Dynamic prediction method and device for key indicators of non-surgical treatment of ectopic pregnancy

Abstract

The application discloses a kind of ectopic pregnancy non-surgical treatment key index dynamic prediction method and device, it is related to ectopic pregnancy treatment prediction technical field, including: by only extracting the clinical feature vector of patient current single visit, treatment plan to be carried out and time interval, input to gradient boosting tree ensemble model, directly output future time blood beta-hCG level interval and the classification probability of adnexal region mass size interval, to realize the dynamic interval prediction without complete history sequence.Thereby, the prediction model in the prior art relies on complete history sequence, cannot adapt to real world medical data missing and non-uniform sampling, and ignores the dynamic interaction of index, and the problem of single static prediction result.

Description

Technical Field

[0001] This application relates to the field of ectopic pregnancy treatment prediction technology, and in particular to a method and device for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy. Background Technology

[0002] Current clinical prediction methods mostly employ traditional statistical approaches (such as logistic regression). This proposed approach collects clinical characteristics of patients at their first visit or baseline (such as age, initial β-hCG, mass size, etc.), uses univariate and multivariate regression analyses to screen independent predictors, and constructs a linear static prediction model to assess the probability of success of non-surgical treatment or the trend of indicator changes.

[0003] However, this solution has many technical flaws: (1) Limitation of strong linearity assumption: Logistic regression assumes that there is a linear relationship between the predictor variables and the outcome, and that each parameter operates in a homogeneous manner, which cannot capture the nonlinear and high-order interaction features that are common in the evolution of TEP (Tubal Ectopic Pregnancy). (2) Static baseline dependence: It relies solely on the baseline data from the first visit and completely ignores the prognostic information implied by the dynamic changes of indicators during the diagnosis and follow-up process, which causes the prediction results to quickly become invalid when the condition fluctuates; (3) Weak generalization ability: It is sensitive to data distribution. When there are missing, noisy or non-uniform sampling in real-world medical data, the model performance drops significantly.

[0004] In recent years, some existing technologies have incorporated machine learning algorithms (such as decision trees, random forests, and support vector machines) or deep learning time series models (such as LSTM (Long Short-Term Memory) networks) for outcome prediction. These approaches typically integrate multiple patient visit records into a time series, inputting it into a recurrent neural network such as LSTM, utilizing its memory units to capture the impact of historical treatment information on the current outcome; or directly using tree ensemble algorithms to perform nonlinear classification and prediction of multimodal clinical data.

[0005] However, this solution still has the following technical shortcomings: (1) The time series model has too high requirements for data integrity: Deep time series networks such as LSTM rely on complete, continuous and uniformly sampled historical sequences. TEP is an obstetric and gynecological emergency with rapid disease progression, short follow-up period and irregular time intervals. In clinical practice, previous medical records are often missing due to patient referrals and cross-hospital visits. LSTM is unable to give full play to the advantages of long sequence modeling and is prone to training instability or prediction bias. (2) Risk of feature redundancy and overfitting: Directly inputting all historical medical records will introduce a lot of noise and redundant information, which is very easy to overfit with a limited sample size; (3) Not aligned with actual clinical workflow: Existing models mostly assume that physicians can obtain complete disease course data, and do not make lightweight designs for the real clinical scenario of "only the patient's latest medical record can be obtained when receiving a patient", resulting in insufficient usability and robustness of the algorithm when it is implemented in clinical practice.

[0006] In summary, existing predictive models rely on complete historical sequences, which cannot adapt to the lack of real-world medical data and non-uniform sampling. Furthermore, they ignore the dynamic interaction relationships between indicators, resulting in static and singular prediction results, which urgently need to be addressed. Summary of the Invention

[0007] This application provides a method and device for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy, in order to solve the problems in the existing technology that the prediction model relies on a complete historical sequence, cannot adapt to the lack of real-world medical data and non-uniform sampling, and ignores the dynamic interaction relationship of indicators, resulting in static and singular prediction results.

[0008] The first aspect of this application provides a method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy, comprising the following steps: acquiring complete clinical record information of multiple patients with tubal pregnancy to construct and preprocess a corresponding total training set, and extracting training clinical feature vectors of a training subset and validation clinical feature vectors of a validation subset from the preprocessed total training set; collecting multiple key control variables corresponding to the training clinical feature vectors, and training a pre-constructed gradient boosting tree model using the training clinical feature vectors and the multiple key control variables to generate a dynamic indicator interval prediction model, and verifying whether the dynamic indicator interval prediction model meets preset performance requirements based on the validation clinical feature vectors, wherein the multiple key control variables include the proposed non-surgical treatment plan category and time interval; in response to meeting the preset performance requirements, acquiring the current clinical information of the current patient with tubal pregnancy, and inputting the current clinical information into the dynamic indicator interval prediction model to output the dynamic interval of key indicators for the current patient with tubal pregnancy, and generating a target clinical prompt report for the dynamic interval of key indicators.

[0009] Optionally, in one embodiment of this application, the step of obtaining complete clinical record information of multiple patients with ectopic pregnancy to construct and preprocess a corresponding total training set, and extracting the training clinical feature vector of the training subset and the validation clinical feature vector of the validation subset in the preprocessed total training set, includes: obtaining complete clinical record information of the multiple patients with ectopic pregnancy, extracting the structured variable set corresponding to the complete clinical record information, and constructing the total training set based on the structured variable set, wherein the structured variable set includes demographic and medical history characteristics, symptoms and signs, laboratory and ultrasound indicators, and diagnostic and treatment information; performing missing value imputation and standardization on the total training set to obtain a standard training dataset, and dividing the standard training dataset into the training subset and the validation subset based on a preset time threshold; deleting relevant data and postoperative follow-up data corresponding to patients with ectopic pregnancy who only have initial visits and no follow-up data in the training subset and the validation subset, and extracting the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset.

[0010] Optionally, in one embodiment of this application, the step of collecting multiple key control variables corresponding to the training clinical feature vector, training a pre-constructed gradient boosting tree model using the training clinical feature vector and the multiple key control variables to generate a dynamic index interval prediction model, and verifying whether the dynamic index interval prediction model meets preset performance requirements based on the verification clinical feature vector, includes: obtaining the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and determining the proposed non-surgical treatment plan category, the time interval, and the prediction label corresponding to the training clinical feature vector, wherein the prediction label includes serum β-hCG. The model is constructed using a horizontal interval label and an appendix block size interval label. Based on a preset cross-entropy loss function, ordered enhancement mechanism, and category boosting algorithm, a gradient boosting tree model is built. The model is iteratively trained using the proposed non-surgical treatment plan category, the time interval, the training clinical feature vector, and the predicted label until the iteration process meets a preset iteration termination requirement. A preset model solidification operation is then performed on the trained gradient boosting tree model to generate the dynamic indicator interval prediction model. The validation clinical feature vector is input into the dynamic indicator interval prediction model, and the matching degree between the predicted result corresponding to the validation clinical feature vector and the true label corresponding to the validation clinical feature vector is calculated. Multiple validation data for the dynamic indicator interval prediction model are determined based on the matching degree, including accuracy, weighted F1 score, area under the receiver operating characteristic curve, recall, and precision. Based on the multiple validation data, a corresponding confusion matrix is ​​generated. The true positive and false negative distributions for each risk stratification interval are statistically analyzed using the confusion matrix. The true positive and false negative distributions are then used to verify whether the dynamic indicator interval prediction model meets the preset discrimination balance requirement in different clinical risk populations.

[0011] Optionally, in one embodiment of this application, the step of obtaining the current clinical information of the current ectopic pregnancy patient and inputting the current clinical information into the dynamic indicator interval prediction model to output the dynamic interval of the key indicators of the current ectopic pregnancy patient and generate a target clinical prompt report of the key indicator dynamic interval includes: collecting the current clinical information of the current ectopic pregnancy patient and inputting the current clinical information into the dynamic indicator interval prediction model, outputting the probability distribution values ​​of the blood β-hCG and mass size of the current ectopic pregnancy patient falling into each preset interval after the time interval, so as to determine the corresponding key indicator dynamic interval; performing a preset variable importance analysis operation on the dynamic indicator interval prediction model based on the probability distribution values ​​to generate the importance ranking result of the associated variables, and generating the target clinical prompt report according to the key indicator dynamic interval and the importance ranking result of the associated variables; pushing the target clinical prompt report to the target clinical information system through a preset API interface or visualization terminal, so that the target physician can perform corresponding diagnosis and treatment operations on the current ectopic pregnancy patient according to the target clinical prompt report.

[0012] A second aspect of this application provides a device for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy, comprising: a preprocessing module for acquiring complete clinical record information of multiple patients with tubal pregnancy to construct and preprocess a corresponding total training set, and extracting training clinical feature vectors of a training subset and validation clinical feature vectors of a validation subset from the preprocessed total training set; a training module for collecting multiple key control variables corresponding to the training clinical feature vectors, training a pre-constructed gradient boosting tree model using the training clinical feature vectors and the multiple key control variables, generating a dynamic indicator interval prediction model, and verifying whether the dynamic indicator interval prediction model meets preset performance requirements based on the validation clinical feature vectors, wherein the multiple key control variables include the category of proposed non-surgical treatment plan and time interval; and an inference module for acquiring current clinical information of the current patient with tubal pregnancy in response to meeting the preset performance requirements, inputting the current clinical information into the dynamic indicator interval prediction model to output the dynamic interval of key indicators of the current patient with tubal pregnancy, and generating a target clinical prompt report of the dynamic interval of key indicators.

[0013] Optionally, in one embodiment of this application, the preprocessing module includes: an extraction unit, configured to acquire complete clinical record information of the plurality of patients with ectopic pregnancy, extract the structured variable set corresponding to the complete clinical record information, and construct the total training set based on the structured variable set, wherein the structured variable set includes demographic and medical history characteristics, symptoms and signs, laboratory and ultrasound indicators, and diagnostic and treatment information; a partitioning unit, configured to perform missing value imputation and standardization processing on the total training set to obtain a standard training dataset, and partition the standard training dataset into the training subset and the validation subset based on a preset time threshold; and a filtering unit, configured to delete relevant data and postoperative follow-up data corresponding to patients with ectopic pregnancy in the training subset and the validation subset who only have initial visit data and no follow-up data, and extract the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset.

[0014] Optionally, in one embodiment of this application, the training module includes: a determining unit, configured to obtain the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and determine the proposed non-surgical treatment plan category, the time interval, and the prediction label corresponding to the training clinical feature vector, wherein the prediction label includes a blood β-hCG level interval label and an adnexal mass size interval label; an iterating unit, configured to construct the gradient boosting tree model based on a preset cross-entropy loss function, an ordered enhancement mechanism, and a category boosting algorithm, and iteratively train the gradient boosting tree model using the proposed non-surgical treatment plan category, the time interval, the training clinical feature vector, and the prediction label until the iteration process meets a preset iteration termination requirement, and perform a preset model solidification operation on the trained gradient boosting tree model to generate the dynamic index interval prediction model; and a calculation unit, configured to input the verification clinical feature vector into the dynamic index interval prediction model and calculate the verification clinical feature vector. The matching degree between the prediction result corresponding to the vector and the true label corresponding to the validation clinical feature vector is used to determine multiple validation data of the dynamic indicator interval prediction model based on the matching degree. The multiple validation data include accuracy, weighted F1 score, area under the receiver operating characteristic curve, recall, and precision. The validation unit is used to generate a corresponding confusion matrix based on the multiple validation data, to statistically analyze the true positive and false negative distributions of each risk stratification interval based on the confusion matrix, and to verify whether the dynamic indicator interval prediction model meets the preset discrimination balance requirements in different clinical risk populations through the true positive and false negative distributions.

[0015] Optionally, in one embodiment of this application, the inference module includes: a collection unit, used to collect the current clinical information of the current ectopic pregnancy patient, input the current clinical information into the dynamic indicator interval prediction model, and output the probability distribution values ​​of the blood β-hCG and mass size of the current ectopic pregnancy patient falling into each preset interval after the time interval, so as to determine the corresponding key indicator dynamic interval; a parsing unit, used to perform preset variable importance parsing operation on the dynamic indicator interval prediction model based on the probability distribution values, so as to generate the importance ranking result of the associated variable, and generate the target clinical prompt report according to the key indicator dynamic interval and the importance ranking result of the associated variable; and a push unit, used to push the target clinical prompt report to the target clinical information system through a preset API interface or visualization terminal, so that the target physician can perform corresponding diagnosis and treatment operations on the current ectopic pregnancy patient according to the target clinical prompt report.

[0016] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the dynamic prediction method for key indicators of non-surgical treatment of ectopic pregnancy as described in the above embodiments.

[0017] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy.

[0018] A fifth aspect of this application provides a computer program product, including a computer program that is executed to implement the above-described method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy.

[0019] Therefore, the embodiments of this application have the following beneficial effects: The embodiments of this application can construct and preprocess a corresponding total training set by obtaining complete clinical record information from multiple patients with ectopic pregnancy. Training clinical feature vectors of the training subset and validation clinical feature vectors of the validation subset are extracted from the preprocessed total training set. Multiple key control variables corresponding to the training clinical feature vectors are collected, and a pre-constructed gradient boosting tree model is trained using the training clinical feature vectors and multiple key control variables to generate a dynamic indicator interval prediction model. Based on the validation clinical feature vectors, the dynamic indicator interval prediction model is verified to meet preset performance requirements. The multiple key control variables include the proposed non-surgical treatment plan category and time interval. In response to meeting the preset performance requirements, the current clinical information of the current patient with ectopic pregnancy is obtained and input into the dynamic indicator interval prediction model to output the dynamic interval of key indicators for the current patient with ectopic pregnancy and generate a target clinical prompt report for the dynamic interval of key indicators. This application only requires the latest medical records to achieve high-precision dynamic interval prediction, providing clinicians with a quantitative decision support tool that fits their actual workflow. This solves the problems in existing technologies, such as prediction models relying on complete historical sequences, being unable to adapt to missing or non-uniform sampling of real-world medical data, ignoring dynamic interactions between indicators, and having static and singular prediction results.

[0020] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0021] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for dynamically predicting key indicators for non-surgical treatment of ectopic pregnancy according to an embodiment of this application; Figure 2 A schematic diagram of a dataset usage path is provided for one embodiment of this application; Figure 3 A schematic diagram illustrating the ranking of variable importance when using a TREND model to predict blood β-hCG levels, as provided in one embodiment of this application; Figure 4 A schematic diagram illustrating the ranking of variable importance when using a TREND model to predict the size of an attachment region in an embodiment of this application; Figure 5 A schematic diagram illustrating the change in accuracy of a TREND model in predicting changes in blood β-hCG levels over time in a validation set, as provided in an embodiment of this application. Figure 6A schematic diagram illustrating the change in accuracy of a TREND model in predicting changes in the size of an appendix region in a validation set as a function of time intervals, according to an embodiment of this application. Figure 7 This is an example diagram of a dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy according to an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0022] Among them, 10-Dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy, 100-Preprocessing module, 200-Training module, 300-Inference module, 801-Memory, 802-Processor, and 803-Communication interface. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0024] The following describes a method and apparatus for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy according to embodiments of this application, with reference to the accompanying drawings. Addressing the problems mentioned in the background art, this application provides a method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy. In this method, complete clinical record information of multiple patients with tubal pregnancy is acquired to construct and preprocess a corresponding total training set. Training clinical feature vectors of the training subset and validation clinical feature vectors of the validation subset are extracted from the preprocessed total training set. Multiple key control variables corresponding to the training clinical feature vectors are collected, and a pre-constructed gradient boosting tree model is trained using the training clinical feature vectors and multiple key control variables to generate a dynamic indicator interval prediction model. Based on the validation clinical feature vectors, the dynamic indicator interval prediction model is verified to meet preset performance requirements. The multiple key control variables include the proposed non-surgical treatment plan category and time interval. In response to meeting the preset performance requirements, the current clinical information of the current patient with tubal pregnancy is acquired and input into the dynamic indicator interval prediction model to output the dynamic interval of key indicators for the current patient with tubal pregnancy and generate a target clinical prompt report for the dynamic interval of key indicators. This application achieves high-precision dynamic range prediction with only the latest patient visit data, providing clinicians with a quantitative decision support tool that aligns with actual workflows. This solves the problems of existing technologies where prediction models rely on complete historical sequences, cannot adapt to missing or non-uniform sampling of real-world medical data, ignore dynamic interactions between indicators, and produce static and limited prediction results.

[0025] Specifically, Figure 1 This is a flowchart illustrating a method for dynamically predicting key indicators of non-surgical treatment of ectopic pregnancy, as provided in an embodiment of this application.

[0026] like Figure 1 As shown, this method for dynamically predicting key indicators of non-surgical treatment of ectopic pregnancy includes the following steps: In step S101, complete clinical record information of multiple patients with ectopic pregnancy is obtained to construct and preprocess the corresponding total training set, and the training clinical feature vector of the training subset and the validation clinical feature vector of the validation subset are extracted from the preprocessed total training set.

[0027] This application embodiment first collects complete clinical records from different patients who visited the clinic for ectopic pregnancy, and constructs a corresponding total training set based on these records. The training set is then preprocessed and partitioned to obtain training and validation subsets of validation clinical feature vectors. Finally, training and validation clinical feature vectors are extracted from both the training and validation subsets, providing reliable data guidance and basis for predicting the dynamic indicator range of key indicators for non-surgical treatment of ectopic pregnancy (EP) (i.e., the TREND task).

[0028] Optionally, in one embodiment of this application, complete clinical record information of multiple patients with ectopic pregnancy is obtained to construct and preprocess the corresponding total training set, and the training clinical feature vector of the training subset and the validation clinical feature vector of the validation subset in the preprocessed total training set are extracted. This includes: obtaining complete clinical record information of multiple patients with ectopic pregnancy, extracting the structured variable set corresponding to the complete clinical record information, and constructing the total training set based on the structured variable set, wherein the structured variable set includes demographic and medical history characteristics, symptoms and signs, laboratory and ultrasound indicators, and diagnostic and treatment information; performing missing value imputation and standardization on the total training set to obtain a standard training dataset, and dividing the standard training dataset into a training subset and a validation subset based on a preset time threshold; deleting the relevant data and postoperative follow-up data of patients with ectopic pregnancy who only have initial visits and no follow-up data in the training subset and validation subset, and extracting the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset.

[0029] In actual implementation, this embodiment first obtains the complete clinical records of multiple patients who visited TEP through the electronic medical record system, extracts the corresponding structured variable set, and constructs the corresponding total training set based on the structured variable set. The structured variable set includes the following: 1. Demographic and medical history characteristics: age, height, weight, BMI, number of days of amenorrhea, parity, parity, number of previous ectopic pregnancies, menstrual regularity, method of conception, plans for continued fertility, history of previous ipsilateral ectopic pregnancies, location of current ectopic pregnancies (left / right), history of use of pregnancy-supporting drugs, smoking / drinking history.

[0030] 2. Symptoms and signs: abdominal pain, a feeling of pressure in the anus, vaginal bleeding, and shock.

[0031] 3. Laboratory and ultrasound indicators: current serum β-hCG level, serum β-hCG (β-human chorionic gonadotropin) level and rate of increase 48 hours ago, endometrial thickness, HGB level, maximum diameter of adnexal mass, depth of pelvic effusion, yolk sac / fetal bud / fetal heartbeat within the mass, and preliminary ultrasound conclusions.

[0032] 4. Treatment information: Current proposed treatment plan (expectant management or MTX (Methotrexate) treatment) and the timestamp of this visit.

[0033] Secondly, embodiments of this application can perform data preprocessing operations on the aforementioned total training set, the specific process of which is as follows: 1. Missing value imputation: For missing items of serum β-hCG, rise rate and endometrial thickness 48 hours ago, if the gradient boosting tree algorithm is used, the outlier value -9999 is uniformly filled in; if the sequence neural network algorithm is used, -1 is filled in and the corresponding binary missing indicator variable is generated simultaneously.

[0034] 2. Standardized measurement: All continuous variables are standardized and mapped to a distribution with a mean of 0 and a variance of 1 (the validation set strictly reuses the mean and variance parameters of the training set).

[0035] Subsequently, in embodiments of this application, historical data in the preprocessed total training set can be divided into a training subset and recent data into a validation subset according to the time of medical visit.

[0036] Based on this, the embodiments of this application can exclude data with only initial visits and no follow-up records and postoperative follow-up data to construct a TREND model dataset, and perform corresponding feature processing operations on the filtered data to extract the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset.

[0037] Figure 2 Use a path diagram for the dataset. For example... Figure 2 As shown, This is a dataset of outpatient visits for TEP patients with generally stable conditions. This is a training subset of TEP patients with generally stable conditions; A validation subset of TEP patients with stable vital signs; The training set for the TREND model is... Based on this, patients with only a single initial visit record were removed, and for patients with multiple visit records, postoperative follow-up data entries were removed. The validation set for the TREND model is... Based on this, patients with only a single initial visit record were removed, and for patients with multiple visit records, postoperative follow-up data entries were removed.

[0038] Therefore, the embodiments of this application construct a corresponding total training set based on the complete clinical record information of patients with ectopic pregnancy, perform preprocessing and segmentation operations on it, and extract the corresponding clinical feature vectors, thereby effectively improving the quality of the data and enhancing the reliability and robustness of model training.

[0039] In step S102, multiple key control variables corresponding to the training clinical feature vector are collected, and a pre-constructed gradient boosting tree model is trained by training the clinical feature vector and multiple key control variables to generate a dynamic index interval prediction model. Based on the validation clinical feature vector, the dynamic index interval prediction model is verified to meet the preset performance requirements. Among them, the multiple key control variables include the proposed non-surgical treatment plan category and time interval.

[0040] Furthermore, embodiments of this application can obtain multiple key control variables such as the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and perform label mapping on the data of the training subset according to the corresponding prediction target (i.e., prediction of the dynamic indicator interval of key indicators), and assemble the model input vector (i.e., input feature vector).

[0041] Secondly, in this embodiment, CatBoost (Categorical Boosting), i.e. gradient boosting tree model, can be trained by inputting feature vectors to generate a corresponding dynamic indicator interval prediction model (i.e., TREND model). The dynamic indicator interval prediction model can be validated by using validation clinical feature vectors to determine whether the trained model meets the preset performance requirements, thereby improving the accuracy and generalization of model training.

[0042] Optionally, in one embodiment of this application, multiple key control variables corresponding to the training clinical feature vector are collected, and a pre-constructed gradient boosting tree model is trained using the training clinical feature vector and the multiple key control variables to generate a dynamic index interval prediction model. Furthermore, based on the validation clinical feature vector, the dynamic index interval prediction model is verified to meet preset performance requirements. This includes: obtaining the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and determining the proposed non-surgical treatment plan category, time interval, and prediction label corresponding to the training clinical feature vector, wherein the prediction label includes serum β-hCG. The model is developed using horizontal interval labels and appendix block size interval labels. Based on a pre-defined cross-entropy loss function, ordered enhancement mechanism, and category boosting algorithm, a gradient boosting tree model is constructed. This model is iteratively trained using proposed non-surgical treatment plan categories, time intervals, training clinical feature vectors, and predicted labels until the iteration process meets a pre-defined termination requirement. The trained gradient boosting tree model is then subjected to a pre-defined model solidification operation to generate a dynamic indicator interval prediction model. Validation clinical feature vectors are input into the dynamic indicator interval prediction model. The matching degree between the predicted results and the true labels corresponding to the validation clinical feature vectors is calculated. Multiple validation data points for the dynamic indicator interval prediction model are determined based on the matching degree. These validation data points include accuracy, weighted F1 score, area under the receiver operating characteristic curve, recall, and precision. Based on these validation data points, a corresponding confusion matrix is ​​generated. The true positive and false negative distributions for each risk stratification interval are statistically analyzed using the confusion matrix. The true positive and false negative distributions are then used to verify whether the dynamic indicator interval prediction model meets the pre-defined discrimination balance requirement across different clinical risk populations.

[0043] It should be noted that the embodiments of this application aim to predict the clinical intervals in which the patient's blood β-hCG level and the size of the adnexal mass will be located after a specific time interval Δt in the future.

[0044] Therefore, this application embodiment only extracts the clinical feature vector of the patient at the time of the current visit (simulating a scenario where previous medical records are missing in real clinical practice), and adds two key control variables: 1. Proposed non-surgical treatment plan: coded as "expectant care" or "MTX treatment".

[0045] 2. Time interval Δt: Defined as the time span (in days) from the current consultation time to the target prediction time. It should be noted that this embodiment does not rely on the patient's previous historical consultation records; it only uses the latest single consultation data to drive the prediction.

[0046] Secondly, embodiments of this application can divide the prediction target (i.e., the dynamic range of key indicators) into corresponding classification intervals based on commonly used clinical decision thresholds, as shown below: 1. Serum β-hCG level: Tripartite label [<1500 IU / L, 1500~5000 IU / L, ≥5000 IU / L].

[0047] 2. Size of the attachment area: binary label [<3.5 cm, ≥3.5 cm].

[0048] Subsequently, the embodiments of this application may define a binary outcome (i.e., perform label definition) based on the actual clinical outcome, as described below: 1. Expectant treatment: Success means no further intervention is received and β-hCG levels drop to negative; Failure means subsequent treatment is switched to surgery or MTX.

[0049] 2. MTX treatment: Success means β-hCG levels drop to negative; failure means subsequent treatment involves surgery.

[0050] Furthermore, in this embodiment, a pre-built gradient boosting tree model can be used as the core prediction engine, and the gradient boosting tree model can be trained based on the input features consisting of the current medical visit feature vector and the proposed non-surgical treatment plan category.

[0051] In the specific implementation process, the process of training the gradient boosting tree model in this embodiment of the application is as follows: 1. Feature Encoding: The embodiments of this application can utilize the built-in Ordered Boosting mechanism of CatBoost to automatically process category features and reduce target leakage bias.

[0052] 2. Iterative optimization: The embodiments of this application can use the cross-entropy loss function for gradient descent optimization, and adopt the algorithm's default recommended hyperparameter initialization.

[0053] 3. Overfitting prevention control: The embodiments of this application can monitor the F1 score of the validation set in real time during the training process and configure an early stopping mechanism; when the optimal F1 score no longer decreases within 5 consecutive rounds, the training is automatically terminated and the snapshot of the model with the optimal weights is saved.

[0054] 4. Model solidification: The trained CatBoost tree ensemble structure is serialized and deployed on the server to perform online prediction operations.

[0055] As one possible approach, embodiments of this application can input the filtered validation subset into a fixed gradient boosting tree model (i.e., a dynamic index interval prediction model), perform corresponding forward inference operations to calculate the matching degree between the prediction results and the true labels, and output accuracy, weighted F1 score, area under the receiver operating characteristic curve (AUC), recall (i.e., sensitivity), and precision (i.e., positive predictive value). In addition, embodiments of this application can also generate a multi-class / binary-class confusion matrix and statistically analyze the true positive and false negative distributions of each risk stratification interval to verify the model's discriminative balance in different risk populations in clinical practice.

[0056] It should be noted that in actual implementation, those skilled in the art can also use other gradient boosting tree algorithms (such as XGBoost, LightGBM) or ensemble learning models (such as random forest, Bagging) to replace CatBoost, depending on the actual situation. These methods can also achieve nonlinear feature mining and high-dimensional data processing, which will not be elaborated here.

[0057] In step S103, in response to meeting the preset performance requirements, the current clinical information of the current ectopic pregnancy patient is obtained and input into the dynamic index interval prediction model to output the dynamic interval of the key index of the current ectopic pregnancy patient and generate a target clinical prompt report of the dynamic interval of the key index.

[0058] Subsequently, if the discrimination balance of the judgment model meets the corresponding requirements in different clinical risk populations, the embodiments of this application can obtain the current clinical information of the current ectopic pregnancy patient and input it into the dynamic indicator interval prediction model, output the probability distribution values ​​of multiple key indicators corresponding to the current ectopic pregnancy patient falling into each preset interval, so as to determine the dynamic interval of the corresponding key indicators, and sort the multiple key indicators to generate an interpretable clinical suggestion report (i.e., target clinical suggestion report) on the dynamic evolution of key indicators for non-surgical treatment of ectopic pregnancy corresponding to the current patient.

[0059] As one possible approach, embodiments of this application may replace the interval classification task with a continuous value regression task (directly predicting specific β-hCG values ​​or mass centimeters); or add probabilistic prediction targets such as "probability of fallopian tube rupture" and "incidence of persistent ectopic pregnancy," which will not be specifically described here.

[0060] Optionally, in one embodiment of this application, the current clinical information of the current patient with ectopic pregnancy is obtained and input into a dynamic indicator interval prediction model to output the dynamic interval of key indicators for the current patient with ectopic pregnancy, and a target clinical prompt report for the dynamic interval of key indicators is generated. This includes: collecting the current clinical information of the current patient with ectopic pregnancy and inputting the current clinical information into the dynamic indicator interval prediction model, outputting the probability distribution values ​​of the blood β-hCG and mass size of the current patient with ectopic pregnancy falling into each preset interval after a time interval, so as to determine the corresponding dynamic interval of key indicators; performing a preset variable importance analysis operation on the dynamic indicator interval prediction model based on the probability distribution values ​​to generate the ranking results of the importance of related variables, and generating a target clinical prompt report based on the dynamic interval of key indicators and the ranking results of the importance of related variables; and pushing the target clinical prompt report to the target clinical information system through a preset API interface or visualization terminal, so that the target physician can perform corresponding diagnosis and treatment operations on the current patient with ectopic pregnancy based on the target clinical prompt report.

[0061] In actual implementation, this application embodiment collects the current clinical information of the current ectopic pregnancy patient and inputs it into the dynamic index interval prediction model to output the corresponding inference results, namely the probability distribution values ​​of blood β-hCG and mass size falling into each preset interval after time interval Δt; furthermore, this application embodiment can call the feature importance parsing algorithm to perform importance analysis on multiple input features of the dynamic index interval prediction model and output the Top-N key driving variables and their feature contribution ranking results (i.e., the importance ranking results of the associated variables).

[0062] Subsequently, in this embodiment of the application, an interpretable clinical prompt report can be generated based on the reasoning results and feature contribution ranking data, and pushed to the clinical information system through an API interface or a visualization terminal, so that physicians can perform corresponding diagnosis and treatment operations on the current ectopic pregnancy patient based on the interpretable clinical prompt report.

[0063] Therefore, in this embodiment, only the clinical feature vector of the patient's current single visit, the proposed treatment plan, and the time interval Δt are extracted and input into the gradient boosting tree ensemble model CatBoost, which directly outputs the classification probability of the blood β-hCG level interval and the size interval of the adnexal mass at the future time Δt, thus achieving dynamic interval prediction without the need for a complete historical sequence.

[0064] The following section uses outpatient, emergency, and inpatient medical records of patients with stable ectopic pregnancy (TEP) admitted to a tertiary hospital from January 2015 to April 2024 to provide a detailed explanation of the specific construction and application process of the TREND model (key indicator dynamic interval prediction model) in this application.

[0065] 1. Dataset Construction and Preprocessing: This application first extracts complete clinical information of the patient's current visit from the electronic medical record system. Each visit record constitutes an independent data entry, and the included clinical variables include the following: (1) Demographic and medical history characteristics: age, height, weight, BMI, number of days of amenorrhea, parity, parity, number of previous ectopic pregnancies (EP), regularity of menstruation, method of conception, whether there is a plan to continue having children, history of previous ipsilateral TEP, location of this TEP (left / right), whether tocolytic drugs were used in this pregnancy, smoking history, and drinking history.

[0066] (2) Symptoms and signs: presence of abdominal pain, anal tenesmus, vaginal bleeding, and shock.

[0067] (3) Laboratory tests: serum β-hCG level at the time of this visit, serum β-hCG level and increase rate 48 hours ago, HGB level, and endometrial thickness.

[0068] (4) Ultrasound examination: maximum diameter of the mass in the adnexal region, depth of pelvic effusion, whether yolk sac / fetal bud / fetal heartbeat is seen in the mass, and preliminary ultrasound conclusion.

[0069] (5) Medical information: the treatment plan (expectant management, MTX treatment or surgery) at the time of this visit and the outcome of the treatment.

[0070] 2. Data partitioning: In this embodiment, the data from January 2015 to December 2022 can be divided into a training subset and the data from January 2023 to April 2024 can be divided into a validation subset according to the time dimension.

[0071] 3. Missing values ​​and standardization: (1) For the missing blood β-hCG level, increase rate and endometrial thickness 48 hours ago, if the CatBoost algorithm is used, the outlier value -9999 is used for imputation; if the LSTM algorithm is used, -1 is used for imputation and missing indicator binary variables are generated simultaneously.

[0072] (2) For the LSTM algorithm, all continuous variables are standardized to a distribution with a mean of 0 and a variance of 1 before input (the validation subset uses the mean and variance of the training set).

[0073] 4. Dataset filtering: TREND model dataset: Patients with only initial visits and no follow-up data, as well as postoperative follow-up data, were removed, resulting in a training subset (n=1,571 records) and a validation subset (n=405 records).

[0074] 5. Label definition and feature vector assembly: Input features: current visit feature vector, proposed non-surgical treatment plan category, and time interval Δt (i.e., the number of days from the current visit to the target prediction time).

[0075] Output labels: Classification intervals are constructed based on clinical decision thresholds; serum β-hCG is divided into three categories: <1500 IU / L, 1500~5000 IU / L, and ≥5000 IU / L; the size of the adnexal mass is divided into two categories: <3.5 cm and ≥3.5 cm.

[0076] 6. Model Training and Inference: This embodiment of the application can use the CatBoost algorithm to construct a gradient boosting tree model, and uses the CatBoost default recommended hyperparameters, with the cross-entropy loss function. During training, this embodiment of the application can iteratively construct a decision tree ensemble on a subset of training data, and use an ordered augmentation mechanism to process class features, reducing target leakage bias, so as to obtain the trained gradient boosting tree model, i.e., a dynamic index interval prediction model.

[0077] Subsequently, during the inference process, after inputting the current medical data of the new patient and the set time interval Δt (e.g., Δt=3 days), the dynamic indicator interval prediction model can output the probability distribution of each indicator falling into each interval.

[0078] 7. Performance Verification and Result Analysis: Furthermore, in this embodiment of the application, validation testing can be performed on an independent validation set (i.e., a subset of validations), and the model performance data is as follows: 1. Blood β-hCG prediction: accuracy 0.885; weighted F1 score 0.886; AUC 0.973.

[0079] 2. Block size prediction: accuracy 0.907; F1 score 0.779; AUC 0.902.

[0080] The embodiments of this application can perform comparative verification between CatBoost and LSTM models, as shown in Table 1: Table 1

[0081] As shown in Table 1, under the same task, the LSTM model that can rely on complete time-series historical data in the embodiments of this application has an accuracy of 0.792 and 0.810, respectively, which is significantly lower than the CatBoost model that only uses the latest data in the embodiments of this application. This strongly demonstrates the robustness advantage of the embodiments of this application in the scenario of incomplete medical records.

[0082] 8. Variable Importance Analysis: The importance of different input features of the model is ranked based on the feature importance parsing algorithm, as detailed below: (1) The embodiments of this application clearly define “current blood β-hCG level, current mass size, time interval Δt, HGB level, pelvic effusion depth, BMI, treatment plan selection” as the core prediction feature set and their stable weight relationship in the ranking of the importance of model features.

[0083] Figure 3 This diagram illustrates the ranking of variable importance when using the TREND model to predict serum β-hCG levels. Figure 3 As shown, when predicting changes in serum β-hCG, the current serum β-hCG level contributes the most, followed by Δt, number of days since menopause, and HGB level.

[0084] (2) Figure 4 A diagram illustrating the importance ranking of variables when the TREND model predicts the size of the adjacent region packet, as shown below. Figure 4 As shown, when predicting the mass size, the current mass size contributes the most, followed by Δt, pelvic effusion depth, and HGB level.

[0085] 9. Clinical guidance: Subsequently, the embodiments of this application can generate interpretable clinical prompt reports based on the dynamic range of key indicators and the importance ranking results of related variables, so as to remind physicians that the predictive efficacy is best in short-term follow-up (Δt≤3 days), and that the combined changes of HGB and pelvic effusion should be closely monitored.

[0086] Understandably, existing technologies mostly rely on static prognostic assessments based on data from a single initial consultation, failing to reflect the dynamic heterogeneity of disease progression. However, this application innovatively introduces the time interval variable Δt as a core input feature, enabling dynamic prediction of the future evolution intervals of serum β-hCG levels and adnexal mass size during non-surgical treatment. Tested on an independent time validation set (n=405), the TREND model achieved a prediction accuracy of 0.885 for the three-category interval of serum β-hCG levels, a weighted F1 score of 0.886, and an AUC as high as 0.973; and a prediction accuracy of 0.907 for the two-category interval of adnexal mass size, with an AUC of 0.902. Therefore, this application can quantify and predict the trajectory of indicator changes in advance, providing data support for physicians to dynamically adjust monitoring frequency and intervention timing, effectively solving the technical deficiency of traditional threshold methods in capturing nonlinear fluctuations of indicators.

[0087] Furthermore, the embodiments of this application clarify the core driving factors of model decision-making through a feature importance ranking algorithm, eliminating concerns about the algorithm being a black box. In the TREND model, the current indicator value, time interval Δt, number of days of amenorrhea, and HGB level contribute the most, and this transparent design makes the prediction results clinically logically traceable.

[0088] Figure 5 This diagram illustrates the change in the accuracy of the TREND model in predicting changes in blood β-hCG levels over time in the validation set. Figure 6 This diagram illustrates the variation of the accuracy of the TREND model in predicting changes in the size of the adjacent region's packet in the validation set over time. Figure 5 and Figure 6 As shown, the TREND model can maintain a stable accuracy of over 80% in predicting serum β-hCG within Δt=1~3 days, and the accuracy in predicting mass size can also maintain a relatively stable accuracy of over 80% within Δt=1~7 days. This prediction window is highly consistent with the routine clinical follow-up cycle of 3~7 days for TEP patients, enabling this application to be seamlessly integrated into the existing outpatient and emergency follow-up process, and possessing strong clinical translational practicality and operational feasibility.

[0089] The method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy proposed in this application involves acquiring complete clinical records of multiple patients with tubal pregnancy to construct and preprocess a corresponding total training set. Training clinical feature vectors from the training subset and validation clinical feature vectors from the validation subset are extracted from the preprocessed total training set. Multiple key control variables corresponding to the training clinical feature vectors are collected, and a pre-constructed gradient boosting tree model is trained using the training clinical feature vectors and multiple key control variables to generate a dynamic indicator interval prediction model. Based on the validation clinical feature vectors, the dynamic indicator interval prediction model is verified to meet preset performance requirements. The multiple key control variables include the proposed non-surgical treatment plan category and time interval. In response to meeting the preset performance requirements, the current clinical information of the current patient with tubal pregnancy is acquired and input into the dynamic indicator interval prediction model to output the dynamic interval of key indicators for the current patient with tubal pregnancy and generate a target clinical prompt report for the dynamic interval of key indicators. This application only requires the latest medical data to achieve high-precision dynamic interval prediction, providing clinicians with a quantitative decision support tool that fits their actual workflow.

[0090] Secondly, with reference to the accompanying drawings, the dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy according to the embodiments of this application is described.

[0091] Figure 7 This is a block diagram of a dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy according to an embodiment of this application.

[0092] like Figure 7 As shown, the dynamic prediction device 10 for key indicators of non-surgical treatment of ectopic pregnancy includes: a preprocessing module 100, a training module 200, and an inference module 300.

[0093] The preprocessing module 100 is used to acquire complete clinical record information of multiple patients with ectopic pregnancy, so as to construct and preprocess the corresponding total training set, and extract the training clinical feature vector of the training subset and the validation clinical feature vector of the validation subset in the preprocessed total training set.

[0094] The training module 200 is used to collect multiple key control variables corresponding to the training clinical feature vectors, and to train a pre-built gradient boosting tree model by training the training clinical feature vectors and multiple key control variables to generate a dynamic index interval prediction model. Based on the validation clinical feature vectors, it verifies whether the dynamic index interval prediction model meets the preset performance requirements. Among them, the multiple key control variables include the proposed non-surgical treatment plan category and time interval.

[0095] The inference module 300 is used to obtain the current clinical information of the current ectopic pregnancy patient in response to the preset performance requirements, and input the current clinical information into the dynamic index interval prediction model to output the dynamic interval of the key index of the current ectopic pregnancy patient and generate a target clinical prompt report of the dynamic interval of the key index.

[0096] Optionally, in one embodiment of this application, the preprocessing module 100 includes: an extraction unit, a segmentation unit, and a filtering unit.

[0097] The extraction unit is used to obtain complete clinical record information of multiple patients with ectopic pregnancy, extract the structured variable set corresponding to the complete clinical record information, and construct a total training set based on the structured variable set. The structured variable set includes demographic and medical history characteristics, symptoms and signs, laboratory and ultrasound indicators, and diagnostic and treatment information.

[0098] The partitioning unit is used to impute missing values ​​and standardize the total training set to obtain a standard training dataset. Based on a preset time threshold, the standard training dataset is divided into a training subset and a validation subset.

[0099] The filtering unit is used to delete relevant data and postoperative follow-up data of patients with tubal pregnancy who only have initial visits and no follow-up data in the training subset and validation subset, and to extract the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset.

[0100] Optionally, in one embodiment of this application, the training module 200 includes: a determination unit, an iteration unit, a calculation unit, and a verification unit.

[0101] The determining unit is used to obtain the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and to determine the proposed non-surgical treatment plan category, time interval and the prediction label corresponding to the training clinical feature vector. The prediction label includes the blood β-hCG level interval label and the adnexal mass size interval label.

[0102] The iterative unit is used to construct a gradient boosting tree model based on a preset cross-entropy loss function, ordered enhancement mechanism, and category boosting algorithm. It iteratively trains the gradient boosting tree model by proposing non-surgical treatment plan categories, time intervals, training clinical feature vectors, and prediction labels until the iteration process meets the preset iteration termination requirements. The trained gradient boosting tree model is then subjected to a preset model solidification operation to generate a dynamic index interval prediction model.

[0103] The computation unit is used to input the validation clinical feature vector into the dynamic index interval prediction model, calculate the matching degree between the prediction result corresponding to the validation clinical feature vector and the true label corresponding to the validation clinical feature vector, and determine multiple validation data of the dynamic index interval prediction model based on the matching degree. The multiple validation data include accuracy, weighted F1 score, area under the receiver operating characteristic curve, recall, and precision.

[0104] The validation unit is used to generate a corresponding confusion matrix based on multiple validation data, and to statistically analyze the true positive and false negative distributions of each risk stratification interval according to the confusion matrix. The true positive and false negative distributions are then used to verify whether the dynamic indicator interval prediction model meets the preset discrimination balance requirements in different clinical risk populations.

[0105] Optionally, in one embodiment of this application, the inference module 300 includes: a collection unit, a parsing unit, and a push unit.

[0106] The acquisition unit is used to collect the current clinical information of the current ectopic pregnancy patient and input the current clinical information into the dynamic indicator interval prediction model. It outputs the probability distribution values ​​of the blood β-hCG and mass size of the current ectopic pregnancy patient falling into each preset interval after a time interval, so as to determine the corresponding dynamic interval of key indicators.

[0107] The parsing unit is used to perform preset variable importance parsing operations on the dynamic indicator interval prediction model based on probability distribution values, so as to generate the importance ranking results of related variables, and generate a target clinical prompt report based on the dynamic interval of key indicators and the importance ranking results of related variables.

[0108] The push unit is used to push the target clinical prompt report to the target clinical information system through a preset API interface or visual terminal, so that the target physician can perform corresponding diagnosis and treatment operations on the current ectopic pregnancy patient based on the target clinical prompt report.

[0109] It should be noted that the explanation of the above-mentioned embodiment of the dynamic prediction method for key indicators of non-surgical treatment of ectopic pregnancy also applies to the dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy in this embodiment, and will not be repeated here.

[0110] The dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy proposed in this application includes a preprocessing module 100, which is used to acquire complete clinical record information of multiple patients with tubal pregnancy to construct and preprocess the corresponding total training set, and extract the training clinical feature vector of the training subset and the validation clinical feature vector of the validation subset in the preprocessed total training set; a training module 200, which is used to collect multiple key control variables corresponding to the training clinical feature vector, and train a pre-constructed gradient boosting tree model through the training clinical feature vector and multiple key control variables to generate a dynamic indicator interval prediction model, and verify whether the dynamic indicator interval prediction model meets the preset performance requirements based on the validation clinical feature vector, wherein the multiple key control variables include the category of the proposed non-surgical treatment plan and the time interval; and an inference module 300, which is used to acquire the current clinical information of the current patient with tubal pregnancy in response to meeting the preset performance requirements, and input the current clinical information into the dynamic indicator interval prediction model to output the dynamic interval of key indicators of the current patient with tubal pregnancy, and generate a target clinical prompt report of the dynamic interval of key indicators. This application can automatically calculate various key indicators such as lumen radius, cross-sectional area, circumference, stenosis rate, and rate of change, providing objective and quantitative assessment basis for clinical practice. Furthermore, by optimizing sampling strategies and algorithm processes, it can achieve functional imaging and quantitative assessment of the lumen without adding additional hardware sensors, reducing system complexity and cost, and facilitating engineering implementation and clinical application. This application only requires the latest patient data to achieve high-precision dynamic interval prediction, providing clinicians with a quantitative decision support tool that fits their actual workflow.

[0111] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 801, the processor 802, and the computer program stored on the memory 801 and capable of running on the processor 802.

[0112] When the processor 802 executes the program, it implements the dynamic prediction method for key indicators of non-surgical treatment of ectopic pregnancy provided in the above embodiments.

[0113] Furthermore, electronic devices also include: Communication interface 803 is used for communication between memory 801 and processor 802.

[0114] The memory 801 is used to store computer programs that can run on the processor 802.

[0115] The memory 801 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0116] If the memory 801, processor 802, and communication interface 803 are implemented independently, then the communication interface 803, memory 801, and processor 802 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized into address buses, data buses, control buses, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0117] Optionally, in a specific implementation, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, then the memory 801, processor 802, and communication interface 803 can communicate with each other through an internal interface.

[0118] The processor 802 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0119] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy.

[0120] This application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-described method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy.

[0121] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0122] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0123] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0124] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0125] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0126] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0127] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0128] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for dynamic prediction of key indicators in non-surgical treatment of ectopic pregnancy, characterized in that, Includes the following steps: Complete clinical records of multiple patients with ectopic pregnancy were obtained to construct and preprocess the corresponding total training set, and the training clinical feature vectors of the training subset and the validation clinical feature vectors of the validation subset were extracted from the preprocessed total training set. Multiple key control variables corresponding to the training clinical feature vector are collected, and a pre-constructed gradient boosting tree model is trained using the training clinical feature vector and the multiple key control variables to generate a dynamic index interval prediction model. Based on the verification clinical feature vector, the dynamic index interval prediction model is verified to meet the preset performance requirements. The multiple key control variables include the proposed non-surgical treatment plan category and time interval. In response to meeting the preset performance requirements, the current clinical information of the current ectopic pregnancy patient is obtained, and the current clinical information is input into the dynamic index interval prediction model to output the dynamic interval of the key index of the current ectopic pregnancy patient and generate a target clinical prompt report of the dynamic interval of the key index.

2. The method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy according to claim 1, characterized in that, The process involves acquiring complete clinical records from multiple patients with ectopic pregnancies to construct and preprocess a corresponding total training set. The preprocessed total training set includes extracting training clinical feature vectors from the training subset and validation clinical feature vectors from the validation subset, including: Complete clinical record information of the multiple patients with ectopic pregnancy is obtained, and the structured variable set corresponding to the complete clinical record information is extracted. The total training set is constructed based on the structured variable set, wherein the structured variable set includes demographic and medical history characteristics, symptoms and signs, laboratory and ultrasound indicators, and diagnostic and treatment information. The total training set is imputed and standardized to obtain a standard training dataset. Based on a preset time threshold, the standard training dataset is divided into the training subset and the validation subset. The relevant data and postoperative follow-up data of patients with ectopic pregnancy who only had initial visits and no follow-up data in the training subset and the validation subset are deleted, and the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset are extracted.

3. The method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy according to claim 2, characterized in that, The process involves collecting multiple key control variables corresponding to the training clinical feature vector, training a pre-constructed gradient boosting tree model using the training clinical feature vector and the multiple key control variables, generating a dynamic indicator interval prediction model, and verifying whether the dynamic indicator interval prediction model meets preset performance requirements based on the validation clinical feature vector, including: Obtain the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and determine the proposed non-surgical treatment plan category, the time interval, and the predicted label corresponding to the training clinical feature vector, wherein the predicted label includes a blood β-hCG level interval label and an adnexal mass size interval label; Based on the preset cross-entropy loss function, ordered enhancement mechanism and category boosting algorithm, the gradient boosting tree model is constructed. The gradient boosting tree model is iteratively trained using the proposed non-surgical treatment plan category, the time interval, the training clinical feature vector and the prediction label until the iteration process meets the preset iteration termination requirements. The trained gradient boosting tree model is then subjected to a preset model solidification operation to generate the dynamic index interval prediction model. The validation clinical feature vector is input into the dynamic index interval prediction model. The matching degree between the prediction result corresponding to the validation clinical feature vector and the true label corresponding to the validation clinical feature vector is calculated. Based on the matching degree, multiple validation data of the dynamic index interval prediction model are determined. The multiple validation data include accuracy, weighted F1 score, area under the receiver operating characteristic curve, recall, and precision. Based on the multiple validation data, a corresponding confusion matrix is ​​generated. The true positive and false negative distributions of each risk stratification interval are statistically analyzed according to the confusion matrix. The true positive and false negative distributions are then used to verify whether the dynamic indicator interval prediction model meets the preset discrimination balance requirements in different clinical risk populations.

4. The method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy according to claim 3, characterized in that, The process of acquiring the current clinical information of the current ectopic pregnancy patient and inputting the current clinical information into the dynamic indicator interval prediction model to output the dynamic interval of the key indicators of the current ectopic pregnancy patient, and generating a target clinical prompt report of the dynamic interval of the key indicators, includes: The current clinical information of the current ectopic pregnancy patient is collected and input into the dynamic index interval prediction model. The probability distribution values ​​of the blood β-hCG and mass size of the current ectopic pregnancy patient falling into each preset interval after the time interval are output to determine the corresponding dynamic interval of key indicators. Based on the probability distribution value, a preset variable importance analysis operation is performed on the dynamic indicator interval prediction model to generate the importance ranking result of the associated variables, and the target clinical prompt report is generated according to the dynamic interval of the key indicator and the importance ranking result of the associated variables. The target clinical alert report is pushed to the target clinical information system through a preset API interface or visual terminal, so that the target physician can perform corresponding diagnosis and treatment operations on the current ectopic pregnancy patient based on the target clinical alert report.

5. A dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy, characterized in that, include: The preprocessing module is used to obtain complete clinical record information of multiple patients with ectopic pregnancy in order to construct and preprocess the corresponding total training set, and extract the training clinical feature vector of the training subset and the validation clinical feature vector of the validation subset in the preprocessed total training set. The training module is used to collect multiple key control variables corresponding to the training clinical feature vector, and train a pre-constructed gradient boosting tree model through the training clinical feature vector and the multiple key control variables to generate a dynamic index interval prediction model. Based on the verification clinical feature vector, the module verifies whether the dynamic index interval prediction model meets the preset performance requirements. The multiple key control variables include the proposed non-surgical treatment plan category and time interval. The inference module is used to obtain the current clinical information of the current ectopic pregnancy patient in response to the preset performance requirements, and input the current clinical information into the dynamic index interval prediction model to output the dynamic interval of the key index of the current ectopic pregnancy patient and generate a target clinical prompt report of the dynamic interval of the key index.

6. The dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy according to claim 5, characterized in that, The preprocessing module includes: The extraction unit is used to obtain complete clinical record information of the multiple patients with ectopic pregnancy, extract the structured variable set corresponding to the complete clinical record information, and construct the total training set based on the structured variable set. The structured variable set includes demographic and medical history characteristics, symptoms and signs, laboratory and ultrasound indicators, and diagnostic and treatment information. A partitioning unit is used to perform missing value imputation and standardization on the total training set to obtain a standard training dataset, and to divide the standard training dataset into the training subset and the validation subset based on a preset time threshold. The filtering unit is used to delete relevant data and postoperative follow-up data of patients with tubal pregnancy who only have initial visits and no follow-up data in the training subset and the validation subset, and to extract the training clinical feature vector corresponding to the training subset and the validation clinical feature vector corresponding to the validation subset.

7. The dynamic prediction device for key indicators of non-surgical treatment of ectopic pregnancy according to claim 6, characterized in that, The training module includes: A determining unit is used to obtain the proposed non-surgical treatment plan category and time interval corresponding to the training clinical feature vector, and to determine the proposed non-surgical treatment plan category, the time interval, and the predicted label corresponding to the training clinical feature vector, wherein the predicted label includes a blood β-hCG level interval label and an adnexal mass size interval label; An iterative unit is used to construct the gradient boosting tree model based on a preset cross-entropy loss function, ordered enhancement mechanism, and category boosting algorithm. The gradient boosting tree model is iteratively trained using the proposed non-surgical treatment plan category, the time interval, the training clinical feature vector, and the prediction label until the iteration process meets the preset iteration termination requirements. The trained gradient boosting tree model is then subjected to a preset model solidification operation to generate the dynamic index interval prediction model. The calculation unit is used to input the validation clinical feature vector into the dynamic index interval prediction model, calculate the matching degree between the prediction result corresponding to the validation clinical feature vector and the true label corresponding to the validation clinical feature vector, and determine multiple validation data of the dynamic index interval prediction model based on the matching degree, wherein the multiple validation data include accuracy, weighted F1 score, area under the receiver operating characteristic curve, recall and precision. The verification unit is used to generate a corresponding confusion matrix based on the multiple verification data, to statistically analyze the true positive and false negative distributions of each risk stratification interval according to the confusion matrix, and to verify whether the dynamic indicator interval prediction model meets the preset discrimination balance requirements in different clinical risk populations through the true positive and false negative distributions.

8. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy as described in any one of claims 1-4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy as described in any one of claims 1-4.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed to implement the method for dynamic prediction of key indicators for non-surgical treatment of ectopic pregnancy as described in any one of claims 1-4.

Images