CN115602299A - ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning - Google Patents
ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning Download PDFInfo
- Publication number
- CN115602299A CN115602299A CN202211267598.1A CN202211267598A CN115602299A CN 115602299 A CN115602299 A CN 115602299A CN 202211267598 A CN202211267598 A CN 202211267598A CN 115602299 A CN115602299 A CN 115602299A
- Authority
- CN
- China
- Prior art keywords
- data
- intervention means
- time
- prediction
- ehr
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000013135 deep learning Methods 0.000 title claims abstract description 21
- 238000003062 neural network model Methods 0.000 claims abstract description 22
- 230000008859 change Effects 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 11
- 230000003068 static effect Effects 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 9
- 230000001364 causal effect Effects 0.000 claims abstract description 7
- 238000004140 cleaning Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 7
- 239000003814 drug Substances 0.000 claims description 5
- 230000001965 increasing effect Effects 0.000 claims description 5
- 238000011545 laboratory measurement Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 19
- 238000009423 ventilation Methods 0.000 description 4
- 208000004756 Respiratory Insufficiency Diseases 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 201000004193 respiratory failure Diseases 0.000 description 3
- 206010021143 Hypoxia Diseases 0.000 description 2
- GXBMIBRIOWHPDT-UHFFFAOYSA-N Vasopressin Natural products N1C(=O)C(CC=2C=C(O)C=CC=2)NC(=O)C(N)CSSCC(C(=O)N2C(CCC2)C(=O)NC(CCCN=C(N)N)C(=O)NCC(N)=O)NC(=O)C(CC(N)=O)NC(=O)C(CCC(N)=O)NC(=O)C1CC1=CC=CC=C1 GXBMIBRIOWHPDT-UHFFFAOYSA-N 0.000 description 2
- 102000002852 Vasopressins Human genes 0.000 description 2
- 108010004977 Vasopressins Proteins 0.000 description 2
- KBZOIRJILGZLEJ-LGYYRGKSSA-N argipressin Chemical compound C([C@H]1C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](CC(N)=O)C(=O)N[C@@H](CSSC[C@@H](C(N[C@@H](CC=2C=CC(O)=CC=2)C(=O)N1)=O)N)C(=O)N1[C@@H](CCC1)C(=O)N[C@@H](CCCN=C(N)N)C(=O)NCC(N)=O)C1=CC=CC=C1 KBZOIRJILGZLEJ-LGYYRGKSSA-N 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 229940079593 drug Drugs 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 229960003726 vasopressin Drugs 0.000 description 2
- 206010002660 Anoxia Diseases 0.000 description 1
- 241000976983 Anoxia Species 0.000 description 1
- 208000006545 Chronic Obstructive Pulmonary Disease Diseases 0.000 description 1
- 206010022773 Intracranial pressure increased Diseases 0.000 description 1
- 208000019693 Lung disease Diseases 0.000 description 1
- 206010037423 Pulmonary oedema Diseases 0.000 description 1
- 206010038669 Respiratory arrest Diseases 0.000 description 1
- 206010047139 Vasoconstriction Diseases 0.000 description 1
- 230000001154 acute effect Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000007953 anoxia Effects 0.000 description 1
- 230000004872 arterial blood pressure Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 208000006673 asthma Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 230000036772 blood pressure Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 230000001269 cardiogenic effect Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000001146 hypoxic effect Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 210000004072 lung Anatomy 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000005399 mechanical ventilation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 206010028417 myasthenia gravis Diseases 0.000 description 1
- 208000018360 neuromuscular disease Diseases 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
- 230000001314 paroxysmal effect Effects 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 208000005333 pulmonary edema Diseases 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000000241 respiratory effect Effects 0.000 description 1
- 208000026425 severe pneumonia Diseases 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 208000011580 syndromic disease Diseases 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000002627 tracheal intubation Methods 0.000 description 1
- 230000025033 vasoconstriction Effects 0.000 description 1
- 230000009278 visceral effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses an ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning, which belongs to the technical field of data processing and comprises the steps of obtaining an MIMIC-III database and extracting EHR (electric discharge reactor) data of a patient in an intensive care unit; cleaning and processing the EHR data and converting the EHR data into time sequence data; constructing a neural network model consisting of a graph learning layer, an expansion causal convolution, an output module and a full connection layer, and training and updating; and (3) taking the time sequence data as input, generating a prediction judgment value by a neural network model, and providing guidance for an intervention means. The invention constructs a neural network model based on a deep learning algorithm, takes various clinical physiological indexes of a patient in an ICU as characteristic nodes with time sequence change, takes physiological information of the patient as static unchangeable characteristic nodes, and processes the physiological information into a group of structured data suitable for being put into the neural network model, thereby predicting an intervention means.
Description
Technical Field
The invention belongs to the technical field of medical treatment, and particularly relates to an ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning.
Background
With the increased care unit integrated unit (ICU) playing an increasing role in acute medical services, clinicians are required to predict the physiological needs of patients in a fast-paced, data-cluttered environment, making judgments and interventions as soon as possible at the optimal rescue time for the disease. The secondary analysis of medical data is a key step for improving modern medical treatment and increasing judgment accuracy. The traditional analysis based on the vital signs of the patient is basically completed by manpower, and doctors need to perform observation analysis and prediction diagnosis through self experience and a great deal of related field knowledge.
Especially in the context of the ICU, the paroxysmal nature of various diseases and the uncertainty of vital signs are undoubtedly time-consuming and laborious work for medical personnel. Patients in ICU are often afflicted with pulmonary diseases such as asthma, chronic Obstructive Pulmonary Disease (COPD), hypoxic respiratory failure, cardiogenic Pulmonary Edema (CPE); neuromuscular diseases such as myasthenia gravis, guillain, barre syndrome; symptoms of respiratory failure such as increased intracranial pressure occur, and the time from respiratory failure to respiratory arrest is extremely short. Particularly, in the current generation, the tolerance of severe pneumonia patients to anoxia is reduced, the intubation time should be controlled within 2 minutes as much as possible, and if an intervention means is not adopted in time, the death rate is extremely high.
Therefore, the invention provides an ICU auxiliary intervention means prediction method based on deep learning, which at least solves part of technical problems.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: an ICU auxiliary intervention means prediction method based on deep learning is provided to solve at least some technical problems.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
an ICU auxiliary intervention means prediction method based on deep learning comprises the following steps:
step 1, obtaining an MIMIC-III database, and extracting EHR data of a patient in an intensive care unit;
step 2, cleaning and processing the EHR data and converting the EHR data into time sequence data;
step 3, constructing a neural network model consisting of a graph learning layer, an expansion causal convolution, an output module and a full connection layer, and training and updating;
and 4, taking the time sequence data as input, generating a prediction judgment value by the neural network model, and providing guidance for an intervention means.
Further, in the step 1, the EHR data includes static characteristics, time-series characteristics, and intervention means state characteristics; the static characteristics are physiological information with invariable time sequence; the time series characteristic is a clinical physiological index of the patient associated with the prediction, and at least comprises time series change physiological characteristic data, laboratory measurement data and medicine data.
Further, the step 2 specifically includes: step 21, selecting EHR data of a patient with hospitalization time exceeding a first time threshold A from a MIMIC-III database; step 22, setting a second time threshold B with the hospitalization time greater than the first time threshold a, performing primary sampling on EHR data with the hospitalization time less than the second time threshold B, performing secondary sampling on EHR data between B hours before the final point of the hospitalization time and the final point of the hospitalization time, and aligning the sampled data based on time sequence characteristics; step 23, standardizing the data after alignment processing, so that each type of EHR data on each timestamp is represented by values of three channels, wherein the values of the three channels include a mask for identifying whether EHR data exists in the current timestamp, a value of actual EHR data, and an accumulated time when EHR data is observed for the last time; and 24, constructing sample data with dimensionalities of [ the number of data pieces, the number of channels, the number of time sequence features used for prediction contained in the EHR data, and a second time threshold B ].
Further, the step 3 specifically includes: step 31, setting a neural network model comprising a graph learning layer, a plurality of sequentially connected expansion cause-and-effect convolutions and an output module, wherein the expansion cause-and-effect convolutions comprise sequentially connected time sequence convolution layers and graph convolution layers, and a 1 x 1 convolution layer is arranged in front of the first time sequence convolution layer; step 32, inputting one constructed sample data in each training, wherein the sample data promotes the channel number dimension through the 1 x 1 convolution layer, the sample data is input into the graph learning layer to obtain an adjacent matrix, and the adjacent matrix is applied to the input of a plurality of graph convolution layers; step 33, setting residual connection among a plurality of sequentially connected expansion causal convolutions; step 34, the output module sets a prediction window with a window size of N hours, and the output module uses the number of time sequence characteristics used for prediction and acted on EHR data by a plurality of full connection layers to obtain output data with a dimensionality of [ second time threshold B, N ]; and 35, inputting sample data, repeating the steps 32 to 35, and training the neural network model.
Further, the intervention instrument state feature includes an intervention instrument state tag value 1 for identifying off and an intervention instrument state tag value 0 for identifying on.
Further, the step 4 includes the following processes: step S41, setting the sampling interval time of EHR data to be M hours, inputting the EHR data to obtain a prediction judgment value with dimension of [ second time threshold B, N ], and obtaining B prediction windows based on the prediction judgment value, wherein each prediction window comprises an intervention means state label value within N hours; and acquiring a predicted intervention means based on the change rule of the state tag value of the intervention means.
Further, the change rule of the intervention means state tag value is that the intervention means is marked as off if the tag value is changed from 0 to 1, the intervention means is marked as on if the tag value is changed from 1 to 0, the intervention means is marked as off if the tag value is kept to 1, and the intervention means is marked as on if the tag value is kept to 0.
Further, based on the change rule of the intervention means state tag value, an intervention means tag is constructed as [ close, open, close keeping, open ].
Compared with the prior art, the invention has the following beneficial effects:
the invention constructs a neural network model based on a deep learning algorithm, takes various clinical physiological indexes of a patient in an ICU as characteristic nodes with time sequence change, takes physiological information of the patient as static unchangeable characteristic nodes, processes the physiological information into a group of structured data suitable for being put into the neural network model, and further predicts the intervention means required by the patient after a period of time. Compared with the existing algorithm, the neural network model has higher accuracy, applies the idea of the graph structure, has high interpretability and flexibility, and is suitable for wider medical environment.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
Technical term interpretation:
EHR data: electronic health profile data;
and the RF algorithm comprises the following steps: a random forest algorithm;
LR algorithm: a logistic regression algorithm;
CNN algorithm: a convolutional neural network algorithm;
the LSTM algorithm: and (4) long-short term memory network algorithm.
The invention relates to a method for making auxiliary intervention prediction for diagnosis under an ICU background by using a graph neural network in deep learning and based on an MIMIC-III database. The MIMIC-III database is an intensive care data set issued by the Massachusetts institute of technology and technology in a computational physiology laboratory, the data set of the third edition is issued in 2016, the third edition of data set comprises nearly 6 ten thousand records of ICU admission records of patients admitted by intensive care units in a large-scale three-level nursing hospital, the reference amount of the records exceeds 1400 times, and the data set is a high-quality data source for researchers developing medical big data analysis. The MIMIC-III database contains 26 data tables, and besides dictionary tables, the tables are connected by patient number (subject _ id), case number (hadm _ id) and ICU number (icustay _ id).
The invention extracts the EHR data of the patient in the intensive care unit in the MIMIC-III database to train and update the neural network model. The EHR data is at least 20000 bars and comprises static characteristics, time sequence characteristics and intervention means state characteristics; static characteristics are time-invariant physiological information such as gender, age, etc.; the time series characteristic is a clinical physiological index of the patient associated with the prediction, and at least comprises time series change physiological characteristic data, laboratory measurement data and medicine data. The static characteristics of the patient are extracted from icu _ detail, administrations and accustays of the MIMIC-III database, the time sequence characteristics of the adaptive task are extracted from chartevents and labevents, and the corresponding prediction result is extracted from the duration as a label.
The EHR data extracted from the MIMIC-III database has many noises, many deletions, a chaotic measurement method, and the like, and is not time-series data suitable for a deep learning network, so that the EHR data needs to be preprocessed. In the step 2, the MIMIC-Extract is adopted to convert the EHR data into time sequence data, and a prediction baseline is set by adopting a classic machine learning algorithm and a deep learning algorithm, wherein the classic machine learning algorithm is an RF algorithm and an LR algorithm, and the deep learning algorithm is a CNN algorithm and an LSTM algorithm.
The step of converting the EHR data into time sequence data after cleaning and processing the EHR data specifically comprises the following steps: step 21, selecting EHR data of a patient with hospitalization time exceeding a first time threshold A (for example, 24 hours) from the MIMIC-III data set; step 22, setting a second time threshold B (for example, 96 hours) that the hospitalization time is greater than the first time threshold a, for example, the hospitalization time is finally 100 hours, performing primary sampling on the EHR data that the hospitalization time is less than the second time threshold B (namely, performing primary sampling from 0 to 96 hours), performing secondary sampling on the EHR data from B hours before the final point of the hospitalization time to the final point of the hospitalization time (namely, performing secondary sampling from 4 to 100 hours), and performing alignment processing on the sampled data based on the time sequence characteristics; step 23, standardizing the data after the alignment processing, so that each type of EHR data on each timestamp is represented by values of three channels, where the values of the three channels include a mask for identifying whether EHR data exists in the current timestamp, a value of actual EHR data, and an accumulated time for last viewing of EHR data; and 24, constructing sample data with dimensions of [ the number of data, the number of channels, the number of time sequence features used for prediction contained in the EHR data, and a second time threshold B ]. After binarization, the static features are copied as the features with invariable time sequence at each time stamp and are aligned with the physiological feature data with time sequence variation and the state features of the intervention means. The dimension of the sample data obtained in the embodiment is [ 2401, 3, 123, 96], 123 features on 96 timestamps are represented, each feature is identified by data of 3 channels, and 2401 pieces of data are extracted in total.
And constructing a neural network model consisting of a graph learning layer, an expansion causal convolution, an output module and a full connection layer, and training and updating. The method specifically comprises the steps of 31, setting a neural network model comprising a graph learning layer, a plurality of expansion cause-and-effect convolutions and an output module which are sequentially connected, wherein the expansion cause-and-effect convolutions comprise a time sequence convolution layer and a graph convolution layer which are sequentially connected, and a 1 x 1 convolution layer is arranged in front of the first time sequence convolution layer; step 32, inputting one constructed sample data in each training, wherein the sample data promotes the channel number dimension through the 1 x 1 convolution layer, the sample data is input into the graph learning layer to obtain an adjacent matrix, and the adjacent matrix is applied to the input of a plurality of graph convolution layers; step 33, setting residual connection among a plurality of sequentially connected expansion causal convolutions; step 34, setting a window size to be a prediction window of N hours by an output module, and enabling the output module to use the number of the time sequence characteristics used for prediction, which are acted on the EHR data by a plurality of full connection layers, to obtain output data with a dimensionality of [ second time threshold B, N ]; and 35, inputting sample data, repeating the steps 32 to 35, and training the neural network model. The purpose of constructing the graph learning layer is to adaptively learn a graph adjacency matrix so as to capture hidden relations among time series data. The sequential convolutional layer and the graph convolutional layer modules are interleaved to capture spatial and temporal dependencies, respectively. And as the network deepens in the network training, the propagation of the gradient becomes more difficult, so residual connection is introduced between the networks. And finally, matching output scales by using a plurality of full connection layers, enhancing network robustness and improving network learning ability. The graph convolution layer and the graph learning layer do not change the dimensionality of data, only feature fusion and transmission are carried out on feature dimensionality, the number of data channels is reduced to 1 before the data channels are transmitted to an output module, the dimensionality is [96, 123] at the moment, finally the output module acts on the feature dimensionality, the data of [96, 123] are mapped to [96,4] by a plurality of full connection layers, and the data correspond to the label extracted in the step 1.
After the neural network model is constructed and trained, the time sequence data is used as input, and the neural network model generates a prediction judgment value to provide guidance for intervention means. The method specifically comprises the following steps: s41, setting the sampling interval time of EHR data to be M hours, inputting the EHR data to obtain a prediction judgment value with dimension of [ second time threshold B, N ], and obtaining B prediction windows based on the prediction judgment value, wherein each prediction window comprises state label values of N types of intervention means; and acquiring a predicted intervention means based on the change rule of the state tag value of the intervention means. For example, setting the prediction window N to 4 hours, the prediction category N to 4 categories, and the EHR data sampling interval time M to 2 hours, i.e., the feature samples are sampled from 1 hour of the patient's stay and spaced 2 hours apart, i.e., the prediction window starts from 3 hours (i.e., 1+ M, the calculation result is 3 when M is 2) of the patient's stay, the first prediction window contains the intervention means status flag values from 3 hours to 7 hours. The intervention instrument state feature includes an intervention instrument state tag value 1 for identifying off and an intervention instrument state tag value 0 for identifying on. The change rule of the intervention means state tag value is as follows: predicting a class I Oset, recording the intervention means as closed if the label value is changed from 0 to 1, predicting a class II Wean, recording the intervention means as open if the label value is changed from 1 to 0, predicting a class III Stay On, recording the intervention means as kept closed if the label value is kept to 1, predicting a class IV Stay Off, and recording the intervention means as kept open if the label value is kept to 0; based on the change rule of the intervention means state tag value, an intervention means tag is constructed as [ close, open, close keeping, open ], namely the neural network model is based on EHR data, each window ultimately outputs the true intervention state value, represented as a 1 x 4 single heat vector, providing guidance for intervention, such as 1,0,0 is class one, [0,1, 0] is class two, [0,1, 0] is class three, and [0, 1] is class four.
The invention is particularly applicable to intervention prediction of vasopressin and invasive ventilation. Vasopressin is a drug which causes vasoconstriction and has a strong effect of increasing Mean Arterial Pressure (MAP), is a medical intervention means for intermittent fluid intervention, can increase blood pressure, reduce visceral blood flow and increase effective circulating blood volume and cardiac output, and is an indispensable drug in an ICU scene. The invasive ventilation is operated by an invasive respirator, the working principle of the invasive ventilation device is an artificial mechanical ventilation device which is used for assisting or controlling the spontaneous respiratory movement of a patient so as to achieve the function of gas exchange in the lungs, and the advance prediction of the invasive ventilation is particularly important for shortening the rescue time and reducing the death rate of the patient.
Finally, it should be noted that: the above embodiments are only preferred embodiments of the present invention to illustrate the technical solutions of the present invention, but not to limit the technical solutions, and certainly not to limit the patent scope of the present invention; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; the modifications or the substitutions do not cause the essence of the corresponding technical solutions to depart from the scope of the technical solutions of the embodiments of the present invention; that is, the technical problems to be solved by the present invention, which are not substantially changed or supplemented by the spirit and the concept of the main body of the present invention, are still consistent with the present invention and shall be included in the scope of the present invention; in addition, the technical scheme of the invention is directly or indirectly applied to other related technical fields, and the technical scheme is included in the patent protection scope of the invention.
Claims (8)
1. An ICU auxiliary intervention means prediction method based on deep learning is characterized by comprising the following steps:
step 1, obtaining an MIMIC-III database, and extracting EHR data of a patient in an intensive care unit;
step 2, cleaning and processing the EHR data and converting the EHR data into time sequence data;
step 3, constructing a neural network model consisting of a graph learning layer, an expansion causal convolution, an output module and a full connection layer, and training and updating;
and 4, taking the time sequence data as input, generating a prediction judgment value by the neural network model, and providing guidance for intervention means.
2. The deep learning-based ICU-assisted intervention tool prediction method of claim 1, wherein in the step 1, the EHR data comprises static characteristics, time sequence characteristics and intervention tool state characteristics; the static characteristics are physiological information with invariable time sequence; the time series characteristic is a clinical physiological index of a patient associated with the prediction, and at least comprises time series change physiological characteristic data, laboratory measurement data and medicine data.
3. The ICU-assisted intervention tool prediction method based on deep learning of claim 2, wherein the step 2 specifically comprises: step 21, selecting EHR data of a patient with hospitalization time exceeding a first time threshold A from a MIMIC-III database; step 22, setting a second time threshold B with the hospitalization time greater than the first time threshold a, performing primary sampling on EHR data with the hospitalization time less than the second time threshold B, performing secondary sampling on EHR data between B hours before the final point of the hospitalization time and the final point of the hospitalization time, and aligning the sampled data based on time sequence characteristics; step 23, standardizing the data after alignment processing, so that each type of EHR data on each timestamp is represented by values of three channels, wherein the values of the three channels include a mask for identifying whether EHR data exists in the current timestamp, a value of actual EHR data, and an accumulated time when EHR data is observed for the last time; and 24, constructing sample data with dimensionalities of [ the number of data pieces, the number of channels, the number of time sequence features used for prediction contained in the EHR data, and a second time threshold B ].
4. The deep learning-based ICU-assisted intervention tool prediction method of claim 3, wherein the step 3 specifically comprises: step 31, setting a neural network model comprising a graph learning layer, a plurality of expansion cause-and-effect convolutions and an output module which are connected in sequence, wherein the expansion cause-and-effect convolutions comprise a time sequence convolution layer and a graph convolution layer which are connected in sequence, and a 1 x 1 convolution layer is arranged before the first time sequence convolution layer; step 32, inputting one constructed sample data in each training, wherein the number dimension of channels is increased by the sample data through the 1 x 1 convolution layer, the sample data is input into the graph learning layer to obtain an adjacent matrix, and the adjacent matrix is applied to the input of a plurality of graph convolution layers; step 33, setting residual connection among a plurality of sequentially connected expansion causal convolutions; step 34, the output module sets a prediction window with a window size of N hours, and the output module uses the number of time sequence characteristics used for prediction and acted on EHR data by a plurality of full connection layers to obtain output data with a dimensionality of [ second time threshold B, N ]; and 35, inputting sample data, repeating the steps 32 to 35, and training the neural network model.
5. The deep learning based ICU assisted intervention means prediction method of claim 4, wherein the intervention means status signature comprises an intervention means status tag value of 1 identifying off and an intervention means status tag value of 0 identifying on.
6. The deep learning-based ICU-assisted intervention tool prediction method of claim 5, wherein the step 4 comprises the following processes: step S41, setting the sampling interval time of EHR data to be M hours, inputting the EHR data to obtain a prediction judgment value with dimension of [ second time threshold B, N ], and obtaining B prediction windows based on the prediction judgment value, wherein each prediction window comprises an intervention means state label value within N hours; and acquiring a predicted intervention means based on the change rule of the state tag value of the intervention means.
7. The deep learning-based ICU assisted intervention means prediction method of claim 6, wherein the intervention means state tag value change rules are that the intervention means is marked as off if the tag value is changed from 0 to 1, the intervention means is marked as on if the tag value is changed from 1 to 0, the intervention means is marked as off if the tag value is kept at 1, and the intervention means is marked as on if the tag value is kept at 0.
8. The ICU assisted intervention means prediction method based on deep learning of claim 7, wherein intervention means tags are constructed as [ off, on, remain off, remain on ] based on a change rule of the intervention means state tag values.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211267598.1A CN115602299A (en) | 2022-10-17 | 2022-10-17 | ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211267598.1A CN115602299A (en) | 2022-10-17 | 2022-10-17 | ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115602299A true CN115602299A (en) | 2023-01-13 |
Family
ID=84846993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211267598.1A Pending CN115602299A (en) | 2022-10-17 | 2022-10-17 | ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115602299A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235487A (en) * | 2023-10-12 | 2023-12-15 | 北京大学第三医院（北京大学第三临床医学院） | Feature extraction method and system for predicting hospitalization event of asthma patient |
-
2022
- 2022-10-17 CN CN202211267598.1A patent/CN115602299A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235487A (en) * | 2023-10-12 | 2023-12-15 | 北京大学第三医院（北京大学第三临床医学院） | Feature extraction method and system for predicting hospitalization event of asthma patient |
| CN117235487B (en) * | 2023-10-12 | 2024-03-12 | 北京大学第三医院（北京大学第三临床医学院） | Feature extraction method and system for predicting hospitalization event of asthma patient |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huddar et al. | Predicting complications in critical care using heterogeneous clinical data | |
| Arnaud et al. | Deep learning to predict hospitalization at triage: Integration of structured data and unstructured text | |
| CN108399951B (en) | Breathing machine-related pneumonia decision-making assisting method, device, equipment and medium | |
| CN109805898A (en) | Critical illness Mortality Prediction method based on attention mechanism timing convolutional network algorithm | |
| Fakhfakh et al. | ProgNet: Covid-19 prognosis using recurrent and convolutional neural networks | |
| JP7464800B2 (en) | METHOD AND SYSTEM FOR RECOGNITION OF MEDICAL EVENTS UNDER SMALL SAMPLE WEAKLY LABELING CONDITIONS - Patent application | |
| CN116364299B (en) | Disease diagnosis and treatment path clustering method and system based on heterogeneous information network | |
| CN111798954A (en) | Drug combination recommendation method based on time attention mechanism and graph convolution network | |
| Gavrilov et al. | Feature extraction method from electronic health records in Russia | |
| CN115602299A (en) | ICU (intensive care unit) auxiliary intervention means prediction method based on deep learning | |
| CN113160986A (en) | Model construction method and system for predicting development of systemic inflammatory response syndrome | |
| CN112908452A (en) | Event data modeling | |
| Yu et al. | Identification of pediatric respiratory diseases using a fine-grained diagnosis system | |
| Gao et al. | Reconstructing missing ehrs using time-aware within-and cross-visit information for septic shock early prediction | |
| Oei et al. | Towards early sepsis detection from measurements at the general ward through deep learning | |
| CN116543917A (en) | Information mining method for heterogeneous time sequence data | |
| Zhang et al. | Real-time Prediction for Mechanical Ventilation in COVID-19 Patients using A Multi-task Gaussian Process Multi-objective Self-attention Network | |
| Subbe et al. | Using trends in electronic recordings of vital signs to identify patients stable for transfer from acute hospitals | |
| Han et al. | Chinese Q&A community medical entity recognition with character-level features and self-attention mechanism | |
| CN112951412A (en) | Auxiliary diagnosis method and application thereof | |
| Diwakar et al. | Patient Monitoring System | |
| Liu et al. | A Deep Learning Post-Discharge Mortality Prediction Method Considering Diagnosis Information for ICU Patients | |
| Shirwaikar et al. | Design framework for a data mart in the neonatal intensive care unit | |
| Rocheteau | Representation learning for patients in the Intensive Care Unit | |
| Liu et al. | GRU-TV: Time-and velocity-aware GRU for patient representation on multivariate clinical time-series data |
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 |