Dynamic evaluation system, electronic equipment and storage medium

Abstract

The invention discloses a dynamic evaluation system which comprises a database and a deep learning model. The database is used for collecting multiple indexes of a user in real time, the deep learning model performs prediction based on the multiple indexes, and the real-time recommended dosage of sodium bicarbonate used by the user is predicted; the database comprises an acquisition baseline index, an information index and an end index; the deep learning model comprises a preprocessing unit, an XGBOOST calculation unit, an SHAP analysis unit, a regression prediction sub-model and an output unit. Based on the indexes of various physiological parameters of the user, especially the important function of the chloride ion level in metabolic acidosis analysis, accurate prediction is performed in the deep learning model, so that the real-time recommended dosage of sodium bicarbonate is accurately output, and the curative effect is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of metabolic acidosis analysis, and in particular to a dynamic evaluation system, electronic equipment and storage medium. Background Art

[0002] Metabolic acidosis is a clinical laboratory indicator of acid-base imbalance caused by various reasons. The buffering acid-base imbalance in the body of such patients leads to a decrease in serum bicarbonate concentration, which in turn causes a secondary decrease in carbon dioxide level and blood pH.

[0003] The assessment of metabolic acidosis in children and the decision on sodium bicarbonate treatment are mainly based on blood pH, bicarbonate HCO3 - Physiological parameters such as acid-base excess (BE) and BE can, to a certain extent, reflect the acid-base balance in the child's body, providing clinicians with a preliminary basis for diagnosis. Assessment methods based on these parameters have long been used clinically and have become a traditional reference for treatment decision-making.

[0004] However, this assessment method based on limited physiological parameters has many obvious shortcomings: 1) Sodium bicarbonate treatment assessment based solely on pH, bicarbonate, and base excess ignores the complex process of acid-base balance regulation in the human body. A few parameters cannot fully and accurately reflect the true condition of metabolic acidosis in children and its underlying pathophysiological mechanisms; 2) A large amount of clinical practice experience has shown that the use of existing assessment methods for sodium bicarbonate treatment has not brought significant clinical benefits to children, especially no reduction in in-hospital mortality or 28-day mortality. This is because traditional assessment methods fail to fully consider the role of other important physiological factors in the occurrence and development of metabolic acidosis, especially the chloride ion level in the patient's blood electrolytes. Summary of the Invention

[0005] In response to the above two problems, the purpose of the present invention is to propose a dynamic evaluation system, electronic device and storage medium. Based on the indicators of multiple physiological parameters of the user, especially the important role of chloride ion level in metabolic acidosis analysis, training and prediction are carried out in a deep learning model, so as to accurately output the real-time recommended dose of sodium bicarbonate and effectively improve the therapeutic effect. At the same time, a characteristic condition judgment mechanism and an emergency mechanism are set up, and specific data are used to limit different situations for separate processing, making the system operation more accurate and highly applicable.

[0006] This is achieved through the following technical solutions: First, a dynamic evaluation system is proposed, including a database and a deep learning model. The database is used to collect multiple indicators of users in real time. The deep learning model predicts the real-time recommended dose of sodium bicarbonate for users based on multiple indicators. The database includes baseline indicators, information indicators and outcome indicators. The baseline indicators include initial pH, HCO3 - 、Cl - 、Na + , K + , AG and lactic acid, information indicators include sodium bicarbonate dosage and administration method, outcome indicators include user background information, post-treatment outcomes, and pH and HCO3 after administration - and Cl - The deep learning model includes a preprocessing unit, an XGBOOST calculation unit, a SHAP analysis unit, a regression prediction sub-model, and an output unit. The preprocessing unit calculates the derived index I=Cl based on multiple indicators in the database. - / Na + and ΔCl - = Cl after administration - - Initial Cl - , also calculates related physiological indicators II, including expected HCO3 - The rise value and AG correction value = AG - lactate; the XGBOOST calculation unit uses the XGBOOST gradient boosting algorithm to calculate each derived indicator and each related physiological indicator; the SHAP analysis unit is used to analyze the calculation results of the XGBOOST calculation unit and transmit the analysis results to the regression prediction sub-model; the regression prediction sub-model outputs the real-time recommended dosage of sodium bicarbonate based on the analysis results and multiple indicators.

[0007] Preferably, the deep learning model divides multiple indicators in the database into 80% training sets and 20% test sets according to a specific ratio. The training set is used to calculate the analysis results, and the test set is input into the regression prediction sub-model.

[0008] Preferably, the user background information includes age, medical history and renal function status.

[0009] Preferably, before the preprocessing unit calculates each derived indicator and each related physiological indicator, it is necessary to perform KNN interpolation processing, add unknown categories, Z-score standardization processing and outlier processing in sequence; wherein, the KNN interpolation method is used to process missing values ​​in each derived indicator and each related physiological indicator; adding unknown categories is used to process each derived indicator and each related physiological indicator that cannot be interpolated using the KNN interpolation method; Z-score standardization processing is used to unify the magnitude of each data; outlier processing is used to eliminate erroneous data, including eliminating any data with pH <6.8 or pH >7.8.

[0010] Preferably, the database is also provided with a characteristic condition judgment mechanism, and the multiple indicators collected in real time also include arterial carbon dioxide partial pressure PaCO2; after collecting multiple indicators of the user in real time, if the characteristic conditions are met, the characteristic conditions include pH <7.35, HCO3 - <22mmol / L, PaCO2<45mmol / L, the deep learning model starts to run; if the characteristic conditions are not met, the deep learning model does not run.

[0011] Preferably, the database is also provided with an emergency mechanism, which is used to bypass the feature condition judgment mechanism and allow the deep learning model to start running when the user is in danger.

[0012] Secondly, an electronic device is proposed, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to support the operation of the dynamic evaluation system as described above when running the computer program.

[0013] In addition, a storage medium is also proposed, in which a computer program is stored. When the computer program runs in a processor, the processor is used to support the operation of the dynamic evaluation system as described above.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The technical solution of the present invention is based on indicators of multiple physiological parameters of the user, especially the important role of chloride ion levels in metabolic acidosis analysis, and is trained and predicted in a deep learning model, thereby accurately outputting the real-time recommended dosage of sodium bicarbonate and effectively improving the therapeutic effect. At the same time, a characteristic condition judgment mechanism and an emergency mechanism are set up, and specific data is used to limit different situations and handle them separately, making the system operation more accurate and highly applicable. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a structural diagram of a dynamic evaluation system; Figure 2 The figure is a flow chart of the operation of a dynamic evaluation system during operation. DETAILED DESCRIPTION

[0016] The following is a combination of the embodiments of the present invention Figure 1 and 2 , the technical solutions in the embodiments of the present invention are described in detail.

[0017] like Figure 1 As shown in the figure, it is a structural diagram of a dynamic evaluation system; Figure 2 The following is a flow chart of the operation of a dynamic evaluation system; combined with Figure 1 and Figure 2As shown, the system mainly includes a database and a deep learning model. The database is used to collect multiple indicators of the user in real time. The deep learning model is trained and predicted based on multiple indicators to predict the real-time recommended dosage of sodium bicarbonate for the user.

[0018] In this embodiment, the database collects the user's baseline indicators, information indicators and outcome indicators in real time. The baseline indicators include the user's initial pH, HCO3 - 、Cl - 、Na + , K + , AG and lactic acid, which are the various physiological parameters of the user before medication. Information indicators include sodium bicarbonate dosage and administration method. The dosage here is the amount of medication the user takes. The administration methods include oral, injection, inhalation and local administration. At the same dosage, different administration methods may cause different reactions, so they need to be recorded to facilitate subsequent fine-tuning of the sodium bicarbonate dosage. For example, if the user takes a certain drug orally, considering the delayed effect, the predicted sodium bicarbonate dosage will be more conservative, and the predicted result will be reduced by 5%. If the user injects a certain drug intravenously, considering that it will take effect faster, the subsequent prediction of the sodium bicarbonate dosage can be more aggressive, and the predicted result will be increased by 5%. Outcome indicators include user background information, post-treatment outcomes, and pH and HCO3 after administration. - and Cl - , user background information should at least include age, medical history and renal function status, and users of different age groups should have different expectations for HCO3 - Metabolic rates often vary, and the final real-time recommended dose needs to be adjusted appropriately based on age. The same principle applies to medical history and renal function. Post-treatment outcomes are calculated based on the different prognoses caused by patients using sodium bicarbonate for acid correction and those not using sodium bicarbonate for acid correction in the training set of the database. Based on different prognoses, such as whether complications will occur and how long it will take for acid-base balance to return to normal, the sodium bicarbonate dosage can be predicted to improve prediction accuracy.

[0019] In this embodiment, the deep learning model includes a preprocessing unit, an XGBOOST calculation unit, a SHAP analysis unit, a regression prediction sub-model and an output unit. The preprocessing unit is based on multiple indicators in the database, including the calculation of the derived indicator I=Cl before and after treatment. - / Na + and ΔCl - = Cl after administration - - Initial Cl - , I corresponds to before treatment, which can reflect the effect of chloride ion retention on acidosis; ΔCl - After the corresponding treatment, the changes in chloride ions and HCO3 -Then, related physiological indicators II were also calculated, including expected HCO3 - The rise value and AG correction value = AG-lactic acid; where the expected HCO3 - Rise in value = HCO3 - Normal value - HCO3 before acid correction - Value, HCO3 - Normal value is the corresponding physiological reference value of healthy people, HCO3 before acid correction - The value is the corresponding HCO3 before the user uses sodium bicarbonate - Numerical value; AG correction value can eliminate the interference of lactic acidosis and thus improve the accuracy of prediction.

[0020] Before the preprocessing unit calculates each derived indicator and each related physiological indicator, it is necessary to perform KNN interpolation, add unknown categories, Z-score standardization, and outlier processing in sequence. KNN stands for K-Nearest Neighbors Imputation. KNN interpolation is used to handle missing values ​​in each derived indicator and each related physiological indicator. Adding unknown categories is used to process each derived indicator and each related physiological indicator that cannot be interpolated using KNN. In other words, the data that cannot be processed is collected and then treated as a new indicator type for subsequent calculations, so that existing data is not missed. Z-score standardization is used to unify the magnitude of each data. Outlier processing is used to eliminate erroneous data, such as any data with pH < 6.8 or pH > 7.8 is considered erroneous data and eliminated.

[0021] In this embodiment, the XGBOOST computing unit uses the XGBOOST gradient boosting algorithm to calculate each derived indicator and each related physiological indicator. XGBOOST, short for eXtreme Gradient Boosting, is an efficient gradient boosting algorithm used to perform classification and regression tasks. It can perform calculations for multiple categories such as each derived indicator and each related physiological indicator. The current usage of the XGBOOST gradient boosting algorithm is publicly available, so the specific calculation process is not repeated here.

[0022] In this embodiment, the SHAP analysis unit is used to analyze the calculation results of the XGBOOST calculation unit. The full name of SHAP is SHapley AddItIve exPlanatIons. It is a method for explaining the output of a machine learning model. It fairly distributes the contribution of each feature to the model prediction based on the Shapley value in game theory. SHAP provides a unified way to explain models containing different types of parameters, including complex black box models such as XGBoost. The SHAP value obtained by analysis can characterize the importance of each type of parameter for a single prediction and the direction of its influence, that is, how each parameter changes the final prediction result. Then, the analysis results are transmitted to the regression prediction sub-model, waiting for prediction.

[0023] In this embodiment, the regression prediction sub-model outputs a recommended sodium bicarbonate dosage based on the analysis results and multiple indicators. The regression prediction sub-model can utilize widely used multivariate regression analysis, decision trees, or neural networks for prediction. All of these can effectively generate accurate prediction results based on the interrelationships between various input data, their respective relationships with the severity of metabolic acidosis, and their relationship with the therapeutic effect.

[0024] In this embodiment, the deep learning model divides multiple indicators in the database into 80% training sets and 20% test sets according to a specific ratio. The training set is used to calculate the analysis results, and the test set is input into the regression prediction sub-model for testing.

[0025] In this embodiment, the database is also provided with a characteristic condition judgment mechanism, and the multiple indicators collected in real time also include arterial carbon dioxide partial pressure PaCO2; after collecting multiple indicators of the user in real time, if the characteristic conditions are met, the characteristic conditions include pH <7.35, HCO3 - <22mmol / L, PaCO2<45mmol / L, the deep learning model starts to run; if the characteristic conditions are not met, the deep learning model does not run.

[0026] In this embodiment, the database is also provided with an emergency mechanism. The emergency mechanism is used to quickly bypass the characteristic condition judgment mechanism when the user is in danger, such as cardiac arrest or shock, and allow the deep learning model to start running, thereby quickly predicting the recommended dosage of sodium bicarbonate required by the user and rescuing the user in time.

[0027] Secondly, an electronic device is proposed, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to support the operation of the dynamic evaluation system as described above when running the computer program.

[0028] In addition, a storage medium is also proposed, in which a computer program is stored. When the computer program runs in a processor, the processor is used to support the operation of the dynamic evaluation system as described above.

[0029] In summary, the present invention uses indicators of multiple physiological parameters of the user, especially the important role of chloride ion levels in metabolic acidosis analysis, to train and predict in a deep learning model, thereby accurately outputting the real-time recommended dosage of sodium bicarbonate and effectively improving the therapeutic effect. At the same time, a characteristic condition judgment mechanism and an emergency mechanism are set up, and specific data are used to define different situations for separate processing, making the system operation more accurate and applicable, and having significant progress.

[0030] The above embodiments are only for illustrating the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A dynamic evaluation system, characterized in that: It includes a database and a deep learning model; the database is used to collect multiple indicators of users in real time, and the deep learning model makes predictions based on the multiple indicators to predict the real-time recommended dosage of sodium bicarbonate for the user; The database includes baseline indicators, information indicators and outcome indicators. The baseline indicators include initial pH, HCO3 - 、Cl - 、Na + , AG and lactic acid, information indicators include sodium bicarbonate dosage and administration method, outcome indicators include user background information, post-treatment outcomes, and pH and HCO3 after administration - and Cl - ; The deep learning model includes a preprocessing unit, an XGBOOST calculation unit, a SHAP analysis unit, and a regression prediction sub-model; the preprocessing unit calculates the derived index I=Cl based on multiple indicators in the database. - / Na + and ΔCl - = Cl after administration - - Initial Cl - , also calculates relevant physiological indicators II, including expected HCO3 - The rise value and AG correction value = AG - lactate; the XGBOOST calculation unit uses the XGBOOST gradient boosting algorithm to calculate each derived indicator and each related physiological indicator; the SHAP analysis unit is used to analyze the calculation results of the XGBOOST calculation unit and transmit the analysis results to the regression prediction sub-model; the regression prediction sub-model outputs the real-time recommended dosage of sodium bicarbonate based on the analysis results and multiple indicators.

2. A dynamic evaluation system according to claim 1, characterized in that: The deep learning model divides multiple indicators in the database into 80% training sets and 20% test sets according to a specific ratio. The training set is used to calculate the analysis results, and the test set is input into the regression prediction sub-model.

3. A dynamic evaluation system according to claim 1, characterized in that: User background information includes age, medical history, and renal function status.

4. A dynamic evaluation system according to claim 1, characterized in that: Before the preprocessing unit calculates each derived index and each related physiological index, it needs to perform KNN interpolation, add unknown categories, Z-score standardization and outlier processing in sequence; Among them, the KNN interpolation method is used to process missing values ​​in each derived indicator and each related physiological indicator; an unknown category is added to process each derived indicator and each related physiological indicator that cannot be interpolated using the KNN interpolation method; Z-score standardization is used to unify the magnitude of each data; outlier processing is used to eliminate erroneous data, including eliminating any data with pH < 6.8 or pH > 7.

8.

5. A dynamic evaluation system according to claim 1, characterized in that: The database is also equipped with a feature condition judgment mechanism. The multiple indicators collected in real time also include arterial carbon dioxide partial pressure PaCO2. After collecting multiple indicators of the user in real time, if the feature conditions are met, the feature conditions include pH <7.35, HCO3 - <22mmol / L, PaCO2<45mmol / L, the deep learning model starts to run; if the characteristic conditions are not met, the deep learning model does not run.

6. A dynamic evaluation system according to claim 5, characterized in that: The database also has an emergency mechanism, which is used to bypass the feature condition judgment mechanism and allow the deep learning model to start running when the user is in danger.

7. An electronic device, characterized in that: The system comprises a processor and a memory, wherein the memory stores a computer program, and the processor is configured to support the operation of the dynamic evaluation system according to any one of claims 1 to 6 when running the computer program.

8. A storage medium, characterized in that: The storage medium stores a computer program. When the computer program runs in the processor, the processor is used to support the operation of the dynamic evaluation system according to any one of claims 1 to 6.

Images