CN110051324A - A kind of acute respiratory distress syndrome anticipated mortality method and system - Google Patents
A kind of acute respiratory distress syndrome anticipated mortality method and system Download PDFInfo
- Publication number
- CN110051324A CN110051324A CN201910194628.2A CN201910194628A CN110051324A CN 110051324 A CN110051324 A CN 110051324A CN 201910194628 A CN201910194628 A CN 201910194628A CN 110051324 A CN110051324 A CN 110051324A
- Authority
- CN
- China
- Prior art keywords
- respiratory distress
- model
- distress syndrome
- acute respiratory
- mortality
- 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.)
- Granted
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
- A61B5/0826—Detecting or evaluating apnoea events
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7271—Specific aspects of physiological measurement analysis
- A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Physiology (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Pulmonology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention discloses a kind of acute respiratory distress syndrome anticipated mortality method and system, method includes: the medical data for obtaining Patients with ARDS；Model training is carried out using the method for machine learning according to the medical data of acquisition, obtains acute respiratory distress syndrome anticipated mortality model；Object to be predicted is predicted using acute respiratory distress syndrome anticipated mortality model, obtains the prediction result of the acute respiratory distress syndrome death rate.The present invention trains acute respiratory distress syndrome anticipated mortality model by the method for machine learning, the death rate of ARDS patient is predicted using acute respiratory distress syndrome anticipated mortality model again, machine learning is applied in the prediction of ARDS mortality, the death rate of ARDS patient can accurately and be objectively predicted by the model of machine learning training, more effective and feasible predictive information is provided for clinician, can be widely applied to medical data mining field.
Description
Technical field
The present invention relates to medical data mining field, especially a kind of acute respiratory distress syndrome anticipated mortality method
And system.
Background technique
Acute respiratory distress syndrome (Acute Respiratory Distress Syndrome, ARDS) is a kind of normal
See critical illness, refers to Acute Diffuse injury of lungs related with risk factor is exposed to, be frequently accompanied by caused by lung inflammation
Pulmonary vascular permeability increases and gassiness lung tissue is reduced.Clinically the patient of various critical illnesses, which exists, occurs the latent of ARDS
In risk, and the hospital mortality after ARDS occurs between 34.9%-46.1%, seriously threatens the life and shadow of patient with severe symptoms
Ring its life quality.Therefore ARDS mortality prediction model is established, coincident with severity degree of condition can be distinguished, to determine difference
Treatment strategy；Various variables can be probed into the influence factor of mortality, to raising patient using prediction model simultaneously
Survival rate has positive meaning.
There is currently prediction ICU patient survival rate (probability of survival) method, be all based on biography
System analysis method, including APACHE II, OSI, OI and LIS analysis etc..These traditional analysis are usually in one or more
Medical center collect data, then the experience based on disease expert and statistical method (the most commonly used is logistic regressions) obtain it is relevant
Variable goes to construct and verifies prediction model finally by gained variable.However there are the following problems for such methods: (1) by expert
The variable that experience or statistical analysis obtain, can have subjectivity and data deviation；(2) the factor pole that ARDS occurs with develops is influenced
For complexity, it is difficult that multidimensional variable is combined to statistically analyze；(3) these methods not aim at ARDS design, at present not yet there are
The Rating Model of effect is suitable for predicting the death rate of ARDS patient.
Summary of the invention
In order to solve the above technical problems, it is an object of the invention to: a kind of objective and accurate acute respiratory distress is provided
Syndrome anticipated mortality method and system.
One aspect of the present invention is adopted the technical scheme that:
A kind of acute respiratory distress syndrome anticipated mortality method, comprising the following steps:
Obtain the medical data of Patients with ARDS；
Model training is carried out using the method for machine learning according to the medical data of acquisition, it is comprehensive to obtain acute respiratory distress
Levy anticipated mortality model；
Object to be predicted is predicted using acute respiratory distress syndrome anticipated mortality model, obtains acute exhale
Inhale the prediction result of the Distress syndrome death rate.
Further, the step for the medical data for obtaining Patients with ARDS, specifically:
The medical data of Patients with ARDS is downloaded from MIMIC-III database.
Further, the medical data according to acquisition carries out model training using the method for machine learning, obtains acute
It the step for Respiratory Distress Syndrome(RDS) anticipated mortality model, specifically includes:
The medical data of acquisition is pre-processed, the pretreatment includes screening sample and feature extraction；
Model training is carried out using the method for machine learning according to pretreated data, it is comprehensive to obtain acute respiratory distress
Levy anticipated mortality model, the acute respiratory distress syndrome anticipated mortality model include hospital mortality prediction model,
30 days anticipated mortality models and an annual death rate prediction model.
Further, the step for medical data of described pair of acquisition pre-processes, specifically includes:
Screening sample is carried out to the medical data of acquisition according to standard of being included in and exclusion criteria, obtains the sample screened,
The standard of being included in includes being more than or equal to moving in intensive care unit and being acute respiration through Berlin standard diagnostics for 18 one full year of life the age
The patient of Distress syndrome, the exclusion criteria include in MIMIC-III database the incomplete data of data record, age it is small
Patient in 18 one full year of life, the patient using palliative treatment and ICU record any one in patient of the time less than 48 hours；
Characteristics of variables to each sample of the sample extraction screened for modeling.
Further, the step for medical data of described pair of acquisition pre-processes, and also specifically includes the following steps:
Multiple interpolation is carried out to the missing data in the medical data of acquisition.
Further, described that model training is carried out using the method for machine learning according to pretreated data, it obtains acute
It the step for Respiratory Distress Syndrome(RDS) anticipated mortality model, specifically includes:
Classified according to survival of patients number of days to pretreated data, respectively obtains 3 anticipated mortality models
Positive group and negative group, 3 anticipated mortality models include hospital mortality prediction model, 30 days anticipated mortality models
With an annual death rate prediction model；
Positive group to 3 anticipated mortality models and negative group analyze between group respectively, and it is aobvious to filter out group difference
The feature of work；
Random forests algorithm is used to establish the acute respiratory distress syndrome death rate according to the significant feature of group difference pre-
Survey model.
Further, the positive group and negative group of 3 anticipated mortality models specifically: hospital mortality prediction model
Positive group be in hospital in dead patient data, negative group of hospital mortality prediction model be in hospital during the patient survived
Data；Positive group of 30 days anticipated mortality models be in hospital after patient data dead in 30 days, 30 days anticipated mortality moulds
Negative group of type is the patient data survived in 30 days afterwards in hospital；Positive group of one annual death rate prediction model is latter year in hospital
The patient data of interior death, negative group of an annual death rate prediction model are the patient datas of interior survival of latter year in hospital.
Further, described that acute respiratory distress synthesis is established using random forests algorithm according to the significant feature of group difference
The step for levying anticipated mortality model, specifically includes:
The significant feature of group difference is divided by the first training set and test set using K folding cross-validation method；
First training set is divided by the second training set using K folding cross-validation method and verifying collects；
Optimal model parameter, and then basis are searched out using the method for grid optimizing according to the second training set and verifying collection
Optimal model parameter constructs several acute respiratory distress syndrome anticipated mortality models using random forests algorithm；
Each acute respiratory distress syndrome anticipated mortality model is respectively adopted to test test set, obtains each folding
Prediction result；
It averages to the prediction result of each folding, it is corresponding to obtain each acute respiratory distress syndrome anticipated mortality model
Estimated performance result；
According to obtained estimated performance as a result, being selected from each acute respiratory distress syndrome anticipated mortality model pre-
The best model of results of property is surveyed as final acute respiratory distress syndrome anticipated mortality model.
Another aspect of the present invention is adopted the technical scheme that:
A kind of acute respiratory distress syndrome anticipated mortality system, comprises the following modules:
Module is obtained, for obtaining the medical data of Patients with ARDS；
Training module obtains urgency for carrying out model training using the method for machine learning according to the medical data of acquisition
Property Respiratory Distress Syndrome(RDS) anticipated mortality model；
Prediction module, it is pre- for being carried out using acute respiratory distress syndrome anticipated mortality model to object to be predicted
It surveys, obtains the prediction result of the acute respiratory distress syndrome death rate.
Another aspect of the present invention is adopted the technical scheme that:
A kind of acute respiratory distress syndrome anticipated mortality system, comprising:
At least one processor；
At least one processor, for storing at least one program；
When at least one described program is executed by least one described processor, so that at least one described processor is realized
Acute respiratory distress syndrome anticipated mortality method as described in the present invention.
The beneficial effects of the present invention are: acute respiratory distress syndrome anticipated mortality method and system of the present invention, obtain
After the medical data of Patients with ARDS, acute respiratory distress syndrome is trained by the method for machine learning
Anticipated mortality model finally predicts the death of ARDS patient using acute respiratory distress syndrome anticipated mortality model
Machine learning is applied in the prediction of ARDS mortality by rate, can be by the model of machine learning training accurately and objectively
The death rate of ARDS patient is predicted, provides more effective and feasible predictive information as reference for clinician.
Detailed description of the invention
Fig. 1 is acute respiratory distress syndrome anticipated mortality method flow diagram provided in an embodiment of the present invention；
Fig. 2 is the data screening flow chart of the embodiment of the present invention；
Fig. 3 is random forests algorithm schematic diagram；
Fig. 4 is the detailed process that the embodiment of the present invention carries out anticipated mortality model training using the method for machine learning
Figure；
Fig. 5 is pre- for the training of the final three anticipated mortality models of the embodiment of the present invention and applied to clinically new data
The flow chart of survey.
Specific embodiment
First to it is involved in the present invention to noun and term be illustrated:
EHR:electronic health record, personal electric health records.
Berlin standard: Berlin definition, ARDS in 2011 define the general mark of the diagnosis ARDS of working group's proposition
It is quasi-.
PEEP:positive end expiratory pressure, end-expiratory positive pressure.
APPS:plateau pressure, air flue platform pressure.
CPAP:Continuous Positive Airway Pressure, continuous positive pressure ventilation.
SAPS:Simplified Acute Physiology Score simplifies Acute Physiology Score.
LIS:the Lung Injury Score, injury of lungs score.
APACHE:Acute Physiology and Chronic Health Evaluation, acute physiology and chronic
Health status scoring.
OI:Oxygenation index, oxygenation index, calculation method: [PaO2/FiO2].
OSI:Oxygenation Saturation Index, oxygen saturation index, calculation method: [FiO2/SpO2].
ROC:Receiver Operating Characteristic Curve, receiver operating characteristic curve.
AUROC:Area Under the Receiver Operating Characteristic Curve, recipient behaviour
Make area under indicatrix.
RF:random forest, random forest are a kind of sorting algorithms of machine learning (ML).
Death: in-hospital mortality is referred to that ARDS patient is dead during being hospitalized, is remembered with admission information
It records subject to the time.
Death in 30 days: 30-day mortality refers to dead in 30 days after ARDS patient is admitted to hospital.
Death in 1 year: it is dead to refer to that ARDS patient was admitted to hospital in latter year by 1-year mortality.
The present invention is further explained and is illustrated with specific embodiment with reference to the accompanying drawings of the specification.
Referring to Fig.1, the embodiment of the invention provides a kind of acute respiratory distress syndrome anticipated mortality methods, including with
Lower step:
Obtain the medical data of Patients with ARDS；
Model training is carried out using the method for machine learning according to the medical data of acquisition, it is comprehensive to obtain acute respiratory distress
Levy anticipated mortality model；
Object to be predicted is predicted using acute respiratory distress syndrome anticipated mortality model, obtains acute exhale
Inhale the prediction result of the Distress syndrome death rate.
Specifically, the medical data of Patients with ARDS can be from public database (such as MIMIC-III
Database etc.) downloading population statistics, vital sign monitoring etc. medical datas.Object to be predicted is that new ARDS suffers from
Person.
Machine learning is a branch of artificial intelligence, enables computer be automatically performed data point by designing special algorithm
Analysis is to grasp regular (" learning "), and assimilated equations are judged or predicted to unknown data.Machine learning method can pass through
Constantly " learn " processing of information can also be easily completed, compared to statistical analysis technique, machine to analyze, grasp rule
Device study has incomparable advantage in terms of analysis big data quantity and high variable dimension.The method of machine learning includes random
Forest algorithm, algorithm of support vector machine, deep learning algorithm etc..
It is dead to find ARDS patient according to existing ARDS patient medical data for the method that the present embodiment applies machine learning
The rule for dying rate obtains ARDS mortality prediction model, and the new data of next still other (object i.e. to be predicted) are just
The death rate of prediction model automatic Prediction ARDS patient can be allowed according to the previously rule that had learnt.
The present embodiment uses machine learning training pattern, using more more full clinical data as input, by expertise
Or the influence of statistical analysis is small, can acquire deviation to avoid data, more hiding informations be obtained, to improve the accurate of prediction
Rate；In addition, machine learning can be trained in the data of different size or scale, it can be found that some new having are predicted
Value or the variable for not having value, additional inspiration is provided for clinical treatment.
Be further used as preferred embodiment, the medical data for obtaining Patients with ARDS this
Step, specifically:
The medical data of Patients with ARDS is downloaded from MIMIC-III database.
Specifically, MIMIC-III is the laboratory physionet by MIT, BIDMC (Beth Israel Deaconess
MedicalCenter) and the database for intensive care patient built jointly of PHILIPS Co., database contains at present
The medical data (wherein ARDS patient data has 3186) of 53423 adult patients, including basic population statistics,
The data such as vital sign monitoring.
The clinical data that structuring in MIMIC-III is utilized in the present embodiment is trained, and is effectively prevented smudgy
Clinical definition and data acquisition deviation, improve the accuracy rate of ARDS anticipated mortality.
It is further used as preferred embodiment, the medical data according to acquisition is carried out using the method for machine learning
Model training, specifically includes the step for obtaining acute respiratory distress syndrome anticipated mortality model:
The medical data of acquisition is pre-processed, the pretreatment includes screening sample and feature extraction；
Model training is carried out using the method for machine learning according to pretreated data, it is comprehensive to obtain acute respiratory distress
Levy anticipated mortality model, the acute respiratory distress syndrome anticipated mortality model include hospital mortality prediction model,
30 days anticipated mortality models and an annual death rate prediction model.
Specifically, screening sample is and to meet subsequent builds model to filter out and meet current ARDS diagnostic criteria
It is required that and the record content sample that can be realized in hospital ICU and (ensure that model can be with practical application).Feature extraction, be for
The feature for carrying out model training is extracted from sample.
It is further used as preferred embodiment, the step for medical data of described pair of acquisition pre-processes, specifically
Include:
Screening sample is carried out to the medical data of acquisition according to standard of being included in and exclusion criteria, obtains the sample screened,
The standard of being included in includes being more than or equal to moving in intensive care unit and being acute respiration through Berlin standard diagnostics for 18 one full year of life the age
The patient of Distress syndrome, the exclusion criteria include in MIMIC-III database the incomplete data of data record, age it is small
Patient in 18 one full year of life, the patient using palliative treatment and ICU record any one in patient of the time less than 48 hours；
Characteristics of variables to each sample of the sample extraction screened for modeling.
Specifically, Berlin standard diagnostics are the foundation of acute respiratory distress syndrome are as follows: 1) acute attack；2) PEEP (or
CPAP)>=5cm H2O when, OI (PaO2/FiO2)<300mmHg；3) Chest Image bilateral infiltrates shadow；4) respiratory failure can not be used
Heart failure is completely explained.
Using the patient of palliative treatment, refer to the patient for not receiving active treatment.
The characteristics of variables of extraction includes the data that age, gender, APACHE II score of patient etc. can be read directly,
It also include the characteristics of variables such as Mechanical ventilation time, patient physiological information.
It is further used as preferred embodiment, the step for medical data of described pair of acquisition pre-processes, also has
Body the following steps are included:
Multiple interpolation is carried out to the missing data in the medical data of acquisition.
Specifically, multiple interpolation is the method that a kind of pair of missing data is adjusted, using a series of possible data sets
Each missing data value is filled, then goes to analyze multiple interpolation data collection using the standard of complete data, finally to these
It analyzes result and concludes synthesis.SPSS software can be used to complete multiple interpolation operation for the present embodiment, to the ARDS of individual data missing
Data are supplemented.
It is further used as preferred embodiment, it is described to be carried out according to pretreated data using the method for machine learning
Model training, specifically includes the step for obtaining acute respiratory distress syndrome anticipated mortality model:
Classified according to survival of patients number of days to pretreated data, respectively obtains 3 anticipated mortality models
Positive group and negative group, 3 anticipated mortality models include hospital mortality prediction model, 30 days anticipated mortality models
With an annual death rate prediction model；
Positive group to 3 anticipated mortality models and negative group analyze between group respectively, and it is aobvious to filter out group difference
The feature of work；
Random forests algorithm is used to establish the acute respiratory distress syndrome death rate according to the significant feature of group difference pre-
Survey model.
Specifically, SPSS software can be used to carry out in analysis between group.The present embodiment the group between sample characteristics by analyzing
The significant feature of different classes of inter-sample difference is found out as input, data redundancy can be reduced, model calculation amount is reduced, find out
More meaningful feature.The present embodiment is respectively to positive group in three prediction models and negative group analyze between group, specifically
Operation is: the continuous variable for meeting normal distribution examines (Student t-test) to analyze using t, and Non-Gaussian Distribution
Continuous variable uses non-parametric test (Mann-Whitney U test)；Chi-square Test is then passed through for discrete variable
(Chi2test) or Fisher ' s exact test is analyzed.Characteristics of variables of the P value less than 0.05 by above-mentioned inspection can
Think that group difference is significant, then retains this feature；This feature is poor between positive and negative group to be thought greater than 0.05 for P value
It is different unobvious, it is smaller on the classification of prediction model influence, it can leave out.
Random forest is a kind of classifier that using sample more decision trees are trained with simultaneously forecast sample result, decision
For the training process of tree using top-down recursion method, basic thought is with comentropy for measurement one entropy of building
Decline most fast tree, until the entropy of leaf node is zero, the sample of each leaf node belongs to same category at this time.When defeated
When entering new samples, each decision tree judges to vote respectively in random forest, number of votes obtained it is most just as final classification results.With
Machine forest possesses preferable noise resisting ability and is not easy over-fitting, energy by the integrated study and most voting mechanisms of decision tree
Preferably the death rate of ARDS patient is predicted.
It is further used as preferred embodiment, the positive group and negative group of 3 anticipated mortality models specifically:
Positive group of hospital mortality prediction model is that interior dead patient data, negative group of hospital mortality prediction model are in hospital
The patient data survived during in hospital；Positive group of 30 days anticipated mortality models be in hospital after patient's number dead in 30 days
According to negative group of 30 days anticipated mortality models is the patient data survived in 30 days afterwards in hospital；One annual death rate prediction model
Positive group be patient data dead in latter year in hospital, negative group of an annual death rate prediction model is in hospital in latter year
The patient data of survival.
It is further used as preferred embodiment, it is described to be built according to the significant feature of group difference using random forests algorithm
It the step for vertical acute respiratory distress syndrome anticipated mortality model, specifically includes:
The significant feature of group difference is divided by the first training set and test set using K folding cross-validation method；
First training set is divided by the second training set using K folding cross-validation method and verifying collects；
Optimal model parameter, and then basis are searched out using the method for grid optimizing according to the second training set and verifying collection
Optimal model parameter constructs several acute respiratory distress syndrome anticipated mortality models using random forests algorithm；
Each acute respiratory distress syndrome anticipated mortality model is respectively adopted to test test set, obtains each folding
Prediction result；
It averages to the prediction result of each folding, it is corresponding to obtain each acute respiratory distress syndrome anticipated mortality model
Estimated performance result；
According to obtained estimated performance as a result, being selected from each acute respiratory distress syndrome anticipated mortality model pre-
The best model of results of property is surveyed as final acute respiratory distress syndrome anticipated mortality model.
It specifically, can more than one using the acute respiratory distress syndrome anticipated mortality model of random forests algorithm building
It is a, therefore model discrimination can be carried out according to the estimated performance result of each prediction model.
When each folding of K folding cross-validation method is using training set building model, the method for grid optimizing will use to seek
Looking for optimal model parameter, (parameter of RF includes decision tree number n_estimators, branch standard criterion, most leaflet
Subsample number min_sample_leaf etc.).So the present embodiment can be again divided into the first training set the second training set and test
Card collection two parts, the effect of every group of parameter of loop test select the optimal one group of parameter building random forest classification of AUROC effect
Device model (i.e. acute respiratory distress syndrome anticipated mortality model), tests test set, obtains this folding prediction result；
Then it averages to the result of K folding, obtains the estimated performance result of each model；Finally selection estimated performance result is best
Model is as final anticipated mortality model.
In order to improve the accuracy rate of ARDS mortality prediction, real-time and feasible prediction letter is provided for clinician
Breath, this specific embodiment propose acute respiratory distress syndrome anticipated mortality scheme.Below by the specific reality to the program
Existing process and process for using are described.
(1) specific implementation flow
The scheme specific implementation flow of this specific embodiment can be divided into three steps: (1) data collection and data prediction；(2)
Prediction model is established and test；(3) outcome evaluation is compared with.
1. data collection and data prediction
1.1 data source
This specific embodiment is from public database MIMIC-III (Medical Information Mart for
Intensive Care) 3186 ARDS patient datas of middle downloading.MIMIC-III be by MIT the laboratory physionet,
What BIDMC (Beth IsraelDeaconess Medical Center) and PHILIPS Co. built jointly is directed to Intensive Care Therapy
The database of patient, database contains the medical data of 53423 adult patients at present, provides including basic demography
The data such as material, vital sign monitoring.
1.2 data prediction
This specific embodiment is included in standard and exclusion criteria and is screened to ARDS patient data according to shown in Fig. 2, with
It selects and meets current ARDS diagnostic criteria, and meet the requirement of subsequent builds model, and record content can be in chain hospital
ICU realizes the sample of (ensuring that model can be with practical application).
The standard of being included in specifically includes following two and requires and (need to meet two requirements simultaneously to be just included in the data):
1) age moves in intensive care unit patient greater than 18 one full year of life (containing for 18 one full year of life)；
It 2) is ARDS patient, diagnostic criteria: a) acute attack through Berlin standard diagnostics；B) PEEP (or CPAP) >=5cmH2O
When, OI (PaO2/FiO2) < 300mmHg；C) Chest Image bilateral infiltrates shadow；D) respiratory failure can not be with heart failure come completely
It explains.
Exclusion criteria is specially that any of following four requirement (will data row as long as there is 1 requirement to be unsatisfactory for
Except):
1) data record is imperfect, such as receives the patient of invasive ventilation (without mechanical ventilation data)；
2) age is less than 18 one full year of life；
3) palliative treatment does not receive active treatment；
4) ICU recorded the time less than 48 hours.
After having screened out undesirable patient data, it there remains 475 patients for establishing prediction model.In order to guarantee
Variable for modeling is meaningful, this specific embodiment extracts clinically related with ARDS face under the confirmation of clinician
Bed data, totally 101 variable informations, can be read directly including the age of patient, gender, APACHE II score etc.
Data also include input variable of the variables such as Mechanical ventilation time, patient physiological information as modeling.
Mechanical ventilation time (length of mechanical ventilation) in this 101 variables is that patient is true
The mechanical ventilation duration for the first time after examining ARDS.Without mechanical number of days (days free of mechanical
Ventilation the number of days for) not receiving (any type of) mechanical ventilation for patient in ICU record, if patient is in tube drawing
Dead in 24 hours afterwards, no mechanical number of days regards as 0.Physiologic information (physiologic information) is ARDS breaking-out
The physiological data recorded before.Ventilation environment (Ventilator settings) depends on patient and receives mechanical ventilation initial 24
The instrument setting of hour.After ARDS occurs 24 hours, in standard ventilation environment (FiO2 >=0.5；PEEP >=5cm H2O) under calculate
PaO2/FiO2 value, as oxygenation index OI.
Due to instrument and data management, the case where there are individual data missings, this specific embodiment is using multiple
The data of the method supplement missing of interpolation (Multiple imputation).Multiple interpolation is that a kind of pair of missing data is adjusted
Whole method fills each missing data value using a series of possible data sets, then uses the standard of complete data
It goes to analyze multiple interpolation data collection, synthesis finally is concluded to these analysis results.This specific embodiment can be used SPSS software complete
It is operated at multiple interpolation.
Above step is the data prediction process of this specific embodiment, and 101 kinds of changes of every patient finally can be obtained
Measure feature is for establishing prediction model.
2. prediction model is established and test
2.1 sample classification
According to survival of patients number of days, three classes can be divided the data into, and (hospital mortality is predicted using three prediction models
Model, 30 days anticipated mortality models and an annual death rate prediction model) it is predicted respectively:
Model 1 (hospital mortality prediction model): survival during death vs. is hospitalized in being hospitalized, wherein the former is the positive, after
Person is feminine gender；
Model 2 (30 days anticipated mortality models): surviving after vs.30 days dead in 30 days after being hospitalized, the former is the positive, after
Person is feminine gender；
Model 3 (an annual death rate prediction model): surviving after death vs. 1 year in latter year of being hospitalized, the former is the positive, after
Person is feminine gender.
The process of the Feature Selections of these three prediction models, classifier training (i.e. model training) and test, outcome evaluation
It is almost the same.
2.2 Feature Selection
This specific embodiment is analyzed carrying out group to feature using SPSS software, and it is significant to find out different classes of inter-sample difference
Feature as input, can reduce data redundancy, reduce model and calculate, find out more meaningful feature.This specific embodiment
Respectively positive group in classification three obtained prediction model and negative group analyze between group, concrete operations are: for symbol
The continuous variable for closing normal distribution examines (Student t-test) analysis using t, and the continuous variable of Non-Gaussian Distribution uses
Non-parametric test (Mann-Whitney U test)；Then pass through Chi-square Test (Chi2test) or Fisher ' for discrete variable
S exact test is analyzed.Characteristics of variables of the P value less than 0.05 by above-mentioned inspection is believed that group difference is significant, then
Retain this feature；This feature is unobvious in the group difference of positive group and feminine gender group to be thought greater than 0.05 for P value, to prediction
The classification influence of model is smaller, can leave out.
2.3 model trainings and test
This specific embodiment using random forest (Random Forest, RF) respectively to the hospital mortality of ARDS patient,
30 days death rates and an annual death rate establish prediction model.RF is a kind of machine learning algorithm, can generate more decisions at random
It sets (Decision tree), every decision tree is all a classifier, can be carried out by a series of data of decisions to input pre-
It surveys, distributes label, the output result of last RF then passes through decision tree " ballot " generation.This specific embodiment uses scikit-
Learn Python library kit realizes RF algorithm, as shown in Figure 3.It include several decision trees in Fig. 3, to each sample
This x, each tree can all provide the prediction result of oneself, and each tree " ballot " determines final result y.
It establishes prediction model and is broadly divided into two steps: training (training) and test (testing).
In order to guarantee the reliable of prediction model and stablize, this specific embodiment has used eight folding cross-validation method (8-folds
Cross-validation) prediction effect of model is evaluated, overall procedure is as shown in figure 4, specifically include:
Divide the data into close 8 parts of classification ratio first, every folding use wherein 7 parts of data as the trained mould of training set
Type is left 1 part of data and is used for test model effect (i.e. test set).The test set and training set of each folding training are all different, and one
8 times (Loop 2 in such as Fig. 4) is recycled altogether, and the final result of model can be found out on the basis of eight folding cross validations.The process
Feature Selection carried out respectively in each folding training set in machine learning cross-validation process, therefore it is it is possible that every
The different situation of feature used in one folding.
When each folding is using training set structure prediction established model, the method for grid optimizing can be used to find optimal model
(parameter of RF includes decision tree number n_estimators, branch standard criterion, minimum leaf sample number min_ to parameter
Sample_leaf etc.).So training set is again divided into training set and verifying collection two parts, circulation by this specific implementation regular meeting
The effect (Loop 1 in such as Fig. 4) of every group of parameter is tested, the optimal one group of parameter of AUROC effect is selected to construct each classification
Device model (i.e. prediction model), then tests test set, obtains this folding prediction result.
It averages again to the result of 8 foldings, obtains the estimated performance result of each prediction model.
The model for finally selecting estimated performance result best from each prediction model is as final prediction model.Due to
This specific embodiment splits data into three classes and 3 different models has been respectively trained, so every one kind can all obtain one most
Whole prediction result.
3. outcome evaluation is compared with
3.1 classification results (i.e. model prediction result) assessment
The evaluation criteria of this specific embodiment classification results (i.e. model prediction result) is AUROC.The horizontal axis of ROC curve is
False sun rate (False Positive Rate, FPR), the longitudinal axis are kidney-Yang rate (True Positive Rate, TPR), on curve
Point is that the TPR showed under different classification thresholds according to the probability output of sample and FPR determine (to work as output probability
When more than or equal to given threshold, otherwise it is feminine gender which, which is the positive).It is more and more as classification thresholds are gradually reduced
Sample be predicted to be the positive, but real negative sample is equally adulterated in these positives, i.e. TPR and FPR can increase simultaneously
Greatly.When threshold value maximum, corresponding ROC curve coordinate points are (0,0), respective coordinates point (1,1) when threshold value minimum, and dreamboat
It is TPR=1, FPR=0, corresponding ROC curve coordinate points are (0,1), so this specific embodiment selects most in model training
Close to probability value representated by the point on the ROC curve of coordinate points (0,1) as classification thresholds.
AUROC is area under ROC curve, can be used for classification of assessment device performance.A positive sample and feminine gender are selected at random
Sample is input among prediction model, exports the prediction probability of two samples, and positive sample is come feminine gender by descending arrangement
Probability before sample is exactly AUC value (namely positive sample output probability is greater than a possibility that negative sample output probability).
, can be in the hope of the accuracy rate of classification, susceptibility, specificity meanwhile according to optimal classification threshold value, calculation formula is such as
Under:
Accuracy rate: Accuracy=(TP+TN)/(TP+TN+FP+FN)
Susceptibility: Sensitivity=TP/ (TP+FN)
Specificity: Specificity=TN/ (TN+FP)
Wherein, TP:True Positive, true positives, i.e., practical is the positive, is predicted as positive sample.
FP:False Positive, false positive, i.e., practical is feminine gender, is predicted as positive sample.
TN:True Negative, true negative, i.e., practical is feminine gender, is predicted as negative sample.
FN:False Negative, false negative, i.e., practical is the positive, is predicted as negative sample.
Using the machine learning method training anticipated mortality model based on RF, it is as follows result can be obtained after tested:
1) the AUROC value of hospital mortality prediction model is that 0.854 (95% confidence interval is 0.835-0.874, p <
0.001)；
2) the AUROC value of 30 days anticipated mortality models is that 0.817 (95% confidence interval is 0.796-0.839, p <
0.001)；
3) the AUROC value of an annual death rate prediction model is that 0.817 (95% confidence interval is 0.800-0.834, p <
0.001)。
3.2 methods compare
For more of the invention and existing method prediction effect, this specific embodiment has also reappeared in existing research respectively
The model about ARDS anticipated mortality mentioned, is predicted with batch of data.Existing prediction model includes SAPS
As a result II, OI, OSI and APPS show as AUROC, and respectively compared with the prediction result of RF.
Statistical disposition is carried out using SPSS software when prediction.Meet the continuous data of normal distribution with means standard deviation
(mean ± std) is indicated；The continuous data for not meeting normal distribution is then indicated with median (quartile)；Enumeration data is to divide
The performance of several or percents.With the age, gender and APACHE II score are as control variable, using multivariable logistic regression
Method analyzes the relationship of each factor and the death rate, and draws ROC curve.
The result of above-mentioned various prediction techniques is as shown in table 1 below:
Table 1
Obviously, as it can be seen from table 1 the present invention predicts ARDS mortality using the machine learning method based on RF,
Effect is generally better than existing method.
4. the application of prediction model
From the above results, it using the prediction model of machine learning method training, can effectively be hospitalized extremely to ARDS
Die, 30 days it is dead and death in 1 year is predicted, and relative to other existing prediction techniques, the prediction of method of the invention
Accuracy rate increases significantly.The result illustrates that the modelling scheme of this specific embodiment with process is feasible, so
This specific embodiment can utilize existing whole ARDS data, according to above scheme and process, one final mask of training, to new
The data come are predicted, to realize application clinically, detailed process as shown in figure 5, by grouping to available data,
Feature Selection, parameter optimization, available three prediction models are model 1 respectively (for predicting ARDS patient's Death
Probability), model 2 (dead probability in prediction 30 days) and model 3 (predicting probability dead in 1 year).
When needing to predict new data, Feature Selection (three kinds of sieves when according to training are carried out to new data first
Choosing strategy), the same feature is selected, then be separately input in corresponding model, and then realize the prediction to new data.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of acute respiratory distress syndrome death rate is pre-
Examining system comprises the following modules:
Module is obtained, for obtaining the medical data of Patients with ARDS；
Training module obtains urgency for carrying out model training using the method for machine learning according to the medical data of acquisition
Property Respiratory Distress Syndrome(RDS) anticipated mortality model；
Prediction module, it is pre- for being carried out using acute respiratory distress syndrome anticipated mortality model to object to be predicted
It surveys, obtains the prediction result of the acute respiratory distress syndrome death rate.
Suitable for this system embodiment, this system embodiment is implemented content in above method embodiment
Function is identical as above method embodiment, and the beneficial effect reached and above method embodiment beneficial effect achieved
It is identical.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of acute respiratory distress syndrome death rate is pre-
Examining system, comprising:
At least one processor；
At least one processor, for storing at least one program；
When at least one described program is executed by least one described processor, so that at least one described processor is realized
Acute respiratory distress syndrome anticipated mortality method as described in the present invention.
Suitable for this system embodiment, this system embodiment is implemented content in above method embodiment
Function is identical as above method embodiment, and the beneficial effect reached and above method embodiment beneficial effect achieved
It is identical.
It is to be illustrated to preferable implementation of the invention, but the present invention is not limited to the embodiment above, it is ripe
Various equivalent deformation or replacement can also be made on the premise of without prejudice to spirit of the invention by knowing those skilled in the art, this
Equivalent deformation or replacement are all included in the scope defined by the claims of the present application a bit.
Claims (10)
1. a kind of acute respiratory distress syndrome anticipated mortality method, it is characterised in that: the following steps are included:
Obtain the medical data of Patients with ARDS；
Model training is carried out using the method for machine learning according to the medical data of acquisition, it is dead to obtain acute respiratory distress syndrome
Die rate prediction model；
Object to be predicted is predicted using acute respiratory distress syndrome anticipated mortality model, it is embarrassed to obtain acute respiration
Compel the prediction result of the syndrome death rate.
2. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 1, it is characterised in that: described
The step for obtaining the medical data of Patients with ARDS, specifically:
The medical data of Patients with ARDS is downloaded from MIMIC-III database.
3. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 1, it is characterised in that: described
Model training is carried out using the method for machine learning according to the medical data of acquisition, obtains the acute respiratory distress syndrome death rate
The step for prediction model, specifically includes:
The medical data of acquisition is pre-processed, the pretreatment includes screening sample and feature extraction；
Model training is carried out using the method for machine learning according to pretreated data, it is dead to obtain acute respiratory distress syndrome
Rate prediction model is died, the acute respiratory distress syndrome anticipated mortality model includes hospital mortality prediction model, 30 days
Anticipated mortality model and an annual death rate prediction model.
4. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 3, it is characterised in that: described
The step for pre-processing to the medical data of acquisition, specifically includes:
Screening sample is carried out to the medical data of acquisition according to standard of being included in and exclusion criteria, obtains the sample screened, it is described
The standard of being included in includes being more than or equal to moving in intensive care unit and being acute respiratory distress through Berlin standard diagnostics for 18 one full year of life the age
The patient of syndrome, the exclusion criteria include in MIMIC-III database the incomplete data of data record, age less than 18
The patient of one full year of life, using the patient of palliative treatment and ICU record patient of the time less than 48 hours in any one；
Characteristics of variables to each sample of the sample extraction screened for modeling.
5. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 4, it is characterised in that: described
The step for medical data of acquisition is pre-processed, and also specifically includes the following steps:
Multiple interpolation is carried out to the missing data in the medical data of acquisition.
6. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 3, it is characterised in that: described
Model training is carried out using the method for machine learning according to pretreated data, obtains the acute respiratory distress syndrome death rate
The step for prediction model, specifically includes:
Classified according to survival of patients number of days to pretreated data, respectively obtains the positive of 3 anticipated mortality models
Group and negative group, 3 anticipated mortality models include hospital mortality prediction model, 30 days anticipated mortality models and one
Annual death rate prediction model；
Positive group to 3 anticipated mortality models and negative group analyze between group respectively, and it is significant to filter out group difference
Feature；
Acute respiratory distress syndrome anticipated mortality mould is established using random forests algorithm according to group difference significant feature
Type.
7. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 6, it is characterised in that: described
The positive group and negative group of 3 anticipated mortality models specifically: positive group of hospital mortality prediction model is interior dead in hospital
The patient data died, negative group of hospital mortality prediction model are the patient datas that period survives in hospital；The death rate is pre- within 30 days
Survey positive group of model be in hospital after patient data dead in 30 days, negative group of 30 days anticipated mortality models be in hospital after
The patient data survived in 30 days；Positive group of one annual death rate prediction model is patient data dead in latter year in hospital,
Negative group of one annual death rate prediction model is the patient data of interior survival of latter year in hospital.
8. a kind of acute respiratory distress syndrome anticipated mortality method according to claim 6, it is characterised in that: described
According to the significant feature of group difference using random forests algorithm establish acute respiratory distress syndrome anticipated mortality model this
One step, specifically includes:
The significant feature of group difference is divided by the first training set and test set using K folding cross-validation method；
First training set is divided by the second training set using K folding cross-validation method and verifying collects；
Optimal model parameter is searched out using the method for grid optimizing according to the second training set and verifying collection, and then according to best
Model parameter several acute respiratory distress syndrome anticipated mortality models are constructed using random forests algorithm；
Each acute respiratory distress syndrome anticipated mortality model is respectively adopted to test test set, obtains the pre- of each folding
Survey result；
It averages to the prediction result of each folding, it is corresponding pre- to obtain each acute respiratory distress syndrome anticipated mortality model
Survey results of property；
According to obtained estimated performance as a result, being selected from each acute respiratory distress syndrome anticipated mortality model predictive
Can the best model of result as final acute respiratory distress syndrome anticipated mortality model.
9. a kind of acute respiratory distress syndrome anticipated mortality system, it is characterised in that: comprise the following modules:
Module is obtained, for obtaining the medical data of Patients with ARDS；
Training module obtains acute exhale for carrying out model training using the method for machine learning according to the medical data of acquisition
Inhale Distress syndrome anticipated mortality model；
Prediction module, for being predicted using acute respiratory distress syndrome anticipated mortality model object to be predicted,
Obtain the prediction result of the acute respiratory distress syndrome death rate.
10. a kind of acute respiratory distress syndrome anticipated mortality system, it is characterised in that: include:
At least one processor；
At least one processor, for storing at least one program；
When at least one described program is executed by least one described processor, so that at least one described processor is realized as weighed
Benefit requires the described in any item acute respiratory distress syndrome anticipated mortality methods of 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910194628.2A CN110051324B (en) | 2019-03-14 | 2019-03-14 | Method and system for predicting death rate of acute respiratory distress syndrome |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910194628.2A CN110051324B (en) | 2019-03-14 | 2019-03-14 | Method and system for predicting death rate of acute respiratory distress syndrome |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110051324A true CN110051324A (en) | 2019-07-26 |
| CN110051324B CN110051324B (en) | 2022-06-10 |
Family
ID=67316101
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910194628.2A Active CN110051324B (en) | 2019-03-14 | 2019-03-14 | Method and system for predicting death rate of acute respiratory distress syndrome |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110051324B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111508604A (en) * | 2020-04-20 | 2020-08-07 | 深圳大学 | Acute kidney injury patient mortality prediction method, server and storage medium |
| CN111657888A (en) * | 2020-05-28 | 2020-09-15 | 首都医科大学附属北京天坛医院 | Severe acute respiratory distress syndrome early warning method and system |
| CN112992368A (en) * | 2021-04-09 | 2021-06-18 | 中山大学附属第三医院（中山大学肝脏病医院） | Prediction model system and recording medium for prognosis of severe spinal cord injury |
| CN112992346A (en) * | 2021-04-09 | 2021-06-18 | 中山大学附属第三医院（中山大学肝脏病医院） | Method for establishing prediction model for prognosis of severe spinal cord injury |
| CN113066584A (en) * | 2021-03-31 | 2021-07-02 | 中国福利会国际和平妇幼保健院 | Prediction method and system for early septicemia |
| CN113299390A (en) * | 2021-05-20 | 2021-08-24 | 广东省科学院智能制造研究所 | System and method for predicting in-hospital mortality of acute kidney injury patient |
| WO2022227280A1 (en) * | 2021-04-30 | 2022-11-03 | 中国人民解放军总医院第一医学中心 | Smart glasses-based disaster rescue triage and auxiliary diagnosis method |
| TWI787130B (en) * | 2022-05-16 | 2022-12-11 | 臺中榮民總醫院 | Respiratory status classifying method and system thereof |
| CN115662613A (en) * | 2022-09-28 | 2023-01-31 | 中日友好医院(中日友好临床医学研究所) | Barotrauma prediction method and device |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104951894A (en) * | 2015-06-25 | 2015-09-30 | 成都厚立信息技术有限公司 | Intelligent analysis and assessment system for disease management in hospital |
| CN105046406A (en) * | 2015-06-25 | 2015-11-11 | 成都厚立信息技术有限公司 | Inpatient medical management quality assessment method |
| US20170055916A1 (en) * | 2015-09-01 | 2017-03-02 | Xerox Corporation | Methods and systems for predicting mortality of a patient |
| US20170177822A1 (en) * | 2015-12-18 | 2017-06-22 | Pointright Inc. | Systems and methods for providing personalized prognostic profiles |
| CN106947830A (en) * | 2017-05-16 | 2017-07-14 | 中山大学肿瘤防治中心 | Gene methylation panel for diagnosing, predicting therapeutic efficacy for hepatic carcinoma and prognosis |
| CN107657149A (en) * | 2017-09-12 | 2018-02-02 | 中国人民解放军军事医学科学院生物医学分析中心 | System for predicting liver cancer patient prognosis |
| CN107683341A (en) * | 2015-05-08 | 2018-02-09 | 新加坡科技研究局 | method for the diagnosis and prognosis of chronic heart failure |
| CN107887029A (en) * | 2017-12-05 | 2018-04-06 | 深圳大学 | Disease forecasting method and device |
| CN108461144A (en) * | 2017-12-21 | 2018-08-28 | 深圳大学 | Chinese population low-level radiation suffers from the appraisal procedure and system of cancer risk |
| CN108682457A (en) * | 2018-04-17 | 2018-10-19 | 中国医学科学院阜外医院 | Patient's long-term prognosis quantitative forecast and interfering system and method |
| AU2018101578A4 (en) * | 2018-10-22 | 2018-11-29 | Menon, Vineet MR | Machine learning applied to Cervical Cancer Data |
| CN109119167A (en) * | 2018-07-11 | 2019-01-01 | 山东师范大学 | Pyemia anticipated mortality system based on integrated model |
| KR20190009166A (en) * | 2017-07-18 | 2019-01-28 | 사회복지법인 삼성생명공익재단 | Method, Apparatus and Program for Predicting Prognosis of Ovarian Cancer Using Machine Learning |
-
2019
- 2019-03-14 CN CN201910194628.2A patent/CN110051324B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107683341A (en) * | 2015-05-08 | 2018-02-09 | 新加坡科技研究局 | method for the diagnosis and prognosis of chronic heart failure |
| CN105046406A (en) * | 2015-06-25 | 2015-11-11 | 成都厚立信息技术有限公司 | Inpatient medical management quality assessment method |
| CN104951894A (en) * | 2015-06-25 | 2015-09-30 | 成都厚立信息技术有限公司 | Intelligent analysis and assessment system for disease management in hospital |
| US20170055916A1 (en) * | 2015-09-01 | 2017-03-02 | Xerox Corporation | Methods and systems for predicting mortality of a patient |
| US20170177822A1 (en) * | 2015-12-18 | 2017-06-22 | Pointright Inc. | Systems and methods for providing personalized prognostic profiles |
| CN106947830A (en) * | 2017-05-16 | 2017-07-14 | 中山大学肿瘤防治中心 | Gene methylation panel for diagnosing, predicting therapeutic efficacy for hepatic carcinoma and prognosis |
| KR20190009166A (en) * | 2017-07-18 | 2019-01-28 | 사회복지법인 삼성생명공익재단 | Method, Apparatus and Program for Predicting Prognosis of Ovarian Cancer Using Machine Learning |
| CN107657149A (en) * | 2017-09-12 | 2018-02-02 | 中国人民解放军军事医学科学院生物医学分析中心 | System for predicting liver cancer patient prognosis |
| CN107887029A (en) * | 2017-12-05 | 2018-04-06 | 深圳大学 | Disease forecasting method and device |
| CN108461144A (en) * | 2017-12-21 | 2018-08-28 | 深圳大学 | Chinese population low-level radiation suffers from the appraisal procedure and system of cancer risk |
| CN108682457A (en) * | 2018-04-17 | 2018-10-19 | 中国医学科学院阜外医院 | Patient's long-term prognosis quantitative forecast and interfering system and method |
| CN109119167A (en) * | 2018-07-11 | 2019-01-01 | 山东师范大学 | Pyemia anticipated mortality system based on integrated model |
| AU2018101578A4 (en) * | 2018-10-22 | 2018-11-29 | Menon, Vineet MR | Machine learning applied to Cervical Cancer Data |
Non-Patent Citations (8)
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111508604A (en) * | 2020-04-20 | 2020-08-07 | 深圳大学 | Acute kidney injury patient mortality prediction method, server and storage medium |
| CN111657888A (en) * | 2020-05-28 | 2020-09-15 | 首都医科大学附属北京天坛医院 | Severe acute respiratory distress syndrome early warning method and system |
| CN113066584A (en) * | 2021-03-31 | 2021-07-02 | 中国福利会国际和平妇幼保健院 | Prediction method and system for early septicemia |
| CN112992368A (en) * | 2021-04-09 | 2021-06-18 | 中山大学附属第三医院（中山大学肝脏病医院） | Prediction model system and recording medium for prognosis of severe spinal cord injury |
| CN112992346A (en) * | 2021-04-09 | 2021-06-18 | 中山大学附属第三医院（中山大学肝脏病医院） | Method for establishing prediction model for prognosis of severe spinal cord injury |
| WO2022227280A1 (en) * | 2021-04-30 | 2022-11-03 | 中国人民解放军总医院第一医学中心 | Smart glasses-based disaster rescue triage and auxiliary diagnosis method |
| CN113299390A (en) * | 2021-05-20 | 2021-08-24 | 广东省科学院智能制造研究所 | System and method for predicting in-hospital mortality of acute kidney injury patient |
| TWI787130B (en) * | 2022-05-16 | 2022-12-11 | 臺中榮民總醫院 | Respiratory status classifying method and system thereof |
| US11908136B2 (en) | 2022-05-16 | 2024-02-20 | Taichung Veterans General Hospital | Respiratory status classifying method and system thereof |
| CN115662613A (en) * | 2022-09-28 | 2023-01-31 | 中日友好医院(中日友好临床医学研究所) | Barotrauma prediction method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110051324B (en) | 2022-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110051324A (en) | A kind of acute respiratory distress syndrome anticipated mortality method and system | |
| US20220254493A1 (en) | Chronic disease prediction system based on multi-task learning model | |
| CN107887032A (en) | A kind of data processing method and device | |
| CN111370126A (en) | ICU mortality prediction method and system based on penalty integration model | |
| CN112967803A (en) | Early mortality prediction method and system for emergency patients based on integrated model | |
| CN112652391A (en) | System for identifying acute exacerbation of chronic obstructive pulmonary disease | |
| CN117116477A (en) | Construction method and system of prostate cancer disease risk prediction model based on random forest and XGBoost | |
| CN114943629A (en) | Health management and health care service system and health management method thereof | |
| CN114191665A (en) | Method and device for classifying man-machine asynchronous phenomena in mechanical ventilation process | |
| Papageorgiou | A Fuzzy Inference Map approach to cope with uncertainty in modeling medical knowledge and making decisions | |
| CN109192312A (en) | A kind of heart failure patient adverse events intelligent management system and method | |
| CN117116475A (en) | Method, system, terminal and storage medium for predicting risk of ischemic cerebral apoplexy | |
| Wang et al. | Method of non-invasive parameters for predicting the probability of early in-hospital death of patients in intensive care unit | |
| CN115633961A (en) | Construction method and system based on dynamic weighted decision fusion high-fear recognition model | |
| CN116013517A (en) | Method for quantifying health score and health care of old people based on IOT (information of interest) evaluation data | |
| CN114927230A (en) | Machine learning-based severe heart failure patient prognosis decision support system and method | |
| Perng et al. | Analysis of the 72-h mortality of emergency room septic patients based on a deep belief network | |
| Umut et al. | Prediction of sepsis disease by Artificial Neural Networks | |
| Berdaly et al. | Comparative machine-learning approach: study for heart diseases | |
| Khalaf et al. | Predicting Acute Respiratory Failure Using Fuzzy Classifier | |
| Han et al. | Intelligent injury prediction for traumatic airway obstruction | |
| CN118136270B (en) | Data analysis based health monitoring and early warning system and method for chronic diseases | |
| Dai | Evaluating the effect of different Mode's attributes on the subjective classification in the case of TCM | |
| Lind | ASSESSING PREDICTION CONDITIONS ANDSEQUENTIAL CLASSIFICATION IN ICU SEPSISPREDICTION | |
| CN113782210B (en) | Method for predicting treatment failure probability of noninvasive ventilator |
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 |