CN112245728B - Respirator false positive alarm signal identification method and system based on integrated tree - Google Patents
Respirator false positive alarm signal identification method and system based on integrated tree Download PDFInfo
- Publication number
- CN112245728B CN112245728B CN202010492039.5A CN202010492039A CN112245728B CN 112245728 B CN112245728 B CN 112245728B CN 202010492039 A CN202010492039 A CN 202010492039A CN 112245728 B CN112245728 B CN 112245728B
- Authority
- CN
- China
- Prior art keywords
- data
- alarm signal
- data set
- characteristic
- false positive
- 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.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61M—DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
- A61M16/00—Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes
- A61M16/0003—Accessories therefor, e.g. sensors, vibrators, negative pressure
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61M—DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
- A61M16/00—Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes
- A61M16/0051—Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes with alarm devices
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses a method and a system for identifying false positive alarm signals of a breathing machine based on an ensemble tree, which comprises the following steps: s1, data collection: collecting monitoring data of patients from hospital ventilators and monitors; s2, data preprocessing: processing missing values, abnormal values and standardization in the data set, and generating an identification rule of a false positive alarm signal; s3, feature extraction: sorting the features by using a random forest, and selecting the features with good recognition capability; s4, false positive alarm signal identification: and establishing a false positive alarm signal identification method of the breathing machine and the monitor. Experimental results show that the method has excellent false positive alarm signal identification performance and steady identification effect.
Description
Technical Field
The invention relates to a method and a system for identifying false positive alarm signals of a hospital respirator-monitor, in particular to a method and a system for identifying false positive alarm signals of a respirator based on an integration tree.
Background
A ventilator is widely used in modern clinical medicine as a kind of medical equipment for emergency treatment and life support, for example, for treating patients suffering from respiratory failure due to various causes, anesthesia and breathing management during major surgery, respiratory support therapy, and emergency resuscitation. Since the ventilator is mainly used for patients with high risk of illness, it is usually used with a monitor. When the breathing of the patient is abnormal or the equipment fails, the breathing machine-monitor sends out an alarm signal, and medical personnel check the state of the corresponding patient according to the alarm signal and check the running condition of the equipment. The effective alarm signal can help medical personnel to correctly identify and timely process alarm of the breathing machine, and normal work of the breathing machine and safety of patients are guaranteed.
However, during use of the ventilator-monitor, there are often situations where most of the alarm signals are false positive alarm signals. According to statistics, medical staff resources in many developing countries are in short supply, and particularly, during epidemic situations or emergencies, the medical staff is subjected to greater working pressure due to false positive alarm of the breathing machine. Frequent false positive alarms can cause the medical personnel to generate alarm fatigue to the alarm signal, and affect the response speed of the medical personnel to the alarm signal. When a plurality of breathing machine-monitors alarm simultaneously, the real dangerous patient is not checked in time and the best treatment opportunity is missed probably because of the occurrence of false positive alarm condition.
False positive alarm signals are identified based on real human body data monitored by a breathing machine and a monitor, the working pressure of medical personnel can be reduced, and the alertness of the medical personnel to the alarm signals can be improved. In addition, the accurate identification of false positive alarm signals can also strive for more timely medical aid for truly risky patients, relieve the alarm of a breathing machine, and enable the patients to be treated safely and effectively. At present, research on methods for analyzing and processing reasons of ventilator alarms is available, but in general, research on recognition of ventilator-monitor false positive alarm signals by means of a machine learning method has not been developed, and most research contents are discussed around problems caused by ventilator false positive alarms, so that an effective ventilator-monitor false positive alarm signal recognition method and system are not available at present. Therefore, it is very important to develop the identification method of false positive alarm signal for hospital respirator-monitor.
Therefore, there is an urgent need for a new method for identifying false positive alarm signals of a ventilator-monitor, which satisfies the following technical requirements: 1) The interpretation capability of the identification result can be effectively improved, and the sign indexes playing a key role in false positive alarm signals can be found; 2) The method and the system have good classification and identification effects and performance, and are a method and a system for effectively identifying false positive alarm signals of a breathing machine-monitor.
Disclosure of Invention
The invention solves the problems: the method and the system for identifying the breathing machine false positive alarm signal based on the integration tree are provided to solve the problems that the subjective judgment or the identification effect is poor in the current breathing machine-monitor false positive alarm signal identification, and the problems that medical staff is high in pressure and misses the best treatment opportunity due to false positive alarm are solved.
The technical scheme adopted by the invention is as follows:
the invention provides a ventilator false positive alarm signal identification method based on an integrated tree, which comprises the following steps:
step 1) data collection: collecting sample data of a plurality of real breathing machines and a plurality of monitors of a patient from a hospital, and combining the sample data of each monitor and the sample data of each breathing machine to be used as an original data set, wherein the original data set comprises a plurality of characteristic data and corresponding alarm signals;
step 2) data preprocessing: carrying out missing value processing, abnormal value processing and data standardization processing on the characteristic data of the original data set in the step 1), and carrying out identification processing on the alarm signals of the original data set so as to obtain a preprocessed data set, wherein the identification processing is to respectively identify different types of label information for the alarm signals according to a set rule, and the different types of label information are true positive alarm signals or false positive alarm signals;
step 3), feature selection: performing feature screening on the feature data of the preprocessed data set in the step 2) by using a random forest, reserving screened features, and forming a training data set by the screened feature data in the preprocessed data set and corresponding early warning signal label information;
step 4), false positive alarm signal identification: training the parameters of the gradient boosting decision tree classifier by using the training data set in the step 3) to obtain the trained alarm signal label information category identifier, and identifying the category of the corresponding alarm signal label information to be a true positive alarm signal or a false positive alarm signal according to newly input screened feature data and the corresponding early warning signal by the identifier.
Further, in the step 1:
the sample frequency of the real breathing machine-monitor monitoring data collected from the hospital is in seconds, and the data is collected three times per second;
the plurality of features comprises 16 individual feature features, the 16 individual feature features are respectively minute expiratory volume, average pressure, oxygen input port pressure, inspiratory oxygen concentration, respiratory non-positive pressure, spontaneous respiratory frequency, inspiratory tidal volume, expiratory tidal volume, peak pressure, invasive blood pressure mean value, invasive blood pressure high value, invasive blood pressure low value, central venous pressure, blood oxygen concentration and heart rate;
the specific implementation of combining each monitor data sample with each ventilator data sample is to combine each monitor data sample with each ventilator data sample into one sample by using a matching method with a ventilator as a main timestamp.
Further, in the step 2:
specifically, the missing value processing is realized by screening the missing value of each feature data in the original data set by adopting a feature averaging method, and filling the missing value of each feature data in the original data set into the mean value of each feature data, wherein the missing value of the jth sample of the ith feature data in the original data set is processed by the missing value processing to fill the missing value into a numerical value x' missing(i，j) ，
Wherein x is i1 ，x i2 ，...，x in Respectively representing the 1,2, \ 8230under the ith characteristic in the original data set, wherein n samples represent the number of the samples;
the abnormal value processing is specifically realized by adopting a triple standard deviation method, firstly screening an abnormal value of which the difference between each characteristic data in the original data set and the mean value of the characteristic data is more than triple of the standard deviation of the characteristic data, and adjusting the abnormal value to be the sum of the mean value of the characteristic data and triple of the standard deviation of the characteristic data; then, abnormal values in each feature data in the original data set, wherein the difference between the abnormal values and the mean value of the feature data is smaller than three times of the abnormal values, and the abnormal values are adjusted to be the mean value of the feature data and three times of the abnormal values of the feature dataWherein the abnormal value of the ith characteristic data of the jth sample in the original data set is processed by the abnormal value to obtain an adjusted value x' outlier(i，j) ，
Wherein x is ij Represents the value of the jth sample under the ith characteristic data in the original data set, mu i Representing the mean, σ, of the ith characteristic data in the original data set i Representing a standard deviation of ith characteristic data in the original data set;
the normalization processing is realized by replacing the numerical value of each characteristic data in the original data set by the z-score of each characteristic data by using a z-score method, wherein the numerical value of the ith characteristic data of the jth sample in the original data set is replaced by the numerical value x 'after the abnormal value processing' norm(i，j) ，
Wherein x is ij A value, μ, representing the j sample under the i characteristic data in the original data set i Means, σ, representing the ith characteristic data in the raw data set i Representing a standard deviation of ith characteristic data in the original data set;
the specific implementation of the identification processing is that the established identification rule includes: when the respirator and the monitor alarm at the same time, the label information of the alarm signal is a true positive alarm signal; when continuous uninterrupted alarm occurs on the respirator or the monitor and the alarm duration times exceed 3 times, label information of the alarm signal is a true positive alarm signal; when the sign characteristic data exceeds the threshold range of the sign characteristic data set by the respirator, the label information of the alarm signal is a true positive alarm signal; the threshold range of the sign characteristic data comprises: the minute expiration amount is 0.5-180L/min, and the allowable error range is +/-3%; average pressure is-2-12 kPa, and the allowable error range is +/-0.1 kPa; the concentration of the inhaled oxygen is 21-100%, and the allowable error range is +/-3%; the non-positive pressure of breathing is-12 to 12kPa, and the allowable error range is +/-0.05 kPa; the spontaneous respiratory frequency and the respiratory frequency are 1 to 150 times/minute, and the allowable error range is +/-3 percent; the inspiration/expiration tidal volume is-10 to 10L, and the allowable error range is +/-3 percent.
Further, in the step 3, the feature screening is performed by using a random forest, and the specific implementation of retaining the screened features is as follows:
calculating the difference information Gain (L, F) between the information Entropy Encopy (L) of the alarm signal label information and the information Entropy Encopy (L, F) of the alarm signal label under the characteristic F for each characteristic F in the preprocessed data set,
Gain(L，F)＝Entropy(L)-Entropy(L，F)，
if Gain (L, F) > theta, keeping the characteristic F as the screened characteristic, and if Gain (L, F) < theta, deleting the characteristic F, and enabling theta to be a set threshold;
wherein L represents alarm tag information of the preprocessed data set, p i The probability that label information of the ith category of the alarm signal appears in the preprocessed data set is represented;
wherein L represents alarm signal label information of the preprocessed data set, v represents the number of values of the preprocessed data set under the characteristic F, and L j And representing the number of the jth value of the preprocessed data set under the characteristic F.
The screened characteristics comprise peak pressure, heart rate, respiratory rate, spontaneous respiratory rate, expiratory tidal volume, inspiratory tidal volume, minute respiratory volume, average pressure and respiratory unpressurized pressure, and preferably comprise peak pressure, heart rate, respiratory rate and spontaneous respiratory rate.
Further, in the step 4, the setting range of the number of decision trees of the gradient boosting decision tree classifier is [50,150], the step size is 10, the setting range of the tree height is [3,10], the step size is 1, the setting range of the number of leaf nodes is [5,15], and the step size is 1.
The specific implementation of the step 4 is as follows:
41 Using the filtered features obtained in the step 3) as an input feature vector space, and if the early warning signal label information output by the gradient lifting decision tree classifier of the (m-1) th round is F m-1 (x) Then the loss function L (y, F) m-1 (x))＝y-F m-1 (x) Wherein x is a sample, and y is real early warning signal label information of the sample;
42 Through L (y, F) m-1 (x) Pair F) m-1 (x) Derivation of the deviation
Obtaining the optimization direction and the learning rate gamma of the gradient lifting decision tree classifier of the mth round m-1 Controlling the contribution degree of the early warning signal label information output by the gradient lifting decision tree classifier in the m-1 th round, wherein the early warning signal label information output by the gradient lifting decision tree classifier in the m-th round is
43 ) iteratively repeating the steps 41) to 42) until the gradient boosting decision tree classifier of the mth round and the mth-1 round identifies the output early warning signal label information F m (x) And F m-1 (x) When the difference is smaller than a set threshold value, iteration is repeatedly stopped to obtain the trained alarm signal label information category identifier;
44 The identifier identifies the type of the tag information of the corresponding early warning signal as a true positive alarm signal or a false positive alarm signal according to the newly input screened feature data and the corresponding early warning signal.
The invention also provides a ventilator false positive alarm signal identification system based on the integrated tree, which comprises:
the system comprises a data acquisition module, a data preprocessing module and an alarm signal label information category identifier;
the data acquisition module acquires a monitoring data set of the breathing machine-monitor input by a user and sends the monitoring data set to the data preprocessing module, wherein the monitoring data set comprises a plurality of characteristic data and alarm signals, and the plurality of characteristics comprise peak pressure, heart rate, respiratory rate, spontaneous respiratory rate, expiratory tidal volume, inspiratory tidal volume, minute respiratory volume, average pressure and respiratory non-positive pressure, preferably peak pressure, heart rate, respiratory rate and spontaneous respiratory rate;
the data preprocessing module receives the monitoring data set sent by the data acquisition module, performs missing value processing, abnormal value processing and data standardization processing on the characteristic data in the detection data set, and sends the preprocessed monitoring data set to the alarm signal label information category identifier;
the alarm signal label information category recognizer is a trained gradient lifting decision tree classifier, receives the preprocessed monitoring data set sent by the data preprocessing module, and recognizes and outputs whether the category of the alarm signal label information is a true positive alarm signal or a false positive alarm signal.
Compared with the prior art, the invention has the advantages that:
(1) The invention discloses a method for recognizing false positive alarm signals of a hospital respirator and a monitor based on a gradient lifting decision tree method, which comprises the steps of firstly collecting real-time monitoring data of a patient from the hospital respirator and the monitor, then carrying out preprocessing operation on the data and identifying the false positive alarm signals, ensuring the integrity and the effectiveness of the data, then extracting effective characteristics of the data by using a random forest method, and finally recognizing and verifying the false positive alarm signals of the hospital respirator and the monitor by using the gradient lifting decision tree method;
(2) The false positive alarm signal identification method provided by the invention has good interpretation capability and provides a basis for finding key classification indexes;
(3) The method has excellent classification and identification performances, and has the best identification result in the aspects of accuracy, AUC, F1-SCORE and the like compared with other machine learning methods.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate an exemplary embodiment of the invention and, together with the description, serve to explain the invention and to make the aforementioned advantages of the invention more apparent. Wherein the content of the first and second substances,
FIG. 1 is a flow chart of a false positive alarm signal identification method of the present invention;
FIG. 2 is a data acquisition device, a ventilator and a monitor;
FIG. 3 is a comparison of the F1-Score indices for the four methods GBDT, SVM, NB, LR, where (a) the training set test set ratio 80; (b) training set test set ratio 70; (c) Training set test set ratio 60, gbdt represents a gradient boosting decision tree, SVM represents a support vector machine, NB represents a naive bayes classifier, and LR represents a Logistic regression.
Detailed Description
In order to make the objects, technical solutions, implementation steps and advantages of the present invention more apparent, the following description is further detailed with reference to the accompanying drawings and implementation examples. It should be noted that the specific implementation examples of the present disclosure are only used for explaining the present invention, and are not used for limiting the present invention, and the technical solutions formed by combining the respective parts in the implementation examples are within the protection scope of the present invention.
The invention mainly aims at the problem that the working pressure of medical staff is increased rapidly because a breathing machine and a monitor matched with a serious patient frequently generate false positive alarm signals in a hospital environment, and provides an integrated tree-based breathing machine false positive alarm signal identification method for identifying the false positive alarm signals to relieve the working pressure of the medical staff, which comprises the following steps: s1, data collection: collecting monitoring data of a patient from a hospital respirator and a monitor as a raw data set; s2, data preprocessing: carrying out missing value, abnormal value and standardization processing on an original data set, carrying out identification processing on alarm signals of the original data set, and respectively identifying different types of label information, namely true positive alarm signals or false positive alarm signals, for the alarm signals according to a set rule; s3, feature extraction: performing feature screening by using a random forest, and reserving screened features to further construct a training data set; s4, false positive alarm signal identification: training the parameters of the gradient boosting decision tree classifier by using a training set, establishing an alarm signal label information category recognizer, and recognizing and outputting the category of the corresponding alarm signal label information as a true positive alarm signal or a false positive alarm signal according to newly input screened feature data and the corresponding alarm signal. The invention firstly collects the real physical sign data of the patient from the breathing machine and the monitor of the hospital, carries out feature selection based on random forest after preprocessing the data, then realizes the identification work of false positive alarm signals of the breathing machine and the monitor by adopting a gradient lifting decision tree method, and carries out experimental verification. Experimental results show that the method has excellent false positive alarm signal identification performance and steady identification effect.
The process of the method mainly comprises the following steps:
step 1) data collection: collecting sample data of a plurality of real breathing machines and a plurality of monitors of a patient from a hospital, and combining the data sample data of each monitor and the data of each breathing machine as an original data set, wherein the original data set comprises a plurality of characteristic data and corresponding alarm signals;
step 2) data preprocessing: carrying out missing value processing, abnormal value processing and data standardization processing on the characteristic data of the original data set in the step 1), and carrying out identification processing on alarm signals of the original data set so as to obtain a preprocessed data set, wherein the identification processing is to respectively identify different types of label information for the alarm signals according to a set rule, and the different types of label information are true positive alarm signals or false positive alarm signals;
step 3) feature selection: performing feature screening on the feature data of the preprocessed data set in the step 2) by using a random forest, reserving screened features, and forming a training data set by the screened feature data in the preprocessed data set and corresponding early warning signal label information;
further, in step 1):
according to the actual running states of the breathing machine and the monitor, the real state of the patient can be more accurately displayed by considering the data with higher resolution, the physical sign monitoring data of the patient are collected from the breathing machine and the monitor in a hospital, the frequency of the collected samples is in units of seconds, and the samples are collected for three times per second.
The collected information includes 16 individual characteristics, which are respectively minute expiratory volume, average pressure, oxygen input port pressure, inspiratory oxygen concentration, respiratory non-positive pressure, spontaneous respiratory frequency, inspiratory tidal volume, expiratory tidal volume, peak pressure, invasive blood pressure average value, invasive blood pressure high value, invasive blood pressure low value, central venous pressure, blood oxygen concentration and heart rate, and a label information for identifying an alarm signal.
The corresponding set of sign features is denoted as X = { X = } 1 ，x 2 ，...，x 16 Label information of 1 alarm signal is represented as Y = {0,1}, where 0 indicates that no alarm occurs and 1 indicates that an alarm occurs.
Because the data is from two devices, namely a breathing machine and a detector, the problem that the sample time stamps of the two data sources are not uniform exists. And (3) taking a breathing machine as a main time stamp, and combining each monitor data sample and each breathing machine data sample into one sample by adopting a matching method. The concrete implementation is as follows: firstly, based on the acquisition time, respectively sequencing samples of a breathing machine and a monitor, correspondingly combining the monitor samples with the same time to the breathing machine samples, and deleting the samples corresponding to the breathing machine at the time when the breathing machine has the time sample and the monitor does not have the time sample, or vice versa.
Further, in step 2):
a missing value processing step, wherein due to unavoidable factors such as machine faults, the situation that completely random missing data occurs in data set needs to be processed, a feature average method is adopted to fill the missing values, and a calculation formula is as follows:
wherein x is i1 ，x i2 ，...，x in Respectively represent the 1 st, 2 nd, n th samples under the ith characteristic in the original data set, wherein j is an index corresponding to a missing value, and n represents the number of the samples.
In the abnormal value processing step, due to recording errors, machine abnormality and other reasons, obviously different samples appear in data, in order to avoid negative influence of abnormal values in the data on the identification of false positive alarm signals, the abnormal values are processed by using a variance of three times of standard deviation, namely, the data exceeding three times of standard deviation is adjusted to three times of standard deviation, so that the problem of the abnormal values is solved, and a specific calculation formula is as follows:
wherein x is ij The j sample under the ith characteristic data in the original data set is represented as an abnormal value mu i Means, σ, representing the ith characteristic data in the raw data set i Representing the standard deviation of the ith feature data in the raw data set.
A standardization processing step, namely, in order to eliminate dimension problems among different physical sign characteristics and avoid the phenomenon that a classification result excessively deviates to a certain dimension larger characteristic, a z-score method is adopted to carry out standardization processing on the data characteristics, and a specific processing formula is as follows:
wherein x is ij Represents the j sample under the i characteristic data in the original data set, mu i Representing the mean, σ, of the ith characteristic data in the original data set i Representing the standard deviation of the ith feature data in the raw data set.
And a false positive alarm signal marking step, wherein the marking of the false positive alarm signal needs professional background knowledge of experts in medical related fields, and after full comprehensive discussion, a rule of true positive alarm signals is provided, and the true positive alarm signals are marked from all the alarm signals, so that the rest part is automatically divided into the false positive alarm signals. The identification rule includes three aspects: (1) When the breathing machine and the monitor alarm at the same time, the alarm is a true positive alarm signal; (2) When continuous uninterrupted alarm occurs in the breathing machine or the monitor, the alarm is a true positive alarm signal; (3) And when the physical sign data of the patient exceeds a set threshold value of the breathing machine, the alarm signal is a true positive alarm signal. The physical characteristics comprise minute breathing rate, flow range: (0.5-180) L/min, allowable error range: ± 3%, average pressure, pressure range: (-2 to 12) kPa, tolerance range: 0.1kPa, inhaled oxygen concentration, range: 21% -100%, allowable error range: ± 3%, respiratory non-positive pressure, range: 12kPa, allowable error range: ± 0.05kPa, spontaneous and respiratory rate, frequency range: (1-150) times/min, allowable error range: ± 3%, inspiratory/expiratory tidal volume, tidal volume: 10L, allowable error range: 3 percent.
Based on the rules, the identification work of the true positive and false positive alarm signals is completed, and a scientific data basis is provided for a subsequent supervised learning classification method.
Further, in step 3):
the collected original data contains 16 individual characteristic data of the patient, wherein the characteristic with weaker performance for identifying false positive alarm signals is not lacked, so that the part of the content adopts a characteristic selection method of Random Forest (Random Forest) to select the characteristics in the original data set, the number of the characteristics is reduced on the basis of not reducing the identification precision, and the calculation efficiency and the identification accuracy are improved.
The main idea of feature selection of the random forest is based on the idea of information gain in a decision tree. The random forest generates a plurality of decision trees with differences by performing double disturbance on the characteristics and samples of data, the idea of selecting the characteristics of the random forest is the same as or different from that of the decision trees, but the random forest has the advantage of integrating a plurality of decision trees, and can be expressed as follows in a model mode:
calculating the difference information Gain (L, F) between the information Entropy Encopy (L) of the alarm signal label information and the information Entropy Encopy (L, F) of the alarm signal label under the characteristic F for each characteristic F in the preprocessed data set,
Gain(L，F)＝Entropy(L)-Entropy(L，F)，
if Gain (L, F) > theta, keeping the characteristic F as the screened characteristic, and if Gain (L, F) < theta, deleting the characteristic F, and enabling theta to be a set threshold;
information entropy of the alarm signal label information
Wherein L represents alarm tag information of the preprocessed data set, p i Probability, p, of occurrence of label information representing the ith category of alarm signal in said preprocessed data set i The method is obtained by calculating the number proportion of the false positive alarm signal samples in the preprocessed data set samples;
information entropy of alarm signal label under characteristic F
Wherein L represents alarm signal tag information of the preprocessed data set, L j The number of the characteristic F of the preprocessed data set which takes a certain numerical value is represented, v represents the number of different values of the preprocessed data set under the characteristic F, and j represents the index of the jth value of the preprocessed data set under the characteristic F.
The random forest method is realized by the following steps:
step 1) inputting training data of an initial feature set, calculating and outputting an information Entropy (L) of an alarm signal, and providing a basis for calculating information gain in step 3);
step 2) inputting training data of an initial feature set, calculating and outputting alarm signal conditional Entropy (L, F) of a feature F under each tree, and providing basis for calculating information gain in the step 3);
and 3) taking the results of the step 1) and the step 2) as input, calculating the information Gain (L, F) of the characteristic F under each tree, and taking the average value of the information Gain according to the number of the trees. The larger the information gain, the more important the feature is to the classification result. And setting a threshold value theta for selecting the characteristic, and keeping the characteristic when Gain (L, F) exceeds theta, otherwise deleting the characteristic.
Further, in step 4):
gradient boosting decision tree parameter initialization process: the main parameters influencing the classification performance of the gradient lifting decision tree comprise the number of trees, the height of the trees and the number of leaf nodes, the setting range of the number of decision trees of the gradient lifting decision tree classifier is [50,150], the step length is 10, the setting range of the tree height is [3,10], the step length is 1, the setting range of the leaf node number is [5,15] and the step length is 1.
Taking the feature subset obtained in step 3 as the input feature vector space of the step, and assuming that the classifier classification result of the (m-1) th round is F for eliminating the residual since the calculation purpose of gradient lifting is to reduce the residual of the last calculation result m-1 (x) Then the loss function is defined as: l (y, F) m-1 (x))＝y-F m-1 (x) Where x is the sample and y is the true alarm signal value of the sample. Gradient boosting decision tree by pair penalty function L (y, F) m-1 (x) Predicted value F of) m-1 (x i ) Derivation of the deviation
Obtaining the optimized direction of the next round of decision tree and using the learning rate gamma m Controlling the contribution degree of the decision tree of each round to the classification result, the classifier result of the mth round can be expressed as
Iteratively repeating the steps until the early warning signal label information category F identified by the gradient lifting decision tree classifier of the mth round m (x) Early warning signal label information category F identified by the gradient boosting decision tree classifier of round m-1 m-1 (x) When the difference is smaller than a set threshold value, iteration is stopped repeatedly, the gradient lifting decision tree classifier finishes training to obtain the trained alarm signal label information category identifier, wherein F m (x i ) Represents the predicted label result of the ith sample of the mth iteration, y i True tag information, L (y), representing a sample i i ，F m (x i ) Is a loss function, i.e. the error between the true tag value and the predicted tag value.
And the identifier identifies the type of the corresponding early warning signal label information as a true positive alarm signal or a false positive alarm signal according to the newly input screened characteristic data and the corresponding early warning signal.
The invention relates to a false positive alarm signal identification system of a hospital respirator-monitor, which comprises:
a data acquisition module: the method comprises the steps of obtaining a monitoring data set of the breathing machine-monitor input by a user, and sending the monitoring data set to a data preprocessing module, wherein the monitoring data set comprises a plurality of characteristic data and alarm signals, the plurality of characteristics comprise peak pressure, heart rate, respiratory frequency, spontaneous respiratory frequency, expiratory tidal volume, inspiratory tidal volume, minute respiratory volume, average pressure and respiratory unpressurized pressure, and the plurality of characteristics are preferably peak pressure, heart rate, respiratory frequency and spontaneous respiratory frequency;
a data preprocessing module: receiving a monitoring data set sent by the data acquisition module, performing missing value processing, abnormal value processing and data standardization processing on characteristic data in the detection data set, and sending the preprocessed monitoring data set to an alarm signal label information category identifier;
and the alarm signal label information category identifier is a trained gradient lifting decision tree classifier, receives the preprocessed monitoring data set sent by the data preprocessing module, and identifies and outputs the category of the alarm signal label information as a true positive alarm signal or a false positive alarm signal.
In order to verify the performance of the method in the identification of false positive alarm signals of a ventilator-monitor, empirical experiments were performed, collecting real patient monitoring data of a plurality of ventilators and monitors from an intensive care unit of a hospital, the ventilator-monitor being as shown in fig. 2. The data sample size was 15006 patient signs and alarm signal recordings.
For the purpose of verifying performance, three common Machine learning classification methods are selected as comparison methods, namely, logistic Regression (LR), support Vector Machine (SVM) and Naive Bayes classifier (Naive Bayes, NB) are compared with the method provided by the invention, and the adopted judgment indexes of false positive alarm signal identification performance comprise Accuracy (Accuracy), first Type error rate (Type I error), second Type error rate (Type II error), AUC and F1-score. The experimental flow is shown in figure 1:
in order to provide an intuitive understanding of the data sets collected by hospital ventilators and monitors, data samples are used as shown in table 1:
table 1: monitoring information dataset samples
| x 1 | x 2 | x 3 | x 4 | x 5 | x 6 | x 7 | x 8 | x 9 | x 10 | x 11 | x 12 | x 13 | x 14 | x 15 | x 16 | y |
| 8.1 | 7.2 | 56.6 | 50 | 5.4 | 29.9 | 29.9 | 258 | 291 | 15.5 | 79 | 111 | 65 | 30 | 99 | 111 | 1 |
| 7.9 | 7.5 | 56.6 | 50 | 5.4 | 30 | 30 | 257 | 260 | 15.6 | 80 | 112 | 66 | 19 | 100 | 109 | 1 |
| 7.6 | 7.6 | 56.6 | 50 | 5.3 | 30.1 | 30.1 | 272 | 292 | 15 | 80 | 111 | 66 | 19 | 100 | 108 | 1 |
| 7.7 | 7.5 | 56.6 | 50 | 5 | 30.1 | 30.1 | 232 | 230 | 15.4 | 78 | 108 | 65 | 20 | 100 | 107 | 1 |
| 6.7 | 7.8 | 56.6 | 50 | 5.3 | 30.3 | 30.3 | 254 | 288 | 15.2 | 78 | 109 | 65 | 21 | 99 | 105 | 1 |
| 8 | 7.4 | 56.6 | 50 | 5.6 | 29.5 | 29.5 | 255 | 254 | 15.6 | 79 | 110 | 65 | 19 | 100 | 109 | 0 |
| 7.7 | 7.5 | 56.6 | 50 | 5.3 | 28 | 28 | 237 | 257 | 15.3 | 78 | 109 | 64 | 19 | 98 | 110 | 0 |
| 7.2 | 7.7 | 56.6 | 50 | 5.1 | 26.5 | 26.5 | 300 | 297 | 15.3 | 78 | 109 | 64 | 19 | 100 | 109 | 0 |
| 6.9 | 7.5 | 56.6 | 50 | 5.7 | 25.5 | 25.5 | 275 | 256 | 15.4 | 78 | 109 | 64 | 19 | 100 | 109 | 0 |
| 7.3 | 7.3 | 56.6 | 50 | 5.4 | 27.9 | 27.9 | 240 | 255 | 15.6 | 78 | 110 | 64 | 19 | 100 | 109 | 0 |
In Table 1, x 1 …x 16 Correspondingly representing 16 physical sign information of a patient (minute expiratory volume, average pressure, oxygen inlet pressure, inspired oxygen concentration, breath non-positive pressure, spontaneous respiratory frequency, inspiratory tidal volume, expiratory tidal volume, peak pressure, invasive blood pressure average value, invasive blood pressure high value, invasive blood pressure low value, central venous pressure, blood oxygen concentration and heart rate respectively), y represents the type of an alarm signal (y =1 represents a false positive alarm signal, y =0 represents a true positive alarm signal)
In order to avoid randomness possibly caused by one experiment, 30 experiments are carried out in a random sampling mode, wherein the division ratio of the training sample and the test sample is 20%,30% and 40%, and finally the average result and variance of the 30 experiments are taken to judge the performance of the method. The results of the method proposed by the invention and the results of the comparative method are listed in tables 2-4, respectively:
table 2: performance comparison of GBDT method and comparison method (LR, SVM and NB) (training set: test set =80
As can be seen from the classification results in Table 2, the gradient boosting decision tree GBDT has the best recognition effect on false positive alarm signals, and compared with the other four methods, the gradient boosting decision tree GBDT has very good classification results on the accuracy, the second error rate, the AUC and the F1-Score, and the AUC reaches 97.6%. Although the error rate of the gradient boosting decision tree is higher than that of Logistic regression and a support vector machine in the first type of error rate, the invention aims at the field of false positive alarm signal identification, namely, the second type of error rate really reflects the identification success rate of a method for false positive alarm signals. Therefore, even though Logistic regression has a good effect on the first type of error rate, the second type of error rate is very high, and the result shows 56.4%, which indicates that the method has a poor effect on identifying the false positive alarm type. In addition, AUC and F1-Score are comprehensive indexes for measuring the classification effect of two types of signals, and the performance of the two indexes of the gradient boosting decision tree is far superior to that of other methods. The F1-Score pair of the four models is shown in FIG. 3 (a). Therefore, the method provided by the invention has a very good function of identifying the false positive alarm model.
Table 3: performance comparison of GBDT method and comparison method (LR, SVM and NB) (training set: test set =70
| Metrics | GBDT | LR | SVM | NB |
| AUC | 0.972(0.02) | 0.691(0.03) | 0.837(0.03) | 0.837(0.03) |
| Accuracy | 0.997(0.00) | 0.989(0.00) | 0.994(0.00) | 0.941(0.01) |
| TypeIError | 0.002(0.00) | 0.000(0.00) | 0.000(0.00) | 0.060(0.01) |
| TypeIIError | 0.054(0.04) | 0.619(0.06) | 0.325(0.06) | 0.012(0.01) |
| F1-Score | 0.914(0.04) | 0.547(0.06) | 0.794(0.04) | 0.364(0.03) |
Table 3 reflects the recognition effect of each method at different training set and test set ratios, and the overall results are consistent with those of table 2. The proposed gradient lifting decision tree method has robustness in the identification of false positive alarm signals, and the signal identification effect is stable. The F1-Score pair of the four models is shown in FIG. 3 (b).
Table 4: performance comparison of GBDT method and comparison method (LR, SVM and NB) (training set: test set =60
The classification effect reflected in table 4 and tables 2 and 3 is the same, and the signal identification performance of all methods is reduced as a whole due to the reduction of the sample size of the training set, but the whole judgment is not influenced, namely, the gradient boosting decision tree has the optimal identification function of false positive alarm signals. In addition, the variance result of 30 experimental results in parentheses is observed, so that the performance of the gradient lifting decision tree is very stable and no large fluctuation is generated. To further show the experimental results more clearly, reference may be made to the comparison results of the F1-SCORE indexes of the four methods GBDT, LR, SVM and NB provided in (c) of fig. 3.
The invention discloses a method for identifying false positive alarm signals of a breathing machine based on an integrated tree. Experimental results show that the method has excellent identification performance of false positive alarm signals, and the identification effect of the method is stable.
The above examples are provided for the purpose of describing the present invention only and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalent substitutions and modifications can be made without departing from the spirit and principles of the invention, and are intended to be within the scope of the invention.
Claims (7)
1. A respiratory machine false positive alarm signal identification method based on an integrated tree is characterized by comprising the following steps:
step (1), data collection: collecting sample data of a plurality of real breathing machines and a plurality of monitors of a patient from a hospital, and combining the sample data of each monitor and the sample data of each breathing machine to be used as an original data set, wherein the original data set comprises a plurality of characteristic data and corresponding alarm signals;
and (2) data preprocessing: performing missing value processing, abnormal value processing and data standardization processing on the characteristic data of the original data set in the step (1), and performing identification processing on alarm signals of the original data set to further obtain a preprocessed data set, wherein the identification processing is to respectively identify different types of label information for the alarm signals according to a set rule, and the different types of label information are true positive alarm signals or false positive alarm signals;
and (3) feature selection: performing feature screening on feature data of the preprocessed data set in the step (2) by using a random forest, reserving screened features, and forming a training data set by the screened feature data in the preprocessed data set and corresponding early warning signal label information;
step (4), false positive alarm signal identification: training parameters of a gradient boosting decision tree classifier by using the training data set in the step (3) to obtain a trained alarm signal label information category recognizer, and recognizing the category of the corresponding alarm signal label information output as a true positive alarm signal or a false positive alarm signal by the recognizer according to newly input screened feature data and the corresponding alarm signal;
the step (4) is realized by the following steps:
(41) Taking the screened features obtained in the step (3) as an input feature vector space, and if the early warning signal label information output by the gradient lifting decision tree classifier of the (m-1) th round is F m-1 (x) Then the loss function L (y, F) m-1 (x))＝y-F m-1 (x) Wherein x is a sample, and y is real early warning signal label information corresponding to the sample;
(42) By L (y, F) m-1 (x) Pair F) m-1 (x) Derivation of the deviation
Obtaining the optimization direction and the learning rate gamma of the gradient lifting decision tree classifier of the mth round m-1 Controlling the contribution degree of the early warning signal label information output by the gradient lifting decision tree classifier in the m-1 th round, wherein the early warning signal label information output by the gradient lifting decision tree classifier in the m-th round is
(43) Iteratively repeating the steps (41) to (42) until the gradient boosting decision tree classifier of the mth round and the (m-1) th round identifies the output early warning signal label information F m (x) And F m-1 (x) When the difference is smaller than a set threshold value, iteration is repeatedly stopped to obtain the trained alarm signal label information category identifier;
(44) And the identifier identifies the type of the corresponding early warning signal label information as a true positive alarm signal or a false positive alarm signal according to the newly input screened characteristic data and the corresponding early warning signal.
2. The integrated tree based ventilator false positive alarm signal identification method as claimed in claim 1, wherein in the step 1:
the sample frequency of the real breathing machine-monitor monitoring data collected from the hospital is in seconds, and the data are collected for three times per second;
the plurality of features comprises 16 individual feature features, wherein the 16 individual feature features are minute expiratory volume, average pressure, oxygen input port pressure, inspiratory oxygen concentration, respiratory non-positive pressure, spontaneous respiratory frequency, inspiratory tidal volume, expiratory tidal volume, peak pressure, invasive blood pressure mean, invasive blood pressure high value, invasive blood pressure low value, central venous pressure, blood oxygen concentration and heart rate respectively;
the specific implementation of combining each monitor data sample and each ventilator sample data is to combine each monitor data sample and each ventilator sample data into one sample by using a matching method and taking a ventilator as a main timestamp.
3. An integrated tree based ventilator false positive alarm signal identification method as claimed in claim 1, wherein in the step (2):
the missing value processing is specifically realized by adopting a feature mean value method to screen missing values of each feature data in the original data set and fill the missing values into a mean value of each feature data, wherein the value x 'of the missing value of the jth sample of the ith feature data in the original data set after filling the missing value through missing value processing' missing(i,j) ，
Wherein x is i1 ,x i2 ,…,x in Respectively representing 1,2, \ 8230under the ith characteristic in the original data set, n samples, and n represents the number of the samples;
the abnormal value processing is specifically realized by adopting a triple standard deviation method, firstly screening an abnormal value of which the difference between each characteristic data in the original data set and the mean value of the characteristic data is more than triple of the standard deviation of the characteristic data, and adjusting the abnormal value to be the sum of the mean value of the characteristic data and triple of the standard deviation of the characteristic data; then, screening the mean value of each characteristic data in the original data set and the characteristic dataAnd adjusting the abnormal value of which the difference is less than three times of the inverse number of the standard deviation of the characteristic data to be the difference between the mean value of the characteristic data and three times of the standard deviation of the characteristic data, wherein the abnormal value of the jth sample ith characteristic data in the original data set is adjusted to be a value x 'by abnormal value processing' outlier(i,j) ，
Wherein x is ij Represents the value of the j sample under the ith characteristic data in the original data set, mu i Means, σ, representing the ith characteristic data in the raw data set i Representing a standard deviation of ith characteristic data in the original data set;
the normalization processing is realized by replacing the numerical value of each characteristic data in the original data set by the z-score of each characteristic data by using a z-score method, wherein the numerical value of the ith characteristic data of the jth sample in the original data set is replaced by the numerical value x 'after the normalization processing' norm(i,j) ，
Wherein x is ij A value, μ, representing the j sample under the i characteristic data in the original data set i Representing the mean, σ, of the ith characteristic data in the original data set i Representing a standard deviation of ith characteristic data in the original data set;
the specific implementation of the identification processing is that the established identification rule includes: when the respirator and the monitor alarm at the same time, the label information of the alarm signal is a true positive alarm signal; when continuous uninterrupted alarm occurs on the respirator or the monitor and the alarm duration times exceed 3 times, label information of the alarm signal is a true positive alarm signal; when the sign characteristic data exceeds the threshold range of the sign characteristic data set by the respirator, the label information of the alarm signal is a true positive alarm signal; the threshold range of the sign characteristic data comprises: the breath volume per minute is 0.5-180L/min, and the allowable error range is +/-3 percent; average pressure is-2 to 12kPa, and the allowable error range is +/-0.1 kPa; the concentration of the inhaled oxygen is 21-100%, and the allowable error range is +/-3%; the non-positive pressure of breathing is-12 to 12kPa, and the allowable error range is +/-0.05 kPa; the spontaneous respiratory frequency and the respiratory frequency are 1 to 150 times/minute, and the allowable error range is +/-3 percent; the inspiration/expiration tidal volume is-10 to 10L, and the allowable error range is +/-3 percent.
4. The integrated tree-based method for identifying false positive alarm signals of a breathing machine according to claim 1, wherein in the step (3), the feature screening is performed by using a random forest, and the specific implementation of the feature after screening is retained as follows:
calculating the difference information Gain (L, F) between the information Entropy (L) of the alarm signal label information and the information Entropy (L, F) of the alarm signal label under the characteristic F for each characteristic F in the preprocessed data set,
Gain(L,F)＝Entropy(L)-Entropy(L,F)，
if Gain (L, F) > theta, keeping the characteristic F as the screened characteristic, and if Gain (L, F) < theta, deleting the characteristic F, wherein theta is a set threshold;
wherein L represents alarm tag information of the preprocessed data set, p i The probability that label information of the ith category of the alarm signal appears in the preprocessed data set is represented;
wherein L represents alarm signal tag information of the preprocessed data set, v represents the preprocessed data setNumber of values, L, taken under the feature F j And representing the number of jth values of the preprocessed data set under the characteristic F.
5. The integrated tree based ventilator false positive alarm signal identification method of claim 1, wherein in step (3), the filtered features comprise peak pressure, heart rate, respiratory rate, spontaneous respiratory rate, expiratory tidal volume, inspiratory tidal volume, minute respiratory volume, mean pressure, and respiratory non-positive pressure.
6. The integrated tree based ventilator false positive alarm signal recognition method of claim 1, wherein in the step (4), the number of decision trees of the gradient boosting decision tree classifier is set to be in a range of [50,150], a step size is 10, a tree height setting range is [3,10], a step size is 1, a number of leaf nodes is set to be in a range of [5,15], and a step size is 1.
7. A system for implementing the integrated tree based ventilator false positive alarm signal identification method of claim 1, the system comprising:
the device comprises a data acquisition module, a data preprocessing module and an alarm signal label information category identifier;
the data acquisition module acquires a monitoring data set of the breathing machine-monitor input by a user and sends the monitoring data set to the data preprocessing module, wherein the monitoring data set comprises a plurality of characteristic data and alarm signals, and the plurality of characteristics comprise peak pressure, heart rate, respiratory rate, spontaneous respiratory rate, expiratory tidal volume, inspiratory tidal volume, minute respiratory volume, average pressure and respiratory unpressurized pressure;
the data preprocessing module is used for receiving the monitoring data set sent by the data acquisition module, performing missing value processing, abnormal value processing and data standardization processing on the characteristic data in the monitoring data set, and sending the preprocessed monitoring data set to the alarm signal label information category identifier;
the alarm signal label information category recognizer is a trained gradient lifting decision tree classifier, receives the preprocessed monitoring data set sent by the data preprocessing module, and recognizes and outputs whether the category of the alarm signal label information is a true positive alarm signal or a false positive alarm signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010492039.5A CN112245728B (en) | 2020-06-03 | 2020-06-03 | Respirator false positive alarm signal identification method and system based on integrated tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010492039.5A CN112245728B (en) | 2020-06-03 | 2020-06-03 | Respirator false positive alarm signal identification method and system based on integrated tree |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112245728A CN112245728A (en) | 2021-01-22 |
| CN112245728B true CN112245728B (en) | 2022-11-29 |
Family
ID=74224209
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010492039.5A Active CN112245728B (en) | 2020-06-03 | 2020-06-03 | Respirator false positive alarm signal identification method and system based on integrated tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112245728B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112800983B (en) * | 2021-02-01 | 2024-03-08 | 玉林师范学院 | Random forest-based non-line-of-sight signal identification method |
| CN113349746A (en) * | 2021-07-21 | 2021-09-07 | 中南大学湘雅医院 | Vital sign monitoring alarm system |
| CN115399738B (en) * | 2022-08-17 | 2023-05-16 | 中南大学湘雅医院 | Rapid ICU false alarm identification method |
| CN117612725B (en) * | 2024-01-23 | 2024-03-29 | 南通大学附属医院 | Respirator alarm management method and system for intensive care unit |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104414636A (en) * | 2013-08-23 | 2015-03-18 | 北京大学 | Magnetic resonance image based cerebral micro-bleeding computer auxiliary detection system |
| CN106961249A (en) * | 2017-03-17 | 2017-07-18 | 广西大学 | A kind of diagnosing failure of photovoltaic array and method for early warning |
| CN110298085A (en) * | 2019-06-11 | 2019-10-01 | 东南大学 | Analog-circuit fault diagnosis method based on XGBoost and random forests algorithm |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2539197B1 (en) * | 2010-02-26 | 2020-12-16 | Gentex Corporation | Automatic vehicle equipment monitoring, warning, and control system |
| JP7057356B6 (en) * | 2016-11-29 | 2022-06-02 | コーニンクレッカ フィリップス エヌ ヴェ | False alarm detection |
| CN106730209B (en) * | 2017-01-18 | 2019-11-22 | 湖南明康中锦医疗科技发展有限公司 | The method and ventilator of ventilator alarm |
| CN109117956B (en) * | 2018-07-05 | 2021-08-24 | 浙江大学 | Method for determining optimal feature subset |
| CN109087482A (en) * | 2018-09-18 | 2018-12-25 | 西安交通大学 | A kind of falling detection device and method |
| CN109489800A (en) * | 2018-12-14 | 2019-03-19 | 广东世港信息科技有限公司 | A kind of disturbance event recognition methods in distribution optic cable vibration safety pre-warning system |
| CN110544373B (en) * | 2019-08-21 | 2020-11-03 | 北京交通大学 | Truck early warning information extraction and risk identification method based on Beidou Internet of vehicles |
| CN110519128B (en) * | 2019-09-20 | 2021-02-19 | 西安交通大学 | Random forest based operating system identification method |
-
2020
- 2020-06-03 CN CN202010492039.5A patent/CN112245728B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104414636A (en) * | 2013-08-23 | 2015-03-18 | 北京大学 | Magnetic resonance image based cerebral micro-bleeding computer auxiliary detection system |
| CN106961249A (en) * | 2017-03-17 | 2017-07-18 | 广西大学 | A kind of diagnosing failure of photovoltaic array and method for early warning |
| CN110298085A (en) * | 2019-06-11 | 2019-10-01 | 东南大学 | Analog-circuit fault diagnosis method based on XGBoost and random forests algorithm |
Non-Patent Citations (2)
| Title |
|---|
| HOG结合随机森林的新型手势识别框架;吴爽等;《湘潭大学自然科学学报》;20180415;第40卷(第02期);第76-79页 * |
| 基于深度学习的癫痫脑电通道选择与发作检测;曹玉珍等;《天津大学学报(自然科学与工程技术版)》;20200410;第53卷(第04期);第426-432页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112245728A (en) | 2021-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112245728B (en) | Respirator false positive alarm signal identification method and system based on integrated tree | |
| Rose et al. | Cough augmentation techniques for extubation or weaning critically ill patients from mechanical ventilation | |
| RU2515401C2 (en) | System and method of identifying respiratory insufficiency of subject's respiration | |
| Orr et al. | A breathing circuit alarm system based on neural networks | |
| JP2006502481A (en) | Neural network in sedation and analgesia system | |
| US10340039B2 (en) | Managing patient devices based on sensor data | |
| CN112509676A (en) | Hemodialysis center intelligent management system | |
| Kamio et al. | Mechanical ventilation-related safety incidents in general care wards and ICU settings | |
| KR20210066271A (en) | Order system using medical deep learning in the field of anesthesia | |
| Ribeiro et al. | A machine learning early warning system: multicenter validation in Brazilian hospitals | |
| CN116530943B (en) | Anesthesia depth detection device based on blood gas data | |
| Durmuşoğlu et al. | Remembering Medical Ventilators and Masks in the Days of COVID-19: Patenting in the Last Decade in Respiratory Technologies | |
| Mahmud et al. | Res-se-convnet: A deep neural network for hypoxemia severity prediction for hospital in-patients using photoplethysmograph signal | |
| Betancourt et al. | Segmented wavelet decomposition for capnogram feature extraction in asthma classification | |
| Liu et al. | Learning features of brain network for anomaly detection | |
| CN115089140A (en) | Sleep analysis system and method capable of monitoring heart rate and respiration | |
| Oliveira et al. | Feature selection for detecting patients with weaning failures in Intensive Medicine | |
| Armarego et al. | High‐flow nasal cannula therapy for infants with bronchiolitis | |
| WO2021110446A1 (en) | Assistance in the detection of pulmonary diseases | |
| CN113870988A (en) | Anesthesia workstation operation supervisory systems based on big data analysis | |
| CN117612725B (en) | Respirator alarm management method and system for intensive care unit | |
| Ren et al. | 1D-CNNs model for classification of sputum deposition degree in mechanical ventilated patients based on airflow signals | |
| Gummadi et al. | Transfer learning based detection of pneumonia from chest x-ray images | |
| CN113257399B (en) | Automatic emergency first-aid article acquisition method and system based on semantic analysis | |
| Alexie et al. | Investigation on Properties of Capnogram: On Stationarity and Spectral Components of the Signal |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |