutoff frequency to process the aligned data sequence and filter out high-frequency electromagnetic interference noise.

[0010] Preferably, the normalization module extracts the temperature parameters of the variable temperature tank and the nasopharyngeal temperature within the same time slice, calculates the time derivative of the difference between the two to obtain the variable temperature gradient rate, substitutes the variable temperature gradient rate into the built-in fluid dynamics viscosity-temperature compensation equation to generate independent compensation coefficients, and uses these compensation coefficients to construct a diagonal compensation matrix.

[0011] Preferably, the normalization module performs algebraic multiplication of the vector composed of the original physiological parameters with the diagonal compensation matrix, and uses the principle of independent scaling to offset the non-pathological resistance fluctuations caused by sudden temperature changes, generating a thermodynamically normalized physiological vector stripped of physical thermal drift properties.

[0012] Preferably, the normalization module concatenates the original mechanical state column vector representing the system's mechanical input conditions with the thermodynamically normalized physiological vector representing the human body's physiological response in the dimensional direction to form a comprehensive state vector covering global variables.

[0013] Preferably, the covariance module allocates a first-in-first-out queue containing a fixed number of discrete sampling points in system memory as a time sliding window. It then sums and averages all accumulated state vectors within the window to obtain the mean vector and calculates the covariance matrix. This covariance matrix is ​​used to quantitatively characterize the linear coupling relationship between mechanical intervention and the body's physiological response over a specific time span.

[0014] Preferably, the imbalance determination module performs algebraic decomposition on the covariance matrix to obtain the eigenvalue diagonal matrix and eigenvector matrix arranged in descending order of absolute value. The system extracts the eigenvalue with the largest absolute value as the principal eigenvalue and the corresponding column vector as the principal eigenvector, thus extracting the feature parameters of the system state deviation direction in multidimensional space.

[0015] Preferably, during the stable induction period of surgery in the system initialization phase, the imbalance determination module continuously extracts the principal feature value and calculates its time average, storing it in a non-volatile register as a dynamic baseline scalar reflecting individual physiological characteristics. During system operation, the relative percentage deviation between the current principal feature value and the dynamic baseline scalar is calculated in real time.

[0016] Preferably, the imbalance judgment module combines time-dimensional anti-shake logic. When the relative deviation percentage continuously exceeds the preset imbalance judgment threshold constant, and the value of the time dwell counter accumulates to exceed the set anti-shake count threshold, the judgment system falls into implicit multidimensional coupling imbalance and outputs a trigger signal, while simultaneously locking the main feature vector at the current moment.

[0017] Preferably, the gain mapping module extracts normalized mean arterial pressure, normalized central venous pressure, and main pump blood flow parameters, and estimates the systemic circulatory resistance index by combining them with the patient's body surface area. This systemic circulatory resistance index is used as a scheduling variable to address the Jacobian matrix that fits the current physiological steady-state operating point in the built-in local linearization matrix library. The Jacobian matrix is ​​then multiplied by the latched principal eigenvector, mapping the imbalance trend in the algebraic feature space to the actuator pointing vector in the specific mechanical physical space.

[0018] Preferably, the gain mapping module traverses all components of the vector pointed to by the actuator, extracts the component with the largest absolute value and locks it as the target regulator, and uses it as the error input term for the target regulator. When multiple components with equal absolute values ​​collide, the target regulator is locked according to the built-in actuator priority array; when the absolute value of the largest component is less than the set physical resolution dead zone threshold, the error input term is forced to zero.

[0019] Preferably, the safety output module substitutes the error input term into the proportional-integral-derivative controller of the independent target controller for incremental calculation to generate the total amount of original control commands; at the same time, it forces the control adjustment increment of the non-target controller in the current cycle to be zero, keeps the actual output command value of the non-target controller unchanged, and realizes decoupled control of single-variable intervention.

[0020] Preferably, the safety output module extracts the safety operating limit boundaries of the corresponding mechanical components from the non-volatile memory, uses a saturation clamping function to perform forced amplitude truncation on the total amount of the original control commands to generate the final control command value, and then encapsulates the final control command value into a standard data frame format and writes it to the outer circulation bottom layer control board through the physical bus interface to realize closed-loop control.

[0021] This invention provides a digital analysis, processing, and control system for cardiac surgical medical information. It has the following beneficial effects: 1. This invention obtains the temperature gradient rate and constructs a diagonal compensation matrix to transform the original physiological parameters into thermodynamically normalized physiological vectors. This can remove the hydrodynamic baseline drift caused by the temperature change process during extracorporeal circulation and eliminate non-pathological interference caused by sudden temperature changes on physiological feedback data such as vascular resistance.

[0022] 2. This invention calculates the covariance matrix of a comprehensive state vector containing mechanical and physiological parameters, and performs eigenvalue decomposition to extract the principal eigenvector and principal eigenvalue. Combined with the dynamic baseline scalar obtained during the system's steady-state period and time-based anti-jitter logic, the deviation rate is compared. This feature can quantitatively extract the state relationship between mechanical intervention and physiological response in multidimensional data, accurately determine the implicit multidimensional coupling imbalance in the system, and reduce the misjudgment rate of control triggering.

[0023] 3. This invention utilizes the Jacobian matrix corresponding to the body circulation resistance index to map the principal eigenvector to the actuator pointing vector, and extracts the component with the largest absolute value to lock the target regulator as a single error input term. At the same time, it forces the control increment of non-target regulators to be zero, thereby realizing single-variable decoupling intervention under multi-variable input conditions. This avoids cross-coupling oscillations caused by the simultaneous action of multiple actuators. Combined with the amplitude limiting and interception logic of physical threshold, it ensures the stability and safety of the underlying mechanical control. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a comparison curve of mean arterial pressure response under different control architectures of the present invention; Figure 4This is a dynamic evolution diagram of the multidimensional coupling principal eigenvalues ​​of the system state according to the present invention; Figure 5 This is a closed-loop response curve of the target regulator (main pump) during the imbalance correction process of the present invention. Figure 6 This is a forced decoupling curve of the non-target regulator (variable temperature water tank) during the imbalance correction process of the present invention.

[0025] Among them, 10 is the sensing module; 20 is the synchronization module; 30 is the normalization module; 40 is the covariance module; 50 is the imbalance judgment module; 60 is the gain mapping module; and 70 is the safety output module. Detailed Implementation

[0026] The technical solutions in 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.

[0027] See attached document Figure 1 , Figure 1 This is a schematic diagram of a system architecture for a digital analysis and processing control system for cardiac surgical medical information provided in an embodiment of the present invention. The present invention provides a digital analysis and processing control system for cardiac surgical medical information, the system comprising: The system includes a sensing module 10, a synchronization module 20, a normalization module 30, a covariance module 40, an imbalance determination module 50, a gain mapping module 60, and a safety output module 70.

[0028] The aforementioned modules are deployed within the edge computing control node. The edge computing control node is constructed using an industrial control motherboard configured with a real-time operating system. The system allocates computing resources and schedules the working sequence of each module within the system clock domain of the industrial control motherboard.

[0029] The sensing module 10 includes an extracorporeal circulation machine interface unit and a multi-parameter monitor interface unit. The extracorporeal circulation machine interface unit connects to the extracorporeal circulation equipment via a serial communication bus to read the main pump blood flow and variable-temperature water tank temperature parameters. The multi-parameter monitor interface unit connects to a vital signs monitoring device via an Ethernet interface to read the patient's mean arterial pressure, central venous pressure, and nasopharyngeal temperature parameters. The sensing module 10 continuously transmits the acquired parameters to the synchronization module 20.

[0030] The synchronization module 20 has a physically isolated ring buffer. The synchronization module 20 uses a hardware timer to add microsecond-level timestamps to the parameters transmitted by the sensing module 10. The synchronization module 20 performs resampling and interpolation operations on the timestamped parameters to eliminate transmission delay differences caused by different communication buses, achieving data timeline alignment. The aligned data sequence is then processed by a low-pass filter with a preset cutoff frequency to filter out electrical signal noise.

[0031] Normalization module 30 connects to synchronization module 20. Normalization module 30 acquires the aligned temperature parameters of the variable-temperature tank and nasopharyngeal temperature, and calculates the temperature gradient rate. Normalization module 30 generates a diagonal compensation matrix based on the built-in fluid dynamics compensation equation, and multiplies the vector composed of physiological parameters with the diagonal compensation matrix to obtain a thermodynamically normalized physiological vector stripped of physical thermal drift properties. Normalization module 30 concatenates the vector composed of mechanical execution parameters with the thermodynamically normalized physiological vector to output a comprehensive state vector.

[0032] The covariance module 40 receives the synthesized state vector. The covariance module 40 sets a fixed-length time sliding window in system memory. The covariance module 40 calculates the covariance matrix for the set of synthesized state vectors within the time sliding window. The covariance matrix reflects the linear coupling relationship between mechanical execution parameters and physiological parameters within the time sliding window. The covariance module 40 updates and outputs the covariance matrix in real time as the system clock advances.

[0033] The imbalance determination module 50 is connected to the covariance module 40. The imbalance determination module 50 performs eigenvalue decomposition on the input covariance matrix to extract the principal eigenvalues ​​and their corresponding principal eigenvectors. The imbalance determination module 50 calculates the deviation rate of the current principal eigenvalues ​​relative to the dynamic baseline scalar obtained during system initialization. When the deviation rate exceeds a preset imbalance determination threshold constant, the imbalance determination module 50 outputs a trigger signal and latches the current principal eigenvector.

[0034] The gain mapping module 60 starts upon receiving a trigger signal. It extracts the current normalized mean arterial pressure, normalized central venous pressure, and mechanical pump flow rate to calculate an estimated value of the systemic circulatory resistance index. Using this estimated index as a scheduling variable, the module addresses the corresponding Jacobian matrix in the system's built-in read-only memory. It then performs a matrix multiplication operation between the addressed Jacobian matrix and the latched principal eigenvector to generate an actuator pointing vector. Finally, the module extracts the component with the largest absolute value from the actuator pointing vector as the error input and outputs it.

[0035] The safety output module 70 receives the error input and calculates the total amount of the original control commands using its built-in proportional-integral-derivative (PID) controller. The safety output module 70 retrieves the absolute physical upper and lower limits of the corresponding mechanical actuator from the read-only memory. The safety output module 70 uses a saturation clamping function to perform a hard threshold interception comparison on the total amount of the original control commands, generating the final control command value. The safety output module 70 sends the final control command value to the lower-level actuator of the extracorporeal circulation machine through the peripheral component interconnection bus interface, completing the closed-loop control of the system's workflow.

[0036] See attached document Figure 2 , Figure 2 This is a flowchart illustrating a method for digital analysis, processing, and control of cardiac surgical medical information provided by an embodiment of the present invention. The present invention provides a method for digital analysis, processing, and control of cardiac surgical medical information, which includes the following steps: S10, acquire the mechanical actuation parameters and physiological parameters transmitted by heterogeneous medical devices; S20 assigns microsecond-level timestamps to mechanical and physiological parameters and performs resampling and low-pass filtering to align the data time axis. S30: Extract the aligned temperature parameters, calculate the temperature gradient rate to generate a diagonal compensation matrix, and use the diagonal compensation matrix to calculate the thermodynamically normalized physiological vector, then splice them together to generate a comprehensive state vector. S40, within the set time sliding window, calculates and updates the covariance matrix for the comprehensive state vector set in real time; S50: Perform eigenvalue decomposition on the covariance matrix to extract the principal eigenvalues ​​and principal eigenvectors, and latch the principal eigenvectors when the deviation rate of the principal eigenvalues ​​is greater than the judgment threshold. S60, calculate the estimated value of the body circulation resistance index to address the corresponding Jacobian matrix, and perform matrix dot product between the Jacobian matrix and the latched principal eigenvector to generate the actuator pointing vector, and extract the component with the largest absolute value as the error input; S70 calculates the total amount of original control commands based on the error input items, and uses the saturation clamping function to perform physical threshold interception, generating the final control command value and sending it to the lower-level actuator of the extracorporeal circulation machine.

[0037] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0038] In this embodiment, step S10 is executed by the sensing module 10.

[0039] To achieve comprehensive acquisition of medical device parameters, the system typically establishes a data communication link with external medical devices through a physical interface deployed in the edge computing control node. Considering that heterogeneous medical devices generally use different hardware interface forms and underlying communication protocols to output data, the sensing module 10 can establish independent data acquisition channels for different devices to read underlying parameters reflecting the device's operating status and the human body's physiological response. Based on the above principle of data classification and acquisition, step S10 may include the following sub-steps:

[0040] S101 reads the main pump blood flow and variable temperature water tank temperature parameters of the extracorporeal circulation machine through the serial communication bus to construct a mechanical state column vector.

[0041] The extracorporeal circulation machine interface unit within the sensing module 10 connects to the data output port of the extracorporeal circulation device via a serial communication bus. For the pin definitions and underlying electrical characteristics configuration of the serial communication bus, those skilled in the art can use RS232 or RS485 industrial communication standards. The underlying handshake and data stream transmission mechanisms are well-known technologies in the field and will not be elaborated upon here. As a preferred approach, the extracorporeal circulation machine, as the active actuator, outputs mechanical parameters that represent the physical boundary conditions imposed on the patient by the system. Extracting and quantifying these parameters independently helps the subsequent control algorithm clearly distinguish between the cause of mechanical intervention and the physiological outcome in the mathematical space.

[0042] In practice, the extracorporeal circulation machine interface unit reads the values ​​from the internal sensors of the extracorporeal circulation machine according to a set sampling period. The determination of this sampling period typically depends on the rate of change of the physical parameters of the extracorporeal circulation. To avoid losing high-frequency hemodynamic characteristics and control the computing power overhead of the computing nodes, this sampling period can be set between 10 milliseconds and 500 milliseconds. The read values ​​include at least the main pump blood flow parameters and the variable temperature water tank temperature parameters. At discrete time points... The sensing module 10 arranges the extracted mechanical execution parameters in a predetermined order to construct a mechanical state column vector. The algebraic expression for the column vector of mechanical states is as follows: ; In the formula, Representing discrete time points The generated mechanical state column vector; This represents the total dimension of the selected mechanical parameters. Its value range can be determined based on the actual number of connected sensors, and for example, it can be a positive integer from 2 to 10. Indicates a point in time The corresponding main pump blood flow value; Indicates a point in time The corresponding temperature value of the variable temperature water tank; Indicates a point in time The corresponding number The measured mechanical execution parameters; This represents the matrix transpose operation, used to convert a row vector into a column vector.

[0043] S102 uses the Ethernet interface protocol to parse the mean arterial pressure, central venous pressure and patient nasopharyngeal temperature parameters output by the multi-parameter monitor to construct a physiological feedback column vector.

[0044] The multi-parameter monitor interface unit within the sensing module 10 is connected to the network port of the multi-parameter monitor via an Ethernet cable. The multi-parameter monitor broadcasts vital sign data to the local area network in a specific data frame format. The sensing module 10 receives network data packets and performs protocol-level field parsing. For the transmission control of network data packets and the message parsing of the Health Level 7 (HL7) standard protocol, those skilled in the art can write socket receiving programs and identifier matching scripts to achieve this; data unpacking and key-value pair extraction are well-known technologies in the field and will not be elaborated upon here. It should be noted that physiological feedback parameters reflect the patient's passive response to mechanical intervention; the purpose of acquiring this type of data is to provide human-side data support for subsequent multidimensional coupling imbalance determination.

[0045] The multi-parameter monitor interface unit extracts parameters reflecting the patient's hemodynamic parameters and body temperature distribution characteristics from the parsed plaintext data. The extracted parameters include at least mean arterial pressure, central venous pressure, and the patient's nasopharyngeal temperature. At the same discrete time point... The sensing module 10 combines the extracted physiological parameters to construct a physiological feedback column vector. The algebraic expression of the physiological feedback column vector is as follows: ; In the formula, Representing discrete time points The generated physiological feedback column vector; This indicates the total dimension of the selected physiological parameters. The value of this dimension also depends on the types of effective vital signs actually output by the monitor, and can usually be configured as a positive integer from 3 to 15. Indicates a point in time The corresponding mean arterial pressure value; Indicates a point in time The corresponding central venous pressure value; Indicates a point in time The corresponding nasopharyngeal temperature value; Indicates a point in time The corresponding number One measured physiological feedback parameter; This represents the matrix transpose operation.

[0046] After constructing the mechanical state column vector and physiological feedback column vector at each discrete time point, the sensing module 10 continuously outputs the above vector sequence to the subsequent data processing node, thereby forming a continuous data acquisition stream to ensure the real-time operation of the subsequent closed-loop control logic.

[0047] In this embodiment, step S20 is executed by the synchronization module 20.

[0048] Because different medical devices often have independent internal clocks, their communication bus transmission rates and data encapsulation mechanisms differ. These differences can easily cause phase shifts in the data stream acquired by the receiving end on the time axis. Directly feeding unaligned data into subsequent algebraic matrix operations may affect the accuracy of the correlation analysis between mechanical intervention and physiological response. To establish a reliable data foundation, synchronization module 20 needs to unify the time base of heterogeneous data. Therefore, step S20 may include the following sub-steps:

[0049] S201 uses a hardware timer inside the edge computing node to attach a high-precision timestamp (microsecond level) to each data packet entering the circular buffer.

[0050] In this embodiment, the edge computing control node runs a real-time operating system, and the synchronization module 20 allocates a physically isolated circular buffer in the system memory. This buffer is used to temporarily store data packets asynchronously pushed in by the sensing module 10 to avoid sudden large amounts of data consuming core computing resources. When a peripheral data packet arrives at the physical network port or serial port of the edge computing control node and is interrupted, the synchronization module 20 calls a high-precision hardware timer to generate a microsecond-level timestamp. This timestamp is appended to the corresponding data packet to form an original data sequence with an absolute arrival time stamp. The memory pointer management and hardware interrupt service routines for the circular buffer can be implemented by those skilled in the art based on the selected real-time operating system kernel mechanism. The data residency and overflow overwrite mechanisms are well-known technologies in the field and will not be described further here.

[0051] S202 performs interpolation calculations on heterogeneous data with timestamps based on the system's unified clock domain to eliminate communication bus transmission delay differences.

[0052] After acquiring the raw data sequence with timestamps, the synchronization module 20 establishes a discrete system clock sequence with a constant step size. ,in Indicates the first The standard system sampling time. Because the arrival time of external data packets differs from the standard system sampling time. Since the data typically does not overlap, the synchronization module 20 needs to resample the original data sequence. As a preferred approach, considering the continuity of human physiological signals over a short period of time, the synchronization module 20 uses a linear interpolation algorithm to estimate the data values ​​at the standard system sampling time.

[0053] With parameter variables to be aligned For example, the synchronization module 20 retrieves the timestamp located at time [time] in the original data sequence. Two adjacent data points on either side. Let the timestamps of the adjacent data points be respectively... and And satisfy The corresponding original parameter values ​​are respectively and Synchronization module 20 calculates the time at time using the following formula. Interpolation alignment results The specific formula is as follows: ; In the formula, Indicates the time after resampling and alignment at time... Parameter values; and These represent adjacent microsecond-level timestamps of the arrival of the original data packets; and This represents the original collected value at the corresponding timestamp. To avoid algorithmic dead zones caused by data stream interruptions due to network latency or other reasons, resulting in the inability to acquire adjacent data points on both sides, the system sets a packet loss tolerance threshold. When the current system time... When the time difference between the arrival time of the most recent valid data and the arrival time exceeds this threshold, the synchronization module 20 will switch to a zero-order hold strategy, that is, directly use the valid alignment value of the previous moment as the current one. The output continues until the receiving end resumes normal data updates. Synchronization module 20 iterates through all mechanical actuation parameters and physiological feedback parameters, performing the above operations and outputting a synchronization data matrix aligned under a unified system clock domain.

[0054] S203 uses a Butterworth low-pass filter with a preset cutoff frequency to process the aligned data sequence in order to filter out high-frequency electromagnetic interference noise in the medical environment.

[0055] High-frequency electromagnetic interference sources exist in the cardiac surgery environment, such as the transient discharge of a high-frequency electrosurgical unit. This interference often superimposed on the transmitted electrical signal in the form of high-frequency spikes. To avoid oscillations in the subsequent covariance matrix calculation results caused by high-frequency noise, the synchronization module 20 introduces a digital low-pass filtering mechanism. In specific implementation, the synchronization module 20 uses a Butterworth infinite impulse response low-pass filter to process the aligned sequence to preserve the low-frequency baseline characteristics of the patient's hemodynamic waveform.

[0056] Synchronization module 20 performs filtering operations in the form of difference equations on the resampled discrete data sequence. Let the continuous data stream sequence output by the interpolation alignment step be... At any given moment The specific value is the one obtained from the aforementioned calculation. The filter calculation formula is as follows: ; In the formula, Indicates time Smoothed parameter values ​​after filtering; Indicates time Alignment data input values; This indicates the order of the filter, and to balance computational power and filtering effect, it is usually configured as a positive integer between 2 and 4. Indicates the feedforward filter coefficients; Represents the feedback filter coefficients. Indicates time Smoothed parameter values ​​after filtering.

[0057] Filter coefficients and The cutoff frequency can be determined based on the preset cutoff frequency and the system's discrete sampling rate. Considering that the changes in core physiological parameters of the human body are relatively gradual, the cutoff frequency of the filter is usually set in the range of 2 Hz to 10 Hz. For the derivation of the coefficients of the continuous domain transfer function to the discrete domain difference equation, those skilled in the art can use the bilinear transform method for conventional calculation. The coefficient solution process is a well-known technique in the field of digital signal processing and will not be elaborated here. After filtering, the synchronization module 20 outputs a smooth and time-strictly aligned data stream.

[0058] In this embodiment, step S30 is executed by the normalization module 30.

[0059] In extracorporeal circulation surgery, the system typically requires cooling or rewarming the human blood using a temperature-controlled water tank. Changes in blood temperature often alter blood rheological properties, particularly physical fluctuations in blood viscosity. These non-pathological changes in physical properties lead to significant fluctuations in peripheral vascular resistance, which manifests as drift in physiological feedback parameters on multi-parameter monitors. Directly using data containing physical thermal drift in subsequent coupling analysis may mask true pathological changes or lead to control misjudgments. To eliminate the interference of overt physical thermal drift and extract physiological features that tend to reflect the relationship between pathology and mechanical coupling, normalization module 30 introduces a dynamic compensation mechanism based on temperature gradient. Therefore, step S30 may include the following sub-steps:

[0060] S301, extract the temperature parameters of the variable temperature tank and the nasopharyngeal temperature in the same time slice and calculate the time derivative of the difference between the two.

[0061] In this embodiment, the normalization module 30 receives the data stream after time-axis alignment processing. At discrete time points The normalization module 30 extracts the variable temperature water tank water temperature parameters from the mechanical state column vector. Simultaneously, the patient's nasopharyngeal temperature parameters are extracted from the physiological feedback column vector. These two temperature parameters represent the thermodynamic input conditions applied to the blood by the system and the current response state of the body's deep core temperature, respectively. To quantitatively assess the transient impact of this heat exchange process on the physiological system, the normalization module 30 calculates the temperature difference between the two and differentiates it with respect to time to obtain the current temperature gradient rate. Considering that data is discretely sampled in actual engineering, the normalization module 30 uses a backward difference algorithm to approximate the derivative during the system's stable operation phase. Simultaneously, to avoid an algorithmic dead zone due to a lack of historical data from the previous moment during the first computation cycle after system startup, the temperature gradient rate is configured to zero by default during the initialization phase. The specific difference calculation formula is as follows: ; In the formula, Representing discrete time points The temperature gradient rate; and Representing time respectively The temperature of the variable temperature water tank and the temperature of the nasopharynx; and These represent the temperature parameters corresponding to the previous sampling time; This represents the time step between two discrete sampling points.

[0062] S302 substitutes the temperature gradient rate into the built-in fluid dynamics viscosity-temperature compensation equation to generate a diagonal matrix of independent compensation coefficients for each physiological parameter.

[0063] After calculating the temperature gradient rate, the system needs to convert it into mathematical weights that can directly affect physiological parameters. As a preferred approach, the normalization module 30 has a pre-installed fluid dynamics viscosity-temperature compensation equation in its internal memory. This equation, based on blood rheological properties, describes the degree of nonlinear influence of the temperature gradient on different physiological parameters. For the first... For each physiological parameter, the normalization module 30 calculates its corresponding independent compensation coefficient. : ; In the formula, Indicates time Regarding the first Independent compensation coefficients for each physiological parameter; For the natural constant An exponential function with base 0; Indicates the corresponding number Empirical damping constants for a physiological parameter. The value of this constant depends on the physical sensitivity of the specific physiological parameter to temperature changes, and its value range is usually set between 0.01 and 0.15. For example, for mean arterial pressure, which is sensitive to blood viscosity, its empirical damping constant can be configured to 0.12; while for central venous pressure, which is less directly affected by temperature, its empirical damping constant is set to a smaller value, such as 0.02.

[0064] To prevent excessively large temperature gradient rates under extreme temperature conditions, which could lead to numerical explosion in the exponential function calculation and cause system overflow, the normalization module 30 sets a safety limit range for the generated independent compensation coefficients. This limit range is typically limited to between 0.7 and 1.3. If the calculated... If the value exceeds this boundary, it will be forcibly clamped to the corresponding boundary limit value.

[0065] After calculating all After determining the independent compensation coefficients for each physiological parameter, the normalization module 30 constructs a dimension of... diagonal compensation matrix The elements on the main diagonal of this matrix are the compensation coefficients for each physiological parameter, while the off-diagonal elements are all zero. The algebraic expression of this matrix is ​​as follows: ; In the formula, Indicates time The generated diagonal compensation matrix; Represents the algebraic operations used to construct a diagonal matrix; to These are the independent compensation coefficients for each physiological parameter.

[0066] S303 performs matrix multiplication on the original physiological vector and the diagonal compensation matrix to remove the physical baseline drift caused by drastic changes in blood temperature.

[0067] Using the compensation matrix constructed above, the normalization module 30 performs an algebraic transformation on the current physiological feedback column vector. Specifically, the normalization module 30 transforms the physiological feedback column vector at discrete time points... Multiply by the diagonal compensation matrix .because It is a diagonal matrix, and the matrix multiplication operation is physically equivalent to multiplying a diagonal matrix. Each physiological component is independently scaled to mathematically approximate the non-pathological resistance fluctuations caused by sudden temperature changes. The calculation process is as follows: ; In the formula, Indicates time The thermodynamically normalized physiological vector, stripped of overt physical thermal drift, maintains the same dimension as the original physiological feedback column vector. ; It is a diagonal compensation matrix; This is the original physiological feedback column vector. After the above processing, the output normalized physiological vector eliminates purely physical interference and retains only the data characteristics that reflect the true pathological regulation and purely mechanical work response of the human body.

[0068] S304 concatenates the original mechanical state column vector with the normalized physiological column vector in the dimensional direction to form a unified variable for subsequent algebraic operations.

[0069] To analyze the coupling relationship between mechanical state and physiological feedback within the same algebraic space, the system needs to establish a mathematical framework encompassing global information. The normalization module 30 extracts data from the same discrete time point. Mechanical state column vector With thermodynamically normalized physiological vector The two are then concatenated in the vertical dimension. The result of this concatenation is the integrated state vector. Its matrix expression is as follows: ; In the formula, Indicates time The generated composite state vector; The dimension is , The dimension is The combined state vector generated by splicing For a dimension a column vector, where Therefore, this comprehensive state vector encompasses the mechanical intervention conditions applied to the system at the current moment, as well as the thermodynamically compensated physiological response of the human body, providing a standardized input benchmark for subsequent covariance matrix calculation and eigenvalue decomposition.

[0070] In this embodiment, step S40 is executed by the covariance module 40.

[0071] In medical control systems, there is often a continuous dynamic coupling process between mechanical parameter intervention and human physiological response. A single-moment state vector is insufficient to reflect the system's changing trends and the deep correlations between heterogeneous variables. To quantitatively analyze the impact of the cardiopulmonary bypass machine's mechanical operation on the patient's hemodynamics, the system needs to extract data distribution characteristics over a period of time. Therefore, step S40 may include the following sub-steps:

[0072] S401 allocates a queue containing a fixed number of discrete sampling points in system memory as a time sliding window.

[0073] In this embodiment, the covariance module 40 utilizes the operating system's memory allocation mechanism to establish a circular data queue with first-in, first-out (FIFO) characteristics. This data queue serves as a time-sliding window, used to buffer the continuously input comprehensive state vector. Let the capacity of the time-sliding window, i.e., the fixed number of discrete sampling points, be... Considering that the low-frequency regulatory mechanisms of the human cardiovascular system (such as respiratory waves or vasomotor waves controlled by the autonomic nervous system) typically have a complete cycle on the order of 10 to 30 seconds, the physical time span of the sliding window needs to cover the aforementioned physiological cycles to extract effective statistical features. Therefore, The value of this parameter usually needs to be configured comprehensively in conjunction with the actual discrete sampling period of the system. For example, when the sampling period is 100 milliseconds, the number of sampling points within the sliding window... It can be set to a positive integer between 100 and 500.

[0074] Furthermore, during system operation, the covariance module 40 pushes the synthesized state vector generated at the current moment into the head of the queue according to the discrete clock cycle. When the amount of data in the queue reaches the set capacity... When new data is pushed in, it synchronously overwrites the earliest historical sample pushed into the queue, thus maintaining a constant window length. To avoid calculating excessively biased random variance due to insufficient data within the window during the initialization phase of system startup, the covariance module 40 is configured to temporarily suspend subsequent covariance outputs before the queue is full, or to use stable baseline data recorded during system initialization for placeholder filling. For the memory address read / write control of the first-in-first-out queue, those skilled in the art can use pointer increment and modulo operations to implement it. The pointer loop and overflow overwrite mechanism are well-known technologies in the field and will not be elaborated here.

[0075] S402 calculates the mean vector by summing and averaging all the accumulated state vectors within the time sliding window.

[0076] Once the data within the time-sliding window reaches a stable update state, the system needs to calculate the statistical baseline of the data within the window. The covariance module iterates through the data retained within the time-sliding window 40 times. A comprehensive state vector is used, and summation and arithmetic mean operations are performed. This mean vector reflects the overall average level of the system's mechanical working state and physiological response state within the sliding time window. Let the current discrete time point be... The specific formula for calculating the mean vector is as follows: ; In the formula, Indicates the current time point The calculated mean vector has the same dimension as the comprehensive state vector. ; This represents the total number of discrete sampling points within the time sliding window. The index reciprocal of the historical data within the window; This represents the time step for discrete sampling in the system. Indicates the time within the window is The comprehensive state column vector.

[0077] S403 uses discrete vector samples and mean vectors within a window to calculate the covariance matrix, which is used to quantitatively characterize the degree of linear coupling between mechanical action and physiological response.

[0078] After obtaining the mean vector, the system further calculates the collaborative changes in deviations from the mean among the variables within the window. As a preferred approach, the covariance module 40 constructs a covariance matrix for the comprehensive state vector set within the time sliding window. From a physical perspective, this matrix is ​​a matrix with dimension... The covariance matrix is ​​a symmetric matrix whose main diagonal elements contain the independent variances of both mechanical and physiological parameters, characterizing the degree of fluctuation of a single parameter; while the off-diagonal elements directly reflect the mutual covariance between heterogeneous parameters. This algebraic structure can quantitatively and intuitively characterize the tightness of the linear coupling between extracorporeal circulation intervention and the body's physiological response. The formula for calculating the covariance matrix is ​​as follows: ; In the formula, Indicates at time Real-time generated covariance matrix; This is the Bessel correction denominator used for unbiased estimation; These are the discrete integrated state vectors within the time sliding window; This is the mean vector calculated in the previous steps; This represents the transpose operation of a column vector to a row vector, such that the product of the vector and its transpose generates a square matrix of the corresponding dimensions.

[0079] Furthermore, in practical engineering deployments, to reduce the computational overhead of performing full matrix multiplication in each discrete sampling period and to avoid algorithm execution lag or even crashes due to computational overload of the core processor, the covariance module 40 can adopt a recursive update mechanism. Specifically, the system only needs to use the covariance matrix from the previous time step, add the vector outer product of the newly entered window, and subtract the earliest vector outer product of the moved-out window to achieve incremental updates of the covariance matrix at the current time step.

[0080] Furthermore, as the system clock continues to advance, the covariance module 40 repeats the aforementioned sliding window queue update and matrix solving process at each discrete time point or within a set calculation cycle. The real-time output dynamic covariance matrix sequence provides a low-level algebraic foundation, including temporal correlation, for subsequent eigenvalue decomposition and system imbalance state determination.

[0081] In this embodiment, step S50 is executed by the imbalance determination module 50.

[0082] After generating the covariance matrix containing historical dynamic information, it is difficult to directly determine the variation trend of the overall system state by observing only a single element of the matrix due to the common collinearity among variables in high-dimensional data. To extract the main direction of system state changes from multidimensional coupled data, the imbalance determination module 50 introduces the algebraic dimensionality reduction idea of ​​principal component analysis, projecting the covariance matrix onto an orthogonal feature space. By examining the dynamic evolution of the principal eigenvalues, the system can quantitatively assess the degree of deviation of the overall state. Therefore, step S50 may include the following sub-steps:

[0083] S501 performs algebraic decomposition on the real-time generated covariance matrix to solve for the eigenvalue diagonal matrix and eigenvector matrix arranged in descending order of absolute value.

[0084] In this embodiment, for each discrete time point The generated symmetric covariance matrix The imbalance determination module 50 calls the underlying algebraic operation library to perform eigenvalue decomposition. The algebraic constraint equations for the decomposition process are as follows: ; In the formula, The dimension input at the current moment is The covariance matrix; For inclusion An eigenvector matrix of orthogonal eigenvectors; It is a diagonal matrix, and its main diagonal elements are obtained by decomposition. The covariance matrix consists of 1,000 eigenvalues. Since the covariance matrix is ​​a positive semi-definite matrix, all the eigenvalues ​​obtained by solving are non-negative.

[0085] To pinpoint the direction of the most significant variation in the system within the multidimensional space, the imbalance determination module 50 sorts the eigenvector matrix and the eigenvalue diagonal matrix according to the magnitude of the eigenvalues, ensuring that they satisfy... After sorting, the system extracts the largest feature value. As the principal feature value at the current moment, the corresponding column vector is also extracted as the principal feature vector. For solving the eigenvalues ​​of real symmetric matrices, those skilled in the art can use the Jacobi iteration method or singular value decomposition algorithm. The numerical solution process is a well-known technique in the field of computational mathematics and will not be elaborated here.

[0086] Furthermore, the direction of the eigenvectors is uncertain during the algebraic solution process, that is... and Both are mathematically valid. To avoid random sign flipping of the principal feature vector within adjacent discrete periods, which could lead to errors in subsequent tracing logic, the imbalance determination module 50 performs sign consistency alignment on the principal feature vector. As a preferred method, the system mandates that the first non-zero element of the principal feature vector must be positive; if the first non-zero element obtained from the decomposition is negative, then the entire principal feature vector is... Multiply by -1 to flip. In addition, if the eigenvalue cannot be obtained within the set number of iterations in the current cycle, the imbalance determination module 50 will use the principal eigenvalue and principal eigenvector of the previous cycle to ensure the continuity of the calculation.

[0087] S502 continuously extracts the principal eigenvalue with the largest absolute value during the stable induction period of surgery and calculates the time average value and stores it in the register as a dynamic baseline scalar.

[0088] Individual differences in body surface area and vascular compliance among patients can lead to variations in baseline covariance eigenvalues ​​under the same set of mechanical parameter inputs. To establish a comparative benchmark that aligns with individual physiological characteristics, the imbalance assessment module 50 performs baseline calibration during the initial stable induction period of cardiopulmonary bypass. This stable induction period is typically defined as the first 5 to 15 minutes after the cardiopulmonary bypass machine takes over blood flow and reaches the target flow rate. During this phase, the patient's core vital signs are usually controlled within a manually set stable zone.

[0089] During the induction period, the imbalance determination module 50 continuously collects the principal eigenvalues ​​at each discrete time point and calculates the arithmetic mean at the end of the induction period: ; In the formula, This represents the generated dynamic baseline scalar; This represents the total number of discrete sampling points included during the stationary induction period; Indicates the time during the induction period Extracted principal feature values. Calculated. The value is written into the system's non-volatile register. To prevent the calculated baseline value from approaching zero due to unexpected sensor disconnection, which could lead to division-by-zero errors in subsequent algorithms, the system sets an engineering minimum limit for the baseline scalar. This limit is typically set based on the sensor's inherent thermal noise variance, for example, a value of 10. -6Order of magnitude. When the calculated result is below this limit, the imbalance determination module 50 will forcibly assign a system noise floor constant of the same order of magnitude as a dynamic baseline scalar.

[0090] S503 calculates the absolute value of the difference between the main feature value extracted by the current sliding window and the dynamic baseline scalar in real time, and calculates the relative deviation percentage.

[0091] Once the system enters the maintenance period, the imbalance determination module 50 assesses the relative degree to which the current system state deviates from the baseline characteristics in each calculation cycle. Specifically, the imbalance determination module 50 extracts the principal eigenvalues ​​obtained at the current moment. The dimensionless percentage of relative deviation is generated by comparing it with a dynamic baseline scalar. The calculation formula is as follows: ; In the formula, Indicates time The relative percentage deviation is used to assess the extent to which the coupled fluctuations of mechanical intervention and physiological response deviate from the steady-state baseline within the current sliding window. Because it eliminates dimensional interference, this indicator provides a basis for a unified assessment of imbalance.

[0092] S504: When the relative deviation percentage continuously exceeds the preset constant, the system is determined to be in a hidden multidimensional coupling imbalance and the main feature vector at the current moment is locked.

[0093] In clinical settings, transient tubing compression or electromagnetic interference can cause abrupt changes in individual data points, leading to an increase in the relative deviation percentage of a single calculation. To improve the algorithm's robustness against interference, the imbalance determination module 50 incorporates anti-jitter logic that incorporates a time dimension. As a preferred approach, the system pre-sets an imbalance determination threshold constant and a time-dwelling counter. Based on the fault tolerance requirements of extracorporeal circulation control specifications, this imbalance determination threshold constant can be configured as a real number between 20% and 40%.

[0094] When the imbalance determination module 50 detects the relative deviation percentage of the current period When the time dwell time exceeds the above threshold, the time dwell time counter is incremented by one; if If the value falls below the threshold, the counter is reset to zero. When the accumulated value of the dwell counter exceeds the set anti-shake count threshold (e.g., the number of discrete beats corresponding to 2 to 5 seconds of physical time), the imbalance judgment module 50 confirms that this deviation has a continuous development trend, and thus outputs a logic high level, determining that the system has fallen into a hidden multidimensional coupling imbalance state.

[0095] Upon triggering the imbalance determination flag, the imbalance determination module 50 immediately latches the principal feature vector corresponding to the trigger moment. The latched principal feature vector is a vector with dimension 1. The column vector, whose elements have magnitudes and signs, is typically used to characterize the projected weights and contributions of corresponding mechanical or physiological parameters in the current system imbalance event. This latched vector provides a reference direction for subsequent retrospective analysis and closed-loop regulation.

[0096] In this embodiment, step S60 is executed by the gain mapping module 60.

[0097] In the preceding steps, the system latches the principal eigenvector representing the multidimensional coupling imbalance. However, this vector resides in an algebraic eigenspace, where each component is a linear combination of parameters with different physical dimensions, and cannot be directly used as a control signal for the underlying mechanical actuator. To achieve closed-loop regulation of the system, this eigenvector needs to be mapped back to a specific mechanical physical space to find a more effective control variable under the current operating conditions. Therefore, step S60 may include the following sub-steps:

[0098] S601 estimates the current vascular resistance operating point status of the patient by combining normalized mean arterial pressure, normalized central venous pressure, and main pump blood flow parameters.

[0099] The systemic vascular resistance index is a physiological indicator for assessing a patient's vasomotor status; its value directly reflects the afterload of the cardiovascular system to external mechanical pumping. The gain mapping module 60 extracts the normalized mean arterial pressure parameter from the thermodynamically normalized physiological vector generated in the previous step, simultaneously extracting the normalized central venous pressure parameter, and extracts the current main pumping blood flow from the mechanical state column vector. Combined with the patient's body surface area, the system calculates the current estimated value of the systemic vascular resistance index. The specific calculation formula is as follows: ; In the formula, Indicates time The estimated value of the systemic circulation resistance index; and Representing time respectively Mean arterial pressure and central venous pressure after thermodynamic normalization; This is a universal unit conversion constant; Main pump blood flow; The patient's body surface area is typically calculated using height and weight and written to memory during system initialization. This is to avoid issues arising when the system pump stops or operates at extremely low flow rates. To prevent algorithmic dead zones caused by division by zero, the gain mapping module 60 sets a flow protection lower limit (e.g., 0.5 liters / minute). When the flow rate collected by the sensor is lower than this lower limit, the blood flow parameter in the denominator will be forced to use this lower limit value for calculation.

[0100] S602, the estimated resistance state is used as a scheduling variable to match and retrieve the corresponding Jacobian matrix from the local linearization matrix library preset in the system read-only memory.

[0101] The human cardiovascular system is a highly nonlinear, time-varying system, and a single sensitivity matrix cannot cover all operating conditions of cardiopulmonary bypass. As a preferred approach, the gain mapping module 60 employs a gain scheduling strategy, using the calculated systemic resistance index as the scheduling variable. A locally linearized matrix library can be pre-built in the system's read-only memory. This matrix library is constructed through system identification based on a large amount of offline historical clinical data, covering the dynamic characteristics of different vascular resistance ranges. In one embodiment, the matrix library is represented at the underlying data structure as a series of interrelated interval boundary key-value pairs and multidimensional matrix data blocks. Specifically, researchers typically use core hemodynamic indicators such as the systemic resistance index as classification dimensions to divide the overall vasomotor state that a patient may experience during surgery into multiple continuous and well-defined working intervals, such as severe vasodilation, mild vasodilation, normal steady state, mild vasoconstriction, and severe vasoconstriction. Each physical interval in the library corresponds to a fixed Jacobian matrix, which represents a locally linearized approximate model of the cardiovascular system within that specific resistance range. Regarding the specific construction process of the aforementioned matrix library, as a preferred approach, its internal data is derived from a large amount of offline historical clinical data through system identification methods. During the offline preparation phase before the equipment leaves the factory, engineers extract multidimensional time-series data from past extracorporeal circulation surgeries from medical databases. This historical data comprehensively includes the step changes in the input of mechanical actuators (such as the main pump and variable-temperature water tank) at different time points, and the resulting deviations in the patient's physiological response. Engineers categorize the massive number of samples into preset working intervals according to the vascular resistance state at the time of collection. Based on this, using known system identification techniques such as multivariate least squares or subspace identification, the optimal mapping matrix that minimizes the sum of squared input-output prediction errors within that interval is calculated and stored in memory. Each Jacobian matrix stored in the library essentially represents a set of partial derivatives. These partial derivatives objectively reflect the sensitivity ratio and transmission direction of the influence of small perturbations in the underlying mechanical parameters on various physiological characteristics near a specific physiological steady-state working point. By combining offline segmentation identification with online dynamic addressing, the system can solve nonlinear differential equations in real time without consuming the core processor's computing power. The device only needs to look up a table to quickly access a locally linear matrix that is highly adapted to the current physiological characteristics based on the currently estimated resistance state, thus achieving a segmented and accurate approximation of the nonlinear physiological system while ensuring real-time computation.

[0102] Furthermore, in actual configuration, the gain mapping module 60 divides the resistance state into multiple intervals based on clinically accepted hemodynamic boundaries. For example, it divides the resistance state into multiple intervals. Less than 1500 dyn·s·cm -5 ·m 2 Set as the vasodilatory zone, 1500 to 2500 dyn·s·cm -5 ·m 2 Set as the normal range, Greater than 2500 dyn·s·cm -5 ·m 2 Set as the vasoconstriction zone. Gain mapping module 60 will then calculate the current... The numerical range is compared with the index boundaries of the matrix library. If the current resistance estimate falls within a specific resistance range, the gain mapping module 60 directly addresses and retrieves the local Jacobian matrix corresponding to that range. This retrieved Jacobian matrix characterizes the partial derivative response of small changes in the system's mechanical intervention parameters to the projection of physiological characteristics near the current physiological steady-state operating point.

[0103] S603 performs a matrix multiplication operation between the retrieved Jacobian matrix and the latched principal eigenvector to complete the dimensionality reduction mapping from the algebraic eigenspace to the specific mechanical physical space.

[0104] After obtaining the matching local Jacobian matrix, the gain mapping module 60 uses this matrix to perform an algebraic transformation on the latched principal eigenvector to solve for the adjustment direction required to restore system equilibrium. Specifically, the gain mapping module 60 performs matrix multiplication on the retrieved Jacobian matrix and the latched principal eigenvector at the time of imbalance. The algebraic mapping formula is as follows: ; In the formula, This represents the generated actuator pointing vector, whose dimension is the same as the number of controllable mechanical actuators at the system's underlying level, denoted as . Column vectors; To match the current resistance range with the Jacobian matrix, in order to achieve the transition from... 3D state space to The projection transformation of the mechanical execution space, the dimension of which is set to 1. ; To trigger the imbalance at the moment The dimension latched by the imbalance determination module is The main eigenvectors. Through the above linear mapping, the system transforms the algebraic trend of multidimensional state deviations into the theoretical adjustment direction and relative adjustment amplitude corresponding to each specific mechanical actuator.

[0105] S604 iterates through all components of the vector pointed to by the actuator, extracts the value of the component with the largest absolute value and its polarity, locks the target regulator, and uses it as the error input source for the subsequent controller.

[0106] In the control of multi-input multi-output extracorporeal circulation devices, simultaneous adjustment of multiple mechanical parameters can easily trigger cross-coupling oscillations in the patient's physiological feedback. To ensure the stability of system operation, the gain mapping module 60 adopts a decoupled single-variable priority intervention strategy.

[0107] In this embodiment, the gain mapping module 60 traverses the generated actuator pointing vector. Internal Given ... Gain mapping module 60 finds the condition that satisfies The amount and the index number of its dimension Considering the possibility of mathematical collisions where multiple components have completely equal absolute values ​​in discrete digital computation, a one-dimensional actuator priority array is built into the gain mapping module 60 to avoid addressing anomalies. When a mathematical collision occurs where components have equal absolute values, the system, based on the built-in actuator priority array, prioritizes the index number corresponding to the mechanical component with the more critical physical response (e.g., prioritizing intervention on the main pump speed, followed by the variable temperature water tank temperature). This index number... It directly targets the regulator that is most sensitive to stabilizing imbalances under the current operating conditions.

[0108] Extracted components The polarity indicates the physical direction in which the target regulator should increase or decrease the output power, while its absolute value is directly assigned as the error input term for the closed-loop control of the target regulator. To prevent frequent vibrations of mechanical components caused by minute execution signals generated during calculation, the gain mapping module 60 sets a physical resolution dead zone threshold for the error input item. This threshold is configured independently based on the mechanical characteristics of different actuators; for example, the dead zone threshold for the main pump speed can be set to 5 revolutions per minute. If the extracted maximum absolute value component is still less than the physical motion resolution limit of the actuator, the system will reset the error input item. Forced zeroing maintains the current mechanical output, thus balancing the sensitivity of system control with the reliability of hardware execution.

[0109] In this embodiment, step S70 is executed by the security output module 70.

[0110] After obtaining the specific error input term for the target regulator, the system needs to convert it into an electrical drive signal that can be recognized by the hardware, and control the underlying mechanism while ensuring patient safety. Through control law calculation and boundary constraints, the system can achieve closed-loop correction of the imbalance state. Therefore, step S70 may include the following sub-steps:

[0111] S701 substitutes the error input source into the proportional-integral-derivative controller of the independent target regulator to calculate the original control adjustment quantity used to counteract coupling drift.

[0112] In this embodiment, the system assigns independent control channels to each underlying target regulator (such as the main pump motor, variable temperature water tank heating rod, oxygenator valve, etc.). The safety output module 70 acquires the error input items output from the preceding steps. This value is then substituted into the discrete proportional-integral-derivative (DI-D) controller of the corresponding target regulator. To avoid integral saturation caused by large state fluctuations in the system during full-position calculations, the safety output module 70 employs an incremental algorithm as a preferred approach. This algorithm calculates the error change between the current cycle and the previous cycle and outputs the adjustment increment of the target regulator. Its discrete calculation formula is as follows: ; In the formula, Representing discrete time The original control adjustment increment calculated for the current target regulator; The error input item is for the current time. and These represent the historical error input items from the previous one and the previous two control cycles, respectively. This is the proportional gain coefficient, used to adjust the system's response strength to the current error; This is the integral gain coefficient, used to accumulate and eliminate steady-state deviation; This is the differential gain coefficient, used to predict error trends and suppress overshoot.

[0113] For the tuning of the aforementioned proportional, integral, and differential gain coefficients, those skilled in the art can perform offline calibration using the critical proportional gain method. Considering the safety of the extracorporeal circulation system, to prevent mechanical overshoot from causing drastic fluctuations in the patient's physiological parameters such as blood pressure, the system is usually configured in an overdamped or critically damped state. The tuning process of these coefficients is well-known in the art and will not be described in detail here.

[0114] After obtaining the increment, the system adds it to the actual output command value of the previous time step to generate the total amount of original control commands at the current time step. Its algebraic expression is: ; In the formula, This represents the actual control command value issued by the system to the lower level at the previous discrete moment.

[0115] Furthermore, in this embodiment, for other underlying mechanical components that are not locked as the target regulator within the current control cycle, the safety output module 70 will skip the current cycle incremental calculation of their internal PID controller and directly force their current cycle control adjustment increment. In other words, these non-target regulators will maintain the actual output command value of the previous cycle unchanged, thereby strictly implementing the decoupled single-variable priority intervention strategy at the physical level and preventing the cross-coupling oscillation of the patient's physiological parameters caused by the simultaneous opposite movements of multiple mechanical components.

[0116] S702 extracts the safe operating limit boundary of the corresponding mechanical component pre-stored in the memory, and uses the saturation function to perform forced amplitude limiting and truncation on the original control adjustment quantity.

[0117] In the actual operation of medical equipment, to reduce the risk of mechanical failure or physiological damage, the control commands calculated by the algorithm need to undergo physical boundary review before being issued. The safety output module 70 accesses the system's internal non-volatile memory to retrieve the safe operating limit boundaries corresponding to the current target regulator. These boundary parameters are typically set by the equipment manufacturer based on the physical tolerance limits of the mechanical components and clinical operating procedures. For example, the upper limit of the main pump speed may be limited to 4000 rpm, and the upper limit of the variable temperature water tank temperature may be limited to 42 degrees Celsius to prevent physical damage or thermal injury to the blood.

[0118] After obtaining the upper and lower limit thresholds, the safety output module 70 uses a saturation clamping function to adjust the total amount of the original control commands calculated above. Forced clipping is applied. The mathematical expression for this saturation clamping mechanism is as follows: ; In the formula, This represents the final control command value generated after physical threshold interception; and These represent the upper and lower physical operating limits of the corresponding target regulator, respectively. Through this hard threshold interception mechanism, the system can clamp the command output within a safe physical range under conditions of sudden sensor changes or accumulated errors in calculations, thus preventing mechanical loss of control.

[0119] Furthermore, to avoid desaturation hysteresis in the controller—that is, when the control command triggers the amplitude limiting boundary, if subsequent cycles continue to accumulate based on the theoretical value exceeding the boundary, resulting in a slow system response when the error reverses—the safety output module 70, after completing the clamping calculation, will forcibly convert the actual output value of the current cycle to the actual output value. Feedback is then written to the controller's internal status register, serving as a historical output item in the incremental superposition formula for the next calculation cycle. This feedback mechanism ensures strict alignment between the algorithm's accumulated state and the actual physical execution state.

[0120] The S703 encapsulates the control quantities that have been intercepted by the hard threshold into a standard data frame format and writes them to the underlying control board of the extracorporeal circulation via the physical bus interface to complete closed-loop mechanical correction.

[0121] After generating the final control command value, the system needs to convert the digitized command into a communication message that the lower-level machine can parse. The safety output module 70 extracts... The numerical values ​​are encapsulated into the payload segment of a standard data frame according to pre-defined communication protocol rules. This data frame typically includes a frame header identifier, target regulator address encoding, command data bits, and a cyclic redundancy check code to meet the anti-interference requirements in industrial environments.

[0122] After completing the data encapsulation operation, the safety output module 70 calls the underlying hardware driver to send the generated data frame to the underlying control board of the extracorporeal circulation machine via a physical bus interface (such as a controller area network bus or industrial serial bus). After receiving and verifying the data frame, the microprocessor on the underlying control board drives the corresponding power amplifier circuit or stepper motor driver to change the operating state of the mechanical components, thereby completing the closed-loop adjustment operation. For the specific data frame encapsulation format and the underlying bus communication verification process, those skilled in the art can refer to relevant industrial communication standard protocols. The communication handshake and verification mechanisms are well-known technologies in the field and will not be elaborated upon here.

[0123] To enable those skilled in the art to more clearly understand the purpose, technical solution, and advantages of this invention, the present invention will be further described in detail below with reference to specific application embodiments, real experimental test data, and corresponding drawings. It should be noted that the embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0124] I. Specific Application Examples

[0125] 1. Initial steady state (corresponding to baseline extraction S502): Body surface area (BSA) is 1.8m². 2 The patient is undergoing cardiopulmonary bypass surgery. During the stable induction period, sensing module 10 continuously collects data: Mechanical parameters: Main pump blood flow L / min, water temperature of variable temperature water tank ℃.

[0126] Physiological parameters: mean arterial pressure mmHg, central venous pressure mmHg, nasopharyngeal temperature ℃.

[0127] The imbalance determination module 50 extracts and calculates the dynamic baseline scalar during this period. .

[0128] 2. Sudden imbalance event (corresponding to S10-S50): During the maintenance period of surgery (45 minutes), the patient experiences sudden peripheral vasodilation due to the metabolism of anesthetic drugs or systemic inflammatory response. It dropped rapidly to 55 mmHg within 30 seconds.

[0129] S30 (normalized): Temperature gradient rate at this point When the value is close to 0, the normalization module 30 determines that there is no significant thermodynamic drift interference and directly outputs the true physiological vector of low blood pressure.

[0130] S40-S50 (Covariance and Imbalance Determination): With... The sudden drop, the covariance matrix of mechanical input and physiological output Distortion has occurred. Principal eigenvalues It quickly surged to 1.85.

[0131] The system calculates the relative deviation percentage. The threshold (30%) has been exceeded. The imbalance detection module 50 immediately triggers a high level and latches the current main feature vector. .

[0132] 3. Intelligent mapping and decoupled execution (corresponding to S60-S70) S601-S602 (State Estimation and Matrix Addressing): Gain Mapping Module 60 calculates the current estimated value of the systemic circulation resistance index. dyn·s·cm -5 ·m 2 The system read-only memory was consulted, and the resistance value fell within the vasodilation range. The system then retrieved the corresponding local Jacobian matrix. .

[0133] S603-S604 (Dimensionality Reduction and Locking): Matrix Dot Multiplication Then, the executor pointing vector is obtained. (Dimension 1 is the main pump speed increment, and dimension 2 is the water temperature increment). Comparing the absolute values ​​reveals |210|≫|−0.5|, therefore the system locks the main pump as the target regulator, and the error input term... .

[0134] S70 (Control Release): Safety output module 70 will... Substituting the values ​​into the incremental PID controller of the main pump, it is calculated that the main pump speed should increase by 30 rpm in this cycle to raise the blood flow to 4.8 L / min to compensate for the drop in blood pressure. Simultaneously, the control increment of the forced temperature variable water tank... The water temperature is maintained at a constant 36.0℃. This avoids system oscillations caused by the simultaneous action of multiple variables.

[0135] II. Experimental Verification and Effect Comparison

[0136] To further verify the effectiveness and technical advantages of the present invention, a composite interference test was conducted in a hardware-in-the-loop simulation platform, and response data of the comparative example (traditional multi-loop independent PID control system) and the embodiment of the present invention (digital analysis and processing control system) were extracted.

[0137] The test results are as follows: according to Figure 3 It can be seen that when a purely physical cooling interference is injected at the 100th second, the comparative curve (thin dashed line) fails to isolate thermodynamic drift, resulting in a false increase in mean arterial pressure, causing the system to erroneously reduce the main pump speed. Furthermore, when a real pathological vasodilation occurs at the 110th second, the blood pressure in the comparative curve drops sharply, accompanied by prolonged and violent oscillations (the stabilization time is as long as 42.5 seconds). In contrast, the curve (thick solid line) of the present invention shows no fluctuation at the 100th second, proving that the normalization module successfully isolates the purely physical thermal drift, with a false imbalance false alarm rate of 0%. When faced with a real pathological imbalance at the 110th second, the present invention can quickly respond and smoothly raise the blood pressure back to the target level (the stabilization time is shortened to 18.1 seconds), significantly improving the system's ability to compensate for real pathological conditions under complex operating conditions.

[0138] according to Figure 4 It can be seen that the multidimensional coupled principal eigenvalues ​​of the system state exhibit highly sensitive dynamic evolution over time. During the period from 100 to 110 seconds, despite the physical cooling disturbance, the principal eigenvalues ​​remain stable at the dynamic baseline scalar (…). The pathological imbalance occurred near the 110-second mark; however, after the sudden onset of the imbalance, the principal eigenvalue curve experienced a step increase with an extremely high slope, breaking through the preset imbalance judgment threshold (+30%) within 3.2 seconds. The system then triggered the imbalance judgment and successfully latched the principal eigenvector at the trigger point (the black dot marked in the figure). This fully demonstrates that the covariance matrix and principal component eigenvalue decomposition algorithm used in this invention can quickly capture the core variation direction from massive heterogeneous data, and its pathological imbalance recognition delay is reduced by approximately 79% compared to the traditional scheme (15.4 seconds).

[0139] according to Figure 5 and Figure 6It can be seen that in the underlying execution stage of imbalance correction, the main pump speed and the variable temperature water tank temperature in the comparative example both exhibited high-frequency cross-coupling oscillations (up to 6 times) after being disturbed. This multivariate disordered intervention is highly likely to cause secondary physiological damage to the patient. However, in the embodiment of this invention, after triggering closed-loop control, Figure 5 The target regulator (main pump, thick solid line) in the model exhibits a step-like, smooth acceleration compensation trend; crucially, Figure 6 The control command curve of the non-target regulator (variable temperature water tank, thick solid line) in the middle is strictly maintained as a horizontal straight line, that is, the physical action increment is forced to be zero. This confirms that the single-variable priority intervention strategy of actuator gain mapping and decoupling proposed in this invention can be accurately implemented by the system safety output module, completely eliminating the risk of hemodynamic cross-oscillation caused by the simultaneous action of multiple mechanical components.

[0140] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various 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 digital analysis processing control system for cardiac surgical medical information, characterized by, include: The sensing module is used to acquire mechanical and physiological parameters transmitted by heterogeneous medical devices; The synchronization module is used to align the mechanical execution parameters and physiological parameters on the time axis and output the aligned data sequence. The normalization module is used to extract the aligned temperature parameters, calculate the thermodynamically normalized physiological vector, and concatenate them to generate a comprehensive state vector. The covariance module is used to calculate and update the covariance matrix in real time for the comprehensive state vector set; The imbalance determination module is used to perform eigenvalue decomposition on the covariance matrix to extract the principal eigenvalues ​​and principal eigenvectors, and output a trigger signal and latch the principal eigenvectors when the deviation rate of the principal eigenvalues ​​is greater than the determination threshold. The gain mapping module is used to map the latched main feature vector to generate the actuator pointing vector and extract the error input term after receiving the trigger signal. The safety output module is used to calculate the total amount of the original control commands based on the error input items, perform physical threshold interception, and generate the final control command value to be sent to the lower-level execution mechanism of the extracorporeal circulation machine. The sensing module is specifically used for: The main pump blood flow and variable temperature water tank temperature parameters of the extracorporeal circulation machine are read through the serial communication bus to construct a mechanical state column vector, which serves as the mechanical execution parameter. Physiological feedback column vectors are constructed by parsing the mean arterial pressure, central venous pressure, and patient's nasopharyngeal temperature parameters output by the multi-parameter monitor via the Ethernet interface, and are used as the physiological parameters. The gain mapping module, when mapping the latched principal feature vector to generate the executor pointing vector, is specifically used for: The estimated value of the body circulation resistance index is calculated to address the corresponding Jacobian matrix, and the Jacobian matrix is ​​multiplied by the latched principal eigenvector to generate the actuator pointing vector. Normalized mean arterial pressure and normalized central venous pressure are extracted from the thermodynamically normalized physiological vector, and main pump blood flow parameters are extracted from the mechanical actuation parameters. The estimated value of systemic circulatory resistance index is calculated by combining the patient's body surface area. The estimated value of the body circulation resistance index is used as a scheduling variable. The numerical range of the estimated value of the body circulation resistance index is compared with the index boundary of the preset local linearization matrix library in the system's built-in read-only memory to address the corresponding Jacobian matrix. The retrieved Jacobian matrix and the latched principal eigenvector are multiplied by a matrix to achieve dimensionality reduction mapping, generating the executor pointing vector. When the safety output module calculates the total number of original control commands based on the error input item and executes physical threshold interception, it is specifically used for: Substitute the error input term into the proportional-integral-derivative controller of the independent target regulator to calculate the original control adjustment amount used to offset coupling drift as the total amount of original control command; For underlying mechanical components that are not locked as target regulators within the current control cycle, the control adjustment increment for this cycle is forced to be zero in order to maintain the actual output command value unchanged. The physical threshold interception is performed using a saturation clamping function. The safe operating limit boundary of the corresponding target regulator is extracted from the memory. The absolute physical upper limit value and absolute physical lower limit value of the corresponding target regulator are retrieved. The total amount of the original control command is forcibly truncated using the saturation clamping function to generate the final control command value. The final control command value is then encapsulated into a standard data frame format and written to the outer circulation bottom layer control board through the physical bus interface.

2. The digital analysis processing control system for cardiac surgical medical information according to claim 1, characterized by, The synchronization module, when aligning data timelines, is specifically used for: The mechanical and physiological parameters are assigned microsecond-level timestamps, and resampling and low-pass filtering are performed. Each data packet entering a physically isolated circular buffer is timestamped at the microsecond level. Based on the unified clock domain of the system, a linear interpolation algorithm is used to perform interpolation calculations on heterogeneous data with timestamps to estimate the data values ​​at the sampling time of the standard system in order to eliminate the transmission delay difference of the communication bus. The aligned data sequence is processed using a Butterworth low-pass filter with a preset cutoff frequency, and the output is a smooth and time-strictly aligned data stream.

3. The digital analysis and processing control system for cardio-surgical medical information according to claim 1, characterized in that, The temperature parameters include the temperature parameters of the variable temperature water tank and the nasopharyngeal temperature; the normalization module, when calculating the thermodynamically normalized physiological vector, specifically includes: The temperature gradient rate is calculated to generate a diagonal compensation matrix, and the diagonal compensation matrix is ​​used to calculate the thermodynamically normalized physiological vector. Extract the time-varying derivative of the difference between the variable temperature tank water temperature parameter and the nasopharyngeal temperature parameter within the same time slice to obtain the variable temperature gradient rate; Substituting the temperature gradient rate into the built-in fluid dynamics viscosity-temperature compensation equation generates independent compensation coefficients for each of the physiological parameters. A diagonal compensation matrix is ​​constructed using the independent compensation coefficients of each of the physiological parameters, and the vector composed of the physiological parameters is multiplied by the diagonal compensation matrix to obtain a thermodynamically normalized physiological vector stripped of physical thermal drift properties.

4. The digital analysis and processing control system for cardio-surgical medical information according to claim 3, characterized in that, The normalization module, when generating the comprehensive state vector, specifically includes: The original mechanical state column vector composed of the mechanical execution parameters is concatenated with the normalized thermodynamic normalized physiological vector in the dimensional direction to form a comprehensive state vector for subsequent algebraic operations.

5. The digital analysis and processing control system for cardiac surgical medical information according to claim 1, wherein, The covariance module, when calculating and updating the covariance matrix in real time for the comprehensive state vector set, is specifically used for: Within a set time sliding window, the covariance matrix is ​​calculated and updated in real time for the comprehensive state vector set; A queue containing a fixed number of discrete sampling points is allocated in the system memory as the time sliding window; The summation and arithmetic mean of all the accumulated comprehensive state vectors within the time sliding window are performed to obtain the mean vector. The covariance matrix is ​​calculated using discrete vector samples within the time sliding window and the mean vector. The covariance matrix reflects the linear coupling relationship between mechanical actuation parameters and physiological parameters within the time sliding window.

6. The digital analysis and processing control system for cardiac surgical medical information according to claim 1, wherein, When the imbalance determination module outputs a trigger signal and latches the main feature vector, it is specifically used for: The covariance matrix generated in real time is subjected to algebraic decomposition to obtain the eigenvalue diagonal matrix and eigenvector matrix arranged in descending order of absolute value. The eigenvalue with the largest absolute value is extracted as the principal eigenvalue, and the corresponding column vector is extracted as the principal eigenvector. Extract the time average of the principal eigenvalue with the largest absolute value during the stable induction period of surgery in the initial stage of the system and store it in a register as a dynamic baseline scalar. Calculate the absolute value of the difference between the current principal feature value and the dynamic baseline scalar in real time, and calculate the relative deviation percentage; When the relative deviation percentage continuously exceeds the preset imbalance judgment threshold constant, and the value of the time dwell counter accumulates to exceed the set anti-shake number threshold, the judgment system falls into implicit multidimensional coupling imbalance and outputs a trigger signal, while simultaneously latching the main feature vector at the current moment.

7. The digital analysis and processing control system for cardio-surgical medical information according to claim 1, characterized in that, The gain mapping module is specifically used for: extracting error input terms. Extract the component with the largest absolute value as the error input; Iterate through all components of the vector pointed to by the actuator, compare and extract the value and polarity of the component with the largest absolute value to lock the target regulator; when a mathematical collision occurs where the absolute values ​​of the components are equal, lock the target regulator according to the index number of the built-in actuator priority array. The extracted absolute value of the maximum component is used as the error input term of the target regulator; when the extracted absolute value of the maximum component is less than the physical resolution dead zone threshold set by the corresponding target regulator, the error input term is forced to zero.

Images