CN114758786A - Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters - Google Patents

Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters Download PDF

Info

Publication number
CN114758786A
CN114758786A CN202210392852.4A CN202210392852A CN114758786A CN 114758786 A CN114758786 A CN 114758786A CN 202210392852 A CN202210392852 A CN 202210392852A CN 114758786 A CN114758786 A CN 114758786A
Authority
CN
China
Prior art keywords
subsystem
early warning
data
post
hemorrhagic shock
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.)
Pending
Application number
CN202210392852.4A
Other languages
Chinese (zh)
Inventor
张广
袁晶
余明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Medical Support Technology of Academy of System Engineering of Academy of Military Science
Original Assignee
Institute of Medical Support Technology of Academy of System Engineering of Academy of Military Science
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Institute of Medical Support Technology of Academy of System Engineering of Academy of Military Science filed Critical Institute of Medical Support Technology of Academy of System Engineering of Academy of Military Science
Priority to CN202210392852.4A priority Critical patent/CN114758786A/en
Publication of CN114758786A publication Critical patent/CN114758786A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Theoretical Computer Science (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Pathology (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention relates to a dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters, which comprises: the system comprises a physiological parameter sensing subsystem, an early warning model updating subsystem, a dynamic early warning subsystem, a post-traumatic hemorrhagic shock onset judgment subsystem and a silencing subsystem; the dynamic early warning subsystem is used for calculating the occurrence probability of post-traumatic hemorrhagic shock of the patient, and the post-traumatic hemorrhagic shock onset judgment subsystem corrects the prediction probability value according to the medical environment data and the like of the site where the post-traumatic hemorrhagic shock onset judgment subsystem is located. And the silencing subsystem controls the early warning result of the dynamic early warning subsystem controlled by the silencing subsystem according to the difference among the test data, the training data and the label data. The invention only uses common non-invasive parameters, does not need laboratory data, can be used in remote areas, emergent public health events, first-line battlefield conditions and other scenes, eliminates the damage to individual patients caused by frequently acquiring laboratory parameters, and reduces the use cost of the system and the early warning error probability.

Description

Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters
Technical Field
The invention relates to the field of artificial intelligence technology and medical health, in particular to a dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters.
Background
Hemorrhagic shock is a hypovolemic shock that results in insufficient oxygen delivery at the cellular level due to severe blood loss. If blood loss is not controlled, the patient can die quickly, and the number of hemorrhagic shock deaths caused by trauma accounts for 10-40% of the total number of deaths caused by trauma. Retrospective studies on multiple databases have shown that patients admitted to the hospital after a recent 1/4 trauma develop symptoms of hemorrhagic shock during admission. For most patients, the physical effects of shock are initially reversible, but repeated or prolonged hypotension, the use of large doses of vasopressors drugs may worsen the prognosis.
Medical personnel intermittently assess monitored vital signs of a patient and rely on individual physiological parameters to identify patients at risk for deterioration. These early warning systems do not use the patient's full information and therefore the alarm is often inaccurate, leading to alarm fatigue. The low-precision early warning system is not beneficial to the identification and interpretation of relevant information by clinicians, nurses or doctors are difficult to continuously supervise or evaluate ICU patients, and the low-frequency low-precision early warning system is difficult to adapt to patients with acute changes of illness states. Due to the lack of predictive tools of sufficient accuracy, clinicians may resort to subjective judgments, which increases the risk of a physician taking action based on relevant information. Prophylactic blood transfusion or the use of 100% mechanical ventilation has a significant effect on reducing the mortality of hemorrhagic shock. If an early warning evaluation system can dynamically and early predict the occurrence probability of the hemorrhagic shock of the patient after the trauma, a timely preventive treatment plan can be provided for the high-risk patient, so that the death rate and the medical expense are greatly reduced. Doctors in intensive care units can obtain a large amount of patient physiological data and measurement indexes from a plurality of monitoring systems, but the limited ability of human beings for processing complex information hinders early identification of patient disease deterioration, and the conventional identification early warning method is difficult to monitor the patient disease in real time. Deep learning techniques perform well in analyzing complex signals in data-rich environments. The large amount of data collected in the ICU and the disclosure of the Medical Information Mart for Intensive Care III (MIMIC III) database, elcu, AmsterdamUMC, etc. large Medical critical data sets are the key to developing machine learning in this environment.
The patent CN201910570791 provides a time series prediction method for a traumatic hemorrhagic shock injury, which can extract traumatic hemorrhagic shock injury data from a database and perform data processing on the traumatic hemorrhagic shock injury data, wherein the time series prediction method has the functions of processing data abnormal values and performing linear supplementation and clustering supplementation on the data; designing a step index for the processed data; and constructing a prediction model by using the index step result and the classifiers of different types, and predicting the result after the preset duration through the prediction model. The invention can implement real-time dynamic prediction and early warning based on time sequence on the traumatic hemorrhagic shock by using the index capable of being monitored in real time, but clinical intervention time is not reserved, and interpretability analysis is not carried out on the model.
In addition, in the field of medical health big data, clinical data of a patient can be divided into section data with only one section and time sequence data with a plurality of sections, and the time sequence prediction precision is higher than that of section prediction due to the fact that the section data has the characteristics of large information amount, trend change and the like, and sliding prediction and real-time monitoring and early warning of illness states can be achieved. However, the problems of sparseness and deficiency of medical data are directly caused to be serious due to the fact that wounded measurement indexes are different, the wounded index measurement time is different, most of test indexes cannot be measured for multiple times in a short period, and the like. The prior art at least has the following problems:
The method has the advantages that (I) no more mature gap filling technical system exists in the aspect of data gap filling, most of the data gap filling methods adopt mean gap filling or linear gap filling, the gap filling method is single, and the problems of poor data quality, large difference with real data and the like still exist after gap filling;
secondly, the prior art method mostly adopts a section prediction mode, for example, the data is subjected to prediction after being averaged to obtain a section, the obtained result is 'final', and rolling prediction and real-time disease monitoring cannot be realized;
the existing method utilizes a small amount of time sequences to predict, only selects vital sign indexes with low measurement cost and many times, such as heart rate, blood pressure and the like, and has poor prediction effect;
the existing early warning method does not adopt related treatment for reducing the false alarm rate, and frequent model false alarm causes alarm fatigue of medical staff, thereby influencing active clinical treatment of patients in critical state.
The related technology disclosed in the prior art can realize the possible prediction of the post-traumatic hemorrhagic shock of a patient by utilizing physiological parameters of the patient, but lacks the clinical intervention time between the running time point of a reserved early warning model and the early warning time point of the attack, and lacks a method for effectively reducing the false alarm rate.
Disclosure of Invention
The invention discloses a dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters, which aims to solve the problems of the existing dynamic early warning system for post-traumatic hemorrhagic shock, realizes the prediction of the possibility of the onset of the post-traumatic hemorrhagic shock of a patient in multiple time scales in the future, reduces the false alarm rate and realizes effective early warning.
The invention discloses a dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters, which comprises: a physiological parameter sensing subsystem, an early warning model updating subsystem, a dynamic early warning subsystem, a post-traumatic hemorrhagic shock onset judgment subsystem and a silencing subsystem;
the physiological parameter perception subsystem is connected with the dynamic early warning subsystem, the dynamic early warning subsystem is respectively connected with the early warning model updating subsystem and the post-traumatic hemorrhagic shock onset judgment subsystem, and the silence subsystem is respectively connected with the post-traumatic hemorrhagic shock onset judgment subsystem, the dynamic early warning subsystem and the early warning model updating subsystem.
The physiological parameter sensing subsystem is used for monitoring and acquiring physiological parameter data information of a patient in real time, preprocessing the acquired physiological parameter data of the patient and then sending the preprocessed data to the dynamic early warning subsystem.
The physiological parameter data information of the patient comprises the conventional noninvasive physiological parameters of the patient and physiological parameter time sequence data in a learning window of the patient.
The method comprises the steps of utilizing discrete sampling values of acquired physiological parameters of patients to form sampling vectors, calculating cross-correlation matrixes of the sampling vectors, utilizing a characteristic extraction method to process the cross-correlation matrixes to obtain characteristic value vectors of the sampling vectors, utilizing the characteristic value vectors to weight the sampling vectors to obtain smooth values of the physiological parameters of the patients, and using the smooth values as preprocessed data, so that interference of the acquired error values such as wild values on a follow-up early warning judgment process is effectively reduced, and false alarm probability of the early warning judgment process is reduced.
The preprocessing of the acquired physiological parameter data of the patient comprises that a sampling vector formed by discrete sampling values of the physiological parameter of the patient acquired within a period of time is represented as [ x1,x2,…,xN]And N is the number of discrete sampling values of the physiological parameters of the patient acquired within a period of time, a mutual matrix C of the sampling vector is obtained by calculation, and the characteristic value decomposition is carried out on the mutual matrix C to obtain:
C＝VDVH，
And C is a characteristic vector matrix, D is a characteristic value matrix, the diagonal elements of the matrix D are normalized and used as weight vectors, and the sampling vectors are subjected to weighted summation to obtain a smooth value of the physiological parameters of the patient as preprocessed data.
The dynamic early warning subsystem is used for receiving and cleaning the conventional noninvasive physiological parameters of the patient and the time sequence data of the physiological parameters in the learning window of the patient, which are acquired by the physiological parameter perception subsystem, calculating the original probability of the hemorrhagic shock of the patient after trauma in the prediction time window, then calculating the occurrence probability of the hemorrhagic shock of the patient after trauma in the future prediction time window range according to the conventional noninvasive physiological parameters and the time sequence data of the physiological parameters in the learning window of the patient, acquiring the prediction probability value, sending the prediction probability value to the hemorrhagic shock onset judgment subsystem after trauma, correcting the prediction probability value according to the medical environment data of the position of the patient and the doctor treatment experience evaluation value, and if the corrected prediction probability value is larger than a first prediction probability threshold value, the dynamic early warning system for post-traumatic hemorrhagic shock based on the noninvasive parameters judges that the post-traumatic hemorrhagic shock will occur in the future prediction time window range of the patient, and if the corrected prediction probability value is less than or equal to a second prediction probability threshold value, the system judges that the post-traumatic hemorrhagic shock will not occur in the future prediction time window range of the patient.
The early warning model updating subsystem monitors the medical environment where the system is located, and if the early warning model updating subsystem monitors that the medical environment where the system is located changes, the subsystem updates the weight of the deep learning model in the dynamic early warning subsystem by using an incremental learning method according to the conventional noninvasive physiological parameters of the patient and the prognosis information including the death condition of the patient, the hemorrhagic shock occurrence condition after trauma and the hospitalization duration, which are acquired by the physiological parameter sensing subsystem, so that the early warning result of the system adapts to the medical environment where the system is located.
The silence subsystem judges the difference among test data, training data and label data of post-traumatic hemorrhagic shock dynamic early warning of the intelligent dynamic early warning subsystem, if the three types of data are remarkably different, the silence subsystem controls the early warning result of the dynamic early warning subsystem not to be output, collects the early warning result data of the dynamic early warning subsystem within a period of time, extracts the early warning result data with the maximum prediction probability within the period of time, and outputs the early warning result data to the post-traumatic hemorrhagic shock onset judgment subsystem.
The silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
regarding data of an updating subsystem of the early warning model as a stable random process, respectively establishing corresponding autoregressive-moving average models, namely ARMA models, aiming at test data, label data and training data to respectively obtain a first ARMA model, a second ARMA model and a third ARMA model, calculating cross correlation matrixes of the coefficients of the three ARMA models, calculating the cross correlation matrixes to obtain maximum characteristic values, and distinguishing the differences of the test data, the label data and the training data by using the maximum characteristic values.
And when the maximum characteristic value is larger than the difference judgment threshold value, judging that the significance difference occurs among the test data, the label data and the training data of the early warning model updating subsystem.
The silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
the silence subsystem predicts to obtain a label data predicted value by using the training data and a deep learning model, calculates the credibility among the test data, the training data and the label data predicted value by using a credibility assessment method, and judges that the early warning data, the label data and the training data of the early warning model updating subsystem have significant difference when the credibility is lower than a credibility threshold value.
The silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
forming a sample set by using training data and a label data predicted value, dividing the sample set into a first reference sample set and a second reference sample set corresponding to the test data according to the label category to which each test data belongs, wherein the first reference sample set is a set formed by samples corresponding to the label category to which the test data belongs, the second reference sample set is a set formed by other samples except the first reference sample set in the sample set, calculating the distance between each sample in the first reference sample set and the reference sample set by using a consistency metric calculation function to obtain a first sample set distance value vector, calculating the distance between each test data and the corresponding first reference sample set and the corresponding second reference sample set by using the consistency metric calculation function to obtain a first set distance value and a second set distance value of the test data correspondingly, sequencing the first set distance values of the test data in an ascending manner in the first sample set distance value vector to obtain sequence number values, calculating the ratio of the sequence number values to the total number of samples of the first reference sample set to be used as the reliability of the test data, and judging that the early warning data, the label data and the training data of the early warning model updating subsystem have significance differences if the reliability of the test data is lower than a reliability threshold; the confidence level of each test datum is calculated and compared to a confidence threshold.
The deep learning model can adopt a stack type deep learning model, a convolution neural network module and the like.
The stack type deep learning model comprises a convolution layer, a bidirectional long-short term memory layer and a self-attention layer.
The dynamic early warning subsystem interpolates missing patient physiological parameter time sequence data in a minimum time unit by adopting a self-adaptive interpolation method, and the method comprises the following specific steps of:
s1, calculating a median Midparam and a quartile IQRIdparam of sampling frequencies of various physiological parameters of each patient during the period from admission to discharge;
s2, selecting physiological parameters with missing values of a patient with missing data, recording the parameters as param parameters, and selecting the time positions of null values or abnormal values in the collected param parameters as selected positions according to the ascending sequence of the collection time;
s3, interpolating data of the selected position in the following priority order, first using the median of the effective values of the preceding (Midparam + IQRidparam) sampling time lengths of the selected position as the data interpolation value of the selected position, second using the median of the effective values of the preceding (Midparam + 2. i qridparam) sampling time lengths of the selected position as the data interpolation value of the selected position, second using the median of the effective values of the collected param parameter as the data interpolation value of the selected position, and finally using the median of the param parameter recorded in the database as the data interpolation value of the selected position.
The post-traumatic hemorrhagic shock morbidity judgment subsystem receives the original probability of the post-traumatic hemorrhagic shock morbidity of a patient in a prediction time window, and obtains a classification threshold value D of a stacked deep learning model according to the sensitivity and the specificity of the stacked deep learning model in the dynamic early warning subsystem on test set data, so that when the classification threshold value is D, the difference value between the sensitivity and the specificity of the stacked deep learning model on the test set data is minimum.
The post-traumatic hemorrhagic shock onset judgment subsystem corrects the occurrence probability value of the post-traumatic hemorrhagic shock according to the medical environment data where the subsystem is located and doctor treatment experience evaluation values, and the calculation formula of the corrected post-traumatic hemorrhagic shock onset probability is as follows:
Figure BDA0003596226490000061
where C is the probability threshold.
After the early warning model updating subsystem monitors that the medical environment of the place where the early warning model updating subsystem is located is changed, physiological parameters and prognosis information of a patient are collected, or the input physiological parameters and prognosis information of the patient are used, the model parameters are updated on the basis that the structure of the stack-type deep learning model is not changed by combining an increment training method, and the physiological parameters of the patient in the range of a learning window are transmitted to the dynamic early warning subsystem after the parameters of the stack-type deep learning model are updated in real time.
The beneficial effects of the invention are as follows:
the invention discloses a dynamic early warning deep learning system for post-traumatic hemorrhagic shock based on noninvasive parameters, which only uses common noninvasive parameters without laboratory data, enlarges the application range of the system, makes the use of the system possible in remote areas, emergent public health events and first-line situations of battlefields, eliminates the damage to individual patients caused by frequent acquisition of laboratory parameters and reduces the use cost of the system. Compared with a traditional linear addition regression model, the automatic and streamlined intelligent dynamic early warning subsystem has higher computational complexity, is more suitable for common nonlinear problems in practical problems, and can provide better post-traumatic hemorrhagic shock onset early warning capability. The pre-warning interval reserved for clinical intervention can provide sufficient time for a physician to design a patient treatment regimen. The analysis result of the system is consistent with the clinical research result, the model performance is excellent, the system can automatically update the model weight according to the environmental change so as to adapt to different clinical environments, and meanwhile, the false probability of predicting the hemorrhagic shock after trauma is effectively reduced.
Drawings
FIG. 1 is a block diagram of a dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters, which is disclosed by the present invention;
FIG. 2 is a diagram showing the relationship among an observation window, a delay window and an early warning window in the dynamic early warning subsystem of the present invention;
FIG. 3 is a flow chart of the mute subsystem operation of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive step based on the embodiments of the present invention, are within the scope of protection of the present invention.
The terms "first," "second," and the like in the description and claims of the present invention and in the above-described drawings are used for distinguishing between different objects and not necessarily for describing a particular sequential or chronological order. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, apparatus, product, or apparatus that comprises a list of steps or elements is not limited to those listed but may alternatively include other steps or elements not listed or inherent to such process, method, product, or apparatus.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the invention. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
FIG. 1 is a block diagram of a dynamic early warning system for post-traumatic hemorrhagic shock based on non-invasive parameters, which is disclosed by the present invention; FIG. 2 is a diagram showing the relationship among an observation window, a delay window, and an early warning window in the dynamic early warning subsystem according to the present invention. In fig. 2, the learning window: and the time range with available data is used for inputting the intelligent dynamic early warning submodule. And (3) prediction window: a period of time to determine whether post-traumatic hemorrhagic shock has occurred. Delay window: the time difference between the prediction window and the learning window. Wherein T0 is the time point when the patient stays in the ICU, T1 is set as the current time point, T1 is more than or equal to T0, x1 is the time length of the learning window, T1-x1 is more than or equal to T0, x2 is the length of the delay window, the time period of the prediction window is [ T1+ x2, T1+ x2+ x3], and x3 is the length of the prediction window. FIG. 3 is a flow chart of the mute subsystem operation of the present invention.
The non-invasive parameters in the invention refer to parameters which can be measured without causing trauma to the body of a patient or laboratory environment, such as heart rate, non-invasive blood pressure, respiratory rate, sex, age and the like. Noninvasive parameters are distinguished from laboratory parameters.
The first embodiment is as follows:
the invention discloses a dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters, which comprises: a physiological parameter sensing subsystem, an early warning model updating subsystem, a dynamic early warning subsystem, a post-traumatic hemorrhagic shock onset judgment subsystem and a silencing subsystem;
the physiological parameter perception subsystem is used for monitoring and acquiring physiological parameter data information of a patient in real time, preprocessing the acquired physiological parameter data of the patient and then sending the preprocessed data to the dynamic early warning subsystem.
The physiological parameter perception subsystem is connected with the dynamic early warning subsystem, the dynamic early warning subsystem is respectively connected with the early warning model updating subsystem and the post-traumatic hemorrhagic shock onset judgment subsystem, and the silence subsystem is respectively connected with the post-traumatic hemorrhagic shock onset judgment subsystem, the dynamic early warning subsystem and the early warning model updating subsystem.
The physiological parameter data information of the patient comprises the conventional noninvasive physiological parameters of the patient and physiological parameter time sequence data in a learning window of the patient.
The method comprises the steps of utilizing discrete sampling values of acquired physiological parameters of patients to form sampling vectors, calculating cross-correlation matrixes of the sampling vectors, utilizing a characteristic extraction method to process the cross-correlation matrixes to obtain characteristic value vectors of the sampling vectors, utilizing the characteristic value vectors to weight the sampling vectors to obtain smooth values of the physiological parameters of the patients, and using the smooth values as preprocessed data, so that interference of the acquired error values such as wild values on a follow-up early warning judgment process is effectively reduced, and false alarm probability of the early warning judgment process is reduced.
The preprocessing of the acquired physiological parameter data of the patient comprises that a sampling vector formed by discrete sampling values of the physiological parameter of the patient acquired within a period of time is represented as [ x1,x2,…,xN]And N is the number of discrete sampling values of the physiological parameters of the patient acquired within a period of time, a mutual matrix C of the sampling vector is obtained by calculation, and the characteristic value decomposition is carried out on the mutual matrix C to obtain:
C＝VDVH，
And C is a characteristic vector matrix, D is a characteristic value matrix, the diagonal elements of the matrix D are normalized and used as weight vectors, and the sampling vectors are subjected to weighted summation to obtain a smooth value of the physiological parameters of the patient as preprocessed data.
The dynamic early warning subsystem is used for receiving and cleaning the conventional noninvasive physiological parameters of the patient and the physiological parameter time sequence data in the learning window of the patient, which are acquired by the physiological parameter sensing subsystem, calculating the original probability of the hemorrhagic shock of the patient after trauma in the prediction time window, then calculating the occurrence probability of the hemorrhagic shock of the patient after trauma in the future prediction time window range according to the conventional noninvasive physiological parameters and the physiological parameter time sequence data in the learning window of the patient, which are acquired by the physiological parameter sensing subsystem, obtaining the prediction probability value, sending the prediction probability value to the hemorrhagic shock onset judgment subsystem after trauma, correcting the prediction probability value according to the medical environment data where the hemorrhagic shock onset judgment subsystem is located and the doctor treatment experience evaluation value, and if the corrected prediction probability value is larger than a first prediction probability threshold, the dynamic early warning system for post-traumatic hemorrhagic shock based on the noninvasive parameters judges that post-traumatic hemorrhagic shock will occur in the range of the future prediction time window of the patient, and if the corrected prediction probability value is less than or equal to the second prediction probability threshold value, the system judges that post-traumatic hemorrhagic shock will not occur in the range of the future prediction time window of the patient.
The early warning model updating subsystem monitors the medical environment where the system is located, and if the early warning model updating subsystem monitors the change of the medical environment where the system is located, the subsystem updates the weight of the deep learning model in the dynamic early warning subsystem by using an incremental learning method according to the conventional non-invasive physiological parameters of the patient and the prognosis information including the death condition of the patient, the hemorrhagic shock occurrence condition after trauma and the hospitalization duration, which are acquired by the physiological parameter sensing subsystem, so that the early warning result of the system adapts to the medical environment where the system is located.
The silence subsystem judges the difference among test data, training data and label data of dynamic early warning of the post-traumatic hemorrhagic shock of the intelligent dynamic early warning subsystem, if the three types of data are significantly different, the silence subsystem controls the early warning result of the dynamic early warning subsystem not to be output, collects the early warning result data of the dynamic early warning subsystem within a period of time, extracts the early warning result data with the maximum prediction probability within the period of time, and outputs the early warning result data to the post-traumatic hemorrhagic shock attack judgment subsystem.
The silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
regarding data of an updating subsystem of the early warning model as a stable random process, respectively establishing corresponding autoregressive-moving average models, namely ARMA models, aiming at test data, label data and training data to respectively obtain a first ARMA model, a second ARMA model and a third ARMA model, calculating cross correlation matrixes of the coefficients of the three ARMA models, calculating the cross correlation matrixes to obtain maximum characteristic values, and distinguishing the differences of the test data, the label data and the training data by using the maximum characteristic values.
And when the maximum characteristic value is larger than the difference judgment threshold value, judging that the significance difference occurs among the test data, the label data and the training data of the early warning model updating subsystem.
The silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
the silence subsystem predicts to obtain a label data predicted value by using the training data and a deep learning model, calculates the credibility among the test data, the training data and the label data predicted value by using a credibility assessment method, and judges that the early warning data, the label data and the training data of the early warning model updating subsystem have significant difference when the credibility is lower than a credibility threshold value.
The silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
forming a sample set by using training data and a label data predicted value, dividing the sample set into a first reference sample set and a second reference sample set corresponding to the test data according to the label category of each test data, wherein the first reference sample set is a set formed by samples corresponding to the label category of the test data, the second reference sample set is a set formed by other samples except the first reference sample set in the sample set, calculating the distance between each sample in the first reference sample set and the reference sample set by using a consistency measurement calculation function to obtain a first sample set distance value vector, calculating the distance between each test data and the corresponding first reference sample set and the corresponding second reference sample set by using the consistency measurement calculation function to obtain a first set distance value and a second set distance value of the test data correspondingly, sequencing the first set distance values of the test data in an ascending manner in the first sample set distance value vector to obtain sequence number values, calculating the ratio of the sequence number values to the total number of samples of the first reference sample set to be used as the reliability of the test data, and judging that the early warning data, the label data and the training data of the early warning model updating subsystem have significance differences if the reliability of the test data is lower than a reliability threshold; the confidence level of each test datum is calculated and compared to a confidence threshold.
The deep learning model can adopt a stack type deep learning model, a convolution neural network module and the like.
The stack type deep learning model comprises a convolution layer, a bidirectional long-short term memory layer and a self-attention layer.
The dynamic early warning subsystem interpolates missing patient physiological parameter time sequence data in a minimum time unit by adopting a self-adaptive interpolation method, and the method comprises the following specific steps of:
s1, calculating a median Midparam and a quartile IQRIdparam of sampling frequencies of various physiological parameters of each patient during the period from admission to discharge;
s2, selecting physiological parameters with missing values of a patient with missing data, recording the parameters as param parameters, and selecting the time positions of null values or abnormal values in the collected param parameters as selected positions according to the ascending sequence of the collection time;
s3, performing data interpolation on the selected position in the following priority order, first using the median of the effective values of the preceding (Midparam + IQRidparam) sampling time lengths of the selected position as the data interpolation value of the selected position, second using the median of the effective values of the preceding (Midparam +2 × IQRidparam) sampling time lengths of the selected position as the data interpolation value of the selected position, second using the median of the effective values of the collected param parameters as the data interpolation value of the selected position, and finally using the median of the param parameters recorded in the database as the data interpolation value of the selected position.
The post-traumatic hemorrhagic shock onset judgment subsystem receives the initial probability of the post-traumatic hemorrhagic shock onset of a patient in a prediction time window, and the post-traumatic hemorrhagic shock onset judgment subsystem obtains a classification threshold value D of the stacked deep learning model according to the sensitivity and the specificity of the stacked deep learning model in the dynamic early warning subsystem on the test set data, so that when the classification threshold value is D, the difference between the sensitivity and the specificity of the stacked deep learning model on the test set data is minimum.
The post-traumatic hemorrhagic shock onset judgment subsystem corrects the occurrence probability value of post-traumatic hemorrhagic shock according to the medical environment data where the subsystem is located and doctor treatment experience evaluation values, and the corrected post-traumatic hemorrhagic shock onset probability calculation formula is as follows:
Figure BDA0003596226490000121
where C is the probability threshold.
After monitoring that the medical environment of the place is changed, the early warning model updating subsystem collects the physiological parameters and prognosis information of the patient, or uses the input physiological parameters and prognosis information of the patient, and updates the model parameters by combining an incremental training method on the basis of not changing the structure of the stack-type deep learning model, and transmits the physiological parameters of the patient in the range of the learning window to the dynamic early warning subsystem after updating the parameters of the stack-type deep learning model in real time.
The working principle of the dynamic early warning system for the post-traumatic hemorrhagic shock based on noninvasive parameters is shown in figure 1. The relationship among the observation window, the delay window and the early warning window in the dynamic early warning subsystem is shown in figure 2. In fig. 2, the learning window: and the time range with available data is used for inputting the intelligent dynamic early warning submodule. And (3) prediction window: a period of time to determine whether post-traumatic hemorrhagic shock has occurred. Delay window: the time difference between the prediction window and the learning window. Wherein T0 is the time point when the patient stays in the ICU, T1 is set as the current time point, T1 is more than or equal to T0, x1 is the time length of the learning window, T1-x1 is more than or equal to T0, x2 is the length of the delay window, the time period of the prediction window is [ T1+ x2, T1+ x2+ x3], and x3 is the length of the prediction window.
The physiological parameter perception subsystem is used for monitoring the conventional noninvasive physiological parameters of a patient in real time, and after acquiring the conventional noninvasive physiological parameters of the patient and physiological parameter time sequence data in a learning window of the patient, the dynamic early warning subsystem is used for receiving the conventional noninvasive physiological parameters of the patient and the physiological parameter time sequence data in the learning window of the patient, which are acquired by the physiological parameter perception subsystem, calculating the occurrence probability of post-traumatic hemorrhagic shock of the patient in the range of a prediction window in the future according to the conventional noninvasive physiological parameters and the physiological parameter time sequence data in the learning window of the patient, obtaining the prediction probability value, and sending the prediction probability value to the post-traumatic hemorrhagic shock onset judgment subsystem, and the post-traumatic hemorrhagic shock onset judgment subsystem corrects the prediction probability value according to the medical environment data where the post-traumatic hemorrhagic shock onset judgment subsystem is located and doctor treatment experience evaluation values, if the corrected prediction probability value is larger than 0.5, the dynamic early warning system for the post-traumatic hemorrhagic shock based on the noninvasive parameters judges that the post-traumatic hemorrhagic shock of the patient will occur in the prediction window range in the future, and if the corrected prediction probability value is smaller than or equal to 0.5, the system judges that the post-traumatic hemorrhagic shock of the patient will not occur in the prediction window range in the future. If the early warning model updating subsystem monitors that the medical environment changes, the subsystem updates the weight of the stack type deep learning model in the dynamic early warning subsystem by using an incremental learning method according to the conventional noninvasive physiological parameters of the patient, the death condition of the patient, the hemorrhagic shock occurrence condition after trauma, the hospitalization duration and other clinical prognosis information of the patient, which are acquired by the physiological parameter sensing subsystem, so that the early warning result of the system adapts to the medical environment where the system is located. The stack type deep learning model comprises a convolution layer, a bidirectional long-short term memory layer and a self-attention layer.
The physiological ventilation parameter sensing subsystem has the function of monitoring the physiological parameters of a patient in real time in a noninvasive mode, and can realize the real-time monitoring of parameters such as age, gender, BMI, the status of Mechanical communication, Glasgow Coma Score (gcs), gcs-verall, gcs-mover, gcs-eye, fiO2, PEEP, etco2, tidal volume, ureteooutput, Heart rate, reproduction rate, temperature, SpO2, Non-innovative systemic blood pressure, Non-innovative blood pressure and the like. The physiological ventilation parameter perception subsystem is connected with a local hospital database or manually input data by medical personnel.
The dynamic early warning subsystem combines the routine noninvasive physiological parameters of the patient in the learning window acquired by the physiological ventilation parameter sensing subsystem to calculate the original probability of the patient with the post-traumatic hemorrhagic shock symptoms in the prediction window in the future.
The post-traumatic hemorrhagic shock onset judgment subsystem corrects the original probability output by the dynamic early warning subsystem according to the local medical environment and the doctor treatment experience and feeds the corrected original probability back to the early dynamic early warning system of the medical personnel to judge whether the post-traumatic hemorrhagic shock symptom occurs in a prediction window of the patient in the future.
After an ICU patient is accessed into the dynamic early warning system for the post-traumatic hemorrhagic shock based on noninvasive parameters, the physiological ventilation parameter perception subsystem monitors the physiological parameters of the patient in real time and transmits the parameters of the patient in the learning window range to the dynamic early warning subsystem in real time.
The dynamic early warning subsystem calculates the total urine output of the patient in one hour by taking 1 hour as the minimum time unit and a plurality of effective sampling points in one hour span in the original data, calculates the median of other parameters and then interpolates the missing data value.
The dynamic early warning subsystem interpolates missing physiological parameter time sequence data in a minimum time unit by adopting a self-adaptive interpolation method, and the method specifically comprises the following steps:
s1, calculating a median Midparam and a quartile IQRIdparam of sampling frequencies of various physiological parameters of each patient during the period from admission to discharge;
s2, selecting physiological parameters with missing values of a patient with missing data, recording the parameters as param parameters, and selecting the time positions of null values or abnormal values in the acquired param parameters as selected positions according to the ascending sequence of recording time; the abnormal value is data outside the 95% CI of the parameter in the database;
S3, performing data interpolation on the selected position in the following priority order, first using the median of the effective values of the preceding (Midparam + IQRidparam) sampling time lengths of the selected position as the data interpolation value of the selected position, second using the median of the effective values of the preceding (Midparam +2 × IQRidparam) sampling time lengths of the selected position as the data interpolation value of the selected position, second using the median of the effective values of the param parameter as the data interpolation value of the selected position, and finally using the median of the param parameter recorded in the database as the data interpolation value of the selected position. For example, for a patient with an ID of 32572156, if the heart rate parameter has a missing value at the 8 th hour after admission, the 8 th hour is taken as the selected position, and the median of the heart rate parameter samples of the patient is calculated to be an effective value at 2 hours, and the upper quartile is an effective value at 1 hour, the median of the effective values of the heart rate parameter at the (8-2-1) th to 8 th hours after admission of the patient is calculated as the interpolation value of the missing value at the 8 th hour.
The interpolation data is not regarded as a valid value point.
And the stack type deep learning model in the dynamic early warning subsystem obtains the original probability of hemorrhagic shock attack of the patient after trauma in the prediction window by utilizing the cleaned physiological parameter time sequence data of the patient in the learning window.
The initial probability of the onset of hemorrhagic shock after trauma of a patient in the prediction window is led into a post-trauma hemorrhagic shock onset judgment subsystem, and the post-trauma hemorrhagic shock onset judgment subsystem obtains a classification threshold value D of a stacked deep learning model according to the sensitivity and specificity of the stacked deep learning model in the dynamic early warning subsystem on test set data, so that when the classification threshold value is D, the difference between the sensitivity and the specificity of the stacked deep learning model on the test set data is minimum.
If the corrected probability value of the post-traumatic hemorrhagic shock morbidity probability is larger than 0.5, the early dynamic early warning system judges that the post-traumatic hemorrhagic shock symptom of the patient will appear in the prediction window range in the future, otherwise, judges that the post-traumatic hemorrhagic shock symptom of the patient will not appear.
The silence subsystem, in the classification problem of machine learning, assumes the training data in the sample set as xiE X, i is 1,2, …, n, and the label data in the sample set is yiE Y, i equals 1,2, …, n, then a sample space z is definedi＝(xi,yi) I is 1,2, …, and n is an element in the sample space Z × Y. In general, when a new object x is givenn+1Then, the machine learning model can predict X according to the rule between the sample X and the label Y in the space Z (X X Y) n+1Is given a label of yn+1＝F(x1,x2,…,xn,xn+1,y1,y2,…,yn)。
The mute subsystem aims at evaluating test data xn+1With early warning tag data
Figure BDA0003596226490000151
The silencing system inhibits the post-traumatic hemorrhagic shock attack judgment subsystem from outputting an early warning result for the label with lower reliability. The parameter epsilon (0,1) reflecting the level of significance is set. The method for calculating the reliability between the object and the early warning label by the silence subsystem is a consistency measurement method, namely test data xn+1Similarity to existing samples. In the silent subsystem, the degree of similarity is the minimum euclidean distance of the test data from the set of reference samples.
Taking a binary classification task as an example, a positive reference sample set in a reference sample set is defined first
Figure BDA0003596226490000152
And negative reference sample set
Figure BDA0003596226490000153
And setting a consistency measurement calculation function as FαWhich is used to calculate the minimum euclidean distance of the test data from the set of reference samples. The minimum euclidean distance between the test data and the reference sample set is the minimum euclidean distance between the test data and the elements of the reference sample set.
Calculating to obtain test data xn+1The minimum distance of the set of reference samples, as opposed to their corresponding prediction labels, is labeled as αn+1. The distance from the sample to the set is obtained by a Euclidean distance calculation method. Calculating the metric values of the positive reference sample set and the negative reference sample set in the same way, wherein the metric values are respectively
Figure BDA0003596226490000161
And with
Figure BDA0003596226490000162
The metric value of a reference sample set is the set of distances to the set for each element in the reference sample set. If the test data xn+1If the early warning by the model is positive sample, then the model is aligned to alphai+1Set of metric values alpha in positive samplesposAnd (3) sequencing in a medium ascending order to obtain a sequencing sequence number result, and calculating the ratio of the sequencing sequence number result to the total number of the positive samples, wherein the calculation formula is as follows:
Figure BDA0003596226490000163
and if the p is smaller than the set threshold (such as 0.05), the early warning result of the model is not output temporarily, the prediction probability output result within 15 minutes of the model is collected and sent to the silencing subsystem, and the silencing subsystem returns the maximum prediction probability output result within 15 minutes to the post-traumatic hemorrhagic shock onset judgment subsystem.
For the condition that the clinical medical environment changes, an ICU patient is accessed with an interpretable post-traumatic hemorrhagic shock dynamic early warning system based on noninvasive parameters and based on noninvasive parameters. The physiological ventilation parameter sensing subsystem monitors the physiological parameters of the patient in real time.
The dynamic early warning subsystem calculates the total urine output of the patient in one hour by the urine output and calculates the median of other parameters by taking the hour as the minimum time unit and a plurality of effective sampling points in one hour span in the original data, and then interpolates the missing data value. The early warning model updating subsystem collects the physiological parameters and prognosis information of the patient, or uses the input physiological parameters and prognosis information, combines an increment training method, updates the model parameters on the basis of not changing the structure of the stacked deep learning model, and transmits the physiological parameters of the patient in the range of a learning window to the dynamic early warning subsystem after updating the parameters of the stacked deep learning model in real time.
The dynamic early warning subsystem transmits the cleaned patient data in the learning window to the trained machine learning model, and the original probability of hemorrhagic shock attack of the patient after trauma in the prediction window can be obtained through the model processing.
And (3) leading the initial probability of the hemorrhagic shock after the trauma in the prediction window into a post-trauma hemorrhagic shock onset judgment subsystem, and combining the early warning performance of the machine learning model in the dynamic early warning subsystem on the test set data and the doctor experience to obtain a classification threshold value D, so that when the classification threshold value is D, the sensitivity or specificity of the model on the test set data is an expert suggestion value.
Therefore, the dynamic early warning depth system for the post-traumatic hemorrhagic shock based on the noninvasive parameters disclosed by the invention only uses common noninvasive parameters without laboratory data, enlarges the application range of the system, makes the use of the system possible in remote areas, emergent public health events and first-line situations of battlefields, eliminates the damage to patients caused by frequent acquisition of laboratory parameters, and reduces the use cost of the system. Compared with a traditional linear addition regression model, the automatic and streamlined intelligent dynamic early warning subsystem has higher computational complexity, is more suitable for common nonlinear problems in practical problems, and can provide better post-traumatic hemorrhagic shock onset early warning capability. The pre-warning interval reserved for clinical intervention can provide sufficient time for a physician to design a patient treatment regimen. The analysis result of the system is consistent with the clinical research result, the model performance is excellent, the system can automatically update the model weight according to the environmental change so as to adapt to different clinical environments, and meanwhile, the false probability of predicting the hemorrhagic shock after trauma is effectively reduced.
The above description is only an example of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art to which the present application pertains. Any modification, equivalent replacement, improvement or the like made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.

Claims (10)

1. A dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters, comprising: a physiological parameter sensing subsystem, an early warning model updating subsystem, a dynamic early warning subsystem, a post-traumatic hemorrhagic shock onset judgment subsystem and a silencing subsystem;
the physiological parameter sensing subsystem is used for monitoring and acquiring physiological parameter data information of a patient in real time, preprocessing the acquired physiological parameter data of the patient and then sending the preprocessed data to the dynamic early warning subsystem;
the physiological parameter perception subsystem is connected with the dynamic early warning subsystem, the dynamic early warning subsystem is respectively connected with the early warning model updating subsystem and the post-traumatic hemorrhagic shock onset judgment subsystem, and the silence subsystem is respectively connected with the post-traumatic hemorrhagic shock onset judgment subsystem, the dynamic early warning subsystem and the early warning model updating subsystem.
2. The dynamic early warning system for post-traumatic hemorrhagic shock based on non-invasive parameters as claimed in claim 1, wherein the dynamic early warning subsystem is used for receiving and cleaning the conventional non-invasive physiological parameters of the patient and the time series data of the physiological parameters in the learning window of the patient, which are acquired by the physiological parameter sensing subsystem, calculating the occurrence probability of post-traumatic hemorrhagic shock of the patient in the future prediction time window according to the conventional non-invasive physiological parameters and the time series data of the physiological parameters in the learning window of the patient, which are acquired by the physiological parameter sensing subsystem, obtaining the prediction probability value, and sending the prediction probability value to the post-traumatic hemorrhagic shock occurrence judging subsystem, which corrects the prediction probability value according to the medical environment data and doctor treatment experience evaluation value of the place where the prediction probability value is located, if the corrected prediction probability value is larger than a first prediction probability threshold value, the dynamic early warning system for the post-traumatic hemorrhagic shock based on the noninvasive parameters judges that the post-traumatic hemorrhagic shock of the patient will occur in the future prediction time window range, and if the corrected prediction probability value is smaller than or equal to a second prediction probability threshold value, the system judges that the post-traumatic hemorrhagic shock of the patient will not occur in the future prediction time window range.
3. The dynamic early warning system for post-traumatic hemorrhagic shock based on non-invasive parameters as claimed in claim 1, wherein the physiological parameter data information of the patient comprises the conventional non-invasive physiological parameters of the patient and the physiological parameter time sequence data in the learning window of the patient.
4. The dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters of claim 1,
the method comprises the steps of utilizing discrete sampling values of acquired physiological parameters of patients to form sampling vectors, calculating cross-correlation matrixes of the sampling vectors, utilizing a feature extraction method to process the cross-correlation matrixes to obtain characteristic value vectors of the sampling vectors, and utilizing the characteristic value vectors to weight the sampling vectors to obtain smooth values of the physiological parameters of the patients to serve as preprocessed data.
5. The non-invasive parameter based dynamic early warning system of post-traumatic hemorrhagic shock in accordance with claim 4,
saidPreprocessing acquired patient physiological parameter data, including representing a sampling vector composed of discrete sampling values of patient physiological parameters acquired over a period of time as [ x 1,x2,…,xN]And N is the number of discrete sampling values of the physiological parameters of the patient acquired within a period of time, a mutual matrix C of the sampling vector is obtained by calculation, and the characteristic value decomposition is carried out on the mutual matrix C to obtain:
C＝VDVH，
and C is a characteristic vector matrix, D is a characteristic value matrix, the diagonal elements of the matrix D are normalized and used as weight vectors, and the sampling vectors are subjected to weighted summation to obtain a smooth value of the physiological parameters of the patient as preprocessed data.
6. The dynamic early warning system for post-traumatic hemorrhagic shock with non-invasive parameters according to claim 1, wherein the early warning model updating subsystem monitors the medical environment of the location of the system, and if the early warning model updating subsystem monitors the change of the medical environment of the location, the subsystem updates the weight of the deep learning model in the dynamic early warning subsystem by using an incremental learning method according to the conventional non-invasive physiological parameters of the patient and the prognosis information including the death condition, the occurrence condition of post-traumatic hemorrhagic shock and the hospitalization duration of the patient, which are acquired by the physiological parameter sensing subsystem, so that the early warning result of the system is adapted to the medical environment of the location of the system.
7. The dynamic early warning system for post-traumatic hemorrhagic shock based on non-invasive parameters as recited in claim 1, wherein the silence subsystem discriminates the differences among the test data, the training data and the label data of the dynamic early warning for post-traumatic hemorrhagic shock of the intelligent dynamic early warning subsystem, if the three types of data are significantly different, the silence subsystem controls the early warning result of the dynamic early warning subsystem not to be output, collects the early warning result data of the dynamic early warning subsystem within a period of time, extracts the early warning result data with the maximum prediction probability within the period of time, and outputs the early warning result data to the post-traumatic hemorrhagic shock onset judgment subsystem.
8. The dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters of claim 7,
the silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
the silence subsystem predicts to obtain a label data predicted value by using the training data and a deep learning model, calculates the credibility among the test data, the training data and the label data predicted value by using a credibility assessment method, and judges that the early warning data, the label data and the training data of the early warning model updating subsystem have significant difference when the credibility is lower than a credibility threshold value.
9. The dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters of claim 7,
the silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
forming a sample set by using training data and a label data predicted value, dividing the sample set into a first reference sample set and a second reference sample set corresponding to the test data according to the label category of each test data, wherein the first reference sample set is a set formed by samples corresponding to the label category of the test data, the second reference sample set is a set formed by other samples except the first reference sample set in the sample set, calculating the distance between each sample in the first reference sample set and the reference sample set by using a consistency measurement calculation function to obtain a first sample set distance value vector, calculating the distance between each test data and the corresponding first reference sample set and the corresponding second reference sample set by using the consistency measurement calculation function to obtain a first set distance value and a second set distance value of the test data correspondingly, sequencing the first set distance values of the test data in an ascending manner in the first sample set distance value vector to obtain sequence number values, calculating the ratio of the sequence number values to the total number of samples of the first reference sample set to be used as the reliability of the test data, and judging that the early warning data, the label data and the training data of the early warning model updating subsystem have significance differences if the reliability of the test data is lower than a reliability threshold; the confidence level of each test datum is calculated and compared to a confidence threshold.
10. The dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters of claim 7,
the silence subsystem distinguishes differences of test data, label data and training data of the early warning model updating subsystem, and comprises the following steps:
regarding data of an early warning model updating subsystem as a stable random process, respectively establishing corresponding autoregressive-sliding average models, namely ARMA models, aiming at test data, label data and training data, respectively obtaining a first ARMA model, a second ARMA model and a third ARMA model, calculating the cross-correlation matrix of the coefficients of the three ARMA models, calculating the cross-correlation matrix to obtain a maximum characteristic value, and distinguishing the difference of the test data, the label data and the training data by using the maximum characteristic value;
and when the maximum characteristic value is larger than the difference judgment threshold value, judging that the significance difference occurs among the test data, the label data and the training data of the early warning model updating subsystem.
CN202210392852.4A 2022-04-14 2022-04-14 Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters Pending CN114758786A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210392852.4A CN114758786A (en) 2022-04-14 2022-04-14 Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210392852.4A CN114758786A (en) 2022-04-14 2022-04-14 Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters

Publications (1)

Publication Number Publication Date
CN114758786A true CN114758786A (en) 2022-07-15

Family

ID=82331306

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210392852.4A Pending CN114758786A (en) 2022-04-14 2022-04-14 Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters

Country Status (1)

Country Link
CN (1) CN114758786A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116258276A (en) * 2023-05-06 2023-06-13 中国能源建设集团湖南省电力设计院有限公司 Carbon emission total amount prediction method based on saturated carbon emission density
CN116978570A (en) * 2023-09-25 2023-10-31 之江实验室 Online real-time patient critical degree assessment and vital sign parameter prediction system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150157275A1 (en) * 2013-12-10 2015-06-11 General Electric Company Respiratory stress detection
CN104755019A (en) * 2012-10-30 2015-07-01 德克斯康公司 Systems and methods for providing sensitive and specific alarms
CN111063162A (en) * 2019-12-05 2020-04-24 恒大新能源汽车科技(广东)有限公司 Silent alarm method and device, computer equipment and storage medium
CN113421654A (en) * 2021-07-12 2021-09-21 军事科学院系统工程研究院卫勤保障技术研究所 Dynamic early warning and deep learning system for post-traumatic hemorrhagic shock

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104755019A (en) * 2012-10-30 2015-07-01 德克斯康公司 Systems and methods for providing sensitive and specific alarms
US20150157275A1 (en) * 2013-12-10 2015-06-11 General Electric Company Respiratory stress detection
CN111063162A (en) * 2019-12-05 2020-04-24 恒大新能源汽车科技(广东)有限公司 Silent alarm method and device, computer equipment and storage medium
CN113421654A (en) * 2021-07-12 2021-09-21 军事科学院系统工程研究院卫勤保障技术研究所 Dynamic early warning and deep learning system for post-traumatic hemorrhagic shock

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116258276A (en) * 2023-05-06 2023-06-13 中国能源建设集团湖南省电力设计院有限公司 Carbon emission total amount prediction method based on saturated carbon emission density
CN116978570A (en) * 2023-09-25 2023-10-31 之江实验室 Online real-time patient critical degree assessment and vital sign parameter prediction system
CN116978570B (en) * 2023-09-25 2024-02-06 之江实验室 Online real-time patient critical degree assessment and vital sign parameter prediction system

Similar Documents

Publication Publication Date Title
US11217349B2 (en) System and method for processing human related data including physiological signals to make context aware decisions with distributed machine learning at edge and cloud
JP5841196B2 (en) Residue-based management of human health
CN112365978B (en) Method and device for establishing early risk assessment model of tachycardia event
CN110289096B (en) ICU (intensive Care Unit) intra-hospital mortality prediction method based on deep learning
US20080281170A1 (en) Method for Detecting Critical Trends in Multi-Parameter Patient Monitoring and Clinical Data Using Clustering
CN111261282A (en) Sepsis early prediction method based on machine learning
Silva et al. Mortality assessment in intensive care units via adverse events using artificial neural networks
CN114758786A (en) Dynamic early warning system for post-traumatic hemorrhagic shock based on noninvasive parameters
CN113421654A (en) Dynamic early warning and deep learning system for post-traumatic hemorrhagic shock
CN111081379B (en) Disease probability decision method and system thereof
US11580432B2 (en) System monitor and method of system monitoring to predict a future state of a system
CN116098595B (en) System and method for monitoring and preventing sudden cardiac death and sudden cerebral death
Shamsollahi et al. Using combined descriptive and predictive methods of data mining for coronary artery disease prediction: a case study approach
CN112768074A (en) Artificial intelligence-based serious disease risk prediction method and system
Kannathal et al. Cardiac health diagnosis using data fusion of cardiovascular and haemodynamic signals
CN118486466A (en) Intelligent robot management system and method applied to health monitoring
CN117476217B (en) Chronic heart disease state of illness trend prediction system
CN118173253A (en) System and method for analyzing and managing based on patient data
CN118038548A (en) Abnormal behavior detection method, device, electronic equipment and storage medium
US20060069535A1 (en) System and method for converting data, and system and method for providing warning signals
Rajmohan et al. G-Sep: A deep learning algorithm for detection of long-term sepsis using bidirectional gated recurrent unit
Pathinarupothi Clinically aware data summarization at the edge for Internet of Medical Things
CN113017572A (en) Severe warning method and device, electronic equipment and storage medium
Funkner et al. Motif identification in vital signs of chronic patients
CN113393939B (en) Method and system for predicting number of hospitalization days of intensive care unit patient

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