CN115019957B - Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning - Google Patents
Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning Download PDFInfo
- Publication number
- CN115019957B CN115019957B CN202210632448.XA CN202210632448A CN115019957B CN 115019957 B CN115019957 B CN 115019957B CN 202210632448 A CN202210632448 A CN 202210632448A CN 115019957 B CN115019957 B CN 115019957B
- Authority
- CN
- China
- Prior art keywords
- enhancement
- feature extraction
- dense
- data
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000002787 reinforcement Effects 0.000 title claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 54
- 238000000034 method Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 12
- 230000007246 mechanism Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000004519 manufacturing process Methods 0.000 claims abstract description 3
- 230000004927 fusion Effects 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 8
- 238000005259 measurement Methods 0.000 claims description 6
- 108010074051 C-Reactive Protein Proteins 0.000 claims description 4
- 102100032752 C-reactive protein Human genes 0.000 claims description 4
- 108010048233 Procalcitonin Proteins 0.000 claims description 4
- 206010040047 Sepsis Diseases 0.000 claims description 4
- 210000003815 abdominal wall Anatomy 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- CWCXERYKLSEGEZ-KDKHKZEGSA-N procalcitonin Chemical compound C([C@@H](C(=O)N1CCC[C@H]1C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@H](C(=O)N[C@@H](C)C(=O)N[C@@H]([C@@H](C)CC)C(=O)NCC(=O)N[C@@H](C(C)C)C(=O)NCC(=O)N[C@@H](C)C(=O)N1[C@@H](CCC1)C(=O)NCC(O)=O)[C@@H](C)O)NC(=O)[C@@H](NC(=O)[C@H](CC=1NC=NC=1)NC(=O)[C@H](CC=1C=CC=CC=1)NC(=O)[C@H](CCCCN)NC(=O)[C@H](CC(N)=O)NC(=O)[C@H](CC=1C=CC=CC=1)NC(=O)[C@H](CC(O)=O)NC(=O)[C@H](CCC(N)=O)NC(=O)[C@@H](NC(=O)[C@H](CC=1C=CC(O)=CC=1)NC(=O)[C@@H](NC(=O)CNC(=O)[C@H](CC(C)C)NC(=O)[C@H](CCSC)NC(=O)[C@H]1NC(=O)[C@H]([C@@H](C)O)NC(=O)[C@H](CO)NC(=O)[C@H](CC(C)C)NC(=O)[C@H](CC(N)=O)NC(=O)CNC(=O)[C@@H](N)CSSC1)[C@@H](C)O)[C@@H](C)O)[C@@H](C)O)C1=CC=CC=C1 CWCXERYKLSEGEZ-KDKHKZEGSA-N 0.000 claims description 4
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- JVTAAEKCZFNVCJ-UHFFFAOYSA-M Lactate Chemical compound CC(O)C([O-])=O JVTAAEKCZFNVCJ-UHFFFAOYSA-M 0.000 claims description 2
- 238000010801 machine learning Methods 0.000 abstract description 7
- 108091007065 BIRCs Proteins 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 5
- JVTAAEKCZFNVCJ-UHFFFAOYSA-N lactic acid Chemical compound CC(O)C(O)=O JVTAAEKCZFNVCJ-UHFFFAOYSA-N 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 230000003187 abdominal effect Effects 0.000 description 3
- 208000002623 Intra-Abdominal Hypertension Diseases 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 235000014655 lactic acid Nutrition 0.000 description 2
- 239000004310 lactic acid Substances 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 229940088594 vitamin Drugs 0.000 description 2
- 239000011782 vitamin Substances 0.000 description 2
- 206010058808 Abdominal compartment syndrome Diseases 0.000 description 1
- 206010020772 Hypertension Diseases 0.000 description 1
- 241000713321 Intracisternal A-particles Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 230000036724 intravesical pressure Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 208000011580 syndromic disease Diseases 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/03—Detecting, measuring or recording fluid pressure within the body other than blood pressure, e.g. cerebral pressure; Measuring pressure in body tissues or organs
-
- 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
-
- 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
-
- 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
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Animal Behavior & Ethology (AREA)
- Heart & Thoracic Surgery (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Fuzzy Systems (AREA)
- Epidemiology (AREA)
- Hematology (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Primary Health Care (AREA)
- Databases & Information Systems (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention relates to a dense iterative characteristic intra-abdominal pressure prediction system based on reinforcement learning, which belongs to the field of machine learning and comprises the following components: and the data acquisition and processing module is used for: the method comprises the steps of acquiring basic data and clinical data of a patient, preprocessing the basic data and the clinical data, and manufacturing a sample and a data set; and a data input module: the method comprises the steps of sending a preprocessed sample into a dense iteration enhancement model, and carrying out iteration enhancement through a plurality of feature extraction units to obtain enhanced features; and a prediction module: the method comprises the steps of (1) after GAP and flat are carried out on the enhancement features, sending the enhancement features into a full-connection layer, and obtaining a prediction result; model training module: the loss is calculated according to the true value and the predicted result, a dynamic adjustment learning rate mechanism is used, WEIGHT DECAY and momentum mechanisms are used, model parameters are updated, and the network is trained in a complete end-to-end mode.
Description
Technical Field
The invention belongs to the field of machine learning, and relates to a dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning.
Background
In recent years, with the application of machine learning in the medical field, a number of machine learning methods have been proposed for analyzing statistical significance and feature correlation of clinical patient medical data, and such methods provide reference for clinical diagnosis of doctors. According to international guidelines for diagnosis and treatment of high abdominal pressure (intra-abdominal hypertension, IAH) and interstitial abdominal syndrome (abdominal compartment syndrome, ACS), intra-abdominal pressure (intra-abdominal pressure, IAP) monitoring is required when severe patients have high risk factors for IAH. Given the importance of this medical technology to patients, intra-abdominal pressure detection has gradually become one of the important parts of critical patient condition monitoring. There are many methods for monitoring IAP, among which intra-vesical pressure (intra-viseral pressure, IVP) measurement is a common method for IAP measurement. However, the measurement method is complicated and is easy to cause infection, and the intermittent measurement method can cause the condition of delayed diagnosis of abdominal hypertension. Currently, IAP detection methods currently reported in the art use physical models for monitoring, and the cystometric pressure is indirectly assessed and monitored by physical measurements by mechanical means. While predictive models that establish regression relationships to IAPs based on patient biochemical indices using machine learning are temporarily omitted. But such predictive models have demonstrated better performance and medical diagnostic assistance in other disease diagnostics in the medical field. Most of the prediction models utilize a data driving method to realize the target prediction of the corresponding medical task through a general machine learning model, but the problem that important information is lost in the continuous downsampling process is ignored, so that complete information cannot be extracted, and the prediction effect is poor. And the connection between the complex high-vitamin index characteristics of the patient cannot be well established.
Disclosure of Invention
In view of the above, in order to supplement the performance deficiency of these machine learning models in IAP prediction and solve the deficiency of the prediction model for medical data diagnosis, the present invention provides a dense iterative feature intra-abdominal pressure prediction system based on reinforcement learning.
In order to achieve the above purpose, the present invention provides the following technical solutions:
A reinforcement learning based dense iterative feature intra-abdominal pressure prediction system, comprising:
And the data acquisition and processing module is used for: the method comprises the steps of acquiring basic data and clinical data of a patient, preprocessing the basic data and the clinical data, and manufacturing a sample and a data set;
and a data input module: the method comprises the steps of sending a preprocessed sample into a dense iteration enhancement model, and carrying out iteration enhancement through a plurality of feature extraction units to obtain enhanced features;
and a prediction module: the method comprises the steps of (1) after GAP and flat are carried out on the enhancement features, sending the enhancement features into a full-connection layer, and obtaining a prediction result;
Model training module: the loss is calculated according to the true value and the predicted result, a dynamic adjustment learning rate mechanism is used, WEIGHT DECAY and momentum mechanisms are used, model parameters are updated, and the network is trained in a complete end-to-end mode.
Further, the data acquisition and processing module acquires actual measurement data from a patient which clinically accords with the standard, wherein the actual measurement data comprises basic data and clinical data, and the basic data comprises gender, age, BMI and ICU reasons; the clinical data include ISS score, APACHE II score on day 1 of ICU, SOFA score, sepsis, procalcitonin, lactate, C-reactive protein, abdominal wall tone.
Further, the data acquisition and processing module performs normalization processing on the data set, preliminarily filters abnormal data, converts the data into a processable format file, forms a sample, divides the sample into a training set and a testing set, and takes a real intra-abdominal pressure index of a patient as a label l.
Further, the sample is sent into a dense iteration enhancement model through the data input module to perform feature extraction, the dense iteration enhancement model is composed of a plurality of feature extraction units, and the feature units are combined together in a dense connection mode; the feature extraction unit consists of a feature extraction network and an iteration enhancement module, wherein the first feature extraction unit does not contain the iteration enhancement module; dense connection the single feature extraction unit and its front and back feature extraction units are spliced in a combination mode of Markov chain and multi-stage residual connection, wherein for the nth feature extraction unit, n is not equal to 1, the iteration enhancement module is represented by I n under a dense connection structure, so that an enhanced output feature is obtainedThe following formula is shown:
Wherein z n denotes an input feature of the nth feature extraction unit, Representing the enhancement features output by the 1 st to n-1 st feature extraction units; the method specifically comprises the following steps:
S11: using T epsilon {1,2, …, n-1} represents the upsampling operator, calculate the enhancement feature at time t/>And t-th enhancement feature/>Difference between/>
S12: usingT.epsilon. {1,2, …, n-1} represents the downsampling operator, use/>Updating the enhancement features:
S13: steps S11-S12 are repeated until all input features have been iterated, i.e. t=n-1.
Further, the feature extraction unit in the dense iteration enhancement model includes a CNN backbone network and a three-layer dense connection structure, the dimension of an input sample is 12×1, the deconvolution realizes an up-sampling operator, and the convolution realizes a down-sampling operator, and the method includes the following steps:
S21: constructing an iterative enhancement module using deconvolution layers and convolution layers, including IEM1 and IEM2 modules;
S22: constructing a feature extraction unit by using a convolution Layer and an iteration enhancement module and a residual feature extraction block, wherein the feature extraction unit comprises Layer1, layer2, layer3, layer4 and Layer5 modules;
s23: inputting the sample into a dense iteration enhancement model to obtain original characteristics processed by Layer 1;
S24: the original features are respectively input into a convolution Layer of Layer2 and an IEM1 module, the IEM1 carries out iteration enhancement and fusion on the output features of Layer1 and Conv2, and then the residual error block is utilized to further complete feature extraction;
S25: the features obtained in the steps S23-S24 are respectively input into a convolution Layer of Layer3 and an IEM2 module, the IEM2 carries out iterative enhancement and fusion on the output features of Layer1, layer2 and Conv3, and then the residual error block is utilized to further complete feature extraction.
Further, the prediction module carries out global average pooling processing and flat on the enhanced fusion feature output by the last feature extraction unit, and inputs the enhanced fusion feature into the full-connection layer to obtain a prediction result y.
Further, the model training module calculates the loss by using the real intra-abdominal pressure value, namely the label L, and the predicted result y, and the loss function L adopts the mean square error MSE:
Wherein m is the number of samples of one batch, and l i and y i respectively represent the label and the predicted value of the ith sample;
The model training module trains a dense iteration enhancement model in an end-to-end mode to realize prediction of intra-abdominal pressure of a patient.
The invention has the beneficial effects that: the system establishes an end-to-end mapping relation between complex high-vitamins index characteristics and intra-abdominal pressure, and overcomes the defects of complex, tedious and high diagnosis delay of the traditional method to a certain extent. The system realizes more complete information extraction, and is realized through a dense connection mechanism and iterative enhancement of different scale features. The iteration enhancement module is used, so that dimension reduction of original high-dimensional data is guaranteed, effective fusion of different scale features is realized through up-sampling and down-sampling, and accurate prediction is facilitated.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.
Drawings
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:
FIG. 1 is a diagram of an overall model structure;
FIG. 2 is a block diagram of an iterative enhancement module;
fig. 3 is a model training flow chart.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.
Referring to fig. 1 to 3, the specific implementation details of each part of the present invention are as follows:
The invention provides a dense iterative characteristic intra-abdominal pressure prediction system based on reinforcement learning, which comprises the following components:
And the data acquisition and processing module is used for: is used for acquiring basic data and clinical data of patients, and preprocessing the basic data and the clinical data to prepare samples and data sets. Measured data including basic data (including: gender, age, BMI, cause of the entering ICU) and clinical data (ISS score, APACHE II score on day 1 of the entering ICU, SOFA score, sepsis, procalcitonin, lactic acid, C-reactive protein, abdominal wall tension) were obtained from patients clinically meeting the criteria. And carrying out normalization processing on the data set, preliminarily filtering abnormal data, and converting the data into a processable format file. In particular, the biochemical index of the patient can be increased or decreased according to the correlation between the biochemical index and the prediction task according to the actual requirement, and in this embodiment, the above-mentioned twelve biochemical indexes for patient admission are taken as an example for illustration. After the data are acquired, the data are divided into a training set and a testing set, and the real intra-abdominal pressure index of the patient is used as a label l.
And a data input module: as shown in fig. 1, the method is used for sending the preprocessed sample into a dense iterative enhancement model, and performing iterative enhancement through a plurality of feature extraction units (as shown in fig. 2) to obtain enhanced features.
In this example, 12 indices of the patient: sex, age, BMI, cause of check-in ICU, ISS score, APACHE II score on day 1 of check-in ICU, SOFA score, sepsis, procalcitonin, lactic acid, C-reactive protein and abdominal wall tension, and the formed sample x is sent into a feature dense iteration enhancement model for feature extraction.
The feature dense iteration enhancement model is composed of a plurality of feature extraction units, and the feature units are combined together in a dense connection mode. In particular, the feature extraction units may be composed of a common feature extraction network and an iterative enhancement module (the first feature extraction unit does not contain an iterative enhancement module), while the dense connection splices a single feature extraction unit with its front and back feature extraction units in a combination of a markov chain and a multi-level residual connection. Wherein, for the nth feature extraction unit (n noteq 1), the iterative enhancement module is represented by I n under a dense connection structure, thereby obtaining an enhanced output featureThe following formula is shown:
The method specifically comprises the following steps:
1) Using T epsilon {1,2, …, n-1} represents the upsampling operator, calculate the enhancement feature at time t/>And t-th enhancement feature/>Difference between/>
2) UsingT.epsilon. {1,2, …, n-1} represents the downsampling operator, using/>, calculated in 1)Updating the enhancement features:
3) Repeating steps 1) -2) until all input features have been iterated, i.e. t=n-1.
Optionally, the feature extraction unit of the dense iterative enhancement model in this embodiment may be replaced by a common feature extraction network, such as CNN, RNN, etc., and the dense connection manner may perform custom adjustment of the layer number or connection parameters according to task requirements. In this embodiment, taking CNN as a main trunk network and a dense connection structure of three layers, the dimension of an input sample is 12×1, deconvolution realizes an up-sampling operator, convolution realizes a down-sampling operator as an example, and specifically includes the following steps:
1) The iterative enhancement module was constructed using deconvolution layers and convolution layers, the structure of which is shown in table 1. Wherein: deConv (convolution kernel size 3) represents a deconvolution layer, conv represents a convolution layer (convolution kernel size 3, zero padding=0), and 16×12×1 represents a 12×1 feature of 16 channels.
TABLE 1
2) The feature extraction unit is constructed using a convolution layer and an iterative enhancement module and a residual feature extraction block. The structure of the dense iterative enhancement model is shown in table 2. Wherein: RB denotes the residual block of the double convolution layer, fc denotes the full-connected layer, GAP denotes global average pooling.
TABLE 2
3) And inputting the sample into a dense iteration enhancement model to obtain the original characteristics processed by Layer 1.
4) The original features are respectively input into a convolution Layer of Layer2 and an IEM module, the IEM1 carries out iterative enhancement and fusion on the output features of Layer1 and Conv2, and then the residual error block is utilized to further complete feature extraction.
5) The features obtained in 3) and 4) are input to the convolutional Layer of Layer3 and the IEM module, respectively. The IEM2 carries out iterative enhancement and fusion on the output characteristics of the Layer1, the Layer2 and the Conv3, and then the residual block is utilized to further complete the characteristic extraction.
And a prediction module: carrying out global average pooling treatment and flat on the reinforced fusion features output by the last feature extraction unit, and inputting the reinforced fusion features into a full-connection layer to obtain a prediction result y;
model training module: using the calculated loss of the actual intra-abdominal pressure value (label L) and the predicted result y, the loss function L uses the Mean Square Error (MSE):
Where m is the number of samples of a batch, and l i and y i represent the label and the predicted value of the ith sample, respectively.
The dense iteration enhancement model is trained in an end-to-end mode, prediction of intra-abdominal pressure of a patient is achieved, and deep diagnosis and treatment are conducted on the patient by combining prediction results.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.
Claims (4)
1. Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning is characterized in that: comprising the following steps:
And the data acquisition and processing module is used for: the method comprises the steps of acquiring basic data and clinical data of a patient, preprocessing the basic data and the clinical data, and manufacturing a sample and a data set; the data acquisition and processing module acquires actual measurement data from a patient which clinically accords with the standard, wherein the actual measurement data comprises basic data and clinical data, and the basic data comprises gender, age, BMI and ICU reasons; the clinical data includes ISS score, APACHE II score on day 1 of ICU, SOFA score, sepsis, procalcitonin, lactate, C-reactive protein and abdominal wall tone;
And a data input module: the method comprises the steps of sending a preprocessed sample into a dense iteration enhancement model, and carrying out iteration enhancement through a plurality of feature extraction units to obtain enhanced features; the dense iteration enhancement model consists of a plurality of feature extraction units, and the feature units are combined together in a dense connection mode; the feature extraction unit consists of a feature extraction network and an iteration enhancement module, wherein the first feature extraction unit does not contain the iteration enhancement module; dense connection the single feature extraction unit and its front and back feature extraction units are spliced in a combination mode of Markov chain and multi-stage residual connection, wherein for the nth feature extraction unit, n is not equal to 1, the iteration enhancement module is represented by I n under a dense connection structure, so that an enhanced output feature is obtained The following formula is shown:
Wherein z n denotes an input feature of the nth feature extraction unit, Representing the enhancement features output by the 1 st to n-1 st feature extraction units; the method specifically comprises the following steps:
S11: using Representing the upsampling operator, computing the enhancement feature at time t/>And t-th enhancement feature/>Difference between/>
S12: usingRepresenting downsampling operators, use/>Updating the enhancement features:
s13: repeating the steps S11-S12 until all input features are iterated, namely t=n-1;
the feature extraction unit in the dense iteration enhancement model comprises a CNN backbone network and a three-layer dense connection structure, wherein the dimension of an input sample is 12 multiplied by 1, the deconvolution realizes an up-sampling operator, and the convolution realizes a down-sampling operator, and the feature extraction unit comprises the following steps:
S21: constructing an iterative enhancement module using deconvolution layers and convolution layers, including IEM1 and IEM2 modules;
S22: constructing a feature extraction unit by using a convolution Layer and an iteration enhancement module and a residual feature extraction block, wherein the feature extraction unit comprises Layer1, layer2, layer3, layer4 and Layer5 modules;
s23: inputting the sample into a dense iteration enhancement model to obtain original characteristics processed by Layer 1;
S24: the original features are respectively input into a convolution Layer of Layer2 and an IEM1 module, the IEM1 carries out iteration enhancement and fusion on the output features of Layer1 and Conv2, and then the residual error block is utilized to further complete feature extraction;
S25: inputting the features obtained in the steps S23-S24 into a convolution Layer of Layer3 and an IEM2 module respectively, carrying out iterative enhancement and fusion on the output features of Layer1, layer2 and Conv3 by the IEM2, and further completing feature extraction by using a residual block;
and a prediction module: the method comprises the steps of (1) after GAP and flat are carried out on the enhancement features, sending the enhancement features into a full-connection layer, and obtaining a prediction result;
Model training module: the loss is calculated according to the true value and the predicted result, a dynamic adjustment learning rate mechanism is used, WEIGHT DECAY and momentum mechanisms are used, model parameters are updated, and the network is trained in a complete end-to-end mode.
2. The reinforcement learning based dense iterative feature intra-abdominal pressure prediction system of claim 1, wherein: the data acquisition and processing module performs normalization processing on the data set, preliminarily filters abnormal data, converts the data into a processable format file to form a sample, divides the sample into a training set and a testing set, and takes a real intra-abdominal pressure index of a patient as a label l.
3. The reinforcement learning based dense iterative feature intra-abdominal pressure prediction system of claim 1, wherein: and the prediction module carries out global average pooling treatment and flat on the enhanced fusion characteristics output by the last characteristic extraction unit, and inputs the enhanced fusion characteristics into a full-connection layer to obtain a prediction result y.
4. The reinforcement learning based dense iterative feature intra-abdominal pressure prediction system of claim 1, wherein: the model training module calculates loss by using a real intra-abdominal pressure value, namely a label L, and a predicted result y, and the loss function L adopts a mean square error MSE:
Wherein m is the number of samples of one batch, and l i and y i respectively represent the label and the predicted value of the ith sample;
The model training module trains a dense iteration enhancement model in an end-to-end mode to realize prediction of intra-abdominal pressure of a patient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210632448.XA CN115019957B (en) | 2022-06-06 | 2022-06-06 | Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210632448.XA CN115019957B (en) | 2022-06-06 | 2022-06-06 | Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115019957A CN115019957A (en) | 2022-09-06 |
CN115019957B true CN115019957B (en) | 2024-05-28 |
Family
ID=83072439
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210632448.XA Active CN115019957B (en) | 2022-06-06 | 2022-06-06 | Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115019957B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117253584B (en) * | 2023-02-14 | 2024-07-19 | 南雄市民望医疗有限公司 | Hemodialysis component detection-based dialysis time prediction system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111722139A (en) * | 2020-06-29 | 2020-09-29 | 重庆邮电大学 | Lithium battery health monitoring model self-learning method based on micro-continuous mapping |
CN113284126A (en) * | 2021-06-10 | 2021-08-20 | 安徽省立医院(中国科学技术大学附属第一医院) | Method for predicting hydrocephalus shunt operation curative effect by artificial neural network image analysis |
CN114332994A (en) * | 2021-12-20 | 2022-04-12 | 深圳数联天下智能科技有限公司 | Method for training age prediction model, age detection method and related device |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7395216B2 (en) * | 1999-06-23 | 2008-07-01 | Visicu, Inc. | Using predictive models to continuously update a treatment plan for a patient in a health care location |
US10318889B2 (en) * | 2017-06-26 | 2019-06-11 | Konica Minolta Laboratory U.S.A., Inc. | Targeted data augmentation using neural style transfer |
-
2022
- 2022-06-06 CN CN202210632448.XA patent/CN115019957B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111722139A (en) * | 2020-06-29 | 2020-09-29 | 重庆邮电大学 | Lithium battery health monitoring model self-learning method based on micro-continuous mapping |
CN113284126A (en) * | 2021-06-10 | 2021-08-20 | 安徽省立医院(中国科学技术大学附属第一医院) | Method for predicting hydrocephalus shunt operation curative effect by artificial neural network image analysis |
CN114332994A (en) * | 2021-12-20 | 2022-04-12 | 深圳数联天下智能科技有限公司 | Method for training age prediction model, age detection method and related device |
Non-Patent Citations (1)
Title |
---|
Effects of Intra-Abdominal Pressure on Lung Mechanics during Laparoscopic Gynaecology;N. A. Jalal;《2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)》;20210909;2091-2094 * |
Also Published As
Publication number | Publication date |
---|---|
CN115019957A (en) | 2022-09-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lin et al. | Early diagnosis and prediction of sepsis shock by combining static and dynamic information using convolutional-LSTM | |
CN111062948B (en) | Multi-tissue segmentation method based on fetal four-chamber cardiac section image | |
CN111445481A (en) | Abdominal CT multi-organ segmentation method based on scale fusion | |
CN112766377B (en) | Left ventricle magnetic resonance image intelligent classification method, device, equipment and medium | |
CN111638249B (en) | Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation | |
CN108766557B (en) | Automatic arrhythmia analysis method based on channel signal fusion neural network | |
CN114549469A (en) | Deep neural network medical image diagnosis method based on confidence degree calibration | |
CN115019957B (en) | Dense iteration characteristic intra-abdominal pressure prediction system based on reinforcement learning | |
CN113516210A (en) | Lung adenocarcinoma squamous carcinoma diagnosis model training method and device based on PET/CT | |
CN115798711B (en) | Chronic nephrosis diagnosis and treatment decision support system based on inverse fact contrast learning | |
CN113077900B (en) | Diabetes early risk assessment method, device, computer equipment and medium | |
CN112381846A (en) | Ultrasonic thyroid nodule segmentation method based on asymmetric network | |
CN111028232A (en) | Diabetes classification method and equipment based on fundus images | |
CN114663383A (en) | Blood cell segmentation and identification method and device, electronic equipment and storage medium | |
CN114549541A (en) | Method and system for predicting occurrence types and risks of cardiovascular and cerebrovascular diseases based on fundus images, computer equipment and storage medium | |
CN116350196A (en) | Noninvasive blood pressure value prediction method and system based on deep learning | |
CN114391827A (en) | Pre-hospital emphysema diagnosis device based on convolutional neural network | |
Poulain et al. | Transformer-based multi-target regression on electronic health records for primordial prevention of cardiovascular disease | |
Xu et al. | Pay more attention with fewer parameters: A novel 1-D convolutional neural network for heart sounds classification | |
CN112200810B (en) | Multi-modal automated ventricle segmentation system and method of use thereof | |
CN111862049B (en) | Brain glioma segmentation network system and brain glioma segmentation method based on deep learning | |
Ullah et al. | Exploring clinical time series forecasting with meta-features in variational recurrent models | |
CN116309507A (en) | AIS focus prediction method for performing feature fusion on CTP under attention mechanism | |
CN116309754A (en) | Brain medical image registration method and system based on local-global information collaboration | |
CN116188420A (en) | Multi-mode medical image segmentation method based on convolution transducer |
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 |