Method and system for predicting postoperative pulmonary infection of chest tumor patient based on deep learning

Abstract

The invention provides a chest tumor patient postoperative pulmonary infection prediction method and system based on deep learning, a logic regression model established based on patient data is used as a risk factor for pulmonary infection judgment, and the risk factor is further explained and evaluated by establishing a hybrid expert model. By comprehensively reflecting indexes of the overall evaluation performance, a verification result shows that the accuracy of the hybrid expert model is better than that of a single model, the hybrid expert model is not influenced by distribution of discrete feature and continuous feature data in a training set and a test set, and the hybrid expert model has good generalization ability.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and particularly relates to a method and system for predicting postoperative pulmonary infection in chest tumor patients based on deep learning. Background Art

[0002] Lung cancer is a frequently-occurring malignant tumor in the respiratory system. According to the latest cancer data statistics in China, the incidence rate of lung cancer is 20.37%, and the mortality rate is 26.99%, ranking first among many cancers. Surgical resection is a common treatment method for clinical lung cancer treatment. With the improvement of technical level and the optimization of medical conditions, the success rate of lung cancer surgery has been greatly improved. However, postoperative pulmonary infection in chest surgery patients is one of the most common complications, and the prognosis of infected patients is poor. Chest tumor patients have low self-immunity, and are prone to nosocomial infection due to invasive operations such as surgery and mechanical ventilation.

[0003] The causes of postoperative pulmonary infection in chest tumor patients are affected by multiple factors, such as long-term smoking, age growth, etc. At present, clinical prediction models have been widely used in clinical prognosis and individualized decision-making, etc. The existing pulmonary infection prognosis models for chest tumor patients usually adopt single-factor and multi-factor analysis to screen statistically significant characteristics of patients to obtain risk factors closely related to the degree of pulmonary infection. However, the above analysis can only perform a single statistical analysis on the prognosis infection. If further analysis of the infection risk is required, based on clustering, regression, and classification analysis, higher-dimensional data needs to be collected for weighted analysis, and the information contained in the population characteristics itself cannot be ignored, and the comprehensive influence of unobservable factors such as lifestyle and living area cannot be fully considered. The existence of these unobservable factors makes the sampled data heterogeneous, and different levels or degrees of unobservable factors will make the discriminant model violate the assumption of traditional single statistical distribution. If the comprehensive influence of these unobservable factors can be characterized by latent variables, and part of the data with common errors or the same distribution form is grouped together for analysis, the accuracy of the non-linear discriminant model can be further improved. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for predicting postoperative pulmonary infection in chest tumor patients based on deep learning. By establishing a mixture of experts model based on a multi-logistic regression model, multiple risk prediction models are mixed together to integrate the linear analysis results and non-linear analysis results to obtain a more comprehensive and accurate auxiliary diagnosis result.

[0005] To achieve the above purpose, on the one hand, the present invention provides a method for predicting postoperative pulmonary infection in chest tumor patients based on deep learning, which includes the following steps:

[0006] Obtain patient data, establish multiple logistic regression models to analyze and screen risk factors, and establish an independent risk prediction model;

[0007] Construct a mixture of experts model based on the combined mean and variance, which mixes multiple risk prediction models through a gating network,

[0008] Obtain patient data, establish multiple logistic regression models to analyze and screen risk factors, and establish an independent risk prediction model;

[0009] Construct a mixture of experts model, which mixes multiple risk prediction models through a gating network, where the input of the model is the feature vector x = [x1, x2, …, x m , where m is the number of features, and the weight of each logistic regression model w = [w1, w2, …, w n , where n is the number of logistic regression models, satisfying and w j ≥ 0;

[0010] The construction of the gating network includes:

[0011] Feature transformation: The input feature x is transformed through a fully connected layer to generate an intermediate representation h,

[0012] h = Relu(w g × x + b g )

[0013] where w g is the weight matrix, b g is the bias vector, with dimension d, and Relu is a non-linear function;

[0014] Calculate the attention scores: Pass the intermediate representation h through another fully connected layer to calculate the attention scores a j ,

[0015]

[0016] where, is the weight variable, with dimension h, and c j is the scalar bias;

[0017] Weight normalization: Use the Softmax function to normalize the attention scores a to weights w,

[0018]

[0019] The output of the logistic regression model is:

[0020]

[0021] where \(x\) is the input feature, \(\theta\) j is the parameter vector of the \(j\)-th expert, is the linear combination of the input \(x\) and the parameter vector \(\theta\) j , \(\exp\) is the exponential function, \(P(y = 1|x,\theta\) j ) is the probability that the input \(x\) belongs to the positive class (\(y = 1\));

[0022] The gating network calculates the weights of each expert:

[0023]

[0024] where Score is the attention scoring function of the input feature \(X\) and experts \(j\), \(k\);

[0025] The final prediction is calculated by weighted average:

[0026]

[0027] Based on the prediction results, the mixture of experts model outputs the auxiliary diagnosis of postoperative pulmonary infection in chest tumor patients. Further, the mixture of experts model also extracts non-linear features:

[0028] \(h1 = Relu(W1x + b1)\)

[0029] where: \(W1\) is the weight matrix of the first layer, \(b1\) is the bias vector of the first layer, and \(h1\) is the hidden representation of the first layer;

[0030] Introduce the context information \(c(x)\) and combine the context information with the hidden representation \(h1\):

[0031] \(h2 = ReLU(W2[h1; c(x)] + b2)\)

[0032] where: \(W2\) is the weight matrix of the second layer; \(b2\) is the bias vector of the second layer; \([h1; c(x)]\) is to concatenate \(h1\) and \(c(x)\) together.

[0033] Further, a non-linear expert model is established based on the above non-linear features, and its output is:

[0034] \(P(y = 1|x,\theta\) j ) = \(\sigma(f\) j (x))

[0035] where \(\sigma f\) j (x) is the non-linear transformation of expert \(j\).

[0036] Further, the final density function of the mixture of experts model is:

[0037]

[0038] where \(y\in\{0,1\}\) is the target variable, and \(P(y = 1|x,\theta\) j ) is the positive class probability predicted by expert \(j\) of the linear and non-linear features.

[0039] Furthermore, the loss function of the mixture of experts model is:

[0040]

[0041] Furthermore, the independent risk factors screened by the risk prediction model include: age ≥ 60 years old, having a smoking history, having a history of diabetes, the surgical method being thoracotomy, the intraoperative blood loss ≥ 200 ml, the case stage being stage III, preoperative ALB, preoperative CRP, and CRP / ALB.

[0042] Furthermore, the data set is divided into a training set and a validation set, and through ten-fold cross-validation, the classification accuracy, recall rate, and F value of the independent risk factors are obtained, which are used as indicators comprehensively reflecting the overall evaluation performance.

[0043] On the other hand, the present invention also provides a postoperative pulmonary infection prediction system for chest tumor patients based on deep learning, including:

[0044] A data processing module for obtaining patient data, establishing multiple logistic regression models to analyze and screen risk factors, and establishing an independent risk prediction model;

[0045] An auxiliary diagnosis module for constructing a mixture of experts model based on the joint mean and variance, and the model mixes multiple risk prediction models through a gating network.

[0046] The output of the logistic regression model is:

[0047]

[0048] where \(x\) is the input feature, \(\theta\) j is the parameter vector of the \(j\)-th expert, is the linear combination of the input \(x\) and the parameter vector \(\theta\) j , \(\exp\) is the exponential function, and \(P(y = 1|x,\theta\) j ) is the probability that the input \(x\) belongs to the positive class (\(y = 1\));

[0049] The gating network calculates the weights of each expert:

[0050]

[0051] where Score is the attention scoring function of the input feature \(X\) and experts \(j,k\);

[0052] The final prediction is calculated through weighted average:

[0053]

[0054] Based on the prediction results, the mixture of experts model outputs an auxiliary diagnosis of postoperative pulmonary infection in chest tumor patients.

[0055] On the other hand, the present invention also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to execute the steps of the above method.

[0056] On the other hand, the present invention also provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to execute the steps of the above method.

[0057] The present invention provides a method and system for predicting postoperative pulmonary infection in chest tumor patients based on deep learning. A logistic regression model established based on patient data is used as a risk factor for judging pulmonary infection. By establishing a mixture of experts model, the above risk factors are further explained and evaluated. Through an index comprehensively reflecting the overall evaluation performance, the verification results show that the accuracy of the mixture of experts model is better than that of a single model, and it is not affected by the distribution of discrete feature and continuous feature data in the training set and the test set, and has good generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0059] Figure 1 is a flowchart of a method for predicting postoperative pulmonary infection in chest tumor patients based on deep learning according to an embodiment of the present invention.

[0060] Figure 2 is a system architecture diagram of a system for predicting postoperative pulmonary infection in chest tumor patients based on deep learning according to an embodiment of the present invention.

[0061] Figure 3 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0063] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0065] Figure 1 is a flowchart of a method for predicting postoperative pulmonary infection in chest tumor patients based on deep learning according to an embodiment of the present invention. As Figure 1 shown, the method for predicting postoperative pulmonary infection in chest tumor patients based on deep learning of the present invention includes the following steps:

[0066] S100, obtaining patient data, establishing multiple logistic regression models to analyze and screen risk factors, and establishing an independent risk prediction model.

[0067] S200, constructing a mixture of experts model, which mixes multiple risk prediction models through a gating network.

[0068] Obtain patient data, establish multiple logistic regression models to analyze and screen risk factors, and establish an independent risk prediction model;

[0069] Construct a mixture of experts model, which mixes multiple risk prediction models through a gating network, where the input of the model is the feature vector x of the patient = [x1, x2,..., x m , where m is the number of features, and the weight w of each logistic regression model = [w1, w2,..., w n where n is the number of logistic regression models, satisfying and w j ≥ 0;

[0070] The construction of the gating network includes:

[0071] Feature transformation, the input feature x is transformed through a fully connected layer to generate an intermediate representation h,

[0072] h = Relu(w g × x + b g )

[0073] where w g is the weight matrix, b g is the bias vector with dimension d, and Relu is a non - linear function;

[0074] Calculate the attention scores. Pass the intermediate representation h through another fully - connected layer to calculate the attention scores a for each logistic regression model j ,

[0075]

[0076] where, is the weight variable with dimension h, c j is the scalar bias;

[0077] Weight normalization. Use the Softmax function to normalize the attention scores a to weights w,

[0078]

[0079] Specifically, the output of the logistic regression model is:

[0080]

[0081] where x is the input feature, θ j is the parameter vector of the j - th expert, is the linear combination of the input x and the parameter vector θ j , exp is the exponential function, and P(y = 1|x, θ j ) is the probability that the input x belongs to the positive class (y = 1);

[0082] The gating network calculates the weights of each expert:

[0083]

[0084] where Score is the attention score function for the input feature X and experts j, k;

[0085] The final prediction is calculated through weighted averaging:

[0086]

[0087] Based on the prediction results, the mixture - of - experts model outputs an auxiliary diagnosis of postoperative pulmonary infection in chest tumor patients. The mixture - of - experts model also extracts non - linear features:

[0088] h1 = Relu(W1x + b1)

[0089] Among them: W1 is the weight matrix of the first layer, b1 is the bias vector of the first layer, and h1 is the hidden representation of the first layer;

[0090] Introduce the context information c(x), and combine the context information with the hidden representation h1:

[0091] h2 = ReLU(W2[h1; c(x)] + b2)

[0092] Among them: W2 is the weight matrix of the second layer; b2 is the bias vector of the second layer; [h1; c(x)] is to concatenate h1 and c(x).

[0093] Based on the above non-linear features, a non-linear expert model is established, and its output is:

[0094] P(y = 1|x, θ j ) = σ(f j (x))

[0095] Among them, σf j (x) is the non-linear transformation of expert j.

[0096] The final density function of the mixture of experts model is:

[0097]

[0098] Among them, y ∈ {0, 1} is the target variable, and P(y = 1|x, θ j ) is the positive class probability predicted by expert j of the linear and non-linear features.

[0099] The loss function of the mixture of experts model is:

[0100]

[0101] In a specific embodiment, patients who underwent chest physical therapy after chest tumor surgery were used as the research object. The inclusion criteria were: (1) Diagnosed as primary non-small cell lung cancer by imaging means and biopsy before surgery, and no distant metastasis was found after surgery; (2) The survival period of the patient was still 6 months or more after postoperative evaluation; (3) Lung cancer surgery was performed in our hospital, and radiotherapy and chemotherapy were not carried out before surgery; (4) Patients who completed chest physical therapy after lung cancer surgery. Exclusion criteria: (1) Patients with pulmonary infection or other acute infections in other parts before surgery; (2) Patients who have undergone tracheotomy or other invasive surgeries; (3) Mechanical ventilation time exceeding 24h; (4) Missing clinical data; (5) Complicated with organ function abnormalities, such as liver failure and kidney injury; (6) History of previous lung diseases, such as bronchopneumonia and pulmonary tuberculosis.

[0102] The diagnostic criteria for pulmonary infection refer to the "Diagnostic Criteria for Nosocomial Infection" to evaluate the pulmonary infection situation in patients undergoing lung cancer surgery after chest physical therapy. (1) Body temperature > 38°C; (2) Peripheral blood white blood cell count > 15×10 9 / L; (3) There are obvious clinical symptoms suggesting pulmonary infection, such as coughing and expectoration; (4) There are significant moist rales on pulmonary auscultation; (5) Imaging examinations show manifestations of pulmonary infection. Meeting any four of the above diagnoses is diagnosed as having a pulmonary infection.

[0103] Patient data include: gender, age, smoking history, diabetes history, drinking history, hypertension history, ratio of forced expiratory volume in 1 second to forced vital capacity (FEV1 / FVC); perioperative data: surgical method, operation time, surgical site, intraoperative blood loss, total postoperative thoracic drainage volume, mechanical ventilation time, thoracic drainage time, hospital stay, pathological type, pathological stage; laboratory indicators. 2 mL of fasting cubital venous blood was drawn from the patient before surgery. After centrifugation and separation, the serum albumin (ALB) level was measured by chemical quantitative method, and the C-reactive protein (CRP) level was detected by an automatic analyzer, and the CRP / ALB ratio was calculated. The chest physical therapy time for the patient was 10 days, and whether the patient had a pulmonary infection was recorded during the treatment period.

[0104] In this example, SPSS 22.0 statistical software was used to process the data. All continuous variables were expressed as , and categorical variables were expressed as rates or constituent ratios. The t-test and X 2 test were used for comparison respectively. Multivariate analysis was performed using a logistic regression model. The indicators with P < 0.05 in the univariate analysis results were included in the multivariate regression analysis to screen independent risk factors and establish a risk prediction model. The nomogram was made using R 3.6.2 software. The accuracy of the model was evaluated by comparing the predicted probability and the actual probability of the nomogram. The calibration curve was drawn by repeating the Bootstrap sampling 1000 times, and the bias of the model was evaluated by the Hosmer-Lemeshow (H-L) test. The receiver operating characteristic (Roc) curve was used to evaluate the predictive value of the model. The cartet package and the Bootstrap resampling method were used for internal validation, and the rms package was used to calculate the concordance index (C-index). P < 0.05 was considered statistically significant.

[0105] After logistic regression model analysis, the independent risk factors screened by the risk prediction model include: age ≥ 60 years old, having a smoking history, having a diabetes history, the surgical method being thoracotomy, intraoperative blood loss ≥ 200 ml, case stage III, preoperative ALB, preoperative CRP, and CRP / ALB.

[0106] Based on the above risk factors, the mixture of experts model of this application divides the data set into a training set and a validation set. Through ten-fold cross-validation, the classification accuracy, recall rate, and F-value of independent risk factors are obtained, which are used as indicators comprehensively reflecting the overall evaluation performance. The results are shown in the following table:

[0107]

[0108] Figure 3 It is the system architecture diagram of the prognostic feature visualization system according to an embodiment of the present invention. As Figure 3 shown, a postoperative pulmonary infection prediction system for chest tumor patients based on deep learning of the present invention includes:

[0109] A data processing module 1 for obtaining patient data, establishing multiple logistic regression models to analyze and screen risk factors, and establishing an independent risk prediction model;

[0110] An auxiliary diagnosis module 2 for constructing a mixture of experts model based on combined mean and variance, and the model mixes multiple risk prediction models through a gating network,

[0111] The output of the logistic regression model is:

[0112]

[0113] where x is the input feature, θ j is the parameter vector of the j-th expert, is the linear combination of the input x and the parameter vector θ j , exp is the exponential function, and P(y = 1|x, θ j ) is the probability that the input x belongs to the positive class (y = 1);

[0114] The gating network calculates the weights of each expert:

[0115]

[0116] where Score is the attention scoring function of the input feature X and experts j, k;

[0117] The final prediction is calculated through weighted average:

[0118]

[0119] Based on the prediction results, the mixture of experts model outputs an auxiliary diagnosis of postoperative pulmonary infection in chest tumor patients. Figure 3It is a schematic structural diagram of an electronic device according to an embodiment of the present invention. An electronic device according to an embodiment of the present invention includes one or more input devices 1000, one or more output devices 1000, one or more processors 3000, and a memory 4000.

[0120] In an embodiment of the present invention, the processor 1000, the input device 2000, the output device 3000, and the memory 4000 can be connected via a bus or other means. The input device 2000 and the output device 3000 can be standard wired or wireless communication interfaces.

[0121] The processor 1000 can be a central processing module (Central Processing Unit, CPU), and the processor can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0122] The memory 4000 can be a high-speed RAM memory or a non-volatile memory, such as a disk memory. The memory 4000 is used to store a set of computer programs, and the input device 2000, the output device 3000, and the processor 1000 can call the program code stored in the memory 4000.

[0123] The computer program stored in the memory 4000 includes program instructions, and when the program instructions are executed by the processor, the processor is caused to execute the steps of the patent value evaluation method as described in the above embodiment.

[0124] An embodiment of the present invention also provides a computer-readable storage medium. The computer-readable storage medium can be a high-speed RAM memory or a non-volatile memory, such as a disk memory. The computer-readable storage medium can be connected via an external computing device or a network to read a set of computer programs stored in the computer-readable storage medium. The computer program stored in the computer-readable storage medium includes program instructions, and when the program instructions are executed by the processor, the processor is caused to execute the steps of the patent value evaluation method as described in the above embodiment.

[0125] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning, characterized in that: The following steps are involved: Obtain patient data, establish multiple logistic regression models to analyze and screen risk factors and establish independent risk prediction models; A hybrid expert model is constructed, which mixes multiple risk prediction models together through a gating network, where the input of the model is the patient's feature vector x = [x1, x2, ..., x m ], where m is the number of features, and the weight of each logistic regression model is w = [w1,w2,…,w n ] where n is the number of logistic regression models, satisfying And w j ≥0; The construction of the gating network includes: Feature transformation, the input feature x is transformed through a fully connected layer to generate an intermediate identifier h. h=Relu(in g ×x+b g ) where w g is the weight matrix, b g is the bias vector, dimension is d, Relu is a nonlinear function; Calculate the attention score and pass the intermediate representation h through another fully connected layer to calculate the attention score a for each logistic regression model j , in, is a weight variable with dimensions h, c j is the scalar bias; Weight normalization: Use the Softmax function to normalize the attention score a to the weight w. After screening by the gating network, the output of the logistic regression model is: Where x is the input feature, θ j is the parameter vector of the jth expert, is the input x and parameter vector θ j The linear combination of exp is the exponential function, P(y=1|x,θ j ) is the probability that the input x belongs to the positive class (y = 1); The gating network calculates the weight of each expert: Where Score is the attention score function of the input feature X and experts j, k; The final prediction is calculated by weighted average: The risk factors are verified based on the prediction results, and based on the verification results, the hybrid expert model outputs an auxiliary diagnosis of postoperative lung infection in patients with chest tumors.

2. A method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning as claimed in claim 1, characterized in that: The hybrid expert model also extracts nonlinear features: h1=Relu(W1x+b1) Where: W1 is the weight matrix of the first layer, b1 is the bias vector of the first layer, and h1 is the hidden representation of the first layer; Introduce context information c(x) and combine the context information with the hidden representation h1 set: h2=ReLU(W2[h1;c(x)]+b2) Where: W2 is the second layer weight matrix; b2 is the second layer bias vector; [h1; c(x)] is the concatenation of h1 and c(x).

3. A method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning as claimed in claim 2, characterized in that: Based on the above nonlinear characteristics, a nonlinear expert model is established, and its output is: P(y=1|x,θ j )=σ(f j (x)) Among them, σf j (x) is the nonlinear transformation of expert j.

4. A method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning as claimed in claim 3, characterized in that: The final density function of the mixed expert model is: Among them, y∈{0,1} is the target variable, P(y=1|x,θ j ) is the positive class probability predicted by expert j for linear features and nonlinear features.

5. The method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning as claimed in claim 4, characterized in that: The loss function of the hybrid expert model is:

6. A method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning as claimed in claim 5, characterized in that: The independent risk factors screened by the risk prediction model include: age ≥ 60 years old, smoking history, history of diabetes, thoracotomy, intraoperative blood loss ≥ 200 ml, case stage three, preoperative ALB, preoperative CRP and CRP / ALB.

7. The method for predicting postoperative lung infection in patients with thoracic tumors based on deep learning as claimed in claim 5, characterized in that: The data set was divided into a training set and a validation set. The classification accuracy, recall rate and F value of the independent risk factors were obtained through ten-fold cross validation, which were used as indicators to comprehensively reflect the overall evaluation performance.

8. A deep learning-based prediction system for postoperative lung infection in patients with thoracic tumors, characterized in that: include: Data processing module, used to obtain patient data, establish multiple logistic regression models to analyze and screen risk factors and establish an independent risk prediction model; Auxiliary diagnosis module, used to build a hybrid expert model based on joint mean and variance, which mixes multiple risk prediction models together through a gating network. Obtain patient data, establish multiple logistic regression models to analyze and screen risk factors and establish independent risk prediction models; A hybrid expert model is constructed, which mixes multiple risk prediction models together through a gating network, where the input of the model is the patient's feature vector x = [x1, x2, ..., x m ], where m is the number of features, and the weight of each logistic regression model is w = [w1,w2,…,w n ] where n is the number of logistic regression models, satisfying And w j ≥0; The construction of the gating network includes: Feature transformation, the input feature x is transformed through a fully connected layer to generate an intermediate identifier h. h=Relu(in g ×x+b g ) where w g is the weight matrix, b g is the bias vector, dimension is d, Relu is a nonlinear function; Calculate the attention score and pass the intermediate representation h through another fully connected layer to calculate the attention score a for each logistic regression model j , in, is a weight variable with dimensions h, c j is the scalar bias; Weight normalization: Use the Softmax function to normalize the attention score a to the weight w. The output of the logistic regression model is: Where x is the input feature, θ j is the parameter vector of the jth expert, is the input x and parameter vector θ j The linear combination of exp is the exponential function, P(y=1|x,θ j ) is the probability that the input x belongs to the positive class (y = 1); The gating network calculates the weight of each expert: Where Score is the attention score function of the input feature X and experts j, k; The final prediction is calculated by weighted average: Based on the prediction results, the hybrid expert model outputs auxiliary diagnosis of postoperative lung infection in patients with chest tumors.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

Images