Marker and system for diagnosing abdominal sepsis of children and application

By screening key markers and building machine learning models, the problem of diagnosis of abdominal sepsis in children is solved, high-accurate early diagnosis is achieved, and disease management and patient survival rate are improved.

CN118969242APending Publication Date: 2024-11-15ZHEJIANG UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410997568.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

There is a lack of an effective diagnostic system in the prior art to accurately judge children's abdominal sepsis, which leads to difficulties in early diagnosis and treatment, affecting disease management and patient survival.

Method used

13 key markers were screened out (hypersensitive C-reactive protein, absolute neutrophil value, body weight, body temperature, plasma D-dimer, absolute lymphocyte value, proalbumin, average platelet volume, aspartate acid Aminotransferase, alanine aminotransferase, γ-glutamyltransferase, total bile acid and calcium) and machine learning algorithms are used to construct diagnostic models to assist clinicians in early diagnosis.

Benefits of technology

The accurate diagnosis of abdominal sepsis in children was achieved, with the AUC in the training center reaching 0.982 and the accuracy rate in the test center reaching 92.3%, which improved the accuracy of early diagnosis and the effect of clinical treatment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118969242A_ABST
    Figure CN118969242A_ABST
Patent Text Reader

Abstract

The invention discloses a marker and system for diagnosing abdominal sepsis of children and application, and relates to the technical field of disease medical artificial intelligence. According to the technical scheme, the marker for diagnosing the abdominal sepsis of the children is characterized by comprising a hypersensitive C-reactive protein, a neutrophil absolute value, a body weight, a body temperature, a plasma D-dimer, a lymphocyte absolute value, prealbumin, an average platelet volume, asparaginic acid aminotransferase, alanine aminotransferase, gamma-glutamyltransferase, total bile acid and calcium; the system for diagnosing the abdominal sepsis of children comprises a data input module, a database storage module, a disease prediction module and a model efficiency evaluation module. The marker capable of being used for diagnosing the abdominal sepsis of children is obtained on the basis of laboratory inspection indexes, whether a subject suffers from the abdominal sepsis or not can be accurately diagnosed on the basis of machine learning, clinical doctors can be assisted in diagnosis and early intervention, and huge clinical value is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of disease medical artificial intelligence technology, and more specifically, to a marker, system and application for diagnosing abdominal sepsis in children. Background Art

[0002] Sepsis is a critical illness caused by an imbalance in the body's response to infection, which leads to life-threatening organ dysfunction. It is common in clinical practice and is currently the leading cause of death in intensive care units. Among children with infections, refractory shock and multiple organ dysfunction caused by sepsis are one of the important causes of death.

[0003] The diagnosis of abdominal sepsis is critical for determining surgical indications for abdominal infections, especially for abdominal infections of unknown etiology. A clear diagnosis of sepsis is one of the main bases for clinicians to decide whether to enter the abdomen to explore the primary lesion. Early diagnosis, precise intervention, and timely removal of the source of abdominal infection are the treatment principles for surgical infectious diseases. To date, multiple versions of sepsis diagnosis and treatment guidelines have been released, but these versions are neither unified nor perfect in the diagnostic criteria for sepsis, which brings troubles to clinical diagnosis and treatment.

[0004] Artificial intelligence technology can efficiently integrate huge amounts of data with the powerful computing power of computers, giving users an intuitive result. It has been widely used in multiple digital medical scenarios such as disease diagnosis, auxiliary treatment, and prognosis judgment. Therefore, through artificial intelligence methods, the optimal early diagnostic markers for pediatric abdominal sepsis are identified from the clinical case database, and a clinically applicable and highly reliable diagnostic model is further established, which will help to assist clinical diagnosis as early as possible and further improve the survival rate of children with abdominal sepsis. However, at present, clinical judgments and decisions are mainly based on the clinical experience of doctors. There is no system in the existing technology that is specifically based on markers to accurately diagnose whether a subject has abdominal sepsis.

[0005] To this end, the present invention aims to provide a marker, system and application for diagnosing abdominal sepsis in children to solve the above-mentioned problems. Summary of the invention

[0006] The purpose of the present invention is to provide a marker, system and application for diagnosing abdominal sepsis in children. The present invention obtains a marker that can be used to diagnose abdominal sepsis in children based on laboratory test indicators, and can accurately diagnose whether a subject has abdominal sepsis based on machine learning, thereby assisting clinicians in early diagnosis and treatment.

[0007] The above technical purpose of the present invention is achieved through the following technical scheme: a marker for diagnosing abdominal sepsis in children, including high-sensitivity C-reactive protein, absolute neutrophil value, body weight, body temperature, plasma D-dimer, absolute lymphocyte value, prealbumin, mean platelet volume, aspartate aminotransferase, alanine aminotransferase, γ-glutamyl transferase, total bile acid and calcium.

[0008] The markers are obtained based on the following steps:

[0009] S11, clinical case data of children with abdominal sepsis and other abdominal diseases in the Children's Hospital Affiliated to Zhejiang University School of Medicine in 2019 were obtained, including basic data (gender, age, etc.), nursing signs and test reports within 24 hours of admission. The test reports only included the results of blood routine, blood gas and biochemistry;

[0010] S12, remove indicators with missing values ​​greater than 20%, and fill the missing values ​​in the remaining indicators with the mean value;

[0011] S13, according to the univariate logistics regression algorithm, screened the indicators with significant differences between abdominal sepsis and other abdominal diseases (P<0.05), and then finally screened 13 difference indicators according to the logistics stepwise regression as markers for diagnosing whether the subjects had abdominal sepsis.

[0012] The present invention also provides a system for diagnosing abdominal sepsis in children, which is characterized by comprising a data input module, a database storage module, a disease prediction module and a model effectiveness evaluation module;

[0013] The data input module is used to collect clinical test data of input markers;

[0014] The database storage module is used to store the clinical test data collected and input by the data input module;

[0015] The disease prediction module collects the input clinical test data by calling the data input module, and uses the prediction model to diagnose whether the child suffers from abdominal sepsis.

[0016] The present invention is further configured as follows: among the clinical test data stored in the database storage module, the clinical case data of children with abdominal sepsis are used as positive data, and other children in the abdominal non-sepsis control group are used as negative data.

[0017] The present invention is further configured as follows: the method for constructing the prediction model comprises the following steps:

[0018] S1. The clinical test data of markers in the population samples were randomly stratified and the negative data and positive data were divided into training set and test set according to the ratio of 6:4. Both the training set and the test set included negative and positive data.

[0019] S2. Use the training set data to build a prediction model based on the machine learning algorithm and perform five-fold cross validation;

[0020] S3. In the test set, the initialized prediction model is verified to obtain a standardized prediction model.

[0021] The present invention is further configured as follows: the machine learning algorithms in step S2 include random forest algorithm, support vector machine algorithm, linear regression algorithm, logistic regression algorithm, extreme gradient boosting algorithm and neural network algorithm respectively.

[0022] The present invention is further configured as follows: the prediction model in step S2 is specifically constructed using a random forest algorithm, wherein the parameters are set as: max_depth=4, max_features=0.2, max_sample=0.7, n_estunators=100, class_weight=balanced.

[0023] The present invention also provides an application of a marker for diagnosing abdominal sepsis in children in the preparation of a product for diagnosing abdominal sepsis in children.

[0024] In summary, the present invention has the following beneficial effects:

[0025] 1. The present invention screens out 13 optimal specific markers for early diagnosis of pediatric abdominal sepsis based on clinical test index data: high-sensitivity C-reactive protein, absolute neutrophil count, body weight, body temperature, plasma D-dimer, absolute lymphocyte count, prealbumin, mean platelet volume, aspartate aminotransferase, alanine aminotransferase, γ-glutamyl transferase, total bile acid and calcium. ;

[0026] 2. The present invention established a machine learning model by screening out markers, and the AUC in the training set reached 0.982, in the test set it could reach 0.948, and the accuracy rate in the external validation set also reached 92.3%, indicating that the markers of the present invention can be used to accurately diagnose whether the subject suffers from abdominal sepsis. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a schematic diagram of the process of screening markers in Example 1 of the present invention;

[0028] Figure 2 is a logistic regression forest plot of the screening markers in Example 1 of the present invention;

[0029] Figure 3 It is a schematic diagram of the process of constructing a machine learning model in Example 1 of the present invention;

[0030] Figure 4 is an importance ranking diagram of the markers obtained in Example 1 of the present invention;

[0031] Figure 5 is a ROC curve diagram of the machine learning model in Example 1 of the present invention in the training set and the test set;

[0032] Figure 6 is a confusion matrix diagram of the machine learning model in the external validation set in Example 1 of the present invention;

[0033] Figure 7 This is a diagram of an application page of a system for diagnosing abdominal sepsis in children in Example 2 of the present invention;

[0034] Figure 8 It is a visual illustration of the diagnostic result output and model explanatory properties of the pediatric abdominal sepsis diagnostic system model in Example 2 of the present invention. DETAILED DESCRIPTION

[0035] The following is combined with Figure 1-8 The present invention is described in further detail.

[0036] Example 1: A marker for diagnosing abdominal sepsis in children, specifically including high-sensitivity C-reactive protein, absolute neutrophil count, body weight, body temperature, plasma D-dimer, absolute lymphocyte count, prealbumin, mean platelet volume, aspartate aminotransferase, alanine aminotransferase, γ-glutamyl transferase, total bile acid and calcium.

[0037] The specific process of screening for diagnostic markers for abdominal sepsis in this embodiment is as follows: Figure 1 As shown in the figure, the dataset of the research cohort population came from the electronic medical record database of 3742 children who visited the Children's Hospital Affiliated to Zhejiang University School of Medicine in 2019. Children with abdominal sepsis and other non-septic abdominal diseases were marked by attending physicians or above.

[0038] Among them, attending physicians and above screened 427 cases of abdominal diseases (including 226 cases of sepsis and 201 cases of non-sepsis). The basic data of the children, nursing signs within 24 hours of admission and test report results were included as preliminary indicators. The test report only included five categories: blood routine, blood gas and biochemistry. By removing indicators with missing values ​​greater than 20%, a total of 77 indicators were obtained for subsequent analysis. Finally, the missing values ​​of the 77 indicators were filled with the average value to complete the data processing.

[0039] In the process of screening for diagnostic markers of abdominal sepsis, this example uses univariate logistic regression to perform differential analysis and screen 47 indicators with significant differences (p<0.05) between the abdominal sepsis group and the non-sepsis group. In order to reduce the complexity of the model and improve the calculation efficiency, explanatory power and stability of the model, this example uses the stepwise regression method and finds a total of 13 independent differential indicators, namely, diagnostic markers of abdominal sepsis in children, such as Figure 2 shown.

[0040] The screened abdominal sepsis diagnostic markers were used to establish a machine learning model for the diagnosis of abdominal sepsis in children. The construction process is as follows: Figure 3 As shown in the figure. In the process of constructing the diagnostic machine learning model, the data of the sepsis group and the non-sepsis group were divided into a training set and a test set according to a ratio of 6:4 by random stratified sampling. Among them, there are 256 samples in the training set, including 135 cases of sepsis and 121 cases of non-sepsis; there are 171 samples in the test set, including 91 cases of sepsis and 80 cases of non-sepsis.

[0041] The diagnostic model of pediatric abdominal sepsis was established by using the random forest algorithm on the training set, and different hyperparameter combinations were traversed using grid search. The model was evaluated by 5-fold cross-validation. The area under the receiver operating characteristic curve (AUC) was used to evaluate the model prediction effect. When max_depth = 4, max_features = 0.2, max_sample = 0.7, n_estunators = 100, and class_weight = balanced, as shown in Table 1:

[0042] Table 1 Optimal random forest model parameters

[0043] parameter Parameter Description Select value max_depth Maximum depth of a subtree 4 max_features Each decision tree randomly selects the number of features 0.2 max_samples Sample sampling ratio 0.7 n_estunators The number of trees in the random forest 100 class_weight Sample weight balance random_state Random Seed 0

[0044] At this point, the model achieves the optimal prediction effect.

[0045] Finally, the optimal hyperparameters were determined and a stable diagnostic model for pediatric abdominal sepsis was constructed using all training set data. The importance of the 13 markers was ranked using the feature_importances of the random forest model, as shown in Figure 2. Figure 4 shown.

[0046] In the model performance evaluation, the established pediatric abdominal sepsis diagnosis model was applied to the training set and test set data for sepsis diagnosis. The AUC, sensitivity, and specificity of the receiver operating characteristic curve were also used to evaluate the model prediction performance. Figure 5 As shown, from Figure 5 As can be seen from the figure, the model performed very well in the training set (Sensitivity = 0.941, Specificity = 0.942, AUC = 0.982). At the same time, in the test set, the model can also well identify children with sepsis and non-sepsis (Sensitivity = 0.868, Specificity = 0.875, AUC = 0.948).

[0047] In addition, this embodiment also uses external data to verify the effectiveness of 13 markers. Specifically, data on 299 pediatric abdominal diseases were obtained from the electronic medical record database of the Children's Hospital Affiliated to Zhejiang University School of Medicine, including 93 children with abdominal sepsis and 203 children with other non-sepsis abdominal diseases. In the validation cohort, the children were predicted using the diagnostic model constructed in the above Example 2. The results showed that the model can identify all children with sepsis in the validation set, such as Figure 6 As shown, the confusion matrix Accuracy = (TP + TN) / (TP + FN + FP + TN) = (96 + 180) / (96 + 23 + 0 + 180) = 0.923, that is, the overall prediction accuracy reaches 92.3%.

[0048] Embodiment 2: A system for diagnosing abdominal sepsis in children, comprising a data input module, a database storage module, a disease prediction module and a model effectiveness evaluation module;

[0049] The data input module is used to input the clinical test data of the markers of the first aspect of the present invention in the subjects obtained;

[0050] The database storage module is used to store the clinical test data collected and input by the data input module, wherein the clinical case data of children with abdominal sepsis are taken as positive data, and the other children in the control group with abdominal non-sepsis are taken as negative data;

[0051] The disease prediction module is connected to the data input module and the database storage module respectively, and is used to construct a prediction model using the test results of the markers of the population sample, and diagnose whether the subject suffers from abdominal sepsis based on the marker data of the subject obtained from the data input module;

[0052] The model effectiveness evaluation module is used to evaluate the prediction results of the prediction model.

[0053] The system is built based on a well-established pediatric abdominal sepsis diagnostic model and has the advantages of high reliability and user-friendliness. It helps assist clinicians in the early diagnosis of pediatric abdominal sepsis and further improves the survival rate of children with abdominal sepsis. Figure 7 As shown in Table 2, the system includes the input of basic data of the child, 13 markers and image pictures. In particular, in this embodiment, the data of a virtual child is input. When the patient marker input is as shown in Table 2, Figure 8 The model diagnosis result and model interpretability of the child were given, indicating that the probability that the child had abdominal sepsis was only 0.143, and the model diagnosis result was that the child did not have abdominal sepsis.

[0054] Table 2 Example 4 Patient marker characteristic values

[0055] Markers Select value Weight kg 19.5 Body temperature℃ 36.8 γ-Glutamyltransferase (U / L) 9 Alanine aminotransferase (U / L) 285 High-sensitivity C-reactive protein (mg / L) 1.44 Calcium (mmol / L) 2.23 Absolute value of lymphocytes (10^9 / L) 2.62 Mean platelet volume 11.2 Prealbumin (g / L) 0.158 Aspartate aminotransferase (U / L) 330 Plasma D-dimer (mg / L) 0.05 Neutrophil absolute value (10^9 / L) 2.71 Total bile acid (μmol / L) 3.7

[0056] In this embodiment, constructing a prediction model using markers of population samples includes the following steps:

[0057] S1. The clinical test data of markers in the population samples were randomly stratified and the negative data and positive data were divided into training set and test set according to the ratio of 6:4. Both the training set and the test set included negative and positive data.

[0058] S2. Use the training set data to build a prediction model based on the machine learning algorithm and perform five-fold cross validation;

[0059] S3. In the test set, the initialized prediction model is verified to obtain a standardized prediction model.

[0060] In this embodiment, the machine learning in step S2 adopts any one of the following algorithms: random forest algorithm, support vector machine algorithm, linear regression algorithm, logistic regression algorithm, extreme gradient boosting algorithm and neural network algorithm.

[0061] Preferably, in the present embodiment, the prediction model in step S2 is constructed by using a random forest algorithm, wherein the parameters are set as: max_depth=4, max_features=0.2, max_sample=0.7, n_estunators=100, class_weight=balanced.

[0062] The present invention established a machine learning model through the screened markers, and the AUC reached 0.982 in the training set, 0.948 in the test set, and the accuracy rate in the external validation set also reached 92.3%, indicating that the markers of the present invention can be used to accurately diagnose whether a subject has abdominal sepsis.

[0063] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A marker for diagnosing abdominal sepsis in children, characterized by: These included high-sensitivity C-reactive protein, absolute neutrophil count, body weight, temperature, plasma D-dimer, absolute lymphocyte count, prealbumin, mean platelet volume, aspartate aminotransferase, alanine aminotransferase, γ-glutamyl transferase, total bile acid, and calcium.

2. A system for diagnosing abdominal sepsis in children, characterized by: It includes data input module, database storage module, disease prediction module and model effectiveness evaluation module; The data input module is used to collect clinical test data of input markers; The database storage module is used to store the clinical test data collected and input by the data input module; The disease prediction module collects the input clinical test data by calling the data input module, and uses the prediction model to diagnose whether the child suffers from abdominal sepsis; The model effectiveness evaluation module is used to evaluate the prediction results of the prediction model.

3. A system for diagnosing abdominal sepsis in children according to claim 2, characterized in that: Among the clinical test data stored in the database storage module, the clinical case data of children with abdominal sepsis are used as positive data, and the other children in the control group with abdominal non-sepsis are used as negative data.

4. A system for diagnosing abdominal sepsis in children according to claim 2, characterized in that: The method for constructing the prediction model comprises the following steps: S1. The clinical test data of markers in the population samples were randomly stratified and the negative data and positive data were divided into training set and test set according to the ratio of 6:

4. Both the training set and the test set included negative and positive data. S2. Use the training set data to build a prediction model based on the machine learning algorithm and perform five-fold cross validation; S3. In the test set, the initialized prediction model is verified to obtain a standardized prediction model.

5. A system for diagnosing abdominal sepsis in children according to claim 4, characterized in that: The machine learning algorithms in step S2 include random forest algorithm, support vector machine algorithm, linear regression algorithm, logistic regression algorithm, extreme gradient boosting algorithm and neural network algorithm.

6. A system for diagnosing abdominal sepsis in children according to claim 4, characterized in that: The prediction model in step S2 is specifically constructed using a random forest algorithm, wherein the parameters are set to: max_depth=4, max_features=0.2, max_sample=0.7, n_estunators=100, class_weight=balanced.

7. Use of a marker for diagnosing abdominal sepsis in children in the preparation of a product for diagnosing abdominal sepsis in children.

Citation Information

Cited By

  • Cytarabine syndrome risk prediction model, training method thereof and system adopting same

    CN121260430A