CN120241087A - A physiological parameter monitoring and intelligent early warning system for critically ill patients - Google Patents
A physiological parameter monitoring and intelligent early warning system for critically ill patients Download PDFInfo
- Publication number
- CN120241087A CN120241087A CN202510757533.2A CN202510757533A CN120241087A CN 120241087 A CN120241087 A CN 120241087A CN 202510757533 A CN202510757533 A CN 202510757533A CN 120241087 A CN120241087 A CN 120241087A
- Authority
- CN
- China
- Prior art keywords
- signal
- physiological parameter
- early warning
- module
- physiological
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Measuring devices for evaluating the respiratory organs
- A61B5/085—Measuring impedance of respiratory organs or lung elasticity
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue
- A61B5/14542—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue for measuring blood gases
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/74—Details of notification to user or communication with user or patient; User input means
- A61B5/746—Alarms related to a physiological condition, e.g. details of setting alarm thresholds or avoiding false alarms
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Physiology (AREA)
- Artificial Intelligence (AREA)
- Pulmonology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Optics & Photonics (AREA)
- Cardiology (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a physiological parameter monitoring and intelligent early warning system for an intensive care patient, which relates to the technical field of intelligent medical treatment, and comprises a multi-mode data collection module, a signal alignment module, a physiological theory range updating module, a transition learning model, a classification early warning module, a Bayesian network composite event detection model, a decision support module and a monitoring terminal, wherein the multi-mode physiological parameter data of the intensive care patient are collected through a sensor array of flexible electronic skin, signal sequences with different sampling rates in the multi-mode physiological parameter data are aligned, the physiological theory range updating module is arranged, the physiological theory range of heart rate variability coefficients is updated in real time according to historical medical records of the patient, the classification early warning module is arranged, the Bayesian network composite event detection model is triggered, a classification early warning signal is generated, the decision support module is arranged, a clinical knowledge map is dynamically loaded through a micro-service architecture, and decision support information containing rescue priority labels is pushed to the monitoring terminal; the accuracy and the efficiency of intensive care are remarkably improved.
Description
Technical Field
The invention relates to the technical field of intelligent medical treatment, in particular to a physiological parameter monitoring and intelligent early warning system for severe patients.
Background
The physiological state of a critical patient is highly complex and dynamic, and the condition may deteriorate rapidly in a short period of time, requiring continuous, accurate monitoring and rapid intervention. The traditional monitoring means depend on single parameters such as heart rate and blood oxygen threshold value alarm, but the single parameters are easy to be interfered by noise and cannot comprehensively reflect the overall state of a patient. In addition, fixed thresholds are difficult to accommodate for individual differences, such as underlying disease, age, constitution, often leading to false positives or false negatives. The intelligent monitoring and grading early warning system can integrate multi-mode physiological data such as electrocardio, blood oxygen and respiratory impedance, and combines dynamic threshold adjustment and compound event analysis, thereby remarkably improving the accuracy and timeliness of early warning and providing scientific basis for clinical decision.
In the existing monitoring and early warning technology, the traditional equipment only monitors single parameters (such as electrocardio or blood oxygen), the collaborative analysis of multi-mode data is lacking, abnormal signals of multi-system linkage cannot be captured, or a fixed threshold value is adopted for alarming, and the traditional equipment cannot be dynamically adjusted according to the individual history data of a patient, so that the problem of insufficient sensitivity to a high-risk patient is caused.
Therefore, the invention provides a physiological parameter monitoring and intelligent early warning system for severe patients.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems existing in the prior art. Therefore, the invention provides the physiological parameter monitoring and intelligent early warning system for the severe patients, which remarkably improves the accuracy and efficiency of the severe monitoring.
In order to achieve the aim, the physiological parameter monitoring and intelligent early warning system for the severe patients comprises a multi-mode data collecting module, a signal alignment module, a physiological theory range updating module, a grading early warning module and a decision support module, wherein the modules are electrically connected;
The multi-mode data collection module is used for collecting multi-mode physiological parameter data of a severe patient through a sensor array of the flexible electronic skin, wherein the multi-mode physiological parameter data comprises an electrocardiographic waveform, blood oxygen saturation and respiratory impedance signals, and the multi-mode physiological parameter data is sent to the signal alignment module;
The signal alignment module aligns signal sequences with different sampling rates in the multi-mode physiological parameter data and sends the signal sequences to the grading early warning module;
the physiological theory range updating module is used for constructing a transfer learning model according to the historical medical records of the patient, automatically generating an individual physical sign fluctuation threshold interval by the transfer learning model, updating the physiological theory range of the heart rate variation coefficient in real time, and sending the physiological theory range to the grading early warning module;
The hierarchical early warning module triggers a Bayesian network composite event detection model when at least two physiological parameters in a signal sequence reach the physiological theory range at the same time, the Bayesian network composite event detection model calculates multi-parameter joint anomaly probability, generates a hierarchical early warning signal and sends the hierarchical early warning signal to the decision support module;
The decision support module dynamically loads a clinical knowledge graph through the micro-service architecture, and the clinical knowledge graph correlates the grading early warning signal with a preset treatment plan to push decision support information containing rescue priority labels to the monitoring terminal;
The acquisition of the multi-mode physiological parameter data of the severe patient comprises the following steps:
Step 11, designing and deploying a sensor array of the flexible electronic skin, wherein the sensor array comprises a plurality of signal acquisition channels for respectively acquiring electrocardio waveforms, blood oxygen saturation and respiratory impedance signals;
step 12, calibrating sensitivity parameters of the flexible electronic skin sensor array, wherein the sensitivity parameters comprise an amplitude threshold value of an electrocardio waveform, photoelectric signal strength of blood oxygen saturation and frequency response characteristics of respiratory impedance;
step 13, synchronously collecting multi-mode physiological parameter data of the body surface of a severe patient, wherein the multi-mode physiological parameter data comprises an electrocardiographic waveform, blood oxygen saturation and respiratory impedance signals;
Step 14, carrying out preliminary processing on the multi-mode physiological parameter data, wherein the preliminary processing comprises signal denoising and normalization processing, and outputting a standardized multi-mode physiological parameter data set;
The method for aligning the signal sequences with different sampling rates in the multi-mode physiological parameter data by adopting a dynamic time warping algorithm comprises the following steps:
aligning signal sequences with different sampling rates by adopting a dynamic time warping algorithm for the synchronized multi-mode physiological parameter data;
the construction of the migration learning model according to the historical medical records of the patient comprises the following steps:
Step 31, extracting multi-mode physiological parameter characteristic data from a patient history medical record, and generating a standardized characteristic vector sequence by using a characteristic extraction technology;
The historical medical records are stored in a relational database in a time sequence form, and each record comprises an electrocardiographic waveform sampling point voltage value sequence;
the historical medical record also comprises a blood oxygen saturation photoelectric volume pulse wave signal;
And 32, constructing a characteristic distribution model of the patient group, inputting a standardized characteristic vector sequence, identifying characteristic distribution rules of different patient groups by using a cluster analysis method, and outputting the characteristic distribution model of the patient group.
Step 33, constructing a transfer learning model, inputting a patient group feature distribution model and an individual feature vector sequence, mapping public feature distribution to individual feature distribution by using a transfer learning algorithm, and outputting an individual sign fluctuation threshold interval;
And 34, updating the normal range of the heart rate variability coefficient in real time, inputting the individual sign fluctuation threshold interval and the physiological parameter data monitored in real time, and updating the normal range of the heart rate variability coefficient by using a dynamic adjustment algorithm.
The mode of triggering the Bayesian network composite event detection model is as follows:
the Bayesian network compound event detection model calculates multi-parameter joint abnormality probability based on the at least two items of physiological parameter data, and outputs the multi-parameter joint abnormality probability.
The method for generating the grading early warning signal comprises the following steps:
And generating a grading early warning signal according to the multi-parameter joint anomaly probability and a preset grading rule, wherein the grading early warning signal comprises early warning information of different grades.
The pushing decision support information containing rescue priority labels to the monitoring terminal comprises the following steps:
Step 51, dynamically loading a clinical knowledge graph through a micro-service architecture, and retrieving a treatment plan associated with the early warning signal from the clinical knowledge graph;
Step 52, associating the early warning signal with the retrieved treatment plan to generate a rescue priority label;
And step 53, pushing the decision support information containing the rescue priority label to the monitoring terminal.
Compared with the prior art, the invention has the beneficial effects that:
According to the technical scheme, firstly, multi-mode physiological signals such as electrocardio, blood oxygen and respiratory impedance are synchronously acquired by adopting a flexible electronic skin sensor array, data flows with different sampling rates are aligned through a dynamic time warping algorithm, the time-frequency misalignment problem is eliminated, the time consistency of the signals is ensured, an individual physical sign fluctuation threshold is built based on a migration learning model, the normal range of key parameters such as heart rate variation coefficient is dynamically adjusted by analyzing the historical medical records and population characteristic distribution of a patient, the personalized adaptation of an early warning threshold is realized, then, a Bayesian network compound event detection model is utilized for carrying out multi-parameter joint anomaly probability calculation, a grading early warning mechanism is triggered when at least two physiological parameters exceed the threshold simultaneously, the false report risk caused by single parameter noise interference is reduced, finally, a clinical knowledge map and a micro-service architecture are combined, the early warning signal and a standardized treatment plan are automatically associated by the system, priority labels are generated and are pushed to a monitoring terminal in real time, and the closed-loop support from monitoring to decision is formed, and the precision and efficiency of intensive care are remarkably improved.
Drawings
FIG. 1 is a diagram showing the connection relationship between modules of a physiological parameter monitoring and intelligent early warning system for severe patients in embodiment 1 of the present invention;
FIG. 2 is a flow chart of the acquisition of multimodal physiological parameter data of a critically ill patient in accordance with embodiment 1 of the present invention;
fig. 3 is a flowchart of constructing a migration learning model according to a patient's historical medical record in embodiment 1 of the present invention.
Detailed Description
The technical solutions of the present invention will be clearly and completely described in connection with the embodiments, and it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
As shown in FIG. 1, the physiological parameter monitoring and intelligent early warning system for the severe patients comprises a multi-mode data collection module, a signal alignment module, a physiological theory range updating module, a grading early warning module and a decision support module, wherein the modules are electrically connected;
The multi-mode data collection module is used for collecting multi-mode physiological parameter data of a severe patient through a sensor array of the flexible electronic skin, wherein the multi-mode physiological parameter data comprises an electrocardiographic waveform, blood oxygen saturation and respiratory impedance signals, and the multi-mode physiological parameter data is sent to the signal alignment module;
The signal alignment module aligns signal sequences with different sampling rates in the multi-mode physiological parameter data and sends the signal sequences to the grading early warning module;
the physiological theory range updating module is used for constructing a transfer learning model according to the historical medical records of the patient, automatically generating an individual physical sign fluctuation threshold interval by the transfer learning model, updating the physiological theory range of the heart rate variation coefficient in real time, and sending the physiological theory range to the grading early warning module;
The hierarchical early warning module triggers a Bayesian network composite event detection model when at least two physiological parameters in a signal sequence reach the physiological theory range at the same time, the Bayesian network composite event detection model calculates multi-parameter joint anomaly probability, generates a hierarchical early warning signal and sends the hierarchical early warning signal to the decision support module;
The decision support module dynamically loads a clinical knowledge graph through the micro-service architecture, and the clinical knowledge graph correlates the grading early warning signal with a preset treatment plan to push decision support information containing rescue priority labels to the monitoring terminal;
In an embodiment of the present invention, as shown in fig. 2, the acquiring the multi-modal physiological parameter data of the severe patient includes the following steps:
Step 11, designing and deploying a sensor array of the flexible electronic skin, wherein the sensor array comprises a plurality of signal acquisition channels for respectively acquiring electrocardio waveforms, blood oxygen saturation and respiratory impedance signals;
Specifically, the sensor array of the flexible electronic skin adopts a multi-layer heterogeneous material composite structure, and an electrocardiosignal acquisition module, a blood oxygen optical sensing unit and a respiratory impedance detection circuit are integrated on a flexible polyimide substrate.
The flexible polyimide substrate adopts a polyimide film with a thermal expansion coefficient of 3.1 multiplied by 10 -5/K as a base material, and the characteristic of the flexible polyimide substrate that the dielectric constant is 4.2@1MHz can effectively isolate parasitic capacitance between electrodes.
The electrocardiosignal acquisition module consists of 3 groups of interdigital silver nanowire electrodes, the electrode spacing of each group is 2.5+/-0.1 mm, the interdigital electrodes are designed by adopting an interdigitated comb-shaped structure, the structure increases the effective contact area by 120 percent, the setting of the electrode spacing of 2.5mm is based on the potential attenuation characteristic of human epidermis, and the electrocardiosignal acquisition module can capture effective signals with the voltage of more than 0.5mV in a QRS complex and is connected with a signal conditioning circuit through silver paste wires.
The blood oxygen optical sensing unit comprises a 660nm and 880nm wavelength double-light-source LED and a photodiode receiver, the photoelectric conversion sensitivity is adjusted to 5 mV/mu W through a built-in transimpedance amplifier, wherein 660nm wavelength corresponds to a deoxidized hemoglobin absorption peak, 880nm corresponds to an oxyhemoglobin isosbestic point, the blood oxygen saturation calculation is realized through a beer-lambert law through the double-wavelength design, the setting of the photoelectric conversion sensitivity of 5 mV/mu W is derived from a photodiode quantum efficiency curve, and the signal to noise ratio of >60dB can be ensured under the typical skin transmittance.
The respiratory impedance detection circuit adopts a four-electrode method to measure, a driving electrode and a detection electrode with the distance of 10mm are arranged on a silicon rubber substrate, and a 50kHz sine excitation signal is applied.
When the sensor array is deployed, the medical-grade acrylic adhesive tape is used for fixing the sensor array on the region from the sternum handle to the xiphoid process of a patient, so that the contact impedance of each sensing unit and the skin is ensured to be less than 2k omega.
Step 12, calibrating sensitivity parameters of the flexible electronic skin sensor array, wherein the sensitivity parameters comprise an amplitude threshold value of an electrocardio waveform, photoelectric signal strength of blood oxygen saturation and frequency response characteristics of respiratory impedance;
it will be appreciated that since the sensor array is required to accommodate physiological characteristics of different patients, calibration of sensitivity parameters is required to ensure measurement accuracy and stability of the sensor array;
Specifically, in the calibration process, an NIBP simulator is used for generating a standard electrocardiographic waveform, wherein the NIBP simulator refers to a noninvasive blood pressure simulator, and the standard electrocardiographic waveform generation is based on Lown-Canong-Levine electrocardiographic models. By adjusting the gain of the preamplifier of the silver nanowire electrode, the amplitude of R wave reaches a threshold value of 1.2+/-0.05V, the threshold value of the amplitude of R wave of 1.2V corresponds to the input range of the II lead of clinical ECG equipment, and the voltage value can be adapted to the measuring range of the 12-bit ADC after gain adjustment.
The blood oxygen saturation calibration adopts a simulated liquid containing 35% of reduced hemoglobin, wherein the 35% concentration in the reduced hemoglobin simulated liquid is set to simulate the hemodynamic balance state when the blood oxygen saturation of human arteries is 95%, the bias voltage of a photodiode is adjusted to enable the light intensity ratio of 880nm/660nm to reach 0.48+/-0.02 when the SpO 2% is achieved, the calibration point of the light intensity ratio of 0.48 is derived from the linear section of an oxygen dissociation curve, and the conversion relation between the ratio and the SpO 2 is in accordance with a Nelder-Mead nonlinear fitting algorithm.
The respiratory impedance calibration is realized by applying sinusoidal displacement load of 0.5-2.0Hz through a mechanical arm, optimizing parameters of a phase-locked amplifier to enable the 50Hz power frequency interference rejection ratio to reach-80 dB, enabling the frequency response linearity error to be smaller than 1.5%, and enabling the 50Hz power frequency interference rejection ratio-80 dB to be derived from YY 0505 standard of medical electrical equipment, wherein the index is realized through quality factor Q value adjustment of the phase-locked amplifier.
The gain coefficients for each channel are finally stored in EEPROM, forming a calibration matrix containing 32 sets of compensation parameters.
Step 13, synchronously collecting multi-mode physiological parameter data of the body surface of a severe patient, wherein the multi-mode physiological parameter data comprises an electrocardiographic waveform, blood oxygen saturation and respiratory impedance signals;
It can be appreciated that since the multimodal physiological parameter data needs to be consistent over time, synchronous acquisition is required to ensure the correlation and accuracy of the data;
specifically, a 10MHz main clock signal is generated through an FPGA during synchronous acquisition;
The ADC sampling units distributed to the sensing channels are used for sampling electrocardiosignals at the rate of 1kHz, a right leg driving technology is adopted for eliminating common mode interference, the right leg refers to a Wilson central terminal improved circuit, a common mode voltage is reduced by 40dB through negative feedback, an blood oxygen signal is used for collecting photoelectric volume waves at the rate of 125Hz, crosstalk is eliminated through time division multiplexing, the sampling rate of the photoelectric volume waves is set to meet the sampling requirement of Nyquist theorem on pulse rate signals (usually <5 Hz), the LED light source crosstalk can be avoided at the interval of 8ms of time division multiplexing, respiratory impedance is used for conducting complex impedance measurement at the rate of 50Hz, real components are used for calculating thoracic volume change, the real components reflect tissue impedance in the complex impedance measurement of the respiratory impedance, the imaginary components correspond to the capacitive reactance, and the real components are selected to mainly cause the resistance change rate of 0.15 omega/cm < 3 due to the thoracic volume change.
The three paths of signals are packed into data frames with aligned time stamps after CRC (cyclic redundancy check) check, and are transmitted to an edge node through Bluetooth 5.0, and the time delay jitter is controlled within +/-2 ms.
Step 14, carrying out preliminary processing on the multi-mode physiological parameter data, wherein the preliminary processing comprises signal denoising and normalization processing, and outputting a standardized multi-mode physiological parameter data set;
it will be appreciated that since the multimodal physiological parameter data may be noisy and of different dimensions, preliminary processing is required to improve data quality and comparability.
Specifically, the signal denoising adopts an improved wavelet threshold algorithm to carry out 6 layers of db4 wavelet decomposition on the electrocardiosignal, wherein the db4 wavelet refers to Daubechies 4-order wavelet, and the support length 8 is suitable for capturing the characteristic of the QRS complex.
The soft threshold function is set to σ (2 lnN) (σ is the noise variance, n=1024), and the noise variance σ is estimated by the median of the Level 1 detail coefficients, where n=1024 in the formula σ (2 lnN) corresponds to the number of sampling points of 1 second of electrocardiographic data.
In the normalization treatment, the electrocardiosignal is subjected to baseline correction and 1mV/unit calibration according to the limb lead II standard, the blood oxygen signal is converted into an optical density ratio ODR= (AC 660/DC 660)/(AC 880/DC 880), an AC component in the optical density ratio ODR refers to an alternating current component (corresponding to pulsating blood flow), DC is a direct current component (reflecting tissue absorption), the ratio is constructed so that the blood oxygen saturation calculation error is less than 1.5%, and the respiratory impedance value is mapped to 0-100% relative chest volume.
The final matrix dimension of the normalized dataset is 3000 x 12, comprising the time domain waveform, frequency domain features, and 12-dimensional statistics.
Further, the method for aligning signal sequences with different sampling rates in the multi-mode physiological parameter data by adopting a dynamic time warping algorithm is as follows:
aligning signal sequences with different sampling rates by adopting a dynamic time warping algorithm for the synchronized multi-mode physiological parameter data;
It will be appreciated that since the multi-modal physiological parameter data may be from different sensors, and the sampling rates may be different, direct data processing may result in signal misalignment or analysis errors, and therefore a dynamic time warping algorithm is required to align the signal sequences at different sampling rates, thereby ensuring data consistency and accuracy.
Specifically, in the signal sequence alignment process, the dynamic time warping algorithm adopts an improved local path constraint strategy to define a cost matrixThe method comprises the following steps:
Wherein, the To accumulate a cost matrix representing the smallest accumulated difference path from the start of the signal to point (i, j), i is the time step of the high frequency signal, e.g. ECG 250Hz, j is the time step of the low frequency signal, e.g. respiratory impedance 50Hz, reflecting a globally optimal solution for the alignment of the two signals in the time domain,AndI.e. the i-th sample point of the ECG signal and the j-th sample point of the SpO2 or respiratory signal respectively,Representing a morphological difference measure of the electrocardiographic signal and the blood oxygen signal, the constraint weight parameters (alpha, beta, gamma) are set (0.3,1,0.3) according to clinical experience to conform to the quasi-periodic characteristics of the respiratory impedance signal. The algorithm sets the dynamic window width for the time-frequency difference of the ECG signal (sampling rate 250 Hz) and the respiratory impedance signal (sampling rate 50 Hz)And extracting characteristic points of the QRS complex as alignment anchor points through sliding window Fourier transform, and finally outputting a composite physiological signal matrix with uniform time resolution.
Further, as shown in fig. 3, the construction of the migration learning model according to the historical medical records of the patient includes the following steps:
Step 31, extracting multi-mode physiological parameter characteristic data from a patient history medical record, and generating a standardized characteristic vector sequence by using a characteristic extraction technology;
It can be appreciated that, due to the heterogeneous and noise limitation of the multi-modal physiological parameter data, feature extraction and normalization are required to obtain a feature vector sequence that can be used for subsequent analysis.
The historical medical records are stored in a relational database in a time sequence form, and each record comprises a voltage value sequence of an electrocardiographic waveform sampling pointElectrocardiographic waveform sampling point voltage value sequenceIn particular to electrocardiosignal digital sampling data continuously acquired by a flexible electrode, wherein each elementExpressed in timeA millivolt potential difference generated in the process of depolarization and repolarization of the myocardial cells measured at the moment;
specifically, the historical medical record also comprises blood oxygen saturation photoelectric volume pulse wave signals Wherein the continuous variableRepresenting the time axis, measuring the instantaneous concentration change of oxyhemoglobin in peripheral capillary vessel by using the ratio of red light (660 nm) to infrared light (940 nm) absorbance after normalization treatment, and measuring the respiratory impedance phase differenceIt refers to the thoracic impedance variation measured by four-electrode method in the frequency range of 10-100kHz, and its positive real number property is derived from the modular value calculation of complex impedance of biological tissue.
The characteristic extraction process adopts a multi-scale analysis method to detect QRS complex of the electrocardiosignal and extract the standard deviation of R-R intervalAs heart rate variability index, the R-R interval standard deviationAs an autonomic nervous system function index, the heart beat interval fluctuation intensity of the sinus node pacing point under the double regulation of the sympathetic and vagus nerves is quantified;
frequency of calculating second derivative zero crossing point of blood oxygen signal Characterizing the peripheral circulation state, and the zero crossing frequency of the second derivative of the blood oxygen signalBy calculation ofReflecting the pulse wave propagation velocity change of the peripheral blood vessel due to the peripheral resistance change;
carrying out power spectrum density analysis on the respiratory impedance signal, and taking the energy duty ratio of the frequency band of 0.15-0.4Hz As a respiratory mode stability index, the respiratory impedance signal has an energy duty ratio of 0.15-0.4Hz frequency bandThe low-frequency oscillation component is formed by driving diaphragm movement corresponding to the rhythmic discharge of the respiratory center under the state of human body rest.
After the normalization of each characteristic parameter by min-max, the characteristic vector is formedWherein the asterisks indicate normalized values, the normalization formula is:
In the middle of For the mean value of the feature in the historical dataset,The factor 3 ensures that most of the data falls within the 0,1 interval, which is the standard deviation.
And 32, constructing a characteristic distribution model of the patient group, inputting a standardized characteristic vector sequence, identifying characteristic distribution rules of different patient groups by using a cluster analysis method, and outputting the characteristic distribution model of the patient group.
It can be appreciated that because of the limitations of individual variability and population diversity of the normalized feature vector sequences, cluster analysis is required to obtain a model that can reflect the distribution of the features of different patient populations.
Specifically, the normalized feature vector sequence is input into a Gaussian Mixture Model (GMM) for cluster analysis, and the patient population is divided intoA subclass of individuals.
Defining an observation datasetSolving parameter sets by a expectation maximization algorithmWhereinIs a mixing coefficient representing the probability of occurrence of a group of patients in the population; Is the first Feature mean vector of individual subclasses, the mean vectorTypical numerical combinations comprising such patient's electrocardiographic, oximetry, and respiratory characteristics;
is a covariance matrix, the covariance matrix The linkage variation characteristics among the physiological parameters are characterized. Clustering effectiveness is improved by contour coefficientsEvaluation:
In the middle of For the sampleThe average distance to other samples in the same cluster,For the sampleMinimum average distance to nearest cluster. When (when)And the time judgment subclass division is reasonable.
The final constructed feature distribution model can be expressed as:。
and 33, constructing a transfer learning model, inputting a patient population feature distribution model and an individual feature vector sequence, mapping the public feature distribution to the individual feature distribution by using a transfer learning algorithm, and outputting an individual sign fluctuation threshold interval.
It can be understood that, because of the limitation of the distribution difference between the patient population characteristic distribution model and the individual characteristic vector sequence, migration learning is required to obtain an individual sign fluctuation threshold interval capable of adapting to individual characteristic distribution.
Specifically, the migration learning model adopts a domain adaptive network architecture, and defines a source domain as the characteristic distribution of the patient groupThe target domain is the characteristic sequence of individual patients. Network comprising shared feature extraction layerSum domain discriminatorDistribution alignment is achieved by Maximum Mean Difference (MMD) loss:
In the middle of To regenerate the mapping function of the kernel Hilbert space, gaussian kernel is selectedGaussian kernel parametersThe rate of similarity decay is controlled with the square root of its reciprocal corresponding to the effective comparison range in feature space.
The individuation threshold interval is determined by means of quantile regression, and is setIs characterized by the target domainThe value at the quantile, then the threshold interval is:
where IQR is a quarter bit distance, Representing interval expansion operations, in quantile regressionIn particular to a target patient whose physiological parameters are in the historical monitoring dataIs lower than the critical value, whereinAndRespectively correspond to the lower limit and the upper limit of the normal fluctuation.
And 34, updating the normal range of the heart rate variability coefficient in real time, inputting the individual sign fluctuation threshold interval and the physiological parameter data monitored in real time, and updating the normal range of the heart rate variability coefficient by using a dynamic adjustment algorithm.
It can be appreciated that, because of the time-varying limitation of the individual sign fluctuation threshold interval, dynamic adjustment is required to obtain the normal range of heart rate variability coefficients capable of reflecting the current physiological state.
Specifically, the heart rate variability coefficientThe dynamic update of (1) adopts an index weighted moving average method, is setThe real-time monitoring value of the time isUpdating the formula to;
Wherein the coefficient of smoothingExponentially weighted moving average coefficientEmbodying time attenuation characteristics, when sampling intervalIn the course of the time of the minute,Meaning that the data weight decays to 24 hours ago. The normal range is updated to;
When in real timeTriggering an early warning rechecking mechanism when the interval is exceeded three times continuously, and carrying out joint probability verification through the Bayesian network composite event detection model, wherein the Bayesian network composite event detection model particularly refers to a joint probability reasoning mechanism integrating multidimensional indexes such as abnormal heart rate variation coefficients, change of blood oxygen fluctuation modes, breathing rhythm disorder and the like through a conditional probability table.
It can be appreciated that, since abnormality of a single physiological parameter may not be enough to accurately determine occurrence of a composite event, it is necessary to monitor at least two physiological parameters simultaneously, and perform comprehensive analysis by using a bayesian network composite event detection model, thereby improving accuracy and reliability of detection.
And triggering the Bayesian network compound event detection model when at least two physiological parameters (such as an electrocardiographic waveform amplitude threshold and blood oxygen saturation photoelectric intensity) in the aligned signal sequence reach the real-time updated individual sign fluctuation threshold interval at the same time. The individualized sign fluctuation threshold interval is the individualized sign fluctuation threshold interval output in the step three, and the numerical range updated in real time through a dynamic adjustment algorithm is used for reflecting the abnormal critical value of the current physiological state of the patient.
Specifically, the method for triggering the bayesian network composite event detection model is as follows:
the Bayesian network compound event detection model calculates multi-parameter joint abnormality probability based on the at least two items of physiological parameter data, and outputs the multi-parameter joint abnormality probability.
It can be understood that, because the abnormality of a single parameter may be caused by multiple factors, the calculation of the multi-parameter combined abnormality probability can more comprehensively reflect the occurrence probability of the composite event, thereby providing a more accurate early warning basis.
Specifically, firstly, preprocessing input data, normalizing physiological parameter data such as electrocardiographic waveforms and blood oxygen saturation into a unified dimension, for example, normalizing electrocardiographic amplitude to a [0,1] interval, and converting the blood oxygen saturation into a percentage form, so as to ensure the compatibility of the input data.
And the model triggering mechanism is used for judging whether a plurality of physiological parameters simultaneously cross the threshold value interval through the real-time monitoring module, for example, if the electrocardio amplitude exceeds 0.8 (normalized value) and the blood oxygen saturation is lower than 90%, the model operation is triggered. Then, loading network topology, and calling a predefined Bayesian network structure file, wherein the file comprises nodes (representing abnormal states of physiological parameters) and directed edges (representing causal relations among parameters), for example, node A represents 'electrocardio abnormality', node B represents 'blood oxygen abnormality', and edge A-B represents that the electrocardio abnormality possibly causes blood oxygen abnormality. Then, the Bayesian network composite event detection model is based on a formulaThe multi-parameter joint abnormality probability is calculated, wherein (P (A)) represents the prior probability of single abnormality of the electrocardio parameter, and the prior probability is obtained through historical data statistics (for example, the historical abnormality rate of a patient is 0.15). (P (B|A)) represents the conditional probability of blood oxygen abnormality under the electrocardiographic abnormality condition, and is provided by the clinical knowledge base (e.g., set to 0.6).
Firstly, according to the abnormal state (such as A=1 and B=1) of input parameter, the correspondent joint probability value is retrieved from the conditional probability table of the model, for example, if the joint probability is 0.75 when A and B are abnormal, said value can be directly outputted, if the deviation of real-time data and history distribution exceeds preset threshold (for example exceeds + -10%), the formula can be updated by BayesianThe conditional probability is modified, (P (a|b)) to be the current observed joint anomaly frequency,Is the original blood oxygen abnormality probability.
Further, the method for generating the hierarchical early warning signal is as follows:
And generating a grading early warning signal according to the multi-parameter joint anomaly probability and a preset grading rule, wherein the grading early warning signal comprises early warning information of different grades.
Specifically, according to the multi-parameter joint anomaly probability (P (A, B)), a grading early warning signal is generated according to the following rule that when P is more than or equal to 0.3 and less than or equal to 0.6, a yellow early warning is triggered to prompt medical staff to check equipment connection or patient position, when P is more than or equal to 0.6 and less than or equal to 0.8, an orange early warning is triggered to automatically push recommended treatment measures (such as increasing oxygen flow), and when P is more than or equal to 0.8, a red early warning is triggered to start an emergency call system and display the position of a patient. In the specific implementation of generating the grading early warning signal, on one hand, the red early warning threshold value is adjusted down from 0.8 to 0.7 for high-risk patients in combination with the patient individuation sign fluctuation threshold value interval to improve sensitivity, and on the other hand, the early warning signal is synchronously output through a flashing indicator lamp with the frequency rising along with the risk, a buzzer with the frequency rising along with the risk and a popup window message containing a treatment guide link, so that timely information transmission is ensured.
It should be noted that, in the embodiment of the present invention, the hierarchical early warning signal includes early warning level (level 1-3), trigger parameter combination and timestamp field. The key fields are extracted through a regular expression matching algorithm, wherein the regular expression matching algorithm refers to a text processing method for extracting key information such as early warning level, parameter combination and the like from unstructured text through a predefined pattern character string, and in the embodiment, the field extraction is realized by adopting PCRE specifications. After extraction, it is converted into a unified dimension vector
Further, the pushing the decision support information including the rescue priority label to the monitoring terminal includes the following steps:
Step 51, dynamically loading a clinical knowledge graph through a micro-service architecture, and retrieving a treatment plan associated with the early warning signal from the clinical knowledge graph;
It will be appreciated that since the clinical knowledge graph may need to be dynamically adjusted according to specific pre-warning signals, dynamic loading by the micro-service architecture is required to ensure that the retrieved treatment plan is relevant to the current pre-warning signals.
Specifically, the clinical knowledge graph is stored in a Neo4j graph database, and the Neo4j graph database is a non-relational database for storing data by adopting a node-relational structure, and is used for storing disease entity nodes and associated treatment plan nodes thereof in the system, wherein the disease entity nodes comprise a plurality of cardiovascular and cerebrovascular disease diagnosis standards, and the treatment plan nodes store a plurality of clinically verified first-aid processes. When receivingIN this case, the plan search is performed by using the Cypher query term MATCH (d: disease) - [ r: HAS_PROTOCOL ] - > (P: PROTOCOL) WHERE d.symptom IN P RETURN p.weight.
Wherein the weight parameterΑ=0.6, β=0.4 is a coefficient determined by receiver operation characteristic curve (ROC curve) optimization, and the optimal coefficient combination is determined by ROC curve optimization to balance the weight of the early warning level and the parameter coverage. Indication functionThe value of s is 1 when the parameter exists in the preset keyword set, otherwise, is 0. TOP5 plan set can be returned by the above operations。
Step 52, associating the early warning signal with the retrieved treatment plan to generate a rescue priority label;
It will be appreciated that since the pre-alarm signals and treatment protocols may need to be prioritized according to a particular clinical context, an association operation is required to generate a rescue priority label that meets clinical needs.
Specifically, in the process ofAnd (3) withWhen in association, a multi-objective optimization model is establishedWhereinThe weight coefficient representing the similarity of the parameter matches,A weight coefficient representing historical success rate, the combination of coefficients being determined by a cardiologist based on clinical experience.
Similarity functionImproved algorithm using Jaccard coefficients to calculate trigger parameter setAnd a set of pre-plan keywordsGeometric mean of the intersection ratio of (i) i.e。
Solving to obtain the optimal planAfter that, its priority is markedWhereinThe representation is rounded up to generate a set of labels. Priority taggingIn (a) and (b)The normalized weight value of the optimal plan is represented, the value range [0,1] is multiplied by 10 and then is converted into an integer priority scale of 1-10 through an upward rounding operator ⌈. ⌉, wherein ζ is more than or equal to 8 and corresponds to critical conditions, ζ is more than or equal to 5 and less than or equal to ζ <7 and corresponds to early warning conditions, and ζ is more than or equal to 5 and corresponds to observation conditions.
Step 53, pushing decision support information containing rescue priority labels to a monitoring terminal;
It can be appreciated that since the decision support information needs to be timely transmitted to the monitoring terminal to guide the clinical operation, a pushing operation is required to ensure timeliness and accuracy of the information.
Specifically, the message queue will be through RabbitMQThe information which is packed into the HL7 format by the data packet is pushed to the monitoring terminal and is the observation result information type in the medical information exchange standard, and the information comprises structural fields such as patient ID, equipment code, early warning data and the like.
The terminal analysis module is according toThe value is color coded, namely, zeta is more than or equal to 8 and flashes in red to correspond to critical conditions and require medical staff to intervene immediately, zeta is more than or equal to 5 and is normally bright in yellow to correspond to early warning conditions and prompt and intensive monitoring, zeta is blue to prompt when zeta is less than or equal to 5 and corresponds to observation conditions and vital sign changes need to be recorded. The push response time is less than or equal to 200ms, and the time stamp checking mechanism is adoptedEnsuring real-time performance, wherein epsilon=300 ms is a system tolerance threshold value, epsilon=300 ms in the timestamp checking mechanism is a system response time upper limit set according to cardiac arrest golden rescue time, and when the message is sentAnd time of receptionThe retransmission mechanism is triggered when the absolute difference of (a) exceeds this threshold.
In addition, in the foregoing technical solutions provided in the embodiments of the present application, parts consistent with implementation principles of corresponding technical solutions in the prior art are not described in detail, so that redundant descriptions are avoided.
The purpose, technical scheme and beneficial effects of the invention are further described in detail in the detailed description. It is to be understood that the above description is only of specific embodiments of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
The above preset parameters or preset thresholds are set by those skilled in the art according to actual conditions or are obtained by mass data simulation.
The above embodiments are only for illustrating the technical method of the present invention and not for limiting the same, and it should be understood by those skilled in the art that the technical method of the present invention may be modified or substituted without departing from the spirit and scope of the technical method of the present invention.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510757533.2A CN120241087A (en) | 2025-06-09 | 2025-06-09 | A physiological parameter monitoring and intelligent early warning system for critically ill patients |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510757533.2A CN120241087A (en) | 2025-06-09 | 2025-06-09 | A physiological parameter monitoring and intelligent early warning system for critically ill patients |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN120241087A true CN120241087A (en) | 2025-07-04 |
Family
ID=96187724
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510757533.2A Withdrawn CN120241087A (en) | 2025-06-09 | 2025-06-09 | A physiological parameter monitoring and intelligent early warning system for critically ill patients |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN120241087A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN121080993A (en) * | 2025-10-30 | 2025-12-09 | 江苏盖睿健康科技有限公司 | A method and system for electrocardiogram signal analysis |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110004110A1 (en) * | 2000-05-30 | 2011-01-06 | Vladimir Shusterman | Personalized Monitoring and Healthcare Information Management Using Physiological Basis Functions |
| CN118965181A (en) * | 2024-10-18 | 2024-11-15 | 吉林大学第一医院 | An intelligent monitoring and analysis system for multi-source critical illness parameters |
| CN119235279A (en) * | 2024-12-05 | 2025-01-03 | 吉林大学 | Intelligent monitoring system and method for nursing of critically ill patients |
| CN119453949A (en) * | 2025-01-10 | 2025-02-18 | 吉林大学 | Intelligent nursing management system and method for critically ill patients |
-
2025
- 2025-06-09 CN CN202510757533.2A patent/CN120241087A/en not_active Withdrawn
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110004110A1 (en) * | 2000-05-30 | 2011-01-06 | Vladimir Shusterman | Personalized Monitoring and Healthcare Information Management Using Physiological Basis Functions |
| CN118965181A (en) * | 2024-10-18 | 2024-11-15 | 吉林大学第一医院 | An intelligent monitoring and analysis system for multi-source critical illness parameters |
| CN119235279A (en) * | 2024-12-05 | 2025-01-03 | 吉林大学 | Intelligent monitoring system and method for nursing of critically ill patients |
| CN119453949A (en) * | 2025-01-10 | 2025-02-18 | 吉林大学 | Intelligent nursing management system and method for critically ill patients |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN121080993A (en) * | 2025-10-30 | 2025-12-09 | 江苏盖睿健康科技有限公司 | A method and system for electrocardiogram signal analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105228508B (en) | A system for determining risk scores for classification | |
| Jin et al. | Predicting cardiovascular disease from real-time electrocardiographic monitoring: An adaptive machine learning approach on a cell phone | |
| EP3698709A1 (en) | Electrocardiogram information processing method and electrocardiogram workstation system | |
| Raj et al. | A personalized point-of-care platform for real-time ECG monitoring | |
| CN108309262A (en) | Multi-parameter monitoring data analysing method and multi-parameter monitor | |
| CN108309263A (en) | Multi-parameter monitoring data analysing method and multi-parameter monitoring system | |
| CN106073755A (en) | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically | |
| Moridani et al. | A reliable algorithm based on combination of EMG, ECG and EEG signals for sleep apnea detection:(a reliable algorithm for sleep apnea detection) | |
| CN106901705A (en) | A kind of unaware human Body Physiology Multi-parameter harvester and acquisition method and application | |
| KR102802002B1 (en) | Ecg measurement service providing method and system using potable ecg device | |
| Roy et al. | On-device reliability assessment and prediction of missing photoplethysmographic data using deep neural networks | |
| CN106037720A (en) | Application method of hybrid continuous information analysis technology in medicine | |
| Hu et al. | A novel method for continuous blood pressure estimation based on a single-channel photoplethysmogram signal | |
| CN120241087A (en) | A physiological parameter monitoring and intelligent early warning system for critically ill patients | |
| Reddy et al. | Automated hilbert envelope based respiration rate measurement from PPG signal for wearable vital signs monitoring devices | |
| CN117297564A (en) | A wearable pulse diagnosis bracelet system based on ECG photoelectric pressure pulse signal | |
| CN105266764B (en) | A kind of traditional Chinese medical science pectoral qi assessment device | |
| JYOTHI et al. | AUTOMATIC CLASSIFICATION OF ECG AND PCG SIGNALS USING CONVOLUTION NEURAL NETWORK FOR DETECTING CARDIOVASCULAR DISEASE | |
| CN120514350A (en) | Continuous invasive blood pressure monitoring system, method and invasive blood pressure sensor | |
| Jung et al. | Monitoring senior wellness status using multimodal biosensors | |
| Suleman et al. | Respiratory Events Estimation From PPG Signals Using a Simple Peak Detection Algorithm | |
| Corradi et al. | Real time electrocardiogram annotation with a long short term memory neural network | |
| Kew et al. | Wearable patch-type ECG using ubiquitous wireless sensor network for healthcare monitoring application | |
| CN119033345A (en) | Omnibearing non-invasive heart failure monitoring system and monitoring method | |
| Ahmad et al. | IoT-Enabled Smart E-Healthcare System with Predictive Prescription Algorithm for Automatic Patient Monitoring and Treatment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20250704 |