Dynamic noninvasive cardiac output monitoring system and method

Abstract

The invention belongs to the technical field of noninvasive hemodynamic monitoring, and provides a dynamic noninvasive cardiac output monitoring system and method. High-quality dynamic pulse wave waveform signals are collected based on a sensor array, motion artifacts in the motion state are corrected, the motion artifacts meet input conditions of a cardiac output calculation model, and dynamic pulse wave feature vectors are calculated; calculating an aorta cross-sectional area estimated value as a correction variable based on the personalized characteristic parameters; and based on a cardiac output model of deep learning training, cardiac output is calculated so as to realize real-time monitoring and evaluation of the dynamic cardiac output of the testee. According to the invention, dynamic, real-time and long-term monitoring of cardiac output parameters can be realized, intermittent correction is not needed, and the system is suitable for household rehabilitation of patients with heart diseases.

Description

Technical Field

[0001] The present application relates to the technical field of hemodynamic monitoring, and in particular to a dynamic non-invasive cardiac output monitoring system and method. Background Art

[0002] Non-invasive, long-term cardiac output parameter monitoring is of great significance for home rehabilitation function monitoring of patients with heart failure and structural heart disease. However, existing technologies and products have not yet been able to truly achieve long-term, dynamic monitoring of cardiac output. There are existing patents that can establish models related to cardiac output through pulse wave characteristics and personalized information of patients. Among them, some technologies require the use of a sphygmomanometer to correct the pulse wave characteristics of peripheral arteries at certain time intervals, which is not suitable for long-term monitoring scenarios. Patent CN103784132A can complete the measurement of cardiac output in a resting state, but does not target the extraction method of feature vectors in active or motion states; Patent CN102499669A can perform waveform contour analysis on the pulse wave waveform in motion to obtain the pulse wave arrival time and the pulse wave inflection point area ratio as feature vectors in motion, but fails to accurately calculate cardiac output. Therefore, it is necessary to develop a high-quality dynamic cardiac output monitoring model to meet the needs of long-term monitoring of cardiovascular patients in home rehabilitation. Summary of the invention

[0003] In order to solve the technical problem that the existing technology cannot achieve dynamic, continuous, long-term and high-precision monitoring of cardiac output, the present invention provides a dynamic non-invasive cardiac output monitoring system and method, which can not only directly measure and collect pulse waves through peripheral arteries, but also can be applied to resting states and daily exercise scenes, thereby realizing dynamic measurement of individual cardiac output, which is of great significance for home rehabilitation monitoring of patients with heart failure and structural heart disease.

[0004] The specific plan is as follows: A dynamic non-invasive cardiac output monitoring system, comprising: a signal acquisition module, a filtering module, a pulse wave feature extraction module, a dynamic signal quality monitoring module, and a cardiac output monitoring and evaluation module; The signal acquisition module is used to continuously acquire multi-channel photoplethysmography signals and accelerometer signals at different peripheral artery sites in a resting or active state, and filter and amplify them to obtain a dynamic pulse wave waveform signal; The filtering module can correct the baseline drift and motion artifacts in the dynamic pulse wave waveform signal in combination with the accelerometer signal and the adaptive filtering algorithm to obtain a pulse wave correction signal; The pulse wave feature extraction module is used to perform pulse wave feature point recognition, feature extraction, and screening of pulse wave feature vectors related to cardiac output calculation on the pulse wave correction signal and perform normalization processing to obtain a normalized feature vector; The dynamic signal quality monitoring module is used to monitor the quality of normalized feature vectors in different states, and use the normalized feature vectors that meet the input conditions of the cardiac output calculation model as dynamic pulse wave feature vectors; the dynamic pulse wave feature vectors are extracted by comparing the number and value range of normalized feature vectors extracted from the subject in an active state and a resting state, and outputting normalized feature vectors that are lower than the difference threshold; The cardiac output monitoring and evaluation module is used for dynamically and continuously monitoring the stroke volume and cardiac output of the subject, and includes: a correction variable estimation submodule, a pulse wave feature vector estimation submodule and a cardiac output evaluation submodule; the correction variable estimation submodule corrects the variable based on the personalized feature parameter input by the subject; the pulse wave feature vector estimation submodule sequentially calls the filtering module and the dynamic signal quality monitoring module to process the dynamic pulse wave waveform signal collected by the signal acquisition module to obtain a dynamic pulse wave feature vector; the cardiac output evaluation submodule inputs the correction variable and the dynamic pulse wave feature vector to obtain the stroke volume and cardiac output, and dynamically displays them on the system interface.

[0005] Preferably, the signal acquisition module includes a sensor array, a pressure control submodule, a waveform construction submodule, and an amplification filtering and transmission submodule; The sensor array is used to dynamically and continuously collect photoelectric signals, contact pressure signals and accelerometer signals of peripheral arterial pulse waves at different parts of the subject in motion; the sensor array includes at least two photoelectric sensors for respectively collecting pulse waves at different peripheral arterial parts, a pressure sensor for collecting contact pressure at corresponding parts, and an accelerometer for measuring displacement and velocity information of the subject; The pressure control submodule is used to apply contact pressure to the peripheral artery and control it during the acquisition process, and includes: an airbag that applies contact pressure to the skin, and a pressure control device that controls the contact pressure through the airbag; The waveform construction submodule combines the photoelectric signals and contact pressure signals collected by the sensor array to construct a multi-channel photoplethysmography signal; The amplification, filtering and transmission submodule is used to amplify and filter the multi-channel photoplethysmography signals and accelerometer signals in sequence to obtain a dynamic pulse wave waveform signal, and transmit it to the filtering module of the host computer.

[0006] Preferably, the dynamic signal quality monitoring module comprises: a signal receiving submodule and a signal quality determination submodule; The signal receiving submodule is connected to the output end of the pulse wave feature extraction module, receives the normalized feature vector and marks the corresponding signal channel; The signal quality judgment submodule judges whether the normalized feature vectors of different signal channels meet the input conditions of the cardiac output calculation model based on the signal quality rules, wherein the signal quality rules include: whether there is a lack of pulse wave feature vectors related to the cardiac output calculation; The pulse wave characteristic vector related to the cardiac output calculation includes: pulse wave characteristic points, amplitude changes of each characteristic point of the pulse wave, time intervals between adjacent characteristic points, the area under the pulse wave systolic curve, and the area under the pulse wave diastolic curve; Based on the difference judgment rule, the difference between the pulse wave feature vector value in the active state and the resting state is compared and scored, and the difference includes: amplitude change range difference, time range difference, and area range difference; the difference judgment rule includes: if the feature vector difference of all channels is less than the threshold, the normalized feature vector in the active state is directly used as the dynamic pulse wave feature vector and output; if the normalized feature vector difference measured in at least one part is greater than the threshold, it is necessary to reduce the influence of motion artifacts or other factors on the pulse wave signal, so that the subject reduces the movement speed to the resting state and checks whether the sensor wearing position changes. When the difference is less than a certain threshold, the dynamic pulse wave feature vector is output.

[0007] Preferably, the sensor array is composed of at least two photoelectric sensors and at least one pressure sensor arranged longitudinally or transversely along the peripheral artery, an optical sensor for filtering out ambient light interference, and an accelerometer worn on the wrist of the subject.

[0008] Preferably, the sensor array is arranged as follows: at least one photoelectric sensor and at least one pressure sensor correspond to the ulnar artery at the wrist to obtain the dynamic pulse wave waveform signal of the ulnar artery, and at least one photoelectric sensor and at least one pressure sensor correspond to the radial artery at the wrist to obtain the dynamic pulse wave waveform signal of the radial artery.

[0009] Preferably, another arrangement of the sensor array is: at least two photoelectric sensors for collecting peripheral arterial pulse waves and one pressure sensor for collecting contact pressure corresponding to the ulnar artery or radial artery at the wrist, the sensors are arranged longitudinally along the blood vessel, wherein the pressure sensor is located at the center, forming a hierarchical structure; the photoelectric signal collected by each photoelectric sensor is respectively combined with the contact pressure signal to form a two-channel dynamic pulse wave waveform signal.

[0010] Preferably, the personalized characteristic parameter generates an estimated value of the aortic cross-sectional area based on a relationship model, and the relationship model is: (1) S: estimated aortic cross-sectional area; Age: age; BSA: body surface area; Height: height; Weight: weight; a, b, c, e: correlation coefficients, and f: correction coefficient.

[0011] Preferably, the characteristic points of the pulse wave include: the starting point of the pulse wave, the main peak point of the pulse wave, the dicrotic wave peak point, and the end point of the pulse wave cycle; the pulse wave characteristic vector related to the calculation of cardiac output includes: the amplitude change of each characteristic point of the pulse wave, the time interval between adjacent characteristic points, the area under the curve during the systolic period of the pulse wave, and the area under the curve during the diastolic period of the pulse wave.

[0012] Preferably, the cardiac output calculation model is obtained based on deep learning model training sample data, and the sample data is divided into a training set and a test set. In the training set, the model input includes: correction variables and pulse wave feature vectors related to cardiac output calculation, and the model output is stroke volume. The cardiac output is calculated based on stroke volume and heart rate. The deep learning model selects an artificial neural network model for training; the test set is used to test the trained deep learning model and evaluate it based on evaluation indicators, and change the deep learning model parameters until the deep learning model that meets the test and indicator requirements is used as the final cardiac output calculation model; the correction variable is the aortic cross-sectional area calculated based on the relationship model between the subject's personalized characteristic parameters and the aortic cross-sectional area; the evaluation indicators include: model accuracy, recall rate, precision rate, etc.

[0013] Preferably, cardiac function parameters that can also be calculated based on cardiac output and pulse wave signals include but are not limited to: Contractility Index (CTI), Ventricular Ejection Time (VET), Heart Rate (HR), continuous systolic blood pressure (SBP), continuous diastolic blood pressure (DBP) and continuous mean arterial pressure (MAP), pulse pressure variation (PPV), stroke volume variation (SVV), peripheral vascular resistance (SVR) and peripheral vascular resistance index (PVR).

[0014] A dynamic non-invasive cardiac output monitoring method, Step 1, pulse signal acquisition: based on a sensor array and photoplethysmography, dynamically and continuously acquire multi-channel photoplethysmography signals and accelerometer signals of different peripheral arterial sites of the subject in a resting or active state, and perform filtering and amplification to obtain a dynamic pulse wave waveform signal, wherein the sensor array comprises at least two photoelectric sensors for acquiring pulse waves of different arterial sites, and an accelerometer for measuring displacement and velocity information of the subject; Step 2, pulse signal filtering: performing motion artifact and baseline drift correction on the dynamic pulse waveform signal in step 1 to generate a pulse wave correction signal; Step 3, pulse wave feature extraction: performing pulse wave feature point identification on the pulse wave correction signal described in step 2, performing feature extraction on the pulse wave correction signal based on the pulse wave feature points, screening the pulse wave feature vector related to the cardiac output calculation and performing normalization processing to obtain a normalized feature vector; Step 4, dynamic signal quality monitoring: compare the number and value range of the normalized feature vectors extracted from the subject in the active state and the resting state, and judge whether they meet the input conditions of the cardiac output calculation model based on the set difference threshold, so as to obtain the dynamic pulse wave feature vector that meets the difference threshold; Step 5, generating correction variables: the subject inputs personalized characteristic parameters, and generates an estimated value of the aortic cross-sectional area based on a relationship model; the relationship model is an estimated value of the aortic cross-sectional area calculated based on the subject's age, body surface area, height, and weight: Step 6, training the cardiac output calculation model: dividing the sample data into a training set and a test set. In the training set, based on the personalized feature parameters, the estimated value of the aortic cross-sectional area is calculated as a correction variable; further extracting and screening the pulse wave feature vector related to the cardiac output calculation, and performing dynamic signal quality monitoring based on step 4, taking the dynamic pulse wave feature vector and correction variable that meet the requirements as input, and the stroke volume as output, and training the artificial neural network as a pre-training model; and testing and evaluating the trained model based on the test set; changing the deep learning model parameters until the deep learning model that meets the test and index requirements is used as the final cardiac output calculation model; Step 7, cardiac output monitoring and evaluation: call steps 1-5 to generate the subject's correction variables and dynamic pulse wave feature vector as input to the cardiac output calculation model, call the cardiac output calculation model in step 6, output the stroke volume, and display it on the system interface.

[0015] Preferably, the steps of performing motion artifact correction based on the accelerometer and the adaptive filtering algorithm are: A1: collecting dynamic pulse waveform signals and accelerometer signals of peripheral arteries, wherein the accelerometer signals obtain time-displacement signals of the system in space; A2: Signal preprocessing: remove high-frequency noise through low-pass filtering and retain low-frequency signals; remove low-frequency drift noise through high-pass filtering and perform normalization; A3: Feature extraction: Extract the motion features of the time-displacement signal of the accelerometer in the x, y, and z axis directions based on spectrum analysis; further identify the pattern and amplitude change of the motion, and determine the frequency and amplitude features related to the motion artifact; analyze the frequency and amplitude of the pulse wave signal based on spectrum analysis to determine the frequency related to the motion artifact; A4: Adaptive filtering: Using the accelerometer’s time-displacement signal as a reference signal, the filter coefficients are automatically adjusted to remove motion artifacts captured by the accelerometer by minimizing the error between the filter’s output and the target signal; A5: Post-processing and verification: Reconstruct the signal after artifact removal to obtain the pulse wave correction signal to ensure that the difference between the pulse wave feature vector collected in the moving state and the pulse wave feature vector collected in the resting state is less than a certain threshold, and use evaluation indicators to quantitatively evaluate the artifact removal effect.

[0016] Beneficial effects: The present invention is a dynamic non-invasive cardiac output monitoring system and method. The system combines photoplethysmography technology and deep learning algorithms to obtain high-quality pulse wave signals. First, the photoplethysmography signals of the peripheral arterial pulse waves in static and active states at different parts and the accelerometer signals used to correct motion artifacts are continuously collected by the signal acquisition module to obtain dynamic pulse wave signals; after screening and processing, the features related to the calculation of cardiac output are extracted; because the features related to cardiac output have corrected the influence of motion artifacts, a cardiac output monitoring model is established based on these features, thereby realizing dynamic, continuous and long-term monitoring of cardiac output. Deep learning, especially neural networks, can effectively capture the potential deep structure in the data through multi-level data representation and nonlinear transformation, and provide support for the effective establishment of a dynamic cardiac output monitoring model. After dynamically extracting the pulse wave features related to the calculation of cardiac output, the present invention introduces correction variables, and further establishes a module for estimating the monitoring model of cardiac output based on deep learning, including a dynamic signal quality monitoring module, a deep learning algorithm and a cardiac output detection and evaluation module. The dynamic signal quality monitoring module is used to obtain a high-quality dynamic pulse wave feature vector required for the input of the deep learning model. The deep learning algorithm is used together with the dynamic pulse wave feature vector and the aortic cross-sectional area as an input model to perform pre-training to obtain a cardiac output pre-training model, and the pulse wave feature vector and the aortic cross-sectional area are obtained; the cardiac output is calculated based on the stroke volume according to the appropriate structural relationship of the stroke volume, and a continuous and accurate value of the cardiac output is obtained according to the individual's actual measured value.

[0017] The present invention can measure the pulse wave signal in a motion state, and can extract the features related to the calculation of cardiac output based on the pulse wave signal in the motion state, and establish a monitoring model based on these features and correction features that can estimate the cardiac output without the need for sphygmomanometer correction, thereby realizing dynamic, long-term, and accurate monitoring of cardiac output. That is, it can directly measure and collect the pulse wave of the peripheral artery without correction, and it is not necessary to keep the measurement in a resting state as much as possible. The obtained model can more accurately identify the waveform characteristics of the pulse wave and take into account individual differences, realizing dynamic and high-precision measurement of individual cardiac output, which is of great significance for long-term home rehabilitation monitoring of patients with heart failure and structural heart disease.

[0018] In addition, the system also integrates the patient's personalized information and pulse wave characteristics to build a deep learning model, thereby truly realizing dynamic, long-cycle, and high-precision cardiac output monitoring.

[0019] First, the present invention solves the problem of difficulty in collecting high-precision pulse wave signals in motion by constructing a pulse wave sensor array. The array includes at least two photoelectric sensors corresponding to the relevant arterial sites on the skin, and integrates an acceleration sensor. Compared with a single sensor, the array-based application can obtain high-fidelity pulse wave signals when simultaneously fitting more than two arterial sites. The acceleration sensor effectively reduces the position interference of the pulse wave signal caused by movement, and can assist in monitoring the motion state of the subject. Based on the difference in pulse wave signal quality under different motion states, the present invention can perform long-term and accurate monitoring, especially for the filtering effect after motion artifact interference and the signal measurement inaccuracy caused by sensor displacement, which can significantly reduce the impact of movement and long-term use on the difference in pulse wave signals, thereby reducing interference with the reliability of cardiac output calculation results.

[0020] Secondly, the present invention combines photoplethysmography with a deep learning model to construct a model that can calculate cardiac output in motion. The model is based on a deep learning model and uses the aortic cross-sectional area, which is closely related to the size of cardiac output, as an input variable. At the same time, this variable is also used as a correction variable. The pulse wave characteristics and the aortic cross-sectional area are used as model inputs instead of directly using personalized feature parameters. This reduces the input variables that are not closely related to cardiac output to a certain extent, and enhances the correlation between output and input, thereby reducing the complexity of the model.

[0021] In summary, the present invention focuses on cardiac output monitoring technology under exercise, combines multiple technologies such as photoplethysmography, sensor array, aortic cross-sectional area estimation model and cardiac output calculation model, realizes accurate measurement of cardiac output under daily conditions, and evaluates signal quality. This makes the calculation results reliable in exercise states and long-term monitoring. The present invention improves the shortcomings of the prior art that require continuous correction, realizes continuous dynamic monitoring of cardiac output parameters in a true sense, and is suitable for long-term monitoring applications in home rehabilitation of patients with heart failure and structural heart disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a structural diagram of a dynamic non-invasive cardiac output monitoring system in an embodiment.

[0023] Figure 2 4 is a structural diagram of the signal acquisition module in the embodiment.

[0024] Figure 3 4 is a structural diagram of a dynamic signal quality monitoring module in an embodiment.

[0025] Figure 4 Schematic diagram of the layout effect of the sensor array placed in the radial artery and the ulnar artery in the embodiment.

[0026] Figure 5 Schematic diagram of the layout effect of the hierarchical sensor array in the embodiment.

[0027] Figure 6 It is a flow chart of a dynamic non-invasive cardiac output monitoring method in an embodiment. DETAILED DESCRIPTION

[0028] The present invention will be further described below in conjunction with the embodiments and drawings.

[0029] like Figure 1 , a dynamic non-invasive cardiac output monitoring system, comprising: a signal acquisition module, a filtering module, a pulse wave feature extraction module, a dynamic signal quality monitoring module, and a cardiac output monitoring and evaluation module; The signal acquisition module is used to continuously acquire multi-channel photoplethysmography signals and accelerometer signals from different peripheral arterial sites in a resting or active state, and filter and amplify them to obtain a dynamic pulse wave waveform signal; The filtering module is used to correct the baseline drift and motion artifacts in the dynamic pulse wave waveform signal to obtain a pulse wave correction signal; The pulse wave feature extraction module is used to perform pulse wave feature point recognition, feature extraction, and screening of pulse wave feature vectors related to cardiac output calculation on the pulse wave correction signal, and perform normalization processing to obtain a normalized feature vector; The dynamic signal quality monitoring module is used to monitor the quality of normalized feature vectors in different states, and use the normalized feature vectors that meet the input conditions of the cardiac output calculation model as dynamic pulse wave feature vectors; the dynamic pulse wave feature vectors are obtained by comparing the number and value range of normalized feature vectors extracted from the subject in an active state and a resting state, and outputting a normalized feature vector that is lower than a difference threshold; The cardiac output monitoring and evaluation module is used for dynamically and continuously evaluating the stroke volume and cardiac output of the subject, and includes: a correction variable estimation submodule, a pulse wave feature vector estimation submodule and a cardiac output evaluation submodule; the correction variable estimation submodule is based on the personalized feature parameters input by the subject and the estimated value of the aortic cross-sectional area as correction variables; the pulse wave feature vector estimation submodule sequentially calls the filtering module and the dynamic signal quality monitoring module to process the dynamic pulse wave waveform signal collected by the signal acquisition module to obtain a dynamic pulse wave feature vector; the cardiac output evaluation submodule inputs the correction variable and the dynamic pulse wave feature vector into the cardiac output calculation model to obtain the stroke volume and cardiac output, and dynamically displays them on the system interface.

[0030] Preferably, if Figure 2As shown, in the signal acquisition module, a sensor array, a pressure control submodule, a waveform construction submodule, and an amplification filtering transmission submodule are included; The sensor array is used to dynamically and continuously collect photoelectric signals, contact pressure signals and accelerometer signals of peripheral arterial pulse waves at different parts of the subject in motion; the sensor array includes at least two photoelectric sensors that can respectively collect pulse waves at different peripheral arterial parts, a pressure sensor that collects contact pressure at corresponding parts, and an accelerometer for measuring displacement and velocity information of the subject; The pressure control submodule is used to apply pressure to the peripheral artery during the acquisition process, and includes: an air bag that applies contact pressure to the skin, and a pressure control device that controls the contact pressure through the air bag; The waveform construction submodule may combine the photoelectric signals and contact pressure signals collected by the sensor array to construct a multi-channel photoplethysmography signal; The amplification, filtering and transmission submodule is used to amplify and filter the multi-channel photoplethysmography signals and accelerometer signals in sequence to obtain a dynamic pulse wave waveform signal, and transmit it to the filtering module of the host computer.

[0031] Preferably, the motion artifact is used as a motion reference signal for adaptive filtering to correct multi-channel dynamic pulse waveform signals obtained from different parts.

[0032] Preferably, if Figure 3 As shown, the dynamic signal quality monitoring module includes: a signal receiving submodule and a signal quality judgment submodule; The signal receiving submodule is connected to the output end of the pulse wave feature extraction module, receives the normalized feature vector and marks the corresponding signal channel; The signal quality judgment submodule judges whether the normalized characteristic vectors of different signal channels meet the input conditions of the cardiac output calculation model based on the signal quality rules. The signal quality rules include: Whether there is a lack of pulse wave feature vectors related to cardiac output calculation, wherein the pulse wave feature vectors related to cardiac output calculation include: pulse wave feature points, amplitude changes of each pulse wave feature point, time intervals between adjacent feature points, area under the curve during the systolic phase of the pulse wave, and area under the curve during the diastolic phase of the pulse wave; Based on the difference judgment rule, the difference between the pulse wave feature vector value in the active state and the resting state is compared and scored, and the difference includes: amplitude change range difference, time range difference, and area range difference; the difference judgment rule includes: if the feature vector difference of all channels is less than the threshold, the normalized feature vector in the active state is directly used as the dynamic pulse wave feature vector and output; if the normalized feature vector difference measured in at least one part is greater than the threshold, it is necessary to reduce the influence of motion artifacts or other factors on the pulse wave signal, so that the subject reduces the movement speed to the resting state and checks whether the sensor wearing position changes. When the difference is less than a certain threshold, the dynamic pulse wave feature vector is output.

[0033] Preferably, the sensor array comprises at least two photoelectric sensors and at least one pressure sensor arranged longitudinally or transversely along the peripheral artery, an optical sensor for filtering out ambient light interference, and an accelerometer worn on the wrist of the subject.

[0034] Preferably, if Figure 4 As shown, in one embodiment, the sensor array is arranged as follows: at least one photoelectric sensor and at least one pressure sensor correspond to the ulnar artery at the wrist to obtain the dynamic pulse wave waveform signal of the ulnar artery, and at least one photoelectric sensor and at least one pressure sensor correspond to the radial artery at the wrist to obtain the dynamic pulse wave waveform signal of the radial artery.

[0035] Preferably, if Figure 5 As shown, in another embodiment, another arrangement of the sensor array is: at least two photoelectric sensors for collecting peripheral arterial pulse waves corresponding to the ulnar artery or radial artery at the wrist and one pressure sensor for collecting contact pressure, the sensors are arranged longitudinally along the blood vessel, wherein the pressure sensor is located at the center, forming a hierarchical structure; the photoelectric signal collected by each photoelectric sensor is respectively combined with the pressure signal to form a two-channel dynamic pulse wave waveform signal.

[0036] Preferably, the personalized characteristic parameter generates an estimated value of the aortic cross-sectional area based on a relationship model, and the relationship model is: (1) S: estimated aortic cross-sectional area; Age: age; BSA: body surface area; Height: height; Weight: weight; a, b, c, e: correlation coefficients, and f: disease correction coefficient.

[0037] In the embodiment, a correlation relationship for aortic aneurysm patients obtained based on the above method is: (2) Aortic aneurysm is a common structural heart disease, which usually manifests as partial or complete dilation of the aorta, resulting in an increase in cross-sectional area.

[0038] Preferably, the characteristic points of the pulse wave include: the starting point of the pulse wave, the main peak point of the pulse wave, the dicrotic wave peak point, and the end point of the pulse wave cycle; the pulse wave characteristic vector related to the calculation of cardiac output includes: the amplitude change of each characteristic point of the pulse wave, the time interval between adjacent characteristic points, the area under the curve during the systolic period of the pulse wave, and the area under the curve during the diastolic period of the pulse wave.

[0039] Preferably, the cardiac output calculation model is obtained based on the deep learning model training sample data, and the sample data is divided into a training set and a test set. In the training set, the model input includes: correction variables and pulse wave feature vectors related to cardiac output calculation, and the model output is stroke volume, which is calculated based on stroke volume and heart rate; the deep learning model selects an artificial neural network model for training. The test set is used to test and evaluate the trained deep learning model, and change the deep learning model parameters until the deep learning model that meets the test and index requirements is used as the final cardiac output calculation model; the correction variables are based on the individual characteristic parameters of the subject and the aortic cross-sectional area; the evaluation indicators include: model accuracy, recall rate, and precision rate.

[0040] The artificial neural network model includes an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; a hidden layer and an output layer; in the artificial neural network model, the activation function can be any one of the following: Sigmoid function, Tanh function, ReLU function, Leaky ReLU function, ELU function or PReLU function.

[0041] Preferably, in the sample data, cardiac output and stroke volume are derived from Picco monitoring data, pulse wave is derived from pulse wave data monitored by a PPG sensor, and personalized characteristic parameters are derived from electronic medical records.

[0042] Preferably, cardiac function parameters that can also be calculated based on cardiac output and pulse wave signals include but are not limited to: CTI, VET, HR, continuous SBP, continuous DBP and continuous MAP, PPV, SVV, SVR and PVR.

[0043] CTI is used to evaluate the heart's ability to pump blood. The calculation formula is: (3) Where CO is cardiac output in liters per minute (L / min) and BSA is body surface area in m².

[0044] VET is calculated based on the signal characteristics of the pulse wave signal.

[0045] HR is obtained based on the number of pulse waves collected per minute.

[0046] SBP is the maximum peak value of the aortic pulse wave signal, DBP is the minimum value of the aortic pulse wave signal, and MAP is an indicator that reflects the overall blood pressure level of the heart and vascular system and is calculated based on SBP and DBP.

[0047] PPV is used to assess the dynamic response of the heart and blood vessels, especially the response to the respiratory cycle. It is often used to assess fluid status and vascular compliance. The formula is: (4) in, , and They refer to the maximum, minimum and mean systolic blood pressure respectively.

[0048] SVV is often used to assess blood volume status, especially in cases of fluid overload or dehydration. SVV can be calculated by analyzing changes in the pulse wave signal.

[0049] (5) in, , , They are the maximum, minimum and mean stroke volume, respectively.

[0050] SVR reflects the total resistance of the blood vessels and directly affects the heart's pumping efficiency. It is usually calculated by the relationship between blood pressure and cardiac output.

[0051] Calculation formula: (6) Where MAP is mean arterial pressure and CO is cardiac output, the unit is .

[0052] PVR is the ratio of peripheral vascular resistance to individual body surface area, calculated using the formula: (7) Among them, SVR is peripheral vascular resistance and BSA is body surface area.

[0053] like Figure 6 As shown, a dynamic non-invasive cardiac output monitoring method, Step 1, pulse signal acquisition: based on the sensor array and photoplethysmography, dynamically and continuously acquire multi-channel photoplethysmography signals and accelerometer signals of different peripheral arterial sites of the subject in a resting or active state, and filter and amplify them to obtain a dynamic pulse wave waveform signal; the sensor array comprises at least two photoelectric sensors for respectively acquiring pulse waves of different arterial sites, and an accelerometer for measuring displacement and velocity information of the subject; Step 2, pulse signal filtering: performing motion artifact and baseline drift correction on the dynamic pulse waveform signal in step 1 to generate a pulse wave correction signal; Step 3, pulse wave feature extraction: performing pulse wave feature point identification on the pulse wave correction signal described in step 2, performing feature extraction on the pulse wave correction signal based on the pulse wave feature points, and screening the pulse wave feature vector related to the cardiac output calculation, performing normalization processing, and obtaining a normalized feature vector; Step 4, dynamic signal quality monitoring: compare the number and value range of normalized feature vectors extracted from the subject in the active state and the resting state, and judge whether it meets the input conditions of the cardiac output calculation model based on the set difference threshold, and obtain the dynamic pulse wave feature vector that meets the difference threshold; Step 5, generating correction variables: the subject inputs personalized characteristic parameters, and generates an estimated value of the aortic cross-sectional area based on a relationship model; the relationship model is to calculate the estimated value of the aortic cross-sectional area based on the subject's age, body surface area, height, and weight: Step 6, training the cardiac output calculation model: divide the sample data into a training set and a test set. In the training set, calculate the estimated value of the aortic cross-sectional area based on the personalized feature parameters as a correction variable; further extract and screen the pulse wave feature vector related to the cardiac output calculation, and perform dynamic signal quality monitoring based on step 4; use the dynamic pulse wave feature vector and correction variable that meet the requirements as input, the stroke volume as output, and the artificial neural network as a pre-training model for training; and test and evaluate the trained model based on the test set; change the deep learning model parameters until the deep learning model that meets the test and indicator requirements is used as the final cardiac output calculation model; Step 7, cardiac output monitoring and evaluation: call steps 1-5 to generate the subject's correction variables and dynamic pulse wave feature vector as input to the cardiac output calculation model, call the cardiac output calculation model in step 6, output the stroke volume, calculate the cardiac output based on the stroke volume and heart rate, and display it on the system interface.

[0054] Preferably, the steps of performing motion artifact correction based on the accelerometer and the adaptive filtering algorithm are: A1: collecting dynamic pulse waveform signals and accelerometer signals of peripheral arteries, wherein the accelerometer signals obtain time-displacement signals of the system in space; A2: Signal preprocessing: low-pass filtering removes high-frequency noise and retains low-frequency signals; high-pass filtering removes low-frequency drift noise and performs normalization processing; A3: Feature extraction: Extract the motion features of the time-displacement signal of the accelerometer in the x, y, and z axis directions based on spectrum analysis; further identify the pattern and amplitude change of the motion, and determine the frequency and amplitude features related to the motion artifact; analyze the frequency and amplitude of the pulse wave signal based on spectrum analysis to determine the frequency related to the motion artifact; A4: Adaptive filtering: Using the accelerometer’s time-displacement signal as a reference signal, the filter coefficients are automatically adjusted to remove motion artifacts captured by the accelerometer by minimizing the error between the filter’s output and the target signal; A5: Post-processing and verification: Reconstruct the signal after artifact removal to obtain the pulse wave correction signal to ensure that the difference between the pulse wave feature vector collected in the moving state and the pulse wave feature vector collected in the resting state is less than a certain threshold, and use evaluation indicators to quantitatively evaluate the artifact removal effect.

[0055] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the embodiments of the present invention.

Claims

1. A dynamic non-invasive cardiac output monitoring system, characterized in that: include: Signal acquisition module, filtering module, pulse wave feature extraction module, dynamic signal quality monitoring module, cardiac output monitoring and evaluation module; The signal acquisition module is used to continuously acquire multi-channel photoplethysmography signals and accelerometer signals at different peripheral artery sites in a resting or active state, and filter and amplify them to obtain a dynamic pulse wave waveform signal; The filtering module is used to correct the baseline drift and motion artifacts in the dynamic pulse wave waveform signal by combining the accelerometer signal and the adaptive filtering algorithm to obtain a pulse wave correction signal; The pulse wave feature extraction module is used to perform pulse wave feature point recognition and feature extraction on the pulse wave correction signal, and to screen and normalize the pulse wave feature vector related to the cardiac output calculation to obtain a normalized feature vector; The dynamic signal quality monitoring module is used to monitor the quality of normalized feature vectors in different states, and use the normalized feature vectors that meet the input conditions of the cardiac output calculation model as dynamic pulse wave feature vectors; the dynamic pulse wave feature vectors are obtained by comparing the number and value range of normalized feature vectors extracted from the subject in an active state and a resting state, and outputting the normalized feature vectors that are lower than the difference threshold; The cardiac output monitoring and evaluation module is used for dynamically and continuously estimating the stroke volume and cardiac output of the subject, and includes: a correction variable estimation submodule, a pulse wave feature vector estimation submodule and a cardiac output evaluation submodule; The correction variable estimation submodule calculates the estimated value of the aortic cross-sectional area as the correction variable based on the personalized characteristic parameters input by the subject; the pulse wave feature vector estimation submodule sequentially calls the filtering module and the dynamic signal quality monitoring module to process the dynamic pulse wave waveform signal collected by the signal acquisition module to obtain the dynamic pulse wave feature vector; the cardiac output assessment submodule inputs the correction variable and the dynamic pulse wave feature vector into the cardiac output calculation model trained based on the deep learning model to obtain the stroke volume and cardiac output, and dynamically displays them on the system interface.

2. A dynamic non-invasive cardiac output monitoring system according to claim 1, characterized in that: The signal acquisition module includes a sensor array, a pressure control submodule, a waveform construction submodule, and an amplification filtering and transmission submodule; The sensor array is used to dynamically and continuously collect photoelectric signals, contact pressure signals and accelerometer signals of peripheral arterial pulse waves at different parts of the subject in motion based on photoplethysmography; the sensor array includes at least two photoelectric sensors for respectively collecting pulse waves at different peripheral arterial parts, a pressure sensor for collecting contact pressure at corresponding parts, and an accelerometer for measuring displacement and velocity information of the subject, i.e., body activity data; The pressure control submodule is used to apply contact pressure to the peripheral artery and control it during the acquisition process, including: an airbag that applies contact pressure to the skin, and a pressure control device that controls the contact pressure through the airbag; The waveform construction submodule constructs a multi-channel photoplethysmography signal by combining the photoelectric signal and the contact pressure signal collected by the sensor array; The amplification, filtering and transmission submodule is used to sequentially amplify and filter the multi-channel photoplethysmography signals and accelerometer signals to obtain dynamic pulse wave waveform signals and transmit them to the filtering module of the host computer.

3. A dynamic non-invasive cardiac output monitoring system according to claim 1, characterized in that: The motion artifact is corrected by using an adaptively filtered motion reference signal of an accelerometer signal to correct multi-channel dynamic pulse waveform signals obtained from different parts.

4. A dynamic non-invasive cardiac output monitoring system according to claim 1, characterized in that: The dynamic signal quality monitoring module includes: a signal receiving submodule and a signal quality judgment submodule; The signal receiving submodule is connected to the output end of the pulse wave feature extraction module, receives the normalized feature vector and marks the corresponding signal channel; The signal quality judgment submodule judges whether the normalized characteristic vectors of different signal channels meet the input conditions of the cardiac output calculation model based on the signal quality rules, and the signal quality rules include: Whether there is a lack of pulse wave feature vectors related to cardiac output calculation, wherein the pulse wave feature vectors related to cardiac output calculation include: pulse wave feature points, amplitude changes of each feature point of the pulse wave, time intervals between adjacent feature points, the area under the curve during the systolic period of the pulse wave, and the area under the curve during the diastolic period of the pulse wave; Based on the difference judgment rule, the difference between the pulse wave feature vector value in the active state and the resting state is compared and scored, and the difference includes: amplitude change range difference, time range difference, and area range difference; the difference judgment rule includes: if the feature vector difference of all channels is less than the threshold, the normalized feature vector in the active state is directly used as the dynamic pulse wave feature vector and output; if the normalized feature vector difference measured in at least one part is greater than the threshold, the signal acquisition module is interacted with to make the subject reduce the movement speed to the resting state and check whether the sensor wearing position changes, and when the difference is less than a certain threshold, the dynamic pulse wave feature vector is output.

5. A dynamic non-invasive cardiac output monitoring system according to claim 1 or 2, characterized in that: include: The sensor array is composed of at least two photoelectric sensors and at least one pressure sensor arranged longitudinally or transversely along the peripheral artery, an optical sensor for filtering out ambient light interference, and an accelerometer worn on the wrist of the subject.

6. A dynamic non-invasive cardiac output monitoring system according to claim 5, characterized in that: The sensor array is arranged in such a manner that at least one photoelectric sensor and at least one pressure sensor correspond to the ulnar artery on the skin to obtain a dynamic pulse waveform signal of the ulnar artery, and at least one photoelectric sensor and at least one pressure sensor correspond to the radial artery on the skin to obtain a dynamic pulse waveform signal of the radial artery.

7. A dynamic non-invasive cardiac output monitoring system according to claim 1 or 2, characterized in that: The arrangement of the sensor array is as follows: at least two photoelectric sensors for collecting peripheral arterial pulse waves and one pressure sensor for collecting contact pressure corresponding to the ulnar artery or radial artery on the skin, the sensors are arranged longitudinally along the blood vessel, wherein the pressure sensor is located at the center, forming a sandwich structure; the photoelectric signal collected by each photoelectric sensor is respectively combined with the contact pressure signal to form a two-channel dynamic pulse wave waveform signal.

8. A dynamic non-invasive cardiac output monitoring system according to claim 2, characterized in that: The personalized characteristic parameters calculate the estimated aortic cross-sectional area based on a relationship model, and the relationship model is constructed based on the linear relationship fitting of age, body surface area, height, weight and the estimated aortic cross-sectional area: （1） S: estimated aortic cross-sectional area; Age: age; BSA: body surface area; Height: height; Weight: weight; a, b, c, e are correlation coefficients, and f is the correction factor.

9. A dynamic non-invasive cardiac output monitoring system according to claim 2, characterized in that: The characteristic points of the pulse wave include: the starting point of the pulse wave, the main peak point of the pulse wave, the dicrotic wave peak point, and the end point of the pulse wave cycle; the pulse wave characteristic vector related to the calculation of cardiac output includes: the amplitude change of each characteristic point of the pulse wave, the time interval between adjacent characteristic points, the area under the curve during the systolic period of the pulse wave, and the area under the curve during the diastolic period of the pulse wave.

10. A dynamic non-invasive cardiac output monitoring system according to claim 1 or 2, characterized in that: The cardiac output calculation model is obtained based on deep learning model training sample data, and the sample data is divided into a training set and a test set. In the training set, the model input includes: correction variables and pulse wave feature vectors related to cardiac output calculation, and the model output is stroke volume. The cardiac output is calculated based on stroke volume and heart rate. The deep learning model selects an artificial neural network model for training; the test set is used to test the trained deep learning model and evaluate it based on evaluation indicators, and the deep learning model parameters are changed until the deep learning model that meets the test and indicator requirements is used as the final cardiac output calculation model; the correction variable is the aortic cross-sectional area calculated based on the relationship model between the subject's personalized characteristic parameters and the aortic cross-sectional area; the evaluation indicators include: model accuracy, recall rate, precision rate, and F1 value.

11. A dynamic non-invasive cardiac output monitoring system according to claim 10, characterized in that: In the sample data, cardiac output and stroke volume are derived from Picco monitoring data, pulse wave is derived from pulse wave data monitored by a PPG sensor, and personalized characteristic parameters are derived from electronic medical records.

12. A dynamic non-invasive cardiac output monitoring system according to claim 11, characterized in that: Cardiac function parameters that can also be calculated based on cardiac output and pulse wave signals include but are not limited to: cardiac contractility index, ventricular ejection time, heart rate, continuous systolic pressure, continuous diastolic pressure and continuous mean pressure, pulse pressure variability, stroke volume variability, peripheral vascular resistance and peripheral vascular resistance index.

13. A dynamic non-invasive cardiac output monitoring method based on the system of any one of claims 1 to 12, characterized in that: Step 1, pulse signal acquisition: dynamically and continuously acquiring multi-channel photoplethysmography signals and accelerometer signals of different peripheral arterial sites of the subject in a resting or active state based on a sensor array and photoplethysmography, and filtering and amplifying the signals to obtain a dynamic pulse wave waveform signal, wherein the sensor array comprises at least two photoelectric sensors for respectively acquiring pulse waves of different arterial sites, and an accelerometer for measuring displacement and velocity information of the subject; Step 2, pulse signal filtering: combining the accelerometer signal and the adaptive filtering algorithm to perform motion artifact and baseline drift correction on the dynamic pulse waveform signal in step 1 to generate a pulse wave correction signal; Step 3, pulse wave feature extraction: performing pulse wave feature point recognition and feature extraction on the pulse wave correction signal described in step 2, screening the pulse wave feature vector related to the cardiac output calculation, and performing normalization processing to obtain a normalized feature vector; Step 4, dynamic signal quality monitoring: compare the number and value range of normalized feature vectors extracted from the subject in the active state and the resting state, determine whether they meet the input conditions of the cardiac output calculation model based on the set difference threshold, and obtain the dynamic pulse wave feature vector that meets the difference threshold; Step 5, generating correction variables: the subject inputs personalized characteristic parameters to generate an estimated value of the aortic cross-sectional area based on a relationship model; the relationship model is a calculation model for obtaining an estimated value of the aortic cross-sectional area based on a linear fitting of the subject's age, body surface area, height, and weight; Step 6, training the cardiac output calculation model: dividing the sample data into a training set and a test set, in the training set, calculating the estimated value of the aortic cross-sectional area based on the personalized characteristic parameters as a correction variable; further extracting and screening the pulse wave feature vector related to the cardiac output calculation and performing dynamic signal quality monitoring based on step 4, taking the dynamic pulse wave feature vector and correction variable that meet the requirements as input, the stroke volume as output, and the artificial neural network as a pre-training model for training; and testing and evaluating the trained model based on the test set; Change the deep learning model parameters until the deep learning model that meets the test and index requirements is used as the final cardiac output calculation model; Step 7, cardiac output monitoring and evaluation: call steps 1-5 to generate the subject's correction variables and dynamic pulse wave feature vector as input to the cardiac output calculation model, call the cardiac output calculation model in step 6, output the stroke volume, calculate the cardiac output based on the stroke volume and heart rate, and display it on the system interface.

14. A dynamic non-invasive cardiac output monitoring method according to claim 13, characterized in that: The steps of performing motion artifact correction based on the accelerometer and the adaptive filtering algorithm are as follows: A1: collecting dynamic pulse waveform signals and accelerometer signals of peripheral arteries, wherein the accelerometer signals obtain time-displacement signals of the system in space; A2: Signal preprocessing: Low-pass filtering: remove high-frequency noise and retain low-frequency signals; High-pass filtering: remove low-frequency drift noise and perform normalization; A3: Feature extraction: Extract the motion characteristics of the accelerometer's time-displacement signal in the x, y, and z axes based on spectrum analysis; further identify the motion pattern and amplitude changes, and determine the frequency and amplitude characteristics associated with motion artifacts; Analyzing the frequency and amplitude of the pulse wave signal based on spectrum analysis to determine the frequency associated with the motion artifact; A4: Adaptive filtering: Using the accelerometer’s time-displacement signal as a reference signal, the filter coefficients are automatically adjusted to remove motion artifacts captured by the accelerometer by minimizing the error between the filter’s output and the target signal; A5: Post-processing and verification: Reconstruct the de-artifacted signal to obtain the pulse wave correction signal to ensure that the difference between the pulse wave feature vector collected when the subject is in motion and when the subject is in a resting state is less than a certain threshold, and use evaluation indicators to quantitatively evaluate the de-artifacting effect.

Images