CN116313053A - Postoperative complications prediction model training method and postoperative complications prediction method - Google Patents
Postoperative complications prediction model training method and postoperative complications prediction method Download PDFInfo
- Publication number
- CN116313053A CN116313053A CN202310266597.3A CN202310266597A CN116313053A CN 116313053 A CN116313053 A CN 116313053A CN 202310266597 A CN202310266597 A CN 202310266597A CN 116313053 A CN116313053 A CN 116313053A
- Authority
- CN
- China
- Prior art keywords
- prediction
- prediction model
- task
- training
- complications
- 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
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—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
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention provides a postoperative complication prediction model training method and a postoperative complication prediction method, wherein the complication prediction model training method comprises the following steps: acquiring clinical data; determining a predictive task based on the clinical data; preprocessing clinical data to obtain a training set; constructing an initial postoperative complication prediction model based on the prediction task; based on the training set, performing multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model. According to the invention, a multi-task learning complication prediction model is constructed on the basis of considering possible synergic or antagonistic actions among various complications, and the complications can be predicted more accurately and comprehensively by training the complication prediction model, so that the life quality of a patient can be improved.
Description
Technical Field
The invention relates to the field of postoperative prediction, in particular to a training method of postoperative complications prediction models and a postoperative complications prediction method.
Background
At present, coronary heart disease has become a major disease burden worldwide. It has been counted that more than one fifth of patients with coronary heart disease are selected to receive Percutaneous Coronary Intervention (PCI) treatment, wherein restenosis occurs in 40% to 50% of patients, and vascular related complications may occur in about 14% of patients, with cerebrovascular accidents occurring after surgery in 3% -5% of patients. In addition, the occurrence of complications such as nerve related complications, contrast agent related complications, coronary artery related complications, etc. may lead to the occurrence of adverse patient outcome events. Therefore, possible complications are predicted early, influencing factors causing the complications are identified, personalized intervention on patients can be facilitated early, and occurrence of main adverse cardiovascular events (Major Adverse Cardiovascular Events, MACEs) is reduced.
In the related art, most models for predicting complications are single-objective/task models, which ignore complex interactions existing between complications, and do not consider synergistic or antagonistic actions possibly existing between complications (i.e. one complication may increase or decrease the probability of occurrence of other complications), so that the complications possibly exist cannot be completely covered, personalized intervention measures cannot be formulated for patients, and postoperative complications risks of the patients are improved.
Disclosure of Invention
Therefore, the technical problem to be solved by the invention is to overcome the defects that complications possibly appear and problems accompanying the complications exist in the prior art that the full coverage cannot be predicted, so that a training method for a postoperative complication prediction model and a postoperative complication prediction method are provided.
With reference to the first aspect, the present invention provides a method for training a postoperative complication prediction model, the method comprising:
acquiring clinical data;
determining a prediction task based on the clinical data, wherein the prediction task is used for representing complications of a patient after operation;
preprocessing the clinical data to obtain a training set;
constructing an initial postoperative complication prediction model based on the prediction task;
and based on the training set, performing multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model.
In the mode, a multi-task learning complication prediction model is constructed on the basis of considering possible synergic or antagonistic actions among various complications, and the complications can be predicted more accurately and comprehensively by training the complication prediction model, so that the prognosis of a patient is effectively improved, and the life quality of the patient is improved.
With reference to the first aspect, in a first embodiment of the first aspect, the preprocessing the clinical data to obtain a training set includes:
labeling complications after operation of a patient on the clinical data to obtain a first data set;
expanding the dimension of the first data set based on the prediction task to obtain second data sets with equal complication proportion of each category;
normalizing the second data set to obtain a third data set;
and filling the missing value of the third data set to obtain a training set.
With reference to the first aspect, in a second embodiment of the first aspect, the preprocessing the clinical data to obtain a training set further includes:
and based on the prediction task, carrying out feature screening on the training set to obtain features and feature values of the training set.
With reference to the second embodiment of the first aspect, in a third embodiment of the first aspect, the performing, based on the training set, a multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model includes:
determining a loss function based on the initial postoperative complications prediction model;
calculating to obtain the error between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function, the characteristics and the characteristic values of the training set;
and based on the error, performing multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model.
With reference to the third embodiment of the first aspect, in a fourth embodiment of the first aspect, the calculating, based on the loss function, the feature of the training set, and the feature value, an error between a prediction probability of each prediction task and a label of a postoperative complication of the patient corresponding to the training set includes:
based on the initial postoperative complication prediction model, respectively determining objective functions of each prediction task;
calculating the prediction probability of each prediction task based on the objective function of each prediction task and the characteristics and the characteristic values of the training set;
and calculating and obtaining errors between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function.
In a second aspect of the present invention, the present invention also provides a method of predicting postoperative complications, the method comprising:
acquiring clinical data;
inputting the clinical data into a postoperative complication prediction model, and predicting and obtaining complications and probabilities of occurrence corresponding to the clinical data, wherein the postoperative complication prediction model is trained by using the postoperative complication prediction model training method of any one of the first aspect and optional embodiments thereof.
In a third aspect of the present invention, the present invention further provides a post-operative complication prediction model training apparatus, the apparatus comprising:
a first acquisition unit configured to acquire clinical data;
a task determination unit for determining a prediction task for characterizing complications occurring after a patient operation based on the clinical data;
the preprocessing unit is used for preprocessing the clinical data to obtain a training set;
the construction unit is used for constructing an initial postoperative complication prediction model based on the prediction task;
and the training unit is used for performing multi-task learning training on the initial postoperative complication prediction model based on the training set to obtain a target postoperative complication prediction model.
With reference to the third aspect, in a first embodiment of the third aspect, the preprocessing unit includes:
the marking unit is used for marking complications after the operation of the patient on the clinical data to obtain a first data set;
the dimension expanding unit is used for expanding the dimension of the first data set based on the prediction task to obtain second data sets with equal complication proportion of each category;
the normalization unit is used for performing normalization processing on the second data set to obtain a third data set;
and the filling unit is used for filling the missing value of the third data set to obtain a training set.
With reference to the third aspect, in a second embodiment of the third aspect, the apparatus further includes:
and the feature screening unit is used for carrying out feature screening on the training set based on the prediction task to obtain the features and the feature values of the training set.
With reference to the second embodiment of the third aspect, in a third embodiment of the third aspect, the training unit includes:
a loss function unit for determining a loss function based on the initial postoperative complications prediction model;
the error unit is used for calculating and obtaining the error between the prediction probability of each prediction task and the labeling of the postoperative complications of the patient corresponding to the training set based on the loss function, the characteristics and the characteristic values of the training set;
and the error training unit is used for performing multi-task learning training on the initial postoperative complication prediction model based on the error to obtain a target postoperative complication prediction model.
With reference to the third embodiment of the third aspect, in a fourth embodiment of the third aspect, the error unit includes:
the objective function unit is used for respectively determining objective functions of all prediction tasks based on the initial postoperative complications prediction model;
the first calculation unit is used for calculating the prediction probability of each prediction task based on the objective function of each prediction task, the characteristics and the characteristic values of the training set;
and the second calculation unit is used for calculating and obtaining the error between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function.
In a fourth aspect of the present invention, the present invention also provides a post-operative complication prediction apparatus, the apparatus comprising:
a second acquisition unit configured to acquire clinical data;
the prediction unit is used for inputting the clinical data into a postoperative complication prediction model to predict and obtain complications and probabilities which occur corresponding to the clinical data, wherein the postoperative complication prediction model is trained by using the postoperative complication prediction model training device in the third aspect.
According to a fifth aspect, the embodiment of the present invention further provides a computer device, including a memory and a processor, where the memory and the processor are communicatively connected to each other, and the memory stores computer instructions, and the processor executes the computer instructions, thereby executing the postoperative complications prediction model training method according to any one of the first aspect and the optional embodiments thereof, or executing the postoperative complications prediction method according to the second aspect.
According to a sixth aspect, embodiments of the present invention further provide a computer readable storage medium storing computer instructions for causing the computer to perform the postoperative complications prediction model training method of any one of the first aspect and optional embodiments thereof or the postoperative complications prediction method of the second aspect.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a post-operative complication prediction model training method according to an exemplary embodiment.
Fig. 2 is a schematic diagram of a post-operative complication prediction model framework according to an exemplary embodiment.
Fig. 3 is a schematic diagram illustrating a visualization of a calculation process of a post-operative complication prediction model according to an exemplary embodiment.
Fig. 4 is a flowchart of a method for postoperative complication prediction according to an exemplary embodiment.
Fig. 5 is a schematic diagram of a PCI post-operative complication assessment system according to an exemplary embodiment.
Fig. 6 is a block diagram of a post-operative complication prediction model training apparatus according to an exemplary embodiment.
Fig. 7 is a block diagram illustrating a structure of a post-operative complication prediction apparatus according to an exemplary embodiment.
Fig. 8 is a schematic diagram of a hardware structure of a computer device according to an exemplary embodiment.
Detailed Description
The following description of the embodiments of the present invention will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the invention are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the related art, most of the current models for complication prediction are single-objective/task models, and the models ignore complex interactions existing between complications, and human body is a complex, and interactions of organs and systems are closely related, so that even small adjustment of single risk factors can significantly influence the health condition and clinical result of patients. Taking the common complication of Percutaneous Coronary Intervention (PCI), i.e. thrombosis as an example, the thrombosis can cause ischemic cerebral apoplexy along with blood flowing to the brain, and anticoagulant is needed to be taken to prevent the cerebral apoplexy from aggravating, and the use of anticoagulant increases the risks of cerebral hemorrhage and alimentary tract hemorrhage, so that the balance between cerebral apoplexy and cerebral hemorrhage cannot be solved by a single-task learning model. In the prior art, the possible synergic or antagonistic actions among the complications are not considered, the possible complications which cannot be covered completely are predicted, personalized intervention measures cannot be formulated for the patient, and the postoperative complication risk of the patient is improved.
In order to solve the above-mentioned problems, in the embodiments of the present invention, a post-operative complication prediction model training method is provided and used in a computer device, and it should be noted that an execution body of the post-operative complication prediction model training device may be a part or all of the computer device, where the computer device may be a terminal, a client, or a server, the server may be a server, or a server cluster formed by multiple servers, and the terminal in the embodiments of the present invention may be a smart phone, a personal computer, a tablet computer, a wearable device, and other intelligent hardware devices such as an intelligent robot. In the following method embodiments, the execution subject is a computer device.
The computer device in this embodiment is suitable for use scenarios in which postoperative complications and corresponding probabilities that may occur to a patient are evaluated. According to the training method for the postoperative complication prediction model, the multi-task learning complication prediction model is constructed on the basis of considering possible synergic or antagonistic actions among various complications, and the complications can be predicted more accurately and comprehensively by training the complication prediction model, so that prognosis of a patient is effectively improved, and life quality of the patient is improved.
FIG. 1 is a flowchart of a post-operative complication prediction model training method according to an exemplary embodiment. As shown in fig. 1, the postoperative complications prediction model training method includes the following steps S101 to S105.
In step S101, clinical data is acquired.
In an embodiment of the invention, the clinical data includes data related to a patient in need of PCI surgery.
In an example, the process of acquiring clinical data may include: collecting the patient needing PCI operation in a certain hospital, and collecting information such as demographic indexes (age, sex, occupation and medical insurance), laboratory detection indexes (blood convention and urine convention), imaging examination (heart color ultrasound, dynamic electrocardiogram and coronary angiography), treatment (anticoagulation, antiplatelet and diuretic) and the like. Wherein the patient information inclusion criteria include: STEMI (including anterior and posterior wall myocardial infarction) within 12h of onset or patient information with new occurrences or possible new occurrences of left bundle branch block; shock appears within 36h of onset, and the lesion is suitable for patient information of vascular reconstruction; patient information with evidence of persistent ischemia was found within 12-24 hours of onset. Patient information exclusion criteria included: patient information of previous hemorrhagic diseases; patient information of contrast agent allergies; patient information on allergies to platelet drugs and/or materials of stents; patient information of simple coronary artery spasms; patient information of pre-expansion insufficiency of severe calcified lesions; patient information for diseases with high complications and high mortality.
In step S102, a prediction task is determined based on the clinical data.
In an embodiment of the invention, the predictive task is used to characterize complications that occur post-operatively to the patient. Since PCI surgery is invasive surgery, some complications inevitably occur after surgery, which affect the prognosis of patients. Some complications are associated with patient specificity (e.g., thrombosis, restenosis, etc., disease specificity that may be due to patient individualization); other complications are sporadic phenomena with very little chance (e.g., sporadic surgical instrument-related complications, bleeding, edema, etc. that occur during guidewire advancement). Therefore, occasional complications need to be excluded and only complications associated with patient specificity are modeled as predictive tasks.
In an example, the prediction tasks may include the following tasks:
task 1: intravascular thrombosis or shedding, and the resulting embolism of vital organs (e.g., pulmonary embolism, renal embolism, cerebral embolism, etc.).
Task 2: incidence of heart and cerebral vascular accidents during/after surgery: acute pulmonary edema and heart failure may occur during surgery; myocardial infarction comprises acute Q-wave myocardial infarction and acute non-Q-wave myocardial infarction; cerebrovascular events (including TIA and stroke).
Task 3: the lower limb vein thrombosis and thrombus shedding cause malignant arrhythmia such as acute pulmonary embolism, ventricular tachycardia, ventricular fibrillation, serious atrioventricular block, cardiac arrest and the like.
Task 4: hemorrhage (such as cerebral hemorrhage, and gastrointestinal hemorrhage) induced by antithrombotic drug after operation.
Task 5: insensitivity to anti-embolic drugs due to individual differences in patients can lead to thrombosis in the coronary arteries, possibly leading to myocardial infarction or sudden death.
Task 6: restenosis.
Task 7: death: death due to various causes, surgery-related death.
In step S103, the clinical data is preprocessed to obtain a training set.
In the embodiment of the invention, because clinical data comprises continuous variables such as age, laboratory indexes and the like and classification indexes such as gender and the like, a training set is required to be preprocessed for ensuring subsequent training, so that the model is more convenient to calculate. The pretreatment process may include: labeling complications after operation of a patient on clinical data to obtain a first data set; expanding the dimension of the first data set based on the prediction task to obtain second data sets with equal complication proportion of each category; normalizing the second data set to obtain a third data set; and filling the missing value of the third data set to obtain a training set.
In one example, the data preprocessing process may include the steps of:
1) Training set, test set, validation set partitioning
Clinical data of a patient who performs a PCI operation in one hospital is set as Training data (Training data), and clinical data of a patient who performs a PCI operation in another hospital is set as external verification data (verification data/verification set). The Training data is divided into a Training set (Training set) and a Test set (Test set) by ten-fold cross validation, and a first data set is obtained.
2) SMOTE processing
When constructing a PCI complication prediction model, clinical data has the problem of unbalanced task categories because of different occurrence probabilities of complications of patients and different occurrence types of complications. Taking the cerebrovascular accident in task 2 as an example, of 1000 patients, 30 patients with the cerebrovascular accident are assumed, 970 patients without the cerebrovascular accident are assumed, and the category proportion of the cerebrovascular accident is 3:97, thus the clinical data of cerebrovascular accident in task 2 belongs to the severely unbalanced data. The training set obtained in the last step is oversampled by adopting an SMOTE algorithm, the proportion of the task labels is controlled strictly internally, the class of the task labels is controlled to be 1:1, the weight (loss_weight) of the loss function of each task is distributed in a uniform distribution mode, and a second data set with equal proportion of complications of each class is obtained.
3) normalization/One-hot encoding
Because the clinical data contains two types of continuous variables and classified variables, in order to facilitate the calculation of a subsequent model, the continuous variables need to be normalized by adopting a Z-score:
wherein mean represents the mean value of the current continuous feature, std represents the variance of the current continuous feature, X is the actual measured value, X new Is a post-conversion indicator.
Meanwhile, the sorting variable is processed by adopting One-hot coding, and the processed continuous variable sorting variable is integrated to obtain a third data set.
4) Missing value filling
In order to ensure that the data structure is kept unchanged as much as possible before and after filling, deleting the variables with the deletion proportion of more than or equal to 30% in the third data set obtained in the last step, and filling the variables with the deletion proportion of less than 30% by adopting a random forest. The general ideas of the random forest method include: for a data with n features, where feature T has a missing value, feature T is taken as a tag, and the other n-1 features and the original tag form a new feature matrix. For the feature T, the part which is not missing is the predicted value y_test, and the part of data has a label and a feature; and the missing part, only the feature without the tag, is the part needing prediction.
Thus, the specific process of missing value padding is as follows:
construction
Is carried out by constructing a random forest model by using +.>
For a pair of
Filling;
wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the missing variable X S Values of observed variables other than; />
Representing the missing variable X S Is a measurement of the observed value of (2);
x represents S Other observations than the missing value of (2); />
X represents S Is a missing value of (c).
For continuous variables, when
At minimum, or for a categorical variable, when
At minimum, the iteration is stopped, the filling result at the latest time is->
And (4) the last time->
And (3) when the missing variable is consistent, obtaining a final filling result of the missing variable. Wherein->
And->
Respectively representing values of the variables before and after filling the continuous variable, wherein delta N represents the values of the changes before and after filling the continuous variable; # NA represents the total number of missing values in the discrete variable,
indicating that the value is 1 when the value before and after filling the classification variable is different; Δf represents the class variable padding variation.
In a real-time scenario, different from the single-task learning, in order to ensure the accuracy of model construction and the efficiency of model construction, the preprocessing process for clinical data may further include: and based on the prediction task, carrying out feature screening on the training set to obtain the features and the feature values of the training set.
In one example, the feature screening process may include: and adopting a single factor analysis and clinical expert method to carry out characteristic variable screening work. Firstly, adopting a single factor analysis method to carry out single factor statistical analysis on the tasks, wherein t-test or variance analysis is adopted for carrying out group comparison on continuous variable of normal distribution, and rank sum test is adopted for continuous variable of non-normal distribution; classification variables were checked using chi-square. A variable set of a test level α=0.05, p <0.05 was taken as a candidate variable set for each task described above. Based on single factor analysis, an expert consultation method is adopted to adjust the candidate variable set, and variables which have no statistical significance but important clinical significance in the single factor analysis are included in the variable set to obtain the characteristics and the characteristic values of a training set for training a postoperative complication prediction model.
In step S104, an initial post-operative complication prediction model is constructed based on the prediction task.
In the embodiment of the invention, the mode of giving personalized weight to each prediction task is adopted to construct the postoperative complication prediction model frame in consideration of possible synergic or antagonistic actions among the prediction tasks, so that the problems that part of tasks are strongly related and part of tasks are weakly related can be well solved.
In step S105, based on the training set, a multi-task learning training is performed on the initial postoperative complication prediction model, so as to obtain a target postoperative complication prediction model.
In the embodiment of the invention, the training set obtained by the pretreatment is adopted to carry out multi-task learning training on the initial postoperative complication prediction model, so as to obtain the postoperative complication prediction model with multi-task prediction capability. Through continuous iteration and optimization, the postoperative complications prediction model can more accurately and comprehensively predict the possible postoperative complications of the patient.
Through the embodiment, the multi-task learning complication prediction model is constructed on the basis of considering possible synergic or antagonistic actions among various complications, and the complications can be predicted more accurately and comprehensively by training the complication prediction model, so that the prognosis of a patient is effectively improved, and the life quality of the patient is improved.
The following examples will specifically illustrate the process of constructing and training a post-operative complication prediction model to obtain a target post-operative complication prediction model.
In an embodiment, based on the training set, performing a multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model may include: determining a loss function based on the initial postoperative complications prediction model; calculating to obtain the error between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function, the characteristics and the characteristic values of the training set; based on the errors, the initial postoperative complication prediction model is subjected to multi-task learning training, and the target postoperative complication prediction model is obtained.
In the embodiment of the invention, based on the loss function, the characteristics and the characteristic values of the training set, the error between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set is calculated, and the method comprises the following steps: based on an initial postoperative complication prediction model, respectively determining objective functions of each prediction task; calculating to obtain the prediction probability of each prediction task based on the objective function of each prediction task and the characteristics and the characteristic values of the training set; and calculating and obtaining errors between the prediction probability of each prediction task and the labels of complications after operation of the patient corresponding to the training set based on the loss function.
Fig. 2 is a schematic diagram of a post-operative complication prediction model framework according to an exemplary embodiment. As shown in fig. 2, in one example, the post-operative complication prediction model includes: input layer (Input), expert layer (Expert), gate layer (Gate), task specific layer (power), output layer (Output). The Input layer inputs information to the Expert module and the Gate module; each Expert layer will act from a different perspective to predict tasks; the number of gates is 7 as same as the number of tasks, and the problems that part of tasks are strongly related and part of tasks are weakly related can be well solved by giving personalized weights to each task; the number of the Tower layers is 7 as same as the number of the tasks, the weighted summation of the Gate module and the Expert is taken as input, the prediction result of each task is calculated through the MLP multi-layer perceptron neural network, and the prediction result is Output through the Output layer. In order to prevent gradient from disappearing, the Expert and Gate modules adopt a deep crossover network to calculate, and the specific calculation process is as follows:
1) The first layer is an Input layer (Input) for receiving clinical data of a patient as an initial feature (X);
x represents 100 pieces of preprocessed clinical variable data, including demographic indexes such as age, gender, occupation, medical insurance and the like; laboratory detection indexes such as glycosylated hemoglobin, total cholesterol, triglyceride, NT-ProBNP and the like; heart color Doppler ultrasound, coronary angiography, therapeutic measures, etc.; x is X i The value of the ith feature is shown.
2) The Input layer inputs information into the Expert network layer Expert, which builds several different Expert networks using deep crossover networks (Deep Cross Network, DCN) to prevent gradient vanishing, each Expert module acts from different angles, and outputs f i (x) A. The invention relates to a method for producing a fibre-reinforced plastic composite The specific calculation process is as follows:
fig. 3 is a schematic diagram illustrating a visualization of a calculation process of a post-operative complication prediction model according to an exemplary embodiment. As shown in FIG. 3, X 0 For the original input variable, l represents the number of data transitions, X l For after the first calculationThe feature vector is used to determine the feature vector,
is X l Layer data transpose, W is weight, b is offset. Taking age and sex as examples 0 The first layer of calculation mode is as follows:
the second layer is calculated as follows:
3) At the same time, input is output to Gate, which outputs the probability g that each Expert is selected k (x) The outputs of the three experert are then weighted summed
Output to the power. The number of Gate layers is consistent with the number of tasks, namely each task shares a specific Gate to solve the problem of weak correlation of partial tasks.
4) The fourth layer is a task-specific power layer, which is essentially an MLP (multi-layer perceptron) neural network, and the number of the power layers corresponds to the number of tasks. The input of the Power is
The output is y k ＝h k (f k (x) And), g k (x) Representing a gating network probability function, f i (x) Representing expert module output functions, f k (x) A weighted summation function for the gating network and the expert module, h k (f k (x) A) represents task specific layer Power at f k (x) A function is calculated for the post-input.
5) The last layer is an output layer, different tasks have different objective functions, and the prediction results of the tasks are correspondingly output. And performing multitasking prediction on clinical data of the patient by using the trained model, and continuously iterating and optimizing the model.
In one example, the loss function employs a cross entropy loss function, specifically defined as:
where N represents the number of samples, K represents the number of tasks, W represents the weight of each model loss function, and θ represents the parameter vector of the model. The loss function L is used to evaluate the task K prediction probabilities f (x i θ) and true distribution y i Errors between them.
Fig. 4 is a flowchart of a method for postoperative complication prediction according to an exemplary embodiment. As shown in fig. 4, the method for determining the GPU resource utilization includes the following steps.
In step S401, clinical data is acquired.
In step S402, clinical data is input into a post-operation complication prediction model, and complications and probabilities corresponding to the clinical data are predicted.
Fig. 5 is a schematic diagram of a PCI post-operative complication assessment system according to an exemplary embodiment.
In the embodiment of the invention, the postoperative complication prediction model obtained by training by using the postoperative complication prediction model training method can predict the occurrence and occurrence probability of various postoperative complications of the PCI patient at the same time, effectively improve the prognosis of the patient and is beneficial to improving the life quality of the patient.
Based on the same inventive concept, the invention also provides a postoperative complications prediction model training device.
Fig. 6 is a block diagram of a post-operative complication prediction model training apparatus according to an exemplary embodiment. As shown in fig. 6, the postoperative complication prediction model training apparatus includes a first acquisition unit 601, a task determination unit 602, a preprocessing unit 603, a construction unit 604, and a training unit 605.
A first acquisition unit 601 for acquiring clinical data;
a task determination unit 602, configured to determine a prediction task based on clinical data, where the prediction task is used to characterize a complication occurring after a patient operation;
a preprocessing unit 603, configured to preprocess clinical data to obtain a training set;
a construction unit 604, configured to construct an initial postoperative complication prediction model based on the prediction task;
the training unit 605 is configured to perform a multi-task learning training on the initial postoperative complication prediction model based on the training set, and obtain a target postoperative complication prediction model.
In an embodiment, the preprocessing unit 603 includes: the marking unit is used for marking complications after the operation of the patient on the clinical data to obtain a first data set; the dimension expanding unit is used for expanding the dimension of the first data set based on the prediction task to obtain second data sets with equal complication proportion of each category; the normalization unit is used for performing normalization processing on the second data set to obtain a third data set; and the filling unit is used for filling the missing value of the third data set to obtain a training set.
In another embodiment, the postoperative complication prediction model training device provided by the embodiment of the present invention further includes: and the feature screening unit is used for carrying out feature screening on the training set based on the prediction task to obtain the features and the feature values of the training set.
In yet another embodiment, the training unit 605 includes: a loss function unit for determining a loss function based on the initial postoperative complications prediction model; the error unit is used for calculating and obtaining the error between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function, the characteristics and the characteristic values of the training set; and the error training unit is used for performing multi-task learning training on the initial postoperative complication prediction model based on the error to obtain a target postoperative complication prediction model.
In a further embodiment, the error unit comprises: the objective function unit is used for respectively determining objective functions of all prediction tasks based on the initial postoperative complication prediction model; the first calculation unit is used for calculating the prediction probability of each prediction task based on the objective function of each prediction task, the characteristics and the characteristic values of the training set; the second calculation unit is used for calculating and obtaining errors between the prediction probability of each prediction task and the labels of complications after operation of the patient corresponding to the training set based on the loss function.
The specific limitation and beneficial effects of the postoperative complication prediction model training device can be referred to the limitation of the postoperative complication prediction model training method, and are not described herein. The various modules described above may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
Based on the same inventive concept, the invention also provides a postoperative complications prediction device.
Fig. 7 is a block diagram illustrating a structure of a post-operative complication prediction apparatus according to an exemplary embodiment. As shown in fig. 7, the postoperative complications prediction apparatus includes a second acquisition unit 701 and a prediction unit 702.
A second acquisition unit 701 for acquiring clinical data;
the prediction unit 702 is configured to input clinical data into a post-operation complication prediction model, and predict and obtain complications and probabilities that occur corresponding to the clinical data, where the post-operation complication prediction model is trained by using the post-operation complication prediction model training device.
The specific limitation of the postoperative complication prediction device and the beneficial effects can be referred to the limitation of the postoperative complication prediction method, and the description thereof is omitted herein. The various modules described above may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
Fig. 8 is a schematic diagram of a hardware structure of a computer device according to an exemplary embodiment. As shown in fig. 8, the device includes one or more processors 810 and a memory 820, the memory 820 including persistent memory, volatile memory and a hard disk, one processor 810 being illustrated in fig. 8. The apparatus may further include: an input device 830 and an output device 840.
The processor 810 may be a central processing unit (Central Processing Unit, CPU). The processor 810 may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination thereof. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory 820 is used as a non-transitory computer readable storage medium, including persistent memory, volatile memory, and hard disk, and may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as program instructions/modules corresponding to the post-operative complication prediction model training method or the post-operative complication prediction method in the embodiments of the present application. The processor 810 executes various functional applications and data processing of the server by running non-transitory software programs, instructions, and modules stored in the memory 820, i.e., implementing any one of the post-operative complication prediction model training methods or post-operative complication prediction methods described above.
The input device 830 may receive input numeric or character information and generate key signal inputs related to user settings and function control. The output device 840 may include a display device such as a display screen.
One or more modules are stored in the memory 820 that, when executed by the one or more processors 810, perform the methods illustrated in fig. 1-5.
The product can execute the method provided by the embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method. Technical details which are not described in detail in the present embodiment can be found in the embodiments shown in fig. 1 to 5.
The embodiment of the invention also provides a non-transitory computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the authentication method in any of the method embodiments. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a Flash Memory (Flash Memory), a Hard Disk (HDD), or a Solid State Drive (SSD); the storage medium may also comprise a combination of memories of the kind described above.
It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. While still being apparent from variations or modifications that may be made by those skilled in the art are within the scope of the invention.
Claims (10)
1. A method for training a post-operative complication prediction model, the method comprising:
acquiring clinical data;
determining a prediction task based on the clinical data, wherein the prediction task is used for representing complications of a patient after operation;
preprocessing the clinical data to obtain a training set;
constructing an initial postoperative complication prediction model based on the prediction task;
and based on the training set, performing multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model.
2. The method of claim 1, wherein preprocessing the clinical data to obtain a training set comprises:
labeling complications after operation of a patient on the clinical data to obtain a first data set;
expanding the dimension of the first data set based on the prediction task to obtain second data sets with equal complication proportion of each category;
normalizing the second data set to obtain a third data set;
and filling the missing value of the third data set to obtain a training set.
3. The method of claim 1, wherein the preprocessing the clinical data to obtain a training set further comprises:
and based on the prediction task, carrying out feature screening on the training set to obtain features and feature values of the training set.
4. The method of claim 3, wherein the performing a multi-task learning training on the initial post-operative complication prediction model based on the training set to obtain a target post-operative complication prediction model comprises:
determining a loss function based on the initial postoperative complications prediction model;
calculating to obtain the error between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function, the characteristics and the characteristic values of the training set;
and based on the error, performing multi-task learning training on the initial postoperative complication prediction model to obtain a target postoperative complication prediction model.
5. The method according to claim 4, wherein calculating, based on the loss function, the features of the training set, and the feature values, an error between a prediction probability of each prediction task and a label of a postoperative complication of the patient corresponding to the training set includes:
based on the initial postoperative complication prediction model, respectively determining objective functions of each prediction task;
calculating the prediction probability of each prediction task based on the objective function of each prediction task and the characteristics and the characteristic values of the training set;
and calculating and obtaining errors between the prediction probability of each prediction task and the labeling of the complications after the operation of the patient corresponding to the training set based on the loss function.
6. A method of postoperative complication prediction, the method comprising:
acquiring clinical data;
inputting the clinical data into a postoperative complication prediction model, and predicting to obtain complications and probabilities of occurrence corresponding to the clinical data, wherein the postoperative complication prediction model is trained by using the postoperative complication prediction model training method according to any one of claims 1-5.
7. A post-operative complication prediction model training apparatus, the apparatus comprising:
a first acquisition unit configured to acquire clinical data;
a task determination unit for determining a prediction task for characterizing complications occurring after a patient operation based on the clinical data;
the preprocessing unit is used for preprocessing the clinical data to obtain a training set;
the construction unit is used for constructing an initial postoperative complication prediction model based on the prediction task;
and the training unit is used for performing multi-task learning training on the initial postoperative complication prediction model based on the training set to obtain a target postoperative complication prediction model.
8. A post-operative complication prediction apparatus, the apparatus comprising:
a second acquisition unit configured to acquire clinical data;
the prediction unit is used for inputting the clinical data into a postoperative complication prediction model to predict and obtain complications and probabilities corresponding to the clinical data, wherein the postoperative complication prediction model is trained by using the postoperative complication prediction model training device according to claim 7.
9. A computer device comprising a memory and a processor, the memory and the processor being communicatively coupled to each other, the memory having stored therein computer instructions, the processor executing the computer instructions to perform the post-operative complications prediction model training method of any one of claims 1-5 or the post-operative complications prediction method of claim 6.
10. A computer-readable storage medium storing computer instructions for causing the computer to perform the post-operative complication prediction model training method of any of claims 1-5 or the post-operative complication prediction method of claim 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310266597.3A CN116313053A (en) | 2023-03-16 | 2023-03-16 | Postoperative complications prediction model training method and postoperative complications prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310266597.3A CN116313053A (en) | 2023-03-16 | 2023-03-16 | Postoperative complications prediction model training method and postoperative complications prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116313053A true CN116313053A (en) | 2023-06-23 |
Family
ID=86835689
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310266597.3A Pending CN116313053A (en) | 2023-03-16 | 2023-03-16 | Postoperative complications prediction model training method and postoperative complications prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116313053A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116825356A (en) * | 2023-07-12 | 2023-09-29 | 中国医学科学院基础医学研究所 | Multi-association surgery complication risk assessment method, system and computing equipment |
-
2023
- 2023-03-16 CN CN202310266597.3A patent/CN116313053A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116825356A (en) * | 2023-07-12 | 2023-09-29 | 中国医学科学院基础医学研究所 | Multi-association surgery complication risk assessment method, system and computing equipment |
| CN116825356B (en) * | 2023-07-12 | 2024-02-06 | 中国医学科学院基础医学研究所 | Multi-association surgery complication risk assessment method, system and computing equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Saboor et al. | A method for improving prediction of human heart disease using machine learning algorithms | |
| CN107851464B (en) | Method and system for disease progression modeling and therapy optimization for individual patients | |
| CN109493933A (en) | A kind of prediction meanss of the adverse cardiac events based on attention mechanism | |
| JP2018512659A (en) | Prediction method of remaining life using biological age | |
| Yasnitsky et al. | Dynamic artificial neural networks as basis for medicine revolution | |
| Pal et al. | Deep learning techniques for prediction and diagnosis of diabetes mellitus | |
| CN116313053A (en) | Postoperative complications prediction model training method and postoperative complications prediction method | |
| CN113066574A (en) | Neural network-based aneurysm rupture prediction method, device and storage medium | |
| Gaikwad et al. | Effective study of machine learning algorithms for heart disease prediction | |
| Mamun et al. | FA-1D-CNN implementation to improve diagnosis of heart disease risk level | |
| Kodera et al. | Prospects for cardiovascular medicine using artificial intelligence | |
| Bajpai et al. | Early Prediction of Cardiac Arrest Using Hybrid Machine Learning Models | |
| JP7173482B2 (en) | Health care data analysis system, health care data analysis method and health care data analysis program | |
| CN116682565B (en) | Digital medical information on-line monitoring method, terminal and medium | |
| Tang et al. | A neural network to pulmonary embolism aided diagnosis with a feature selection approach | |
| CN109192306A (en) | A kind of judgment means of diabetes, equipment and computer readable storage medium | |
| Noori et al. | A comparative analysis for diabetic prediction based on machine learning techniques | |
| Wolters | An AI-ECG algorithm for atrial fibrillation risk: steps towards clinical implementation | |
| TW202143248A (en) | System and method of biomedical data prediction risk including a data collection unit, a data processing unit and a judgment unit | |
| CN108346471A (en) | A kind of analysis method and device of pathological data | |
| CN112022140A (en) | Automatic diagnosis method and system for diagnosis conclusion of electrocardiogram | |
| US20220351862A1 (en) | Prediction of the onset of critical limb threatening ischemia (clti) | |
| Kasabe et al. | Cardio Vascular ailments prediction and analysis based on deep learning techniques | |
| Абдуалимов et al. | Prediction of the fact and degree of coronary artery disease using the processing of clinical and instrumental data by artificial intelligence | |
| KR102380312B1 (en) | Method and apparatus for predicting dementia using dementia risk factors according to patient's gender |
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 |