Stomach content model training method, medium, device and application thereof

Abstract

The invention relates to the field of medical treatment, and particularly discloses a stomach content model training method, a medium, a device and application thereof. The invention aims to solve the problem of inaccurate prediction of an existing stomach content model, and particularly provides a stomach content model with higher accuracy, a training method thereof, a medium, a device and application thereof.

Description

Technical Field

[0001] The invention relates to the field of medical prediction, and in particular to a gastric content model training method, medium, device and application thereof. Background Art

[0002] With the advancement of medical technology and the continuous increase in clinical needs, the prevention and management of perioperative complications have become increasingly important. Especially in anesthesia-related operations such as inpatient surgery, day surgery, and outpatient gastroenteroscopy, perioperative reflux aspiration has become one of the most serious complications. Reflux aspiration can lead to a series of serious pulmonary complications such as airway obstruction, pulmonary edema, atelectasis, aspiration pneumonia, and even endanger the patient's life in some cases. According to statistics, the mortality rate caused by aspiration during surgery is as high as 5%, accounting for 9% of all anesthesia-related deaths. Especially in digestive endoscopy, reflux aspiration has become one of the most serious complications, further highlighting the importance of accurately predicting the amount of gastric contents and assessing the risk of reflux.

[0003] The existing gastric content model is mainly based on two variables: age and gastric antral cross-sectional area (CSA), and is predicted through a linear regression model. For example, the model of capacity = 27.0 + 14.6 * right transverse abdominal muscle CSA - 1.28 * age. Although this method provides useful information to a certain extent, it fails to incorporate other complications of the patient and only considers simple linear relationships, resulting in certain deviations in practical applications. Especially when facing patients with complex medical histories, multiple influencing factors, and different types of surgeries, previous models have obvious deficiencies in accuracy and stability, and it is difficult to fully meet clinical needs. Therefore, the development of a new gastric content assessment model has become an urgent need. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of inaccurate prediction of existing gastric content models, and specifically to provide a gastric content model with higher accuracy and its training method, medium, device and application.

[0005] The basic technical solution of the present invention is: a method for training a gastric content model, comprising collecting information, wherein the information includes Perlas classification.

[0006] In recent years, machine learning has attracted much attention as a part of artificial intelligence. Therefore, some technicians in this field have adopted machine learning to train new models of gastric contents. When machine learning is used to train new models, it usually includes several steps such as collecting information data, data processing, feature determination, model training, verification and testing. When collecting information data, previous models usually collect patients' basic information data and the cross-sectional area of ​​the gastric antrum as model training information, and also collect a large number of candidate predictive factors, such as basic information, preoperative management data, clinical information, surgical operation history, etc. In addition to collecting basic information such as age, height, and weight, a series of diseases are also collected, such as asthma, Parkinson's disease, coronary heart disease, hypertension, cirrhosis, neurological complications, cerebral infarction, etc., and gastrointestinal surgery history is also collected; after collecting information data, the data is processed and screened by an algorithm to obtain the screened features, and then trained to obtain a model.

[0007] In model training, it is always difficult to achieve ideal prediction accuracy when using models such as linear regression and random forest. In this regard, the inventors have found through research that after incorporating Perlas classification information, the accuracy of the trained model is significantly better than previous models.

[0008] The ‌Perlas classification‌ is a system for evaluating the status of the gastric antrum, mainly used in ultrasound examinations. It divides the gastric antrum into different levels according to the contents of the gastric antrum to help determine the risk of aspiration. Grade 0‌: No gastric contents are seen in both the supine and right lateral positions, indicating that there is no gastric content and the risk of aspiration is low; Grade 1‌: Only a small amount of fluid is observed in the right lateral position (usually considered to be ≤1.5 mL / kg of fluid), indicating a small amount of fluid in the stomach and a low risk of aspiration‌12; Grade 2‌: Obvious fluid is visible in both the supine and right lateral positions (fluid volume >1.5 mL / kg), indicating that there is a large amount of gastric content and a high risk of aspiration‌.

[0009] The Perlas grade is usually used as a qualitative grade of gastric fluid volume by ultrasound examination. The inventors incorporated its characteristics and used it for quantitative purposes through model training, thereby improving the accuracy of predicting gastric fluid volume and the risk prediction of reflux aspiration.

[0010] Further, the information includes age and / or the cross-sectional area of ​​the gastric antrum in the right lateral decubitus position.

[0011] After the inventors' research, they found that in terms of model performance, the combination of information Perlas classification + right lateral decubitus gastric antrum cross-sectional area (RLD CSA) + age performed best. In addition, Perlas classification + right lateral decubitus gastric antrum cross-sectional area (RLD CSA) performed better than right lateral decubitus gastric antrum cross-sectional area (RLD CSA) alone, and Perlas classification + age also performed better than age alone. Of course, the results of the above parameters are better than previous models. The previous model is: the capacity formula studied by Professor Perlas et al.: gastric volume (ml) = 27.0 + 14.6 × right lateral decubitus gastric antrum cross-sectional area - 1.28 × age.

[0012] Another basic technical solution provided by the present invention is: a gastric content model obtained by the above training method.

[0013] Furthermore, the features collected by the model are Perlas grade; or Perlas grade and age; or Perlas grade and right lateral gastric antral cross-sectional area; or Perlas grade, age and right lateral gastric antral cross-sectional area.

[0014] Furthermore, the model includes any one of a random forest model, an extreme gradient boosting model, an adaptive boosting model, a gradient boosting decision tree model, a neural network model, a K nearest neighbor model, a support vector machine model, and a linear regression model.

[0015] Furthermore, the random forest model uses the following parameters: max_depth=10, max_features=2, min_samples_leaf=2, min_samples_split=5, n_jobs=-1, random_state=90; The parameters used for the extreme gradient boosting model are: colsample_bytree=1.0, gamma=0.2, learning_rate=0.2, max_depth=3, n_estimators=50, subsample=1.0; The adaptive enhancement model uses the following parameters: DecisionTreeRegressor(max_depth=3), random_state=90, learning_rate=0.01, loss='exponential', n_estimators=50; The parameters used for the gradient boosting decision tree model are: learning_rate=0.01, max_depth=3, max_features='sqrt', min_samples_leaf=2, min_samples_split=5, n_estimators=200; The parameters used for the neural network model are: activation='tanh', alpha=0.01, hidden_layer_sizes=(50,), learning_rate='constant', solver='adam'; The parameters used for the K nearest neighbor model are: algorithm='auto', leaf_size=20, n_neighbors=10, p=2, weights='uniform'; The support vector machine model uses the following parameters: C=0.1, gamma='scale', kernel='linear'; The linear regression model uses the formula: .

[0016] The optimal parameters of the model are different when different indicators are selected. The above parameters are the optimal parameters of each model when Perlas grade, age and right lateral gastric antral cross-sectional area are selected. The accuracy is far better than that of previous models.

[0017] The present invention uses the RF method to construct a model for predicting gastric volume: Compared with the previous model that only uses a linear regression model to predict gastric volume, the present model introduces the RF method to construct a model for gastric volume. The ensemble learning method based on the decision tree improves the stability and accuracy of the model by combining the prediction results of multiple decision trees, especially when dealing with nonlinear relationships and outliers.

[0018] A model for predicting gastric volume was constructed using the XGBoost method: a decision tree algorithm based on gradient boosting, which can effectively handle complex nonlinear relationships and prevent overfitting through multiple rounds of iterative optimization and built-in regularization, and is suitable for large-scale high-dimensional data.

[0019] Use the AdaBoost method to build a model for predicting gastric volume: AdaBoost is an integrated learning method that builds a powerful model by combining multiple weak learners (such as decision trees). It can gradually improve the predictive ability and stability of the model and is suitable for complex data environments.

[0020] Use the GBDT method to build a model for predicting gastric volume: GBDT is an integrated learning method that improves prediction accuracy by gradually constructing a residual-based decision tree combination. It is particularly suitable for high-dimensional and nonlinear data and has the advantages of automatically selecting features and avoiding overfitting.

[0021] Use the KNN method to build a model for predicting gastric volume: KNN, as a non-parametric supervised learning algorithm, does not require assumptions about data distribution and predicts the target variable by the distance to neighbors. It is suitable for processing complex features and nonlinear data.

[0022] Use the SVM method to build a model for predicting gastric volume: As a powerful supervised learning algorithm, SVM can effectively process high-dimensional data and performs particularly well in dealing with nonlinear problems. The core idea of ​​SVM is to achieve data classification or regression by finding an optimal hyperplane to maximize the boundary between different categories or data points.

[0023] Use the LR method to build a model for predicting gastric volume: Compared with traditional LR, the LR of the present invention can combine cross-validation and parameter tuning to build a more accurate prediction model.

[0024] Use the MLP method to build a model for predicting gastric volume: MLP is a feedforward neural network that can capture complex feature relationships through multiple layers of nonlinear transformations, and is particularly suitable for processing high-dimensional and nonlinear data. MLP has strong fitting capabilities and optimizes model weights through the back-propagation algorithm to improve prediction accuracy. At the same time, it can effectively avoid overfitting under appropriate regularization.

[0025] Another basic technical solution provided by the present invention is to utilize the storage medium of the above model, that is, to hardware the above model, including various readable carriers of computer-level software such as chips and hard disks.

[0026] Another basic technical solution provided by the present invention is: a device using the above model. In practical applications, the above model can be used in an ultrasonic device to perform gastric antrum examination on a patient.

[0027] Another basic technical solution provided by the present invention is: the application of the above-mentioned gastric content model as a device for predicting gastric content capacity.

[0028] Another basic technical solution provided by the present invention is: the application of the above-mentioned gastric content model as a device for predicting the risk level of reflux aspiration.

[0029] The model of the present invention shows significantly better performance than previous models in predicting gastric content capacity and in predicting medium-high risk and high risk of reflux aspiration, and has stability and consistency in different verification environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is the LASSO screening variable diagram in the first embodiment of the present invention; Figure 2 A flowchart of the prediction model construction in the third embodiment of the present invention; Figure 3 The SHAP global analysis and single sample analysis diagram of the interpretability analysis of the prediction model in Example 3 of the present invention; Figure 4 The following are the prediction performances of each model for the medium and high risk of aspiration in the internal validation set; Figure 5 This is a graph showing the prediction performance of each model for high risk of aspiration in the internal validation set; Figure 6 The following are the prediction performances of each model for the medium and high risk of aspiration in the external validation set; Figure 7 Figure 2 shows the predictive performance of each model for high risk of aspiration in the external validation set. DETAILED DESCRIPTION

[0031] 1. Explanation of terms: CSA: gastric antral cross-sectional area; RLD CSA: right lateral decubitus antral cross-sectional area; Age: age; Previous model: The volume formula studied by Professor Perlas et al.: gastric volume (ml) = 27.0 + 14.6 × right lateral gastric antral cross-sectional area - 1.28 × age; RF: Random Forest Model; XGBoost: Extreme Gradient Boosting Model; AdaBoost: Adaptive boosting model; GBDT: Gradient Boosted Decision Tree; KNN: K nearest neighbor model; MLP: neural network model; SVM: support vector machine model; LR: Linear regression model.

[0032] The following is further described in detail through specific implementation methods: 2. Example 1: Screening and collecting features 1. Information collection: Various clinical data that may be related to gastric emptying during the perioperative period were collected, including the patient's general condition, comorbidities, fasting and drinking conditions, and related information of ultrasound examination. The patient's general condition included age, gender, height, weight, etc. The patient's comorbidities included GOS score, gastritis, antral gastritis, chronic non-atrophic gastritis, gastric ulcer, duodenal ulcer, gastric retention, achalasia, esophagitis, gastroesophageal reflux, esophageal ulcer, esophageal cancer, bile reflux, hiatal hernia, gastroesophageal mucosal ectopy, gastric polyp, gastric stromal tumor, gastric cancer, gastric tumor, duodenal tumor, esophageal leiomyoma, Cirrhosis, portal hypertensive gastropathy, upper gastrointestinal bleeding, liver dysfunction, fatty liver, liver cyst, liver hemangioma, liver mass, primary cholangitis, gallbladder polyps, ulcerative colitis, Crohn's disease, intestinal obstruction, intestinal polyps, pancreatitis, pancreatic tumors, peritonitis, diabetes, hypothyroidism, abdominal surgery history, etc.; patients' fasting and drinking parameters include fasting time, drinking time, drinking dimethicone, etc.; patients' ultrasound parameters include maximum anteroposterior diameter (AP) cm, maximum cranio-caudal diameter (CC) cm, gastric antral cross-sectional area (CSA) (cm²), Perlas score, etc.; patients' outcome information includes gastric content volume; 2. Random division into training set and internal validation set: The data of a hospital in Chongqing (Center 1) were randomly divided into a training set (415 cases) and an internal validation set (179 cases) in a ratio of 7:3.

[0033] 3. Screening features: Using the data from a hospital in Chongqing (Center 1), we screened variables in the training set and used LASSO regression to automatically select the regularization parameter λ that can minimize the mean square error of the model through cross-validation. Figure 1 As shown in the figure, among the initially included characteristics, three key informative variables were finally screened out: age, Perlas grade, and right lateral decubitus gastric antral cross-sectional area (RLD CSA).

[0034] The LASSO regression method is used to screen multiple variables, avoiding the problem of multicollinearity and ensuring the stability and accuracy of the model. This approach significantly improves the model's ability to handle complex data, allowing the present invention to capture deeper variable relationships and surpass the limitations of previous models.

[0035] 4. Model performance evaluation: Grid search combined with five-fold cross-validation was used to optimize the parameters of the AdaBoost model, and the optimal parameters were applied to the model to evaluate its performance on gastric content capacity and its ability to predict aspiration risk.

[0036] 5. Use the AdaBoost model and different features to verify the model performance of the training set and the internal validation set. The data results are shown in Table 1: Table 1: Model performance of different features in the training set and internal validation set (using AdaBoost model)

[0037] It can be seen that in the training set, the feature combination "Perlas classification + RLD CSA + age" performed best, achieving the lowest root mean square error (RMSE: 21.68) and mean absolute error (MAE: 16.35). In addition, the "Perlas classification + RLDCSA" model performed better than RLD CSA alone (RMSE: 22.43 vs. 26.66; MAE: 16.95 vs. 21.19), and the "Perlas classification + age" model also performed better than age alone (RMSE: 21.91 vs. 27.17; MAE: 16.67 vs. 21.47). The results of all features and their feature combinations were better than previous models.

[0038] In the internal validation, the performance of different combinations of age, Perlas grade, and RLD CSA in the model was further explored, and the results were basically consistent with the training set. Therefore, considering the model performance of various factors, the combination of age, Perlas grade, and RLD CSA was considered to be the best choice.

[0039] 6. Using the AdaBoost model, different information variables were used and the internal validation set was used to predict the gastric content capacity. The data results are shown in Table 2: Table 2: Bland-Altman analysis results of different information variables in the internal validation set (using AdaBoost model)

[0040] It can be seen that the performance of different feature combinations of age, Perlas grade, and RLD CSA in the model is basically consistent with the training set. The consistency analysis of different feature combinations in predicting gastric capacity showed that their average bias range (-0.6 to 2.3 mL) was significantly lower than the 23.5 mL of the previous model. Although the average prediction bias of the combination of "Perlas grade + RLD CSA" and "Perlas grade + Age" was slightly smaller than that of "Perlas grade + RLD CSA + Age", the difference was not significant, and in terms of 95% LOA, "Perlas grade + RLD CSA + Age" was better than "Perlas grade + Age" and significantly better than the combination of age, RLD CSA or "Age + RLD CSA" alone. Of course, no matter which combination is used, the model is far superior to the previous model in terms of prediction accuracy. Considering that age may have potential additional explanatory power in prediction, "Perlas + RLD CSA + Age" was finally selected as the best parameter combination to more comprehensively reflect the individual characteristics of patients while ensuring prediction accuracy. In addition, this combination has a more intuitive interpretation in clinical practice, which helps doctors to make more personalized judgments by combining the patient's age factor when evaluating gastric capacity. Therefore, considering the comprehensive model performance and clinical application needs, "Perlas+RLD CSA+Age" is considered the best choice.

[0041] 7. Using the AdaBoost model, different feature variables were used, and the internal validation set was used to predict the medium-high risk and high risk of reflux aspiration. The data results are shown in Tables 3 and 4: Table 3 Prediction of medium and high risk of aspiration by different features in the internal validation set

[0042] Table 4: Prediction of high risk of aspiration by different features in the internal validation set

[0043] It can be seen that in the internal validation of aspiration risk, the prediction AUCs of Perlas grading, age, and RLD CSA for medium- and high-risk aspiration were all above 0.7, and significantly better than the previous model (AUC: 0.629). Among them, the combination of "Perlas grading + RLD CSA" was better than RLD CSA alone in predicting medium- and high-risk aspiration (AUC: 0.774 vs. 0.612), while the combination of "Perlas grading + age" was better than age alone (AUC: 0.717 vs. 0.47). In the prediction of high risk, the performance of each parameter combination was basically consistent with the internal validation results. The prediction AUCs of Perlas grading and its different combinations with age or RLDCSA for high-risk aspiration were all above 0.8, and significantly better than the previous model (AUC: 0.743). Among them, the combination of "Perlas classification + RLD CSA" is better than using RLD CSA alone in predicting high-risk aspiration (AUC: 0.913 vs. 0.764), while the combination of "Perlas classification + age" is better than using age alone (AUC: 0.8 vs. 0.461). The combination of Perlas + RLD CSA + Age has the highest AUC in predicting high-risk aspiration: 0.94.

[0044] In summary, the final optimal feature of the present invention is the combination of "age, Perlas grade and RLD CSA", and it is applied to other learning models and external validation sets. Next, "Perlas grade + RLD CSA" can be selected, and then "Perlas grade + age" can be selected, or Perlas grade can be selected alone. And it is applied to other learning models and external validation sets.

[0045] The feature selection in other models also showed the same results as the AdaBoost model, showing that the best choice was the combination of "age, Perlas grade and RLD CSA", followed by "Perlas grade + RLD CSA", and then "Perlas grade + age", or Perlas grade alone; we will not repeat this.

[0046] 3. Example 2: Model training Take the RF model as an example: 1. We decided to include age, Perlas grade, and RLD CSA as indicators in the random forest (RF) model, and optimized the model hyperparameters through grid search combined with five-fold cross validation. The performance evaluation indicator of the model was the mean square error (MSE). By systematically adjusting the search space of different hyperparameters, we finally determined the optimal parameter combination corresponding to the minimum MSE.

[0047] 2. During the hyperparameter tuning process, adjust the following parameters: n_estimators (the number of trees), max_depth (the maximum depth of the tree), min_samples_split (the minimum number of samples required to split an internal node), min_samples_leaf (the minimum number of samples on a leaf node), and max_features (the maximum number of features considered for each split). Specifically, the candidate values ​​of n_estimators are 50, 100, and 200, the candidate values ​​of max_depth are 10, 20, and no depth limit, the candidate values ​​of min_samples_split are 2 and 5, the candidate values ​​of min_samples_leaf are 1 and 2, and the candidate values ​​of max_features are 2, 3, and "auto".

[0048] After optimization, the optimal hyperparameter combination of the model is: max_depth=10, max_features=2, min_samples_leaf=2, min_samples_split=5, and n_jobs=-1 and random_state=90 are set.

[0049] 3. Apply the above optimal parameters to the RF model, and evaluate the prediction performance of the model on the internal validation set and the external validation set (the data results are consistent with the results of Example 4). In order to quantify the accuracy of the model, the root mean square error (RMSE) and the mean absolute error (MAE) are used as evaluation indicators (the data results are consistent with the results of Example 3). Thus, the RF optimal model is obtained.

[0050] 4. Similarly, the optimal parameter determination method of other models is consistent with this process, ensuring the consistency and comparability of the tuning process.

[0051] 5. Conclusion: The optimal parameters of each model are as follows: (1) The parameters used for the extreme gradient boosting model are: colsample_bytree=1.0, gamma=0.2, learning_rate=0.2, max_depth=3, n_estimators=50, subsample=1.0; (2) The parameters used for the adaptive enhancement model are: DecisionTreeRegressor(max_depth=3), random_state=90, learning_rate=0.01, loss='exponential', n_estimators=50; (3) The parameters used for the gradient boosting decision tree model are: learning_rate=0.01, max_depth=3, max_features='sqrt', min_samples_leaf=2, min_samples_split=5, n_estimators=200; (4) The parameters used for the neural network model are: activation='tanh', alpha=0.01, hidden_layer_sizes=(50,), learning_rate='constant', solver='adam'; (5) The parameters used for the K nearest neighbor model are: algorithm='auto', leaf_size=20, n_neighbors=10, p=2, weights='uniform'; (6) The parameters used in the support vector machine model are: C=0.1, gamma='scale', kernel='linear'; (7) The formula used for the linear regression model is: .

[0052] Example 3: Application of gastric content model to predict gastric content capacity Taking the random forest RF model as an example, the application of the gastric content model of the present invention for predicting gastric content capacity is explained.

[0053] 1. Sample selection: like Figure 2 As shown in the figure, the clinical data of 793 patients were used as the research platform, and the RF model method was used to construct a prediction model for gastric content volume. The data of a hospital in Chongqing (center 1) were randomly divided into a training set and an internal validation set in a ratio of 7:3, with 415 cases in the training set and 179 cases in the internal validation set; the data of a district hospital in Chongqing (center 2) were used as an external validation set, with 199 cases; 2. Collection characteristics: Age, Perlas grade, and RLD CSA parameters were selected, with one-hot encoding used for categorical variables and standardization of continuous features for models that rely on distance metrics or gradient optimization; the median age in the training set was 51 (39, 58), the median age in the internal validation set was 50 (37, 58), and the median age in the external validation set was 54 (47, 64). The proportions of Perlas grade 1 in the training set, internal validation set, and external validation set were 57.8%, 57%, and 57.3%, respectively, and the proportions of Perlas grade 2 were 11.1%, 12.8%, and 8%, respectively; the median RLD CSA in the training set, internal validation set, and external validation set were 5.69 (4.61, 7.37) cm 2 , 5.71 (4.67, 6.93) cm 2 and 5.43 (4.35, 6.72) cm 2 The median gastric content volume in the training set, internal validation set, and external validation set was 25 (10, 50) mL, 20 (10, 45) mL, and 25 (15, 40) mL, respectively.

[0054] 3. Model parameter selection: Age, Perlas grade, and RLD CSA were incorporated into the RF model. The optimal parameter combination was selected by grid search combined with five-fold cross validation, and the prediction performance of the model was evaluated in the internal validation set and the external validation set. By defining the search space of different parameters, the optimal parameters were finally determined to be (max_depth=10, max_features=2,min_samples_leaf=2,min_samples_split=5, n_jobs=-1,random_state=90). These optimal parameters were applied to the RF model, and the prediction performance of the model was evaluated in the internal validation set and the external validation set. In order to quantify the accuracy of the model, the root mean square error (RMSE) and the mean absolute error (MAE) were used as evaluation indicators.

[0055] 4. Model performance: (1) This example provides the gastric content volume observed in the study. The RMSE and MAE of the RF model in the training set are 22.8 and 17.04, respectively, which are much smaller than the previous model (RMSE: 42.06; MAE: 31.46). The RMSE and MAE in the internal validation set are 23.97 and 18.64, respectively, which are much smaller than the previous model (RMSE: 38.54; MAE: 32.46). The RMSE and MAE in the external validation set are 16.79 and 13.15, respectively, which are much smaller than the previous model (RMSE: 32.58; MAE: 24.43). It can be seen that the model performance of this example is far better than the previous model.

[0056] (2) In the prediction of gastric content volume in this embodiment, the prediction model of this embodiment has good consistency, with an average bias of 1 mL in the internal validation set and 23.5 mL in the previous model; the average bias of the external validation set is 0.4 mL, and 14.9 mL in the previous model. It can be seen that the prediction accuracy of this embodiment is far superior to that of the previous model.

[0057] 5. Model explanation: like Figure 3 As shown, the present invention provides a global explanation diagram of the SHAP value to reveal the decision logic of the model as a whole. As can be seen from the figure, with the increase of Perlas score, the increase of RLD CSA and the decrease of age, the predicted gastric volume shows an upward trend.

[0058] In order to further demonstrate how to use the model to make personalized predictions of gastric volume for individual patients, the present invention lists two specific cases. According to the results of the prediction model, a 50-year-old patient with a Perlas score of 0 and an RLD CSA of 7.52 cm² has a predicted gastric volume of 87.25 mL; while another 71-year-old patient with a Perlas score of 2 and an RLD CSA of 10.52 cm² has a predicted gastric volume of only 1.64 mL. It can be seen that compared with the existing foreign models (Volume=27.0+14.6×RLD CSA-1.28×Age), the model of this embodiment shows better risk identification ability.

[0059] 5. Application of other models for predicting gastric content volume 1. The specific implementation method is consistent with that of Example 3, except that the optimal parameters ultimately determined by each model and the predicted results are different.

[0060] 2. The optimal parameters of each model are consistent with those in Example 2.

[0061] 3. Performance data of each model (1) The model performance of each model in the training set, internal validation set and external validation set is shown in Table 5: Table 5 Model performance of each model in the training set, internal validation set and internal validation set

[0062] (2) The prediction performance of each model in the internal validation set and the external validation set for gastric content volume is shown in Table 6: Table 6: Bland-Altman analysis results of each model in the internal validation set and external validation set

[0063] The above results show that the model performance of each model of the present invention in the training set, internal validation set and external validation set, and the prediction performance of gastric content capacity in the internal validation set and external validation set are far superior to previous models.

[0064] VI. Example 4. Application of Gastric Content Model for Prediction of Medium-High Risk and High Risk of Reflux Aspiration Taking the RF model as an example, the application of the gastric content model of the present invention as a medium-high risk and high-risk reflux aspiration is explained.

[0065] 1. Sample selection: like Figure 2 As shown in the figure, the clinical data of 793 patients were used as the research platform, and the RF model method was used to construct a prediction model for gastric content volume. The data of a hospital in Chongqing (center 1) were randomly divided into a training set and an internal validation set in a ratio of 7:3, with 415 cases in the training set and 179 cases in the internal validation set; the data of a district hospital in Chongqing (center 2) were used as an external validation set, with 199 cases; 2. Model indicator selection: Age, Perlas grade, and RLD CSA parameters were selected, with one-hot encoding used for categorical variables and standardization of continuous features for models that rely on distance metrics or gradient optimization; the median age in the training set was 51 (39, 58), the median age in the internal validation set was 50 (37, 58), and the median age in the external validation set was 54 (47, 64). The proportions of Perlas grade 1 in the training set, internal validation set, and external validation set were 57.8%, 57%, and 57.3%, respectively, and the proportions of Perlas grade 2 were 11.1%, 12.8%, and 8%, respectively; the median RLD CSA in the training set, internal validation set, and external validation set were 5.69 (4.61, 7.37) cm 2 , 5.71 (4.67, 6.93) cm 2and 5.43 (4.35, 6.72)cm 2 The median gastric content volume in the training set, internal validation set, and external validation set was 25 (10, 50) mL, 20 (10, 45) mL, and 25 (15, 40) mL, respectively.

[0066] 3. Model parameter selection: Age, Perlas grade, and RLD CSA were incorporated into the RF model, and the optimal parameter combination was selected through grid search combined with five-fold cross validation, and the prediction performance of the model was evaluated in the internal validation set and the external validation set. By defining the search space of different parameters, the optimal parameters were finally determined to be (max_depth=10, max_features=2,min_samples_leaf=2,min_samples_split=5, n_jobs=-1,random_state=90).

[0067] 4. Application of predictive ability to predict aspiration risk The optimal hyperparameters were applied to the random forest (RF) model, and the predicted gastric volume was calculated in the internal and external validation sets. Based on the predicted gastric volume, the performance of the model in predicting moderate-high risk and high risk of aspiration was further evaluated.

[0068] 5. Prediction conclusion Specifically, in the RF model, moderate to high risk of aspiration was defined as gastric volume ≥50 mL or ≥0.8 mL / kg, and high risk of aspiration was defined as gastric volume ≥100 mL or ≥1.5 mL / kg. Figure 4 — Figure 7 As shown, the calculated area under the receiver operating characteristic curve (AUC) shows that in the internal validation set, the RF model has a prediction AUC of 0.736 and 0.877 for medium-high risk and high risk of aspiration, respectively, which is significantly higher than the previous model (AUC for medium-high risk of aspiration: 0.629; AUC for high risk of aspiration: 0.743). In the external validation set, the RF model has a prediction AUC of 0.796 and 0.946 for medium-high risk and high risk of aspiration, respectively, which is also significantly higher than the previous model (AUC for medium-high risk of aspiration: 0.759; AUC for high risk of aspiration: 0.774). The above results show that the RF model constructed in this embodiment performs significantly better than the previous model in predicting the risk of aspiration.

[0069] 7. Application of other models to predict the predictive power of aspiration risk 1. The specific implementation method is consistent with the fourth embodiment, the only difference is that the optimal parameters ultimately determined by each model are different and the predicted results are different.

[0070] 2. The optimal parameters of each model are consistent with those in Example 2.

[0071] 3. The prediction performance of each model for medium-high risk and high risk of aspiration in the internal validation set and the external validation set, such as Figure 4 — Figure 7 The data results are shown in Table 7 and Table 8: Table 7 Prediction performance of the model for medium-high and high risk of aspiration in the internal validation set

[0072] Table 8 Prediction performance of the model for medium-high and high risk of aspiration in the external validation set

[0073] It can be seen that the models of the present invention perform better than the previous models in the medium-high risk and high risk of reflux aspiration.

[0074] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention. These should also be regarded as the scope of protection of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for training a gastric content model, characterized in that: Includes collection information, including Perlas classification.

2. The training method of the gastric content model according to claim 1, characterized in that: The information also includes age and / or right lateral gastric antral cross-sectional area.

3. A gastric content model obtained using the training method of claim 1 or 2.

4. The stomach content model according to claim 3, characterized in that: The collection features are comprised of Perlas grade, or consisting of Perlas grade and age, or consisting of Perlas grade and right lateral gastric antrum cross-sectional area, or consisting of Perlas grade, age, and right lateral gastric antrum cross-sectional area.

5. The stomach content model according to claim 4, characterized in that: Including any one of the random forest model, extreme gradient boosting model, adaptive boosting model, gradient boosting decision tree model, neural network model, K nearest neighbor model, support vector machine model, and linear regression model.

6. The stomach content model according to claim 5, characterized in that: The random forest model uses the following parameters: max_depth=10, max_features=2, min_samples_leaf=2, min_samples_split=5, n_jobs=-1, random_state=90; The extreme gradient boosting model uses the following parameters: colsample_bytree=1.0, gamma=0.2, learning_rate=0.2, max_depth=3, n_estimators=50, subsample=1.0; The parameters used in the adaptive enhancement model are: DecisionTreeRegressor(max_depth=3), random_state=90, learning_rate=0.01, loss='exponential', n_estimators=50; The gradient boosting decision tree model uses the following parameters: learning_rate=0.01, max_depth=3, max_features='sqrt', min_samples_leaf=2, min_samples_split=5, n_estimators=200; The neural network model uses the following parameters: activation='tanh', alpha=0.01, hidden_layer_sizes=(50,), learning_rate='constant', solver='adam'; The parameters used in the K nearest neighbor model are: algorithm='auto', leaf_size=20, n_neighbors=10, p=2, weights='uniform'; The support vector machine model uses the following parameters: C=0.1, gamma='scale', kernel='linear'; The linear regression model uses the formula: 。 7. A storage medium using the stomach content model according to any one of claims 4 to 6.

8. A device using the gastric content model according to any one of claims 4 to 6.

9. Use of the gastric content model according to any one of claims 4 to 6 as a device for predicting gastric content capacity.

10. Use of the gastric content model according to any one of claims 4 to 6 as a device for predicting the risk level of reflux aspiration.

Images