Deep endometriosis postoperative pregnancy outcome prediction and model construction method and device

Abstract

The invention discloses a deep endometriosis postoperative pregnancy outcome prediction and model construction method and device, and belongs to the field of gynaecology and obstetrics data processing, and the method comprises the steps: collecting and dividing retrospective collection data of a first target group before a first time point, and obtaining a first training set and a first verification set; after a plurality of different types of preset prediction models are trained through the first training set, screening a first prediction model from the trained preset prediction models through the first verification set; after the first time point, collecting and dividing prospective collection data of a second target population, and obtaining a second training set and a second verification set; when it is judged that the first prediction model needs to be calibrated according to the second verification set, the first prediction model is finely adjusted according to the second training set; and taking the finally obtained first prediction model as a deep endometriosis postoperative pregnancy outcome model. According to the method, the postoperative pregnancy probability of the deep endometriosis patient can be predicted through various clinical risk factors.

Description

Technical Field

[0001] The present application relates to the field of obstetrics and gynecology data processing, and in particular to a method and device for predicting pregnancy outcomes and model construction after surgery for deep endometriosis. Background Art

[0002] Deep endometriosis (DIE), also known as deep infiltrating endometriosis, is a specific clinical and pathological type of endometriosis. Surgical treatment for patients with DIE and infertility requires complex considerations, and currently, there is no method to accurately predict postoperative pregnancy outcomes in patients with DIE and infertility. Therefore, it is necessary to predict postoperative pregnancy outcomes based on clinical data in patients with DIE and infertility in order to provide personalized treatment plans.

[0003] In order to predict the postoperative pregnancy outcome of patients with DIE combined with infertility, the existing technology collects clinical data of patients with DIE combined with infertility who have already had a pregnancy outcome, and conducts correlation analysis based on the patients' clinical data and the final pregnancy outcome to obtain the influence of different single factors on the pregnancy outcome. However, the results reported so far are inconsistent, and the number of relevant clinical cases of patients with DIE combined with infertility included is too small, which will lead to the accuracy of the results obtained by the above method being too low. At the same time, since the factors that affect the final pregnancy outcome of patients with DIE combined with infertility are usually the coupling results of multiple factors, and the above method only relies on the influence of a single factor on different pregnancy outcomes, the final prediction results of the postoperative pregnancy outcome of patients with DIE combined with infertility will be inaccurate. Therefore, there is currently a lack of an effective model for predicting the postoperative pregnancy outcome of deep infiltrating endometriosis.

[0004] Therefore, how to construct a high-quality and effective prediction model for postoperative pregnancy outcomes in patients with DIE and infertility is a technical problem that needs to be solved at present. Summary of the Invention

[0005] The present application provides a method and device for predicting and modeling the postoperative pregnancy outcome of deep endometriosis, which can solve the current problem of the lack of high-quality postoperative pregnancy outcome prediction models for patients with DIE combined with infertility.

[0006] In some embodiments of the present application, a method for constructing a pregnancy outcome model after surgery for deep endometriosis is provided, comprising:

[0007] Before the first time point, retrospective data of a first target population is collected and divided into a first training set and a first validation set; wherein the first target population comprises patients diagnosed with deep endometriosis and infertility, who have undergone surgery and have achieved pregnancy outcomes;

[0008] After training several different types of preset prediction models using the first training set, a first prediction model is selected from the trained preset prediction models using the first validation set; wherein the preset prediction models include: a Cox regression model, a random forest model, and a LASSO regression model;

[0009] After the first time point, prospectively collected data from a second target population is collected and divided into a second training set and a second validation set; wherein the second target population comprises patients diagnosed with deep endometriosis and infertility, who have undergone surgery and are awaiting postoperative follow-up;

[0010] Determine whether the first prediction model needs to be calibrated based on the second validation set; if necessary, fine-tune the first prediction model based on the second training set to obtain a deep endometriosis postoperative pregnancy outcome model; if not, use the first prediction model as the deep endometriosis postoperative pregnancy outcome model.

[0011] Compared with the prior art, the above embodiments have the following beneficial effects: In response to the problem of the lack of high-quality prediction models, the present application simultaneously adopts multiple statistical methods (such as Cox proportional hazard regression, random forest, LASSO regression) to construct a prediction model, and screens out the optimal first prediction model, which significantly improves the scientificity and reliability of model selection and overcomes the potential limitations of a single model; in addition, the present application not only uses retrospectively collected data for internal training and preliminary verification, but more importantly, through prospective cohort studies, and collects prospectively collected data for external verification and correction of the first prediction model, it ensures the robustness and clinical applicability of the model, and also solves the problem of the lack of high-quality postoperative pregnancy outcome prediction models for patients with DIE combined with infertility.

[0012] Furthermore, the selecting the first prediction model from the preset prediction models using the first validation set includes:

[0013] Using the first validation set, plotting first ROC curves corresponding to the respective preset prediction models, and calculating first areas under the curves corresponding to the respective first ROC curves;

[0014] The preset prediction model corresponding to the largest area under the first curve among the preset prediction models is used as the first prediction model;

[0015] The first prediction model is verified according to a preset performance indicator, and when the first prediction model fails the verification, the first prediction model is fine-tuned according to the first training set.

[0016] Compared to existing technologies, the above embodiment has the following beneficial effects: by plotting the ROC curve and calculating the area under the ROC curve, the true positive rate and false positive rate of the model are comprehensively considered through the area under the curve, thereby more comprehensively measuring the classification ability of the model at different thresholds. At the same time, the first prediction model is screened by the area under the curve, ensuring that the selected first prediction model has the best discrimination in the retrospectively collected data, providing a high starting point model for the calibration of subsequent prospectively collected data, avoiding subjective judgment bias, and improving the scientific nature and reliability of model selection.

[0017] Furthermore, determining whether the first prediction model needs to be calibrated based on the second validation set includes:

[0018] Using the second validation set, plotting a second ROC curve corresponding to the first prediction model, and calculating a second area under the curve corresponding to the second ROC curve;

[0019] When the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the first curve corresponding to the first prediction model, the first prediction model needs to be calibrated.

[0020] Compared with the existing technology, the above embodiment has the following beneficial effects: a dynamic calibration mechanism is introduced to determine whether the model needs to be adjusted and trigger the calibration process by comparing the AUC decrease of the prospective validation set and the slope of the calibration curve, thereby ensuring that the prediction results of the model on new data are both accurate and reliable. At the same time, the dynamic calibration mechanism enhances the adaptability of the model to changes in external data, reduces the prediction error caused by differences in clinical data distribution, and improves the robustness of the model.

[0021] Furthermore, fine-tuning the first prediction model according to the second training set includes:

[0022] Using the second validation set, plotting a calibration curve corresponding to the first prediction model, and calculating a calibration slope of the calibration curve;

[0023] determining whether the calibration slope is abnormal;

[0024] If it is abnormal, then fine-tune the first prediction model by adding nonlinear terms and combining with the second training set; if it is not abnormal, then fine-tune the first prediction model by optimizing feature engineering and combining with the second training set;

[0025] The fine-tuned first prediction model is verified according to the preset performance indicator. When the fine-tuned first prediction model fails the verification, the first prediction model is retrained according to the first training set and the second training set.

[0026] Compared with the existing technology, the above embodiment has the following beneficial effects: for abnormal calibration slope, the expression ability of the model is enhanced by introducing nonlinear features, the ability of the model to capture complex relationships is improved, and the problem that the linear model assumption is not valid is solved; further, when only the area under the curve does not meet the standard, the noise interference is reduced and the generalization ability of the model is improved by screening key features, processing missing values ​​or deriving new features; finally, when the performance of the fine-tuned model still does not meet the standard, the model is retrained using all historical data to avoid local adjustments falling into local optimality.

[0027] Another embodiment of the present application further provides a method for predicting pregnancy outcomes after surgery for deep endometriosis, comprising:

[0028] Collect clinical data of the target to be predicted;

[0029] The clinical data are input into a deep endometriosis postoperative pregnancy outcome model to obtain a deep endometriosis postoperative pregnancy outcome model; wherein, the deep endometriosis postoperative pregnancy outcome model is constructed and obtained by the deep endometriosis postoperative pregnancy outcome model construction method described in any one of the embodiments of the present application.

[0030] Compared with the existing technology, the above embodiment has the following beneficial effects: by constructing a completed deep endometriosis postoperative pregnancy outcome model as a tool for predicting pregnancy outcomes, compared with the traditional list chart that relies on expert experience or reflects the influence of a single indicator, it can integrate multi-dimensional clinical data information, avoid human subjective judgment, improve prediction accuracy, and at the same time improve prediction efficiency.

[0031] Another embodiment of the present application further provides a device for constructing a pregnancy outcome model after deep endometriosis surgery, comprising: a first data acquisition module, a first prediction model acquisition module, a second data acquisition module, and a pregnancy outcome model acquisition module after deep endometriosis surgery;

[0032] The first data collection module is configured to collect retrospectively collected data of a first target population before a first time point, and divide the retrospectively collected data into a first training set and a first validation set; wherein the first target population is patients diagnosed with deep endometriosis and infertility, who have undergone surgery and have achieved pregnancy outcomes;

[0033] The first prediction model acquisition module is configured to train several different types of preset prediction models using the first training set, and then select a first prediction model from the trained preset prediction models using the first validation set; wherein the preset prediction models include: a Cox regression model, a random forest model, and a LASSO regression model;

[0034] The second data collection module is configured to collect prospectively collected data of a second target population after the first time point, and divide the prospectively collected data into a second training set and a second validation set; wherein the second target population comprises patients diagnosed with deep endometriosis and infertility, who have undergone surgery and are awaiting postoperative follow-up;

[0035] The deep endometriosis postoperative pregnancy outcome model acquisition module is used to determine whether the first prediction model needs to be calibrated based on the second validation set; if necessary, fine-tune the first prediction model based on the second training set to obtain the deep endometriosis postoperative pregnancy outcome model; if not, use the first prediction model as the deep endometriosis postoperative pregnancy outcome model.

[0036] Furthermore, the selecting the first prediction model from the trained preset prediction models through the first validation set includes:

[0037] Using the first validation set, plotting first ROC curves corresponding to the respective preset prediction models, and calculating first areas under the curves corresponding to the respective first ROC curves;

[0038] Determine, among the preset prediction models, the preset prediction model corresponding to the largest area under the first curve as the first prediction model;

[0039] The first prediction model is verified according to a preset performance indicator, and when the first prediction model fails the verification, the first prediction model is fine-tuned according to the first training set.

[0040] Furthermore, determining whether the first prediction model needs to be calibrated based on the second validation set includes:

[0041] Using the second validation set, plotting a second ROC curve corresponding to the first prediction model, and calculating a second area under the curve corresponding to the second ROC curve;

[0042] When the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the first curve corresponding to the first prediction model, the first prediction model needs to be calibrated.

[0043] Furthermore, fine-tuning the first prediction model according to the second training set includes:

[0044] Using the second validation set, plotting a calibration curve corresponding to the first prediction model, and calculating a calibration slope of the calibration curve;

[0045] determining whether the calibration slope is abnormal;

[0046] If it is abnormal, then fine-tune the first prediction model by adding nonlinear terms and combining with the second training set; if it is not abnormal, then fine-tune the first prediction model by optimizing feature engineering and combining with the second training set;

[0047] The fine-tuned first prediction model is verified according to the preset performance indicator. When the fine-tuned first prediction model fails the verification, the first prediction model is retrained according to the first training set and the second training set.

[0048] Another embodiment of the present application further provides a device for predicting pregnancy outcomes after deep endometriosis surgery, comprising: a third data acquisition module and a pregnancy outcome acquisition module after deep endometriosis surgery;

[0049] Wherein, the third data acquisition module is used to collect clinical data of the target to be predicted;

[0050] The deep endometriosis postoperative pregnancy outcome acquisition module is used to input the clinical data into the deep endometriosis postoperative pregnancy outcome model to obtain the deep endometriosis postoperative pregnancy outcome; wherein, the deep endometriosis postoperative pregnancy outcome model is constructed and acquired through the deep endometriosis postoperative pregnancy outcome model construction device described in any one of the embodiments of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0052] Figure 1 This is a flow chart of a method for constructing a pregnancy outcome model after surgery for deep endometriosis provided in some embodiments of the present application;

[0053] Figure 2 This is a flow chart of a method for predicting pregnancy outcomes after surgery for deep endometriosis provided in some embodiments of the present application;

[0054] Figure 3 This is a schematic diagram of the structure of a device for constructing a pregnancy outcome model after deep endometriosis surgery provided in some embodiments of the present application;

[0055] Figure 4 This is a schematic structural diagram of a device for predicting pregnancy outcomes after surgery for deep endometriosis provided in some embodiments of the present application. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0058] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0059] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0060] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0061] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0062] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0063] Because the treatment of patients with DIE combined with infertility requires consideration of multiple factors, such as patient age, ovarian reserve function, duration of infertility, previous reproductive and surgical history, the female partner's fallopian tube condition, and the male partner's semen condition, clinical decision-making for patients with DIE combined with infertility is particularly challenging. It is necessary to predict the pregnancy outcome of patients with DIE combined with infertility in advance to assist in clinical decision-making.

[0064] Currently, some scholars have analyzed the ICSI–IVF (Intracytoplasmic Sperm Injection-In Vitro Fertilization) outcomes of 75 patients with deep colorectal infiltrating endometriosis and infertility in their medical center, studied the impact of different factors on the clinical pregnancy rate of ICSI-VF, and designed an intuitive list chart that can be used to predict the ICS-IVF pregnancy rate of DIE patients. However, the list chart included too little clinical data, and by analyzing the correlation between independent factors and ICS-IVF pregnancy rates, it failed to consider the coupling effect when multiple factors occur simultaneously. At the same time, the above scheme is mainly aimed at non-surgical patients with DIE, and the model has not been verified, so it cannot accurately predict pregnancy outcomes.

[0065] Please refer to Figure 1 To address the current lack of a high-quality and effective model for predicting postoperative pregnancy outcomes in patients with DIE and infertility, the present invention provides a method for constructing a postoperative pregnancy outcome model for deep endometriosis, including S101 to S104, specifically including:

[0066] S101: Before a first time point, retrospectively collected data of a first target population is collected, and the retrospectively collected data is divided into a first training set and a first validation set; wherein, the first target population is patients who have been diagnosed with deep endometriosis and infertility, and have undergone surgery and have achieved pregnancy outcomes.

[0067] Preferably, in some embodiments of the present application, prior to the first time point, retrospective data collection of the first target population is performed using a case-control study method, including retrospective collection of data from a large data platform, from January 2009 to January 2022, from patients diagnosed with deep infiltrating endometriosis who had complete medical histories and underwent surgery and who desired pregnancy after surgery, and follow-up collection of pregnancy outcomes and other pregnancy information. The case group consists of patients with deep infiltrating endometriosis who successfully became pregnant after surgery; the control group consists of patients with deep infiltrating endometriosis who did not become pregnant after surgery.

[0068] The retrospectively collected data included the following relevant information of each patient in the first target population: demographic characteristics (age, weight, BMI, menstrual and reproductive history, years of infertility, and past medical history); clinical symptoms (pain, general quality of life, bowel function, urinary symptoms, and sexual function scores); gynecological examination reports (married patients underwent triple consultation and unmarried patients underwent rectal examination); imaging data (ultrasound, MRI, colonoscopy, IVP, etc.); serum markers (tumor markers, ovarian function, etc.); surgical pathology information (lesion characteristics, rAFS stage and score, EFI score, ENZIAN classification, surgical method, surgical time, whether the surgeon was on the same stage, residual lesions, and pathology report); postoperative treatment (whether GnRHa was used and the dosage, etc.); and postoperative follow-up (whether pregnancy occurred, method of pregnancy, pregnancy outcome, time from pregnancy to surgery, postoperative pain, general quality of life, bowel function, urinary symptoms, and sexual function scores).

[0069] Preferably, in some embodiments of the present application, after collecting retrospective data, it also includes: using a univariate Cox regression model (Cox proportional-hazards model) to analyze the impact of a single indicator (demographic characteristics, clinical symptoms, specialist examinations, imaging characteristics, serum markers, surgical conditions, postoperative treatment) on the postoperative pregnancy outcome of deep infiltrating endometriosis. Screen out single variables that have a significant impact on the postoperative pregnancy outcome of DIE. For categorical variables, a confusion matrix is ​​directly established, and the sensitivity, specificity, negative and positive predictive values ​​of a single indicator are calculated using statistical formulas and statistical methods; for continuous variables, the ROC curve (Receiver Operating Characteristic Curve, receiver characteristic curve, such as time-dependent ROC) of survival analysis is used to determine the optimal cutoff value, and patients are divided into "high risk" and "low risk" groups, and then a confusion matrix is ​​established, and the sensitivity, specificity, negative and positive predictive values ​​of a single indicator are calculated using statistical formulas.

[0070] S102: After training several different types of preset prediction models using the first training set, screen a first prediction model from the trained preset prediction models using the first validation set; wherein the preset prediction models include: a Cox regression model, a random forest model, and a LASSO regression model (Least Absolute Shrinkage and Selection Operator).

[0071] Preferably, in some embodiments of the present application, when training the Cox regression model, the method includes: based on the preliminary screening results of the univariate Cox proportional hazard model, performing a multivariate regression model analysis on the single factors with statistical association. The statistical method of multivariate Cox regression analysis is used to further identify independent influencing factors with statistical significance (P < 0.1), and weighting is performed in combination with the β regression coefficient. The R language programming environment is used to construct a parameterized Cox regression model to obtain a pregnancy outcome prediction model.

[0072] Preferably, in some embodiments of the present application, when training the random forest model, it includes: using the R language programming environment and random forest, using retrospectively collected data (demographic characteristics, clinical symptoms, specialist examinations, imaging characteristics, serum markers, surgical conditions, postoperative treatment), to obtain a pregnancy outcome prediction model.

[0073] Preferably, in some embodiments of the present application, when training the LASSO regression model, it includes: using the R language programming environment and the LASSO regression model, using retrospectively collected data (demographic characteristics, clinical symptoms, specialist examinations, imaging characteristics, serum markers, surgical conditions, postoperative treatment), to obtain a pregnancy outcome prediction model.

[0074] Furthermore, in some embodiments of the present application, screening the first prediction model from the preset prediction models using the first validation set includes:

[0075] Using the first validation set, plotting first ROC curves corresponding to the respective preset prediction models, and calculating first areas under the curves corresponding to the respective first ROC curves;

[0076] The preset prediction model corresponding to the largest area under the first curve among the preset prediction models is used as the first prediction model;

[0077] The first prediction model is verified according to a preset performance indicator, and when the first prediction model fails the verification, the first prediction model is fine-tuned according to the first training set.

[0078] Preferably, in some embodiments of the present application, determining the first prediction model corresponding to the largest area under the first curve includes: using the ROC curve to evaluate and measure the overall performance of each trained preset prediction model, obtaining the first ROC curve of each trained preset prediction model, further calculating the first area under the curve (Area Under The Curve, AUC) of each trained preset prediction model, and screening out the prediction model with a first AUC greater than 0.7 and the highest first AUC as the first prediction model.

[0079] By plotting the ROC curve and calculating its area under the curve, the model's true positive rate and false positive rate are comprehensively considered through the area under the curve, thereby more comprehensively measuring the model's classification ability at different thresholds. At the same time, the area under the curve is used to screen the first prediction model, ensuring that the selected first prediction model has the best discrimination in the retrospectively collected data. This provides a high starting point model for calibration of subsequent prospectively collected data, avoids subjective judgment bias, and improves the scientific nature and reliability of model selection.

[0080] Preferably, in some embodiments of the present application, after the first prediction model is determined, the first prediction model is verified according to preset performance indicators, including: cross-validating the first prediction model using R language; drawing a calibration curve to determine the predictive ability of the first prediction model; using decision curve analysis (DCA) to evaluate the clinical practicality of the first prediction model; dividing DIE postoperative patients into high-risk and low-risk groups according to the first prediction model, and using the Kaplan-Meier analysis method to compare the predicted pregnancy risks of the two groups after surgery to complete the evaluation of the prediction model.

[0081] Preferably, in some embodiments of the present application, when any of the above performance evaluation results does not meet the preset standard, the first prediction model is fine-tuned in combination with the first training set through the following steps S1021 to S1023:

[0082] S1021: Analyze the reasons why the performance evaluation results do not meet the preset standards from the data level, feature level, and model level, including:

[0083] For data-level issues, determine whether the sample size is insufficient and check whether each variable meets the rule of having at least ten events, especially when the first prediction model is a Cox regression model or a LASSO regression model; further judge the data quality, such as whether there are too many missing values ​​(missing values ​​account for more than 30%), unbalanced data categories (such as a 9:1 ratio of successful and failed pregnancies), or interference from outliers; finally, determine whether the events and event information are incomplete, such as excessive right-censored data (such as insufficient follow-up time resulting in a large number of unobserved pregnancy outcomes).

[0084] For feature-level issues, determine whether there is selection bias in the predictors, such as omitting key clinical variables (such as compliance with postoperative hormone therapy); further determine whether nonlinear or interactive effects are not successfully captured, such as the nonlinear association between age and ovarian reserve function is not modeled; finally, determine whether there is multicollinearity, such as highly correlated variables (such as CA125 and CA19-9) resulting in unstable coefficients.

[0085] For model-level problems, determine whether the model assumptions are invalid, such as when the first prediction model is a Cox regression model, the Cox regression model violates the proportional hazard assumption; further determine whether the hyperparameters of the first prediction model have not been tuned, such as when the first prediction model is a random forest model, the tree depth of the random forest is inappropriately selected, or when the first prediction model is a LASSO regression model, the lambda value of the LASSO regression model is inappropriately selected; finally, determine whether the first prediction model is overfitting or underfitting, such as when the training set performs well but the validation set performs poorly (overfitting), or both perform poorly (underfitting).

[0086] S1022: Based on the problem diagnosed in S1021, correct and improve it by following the steps below:

[0087] For problems caused by the data level: if the sample size is insufficient, external data should be supplemented, such as incorporating multicenter cohorts or public databases; if the data categories are unbalanced, the problem of category imbalance should be addressed through oversampling (Synthetic Minority Oversampling, SMOTE); if the data contain too many missing values, multiple imputation by chained equations (MICE) should be used instead of simple deletion, and missing values ​​for key variables should be set as a separate category; if the data contain too many outliers, winsorizing truncation or correction based on clinical thresholds should be performed (such as serum CA125 greater than 1000 U / mL should be set as missing); if there are too many right-censored data, the follow-up time should be extended or the follow-up density should be increased to reduce right censoring, and inverse probability of censoring weighting (IPCW) should be used to correct the bias of the censored data.

[0088] For feature-level issues: by performing spline interpolation (natural cubic spline) and binning (e.g., age less than 30, age between 30 and 35, and age greater than 35 years) on continuous variables (e.g., age less than 30, age between 30 and 35, and age greater than 35 years), the problem of the results failing to capture nonlinear relationships was addressed; by manually adding interaction terms based on clinical knowledge (e.g., "postoperative GnRH use × lesion size"), the problem of the results failing to capture interaction effects was addressed; by using multi-method cross-validation [e.g., first using random forest / LASSO to screen variables, then manually reviewing the clinical significance, and using elastic network (α = 0.5) to retain relevant variable groups (e.g., multiple serum markers) for the LASSO model] or domain knowledge-driven [e.g., forcibly incorporating clinical consensus variables (e.g., the Endometriosis Fertility Index EFI)], the problem of omitting key clinical variables was addressed; finally, by performing deep learning feature extraction on image features (e.g., extracting lesion texture features), a composite index (e.g., "inflammation-fibrosis score" = 0.3 × CA125 + 0.7 × ultrasound elastography value) was constructed to optimize feature representation.

[0089] For model-level issues: If the primary prediction model is a Cox regression model, then the introduction of covariates (such as the tt function) or the use of parametric models (Weibull / AFT models) or additive hazard models (Additive Hazards) can address the issue of Cox regression models violating the proportional hazards assumption. Furthermore, nonlinear or interaction terms can be added, and feature selection can be performed using the Lasso-Cox combination to improve discrimination. If the primary prediction model is a random forest model, the number of variables randomly selected for each tree and the minimum sample size for terminal nodes can be adjusted through grid search, while the number of trees (ntree>500) can be increased to reduce variance in order to optimize the parameters of the random forest model. Furthermore, a survival forest (such as the randomForestSRC package) can be used instead of a classification forest, and dynamic prediction indicators can be calculated based on the cumulative hazard function to process survival data. If the first prediction model is the LASSO regression model, the elastic network (α = 0.1-0.9) is used to balance variable selection and collinearity tolerance, and the LASSO is grouped to process categorical variables (such as the surgical type variable group) to achieve regularization optimization; further, Bootstrap resampling is performed 500 times, and variables with a selection frequency greater than 60% are retained to enhance the stability of the LASSO regression model.

[0090] S1023: If the performance evaluation result of the first prediction model still does not meet the preset standard after S1021 to S1022, the first prediction model is corrected by the following method:

[0091] First, try to use a more complex model as the first prediction model, such as using neural networks (MLP / Transformer) for high-dimensional data (such as radiomics), or random forests through gradient boosting machines (XGBoost / LightGBM). This method is particularly suitable for small sample nonlinear data; secondly, try to use the prediction results of Cox, random forest, and LASSO as meta-features, train a logistic regression model as the first prediction model, and for time-dependent predictions, assign different model weights according to time points.

[0092] S103: After the first time point, prospectively collected data of a second target population is collected, and the prospectively collected data is divided into a second training set and a second validation set; wherein the second target population is patients who have been diagnosed with deep endometriosis and infertility, have undergone surgery and are waiting for postoperative follow-up.

[0093] Preferably, in some embodiments of the present application, after the first time point, the prospective collection data of the second target population is collected, including: since January 2022, the relevant clinical information of patients with deep infiltrating endometriosis and infertility who have been treated and received surgical treatment is prospectively collected as prospective collection data, wherein the prospective collection data includes: demographic characteristics (age, weight, BMI, menstrual and reproductive history, years of infertility, past history); clinical symptoms (pain, general quality of life, intestinal function, urinary system symptoms and sexual function scores); gynecological examination report (married patients undergo triple consultation, unmarried patients undergo rectal examination; imaging Data (B-ultrasound, MRI, colonoscopy, IVP, etc.); serum markers (tumor markers, ovarian function, etc.); surgical pathology information (lesion characteristics, rAFS stage and score, EFI score, ENZIAN classification, surgical method, surgical time, whether the surgeon was on the same stage, residual lesions, pathology report); postoperative treatment (whether GnRHa was used and the dosage, etc.). Furthermore, the pregnancy outcomes of the patients were followed up regularly after surgery, and their pregnancy outcomes and other pregnancy conditions were collected, including: whether the pregnancy was present, the pregnancy method, the pregnancy outcome, the time from pregnancy to surgery, postoperative pain, general quality of life, bowel function, urinary symptoms, and sexual function scores. It should be noted that the pregnancy outcomes and other pregnancy conditions collected during regular postoperative follow-up were also part of the prospectively collected data, and the pregnancy outcomes were used to compare the accuracy of the model assessment results.

[0094] S104: Determine whether the first prediction model needs to be calibrated based on the second validation set; if so, fine-tune the first prediction model based on the second training set to obtain a deep endometriosis postoperative pregnancy outcome model; if not, use the first prediction model as the deep endometriosis postoperative pregnancy outcome model.

[0095] Furthermore, in some embodiments of the present invention, determining whether the first prediction model needs to be calibrated based on the second validation set includes:

[0096] Using the second validation set, plotting a second ROC curve corresponding to the first prediction model, and calculating a second area under the curve corresponding to the second ROC curve;

[0097] When the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the first curve corresponding to the first prediction model, the first prediction model needs to be calibrated.

[0098] Preferably, in some embodiments of the present application, when the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the corresponding first curve, the first prediction model needs to be calibrated, including: determining whether the second AUC obtained from the second training set decreases by more than 0.15 compared to the first AUC obtained from the first training set; if so, further calibration of the first prediction model is required.

[0099] As can be seen from the above examples, this application introduces a dynamic calibration mechanism, which performs external validation through prospective cohort studies. By comparing the AUC drop in the prospective validation set and the slope of the calibration curve, it determines whether the model needs to be adjusted and triggers the calibration process, ensuring that the model's prediction results on new data are both accurate and reliable. At the same time, the dynamic calibration mechanism enhances the model's adaptability to changes in external data, reduces prediction errors caused by differences in clinical data distribution, and improves the robustness of the model.

[0100] Furthermore, in some embodiments of the present invention, fine-tuning the first prediction model based on the second training set includes:

[0101] Using the second validation set, plotting a calibration curve corresponding to the first prediction model, and calculating a calibration slope of the calibration curve;

[0102] determining whether the calibration slope is abnormal;

[0103] If it is abnormal, then fine-tune the first prediction model by adding nonlinear terms and combining with the second training set; if it is not abnormal, then fine-tune the first prediction model by optimizing feature engineering and combining with the second training set;

[0104] The fine-tuned first prediction model is verified according to the preset performance indicator. When the fine-tuned first prediction model fails the verification, the first prediction model is retrained according to the first training set and the second training set.

[0105] Preferably, in some embodiments of the present application, the determination of whether the calibration slope is abnormal includes: determining whether the 95% confidence interval of the calibration slope does not contain 1 or the P value is less than 0.05; determining whether the calibration slope is less than 0.85 (i.e., the first prediction model is overconfident and overestimates the risk) or whether the calibration slope is greater than 1.15 (i.e., the first prediction model is underconfident and underestimates the risk); determining whether the difference between the predicted value and the actual incidence rate is greater than 10% in the critical pregnancy probability interval (such as 30% to 70%); if any one of the above three determination conditions is met, it is determined to be abnormal. At this time, it is necessary to fine-tune the first prediction model by adding a nonlinear term and combining the second training set, or to start transfer learning, fix the underlying parameters, and fine-tune the decision layer of the first prediction model.

[0106] Preferably, in some embodiments of the present application, the method of solving feature-level problems in S1022 may be adopted to fine-tune the first prediction model by optimizing feature engineering and combining the second training set.

[0107] Preferably, in some embodiments of the present application, the verification of the fine-tuned first prediction model according to the preset performance indicators includes: first, determining whether the second AUC recalculated using the second validation set for the fine-tuned first prediction model satisfies the requirement of being greater than or equal to 0.75, and whether the decrease is less than or equal to 0.1 compared with the first AUC recalculated from the first training set; then, determining whether the calibration slope of the fine-tuned first prediction model satisfies the requirement of being in the range of 0.85 to 1.15, and whether the Brier score is less than or equal to 0.15; finally, determining whether the decision curve analysis of the fine-tuned first prediction model satisfies the requirement of being in the range of 30% to 50% of the threshold probability, and whether the net benefit is greater than the "all intervention" and "no intervention" strategies; if all three of the above evaluation conditions are met, the first prediction model is determined to have passed the performance verification.

[0108] Preferably, in some embodiments of the application, after obtaining the final deep endometriosis postoperative pregnancy outcome model, the cut-off value of the deep infiltrating endometriosis model system is calculated by using R language, so as to risk stratify deep infiltrating endometriosis patients and realize stratified management of patients.

[0109] In summary, it can be seen that compared with the prior art, the method for constructing a postoperative pregnancy outcome model for deep endometriosis provided in the embodiment of the present application has the following beneficial effects: in response to the problem of the lack of high-quality prediction models, the present application simultaneously uses multiple statistical methods (such as Cox proportional hazard regression, random forest, LASSO regression) to construct a prediction model, and screens out the optimal first prediction model from it, significantly improving the scientificity and reliability of model selection and overcoming the potential limitations of a single model; in addition, the present application not only uses retrospectively collected data for internal training and preliminary verification, but more importantly, through prospective cohort studies, and collects prospective collected data for external verification and correction of the first prediction model, by adopting a comprehensive evaluation system (including discrimination AUC / C-index, calibration curve / Brier score, clinical practicality DCA decision curve analysis and Kaplan-Meier risk stratification) to comprehensively evaluate the predictive performance, calibration and clinical value of the model, ensuring the robustness and clinical applicability of the model, and also solving the problem of the lack of high-quality postoperative pregnancy outcome prediction model for patients with DIE and infertility. This model can improve the accuracy of the prediction of the probability of pregnancy after surgery in patients with deep DIE, thereby improving the accuracy of risk stratification for patients.

[0110] like Figure 2 As shown, a method for predicting pregnancy outcomes after surgery for deep endometriosis provided in an embodiment of the present application includes S201 to S202, specifically:

[0111] S201: Collect clinical data of the target to be predicted.

[0112] Preferably, in some embodiments of the present application, the clinical data include: demographic characteristics of the target to be predicted (age, weight, BMI, menstrual and reproductive history, years of infertility, past history); clinical symptoms (pain, general quality of life, intestinal function, urinary system symptoms and sexual function score); gynecological examination report (married patients undergo triple consultation, unmarried patients undergo anal examination; imaging data (B-ultrasound, MRI, colonoscopy, IVP, etc.); serum markers (tumor markers, ovarian function, etc.); surgical pathology information (lesion characteristics, rAFS staging and scoring, EFI score, ENZIAN classification, surgical method, surgical time, whether the surgeon is on the same table, residual lesions, pathology report); postoperative treatment (whether GnRHa is used and the dosage, etc.).

[0113] S202: Inputting the clinical data into a deep endometriosis postoperative pregnancy outcome model to obtain a deep endometriosis postoperative pregnancy outcome model; wherein, the deep endometriosis postoperative pregnancy outcome model is constructed and obtained by the deep endometriosis postoperative pregnancy outcome model construction method described in any one of the embodiments of the present application.

[0114] In summary, it can be seen that compared with the existing technology, the method for predicting pregnancy outcomes after surgery for deep endometriosis provided in the embodiment of the present application has the following beneficial effects: by constructing a completed pregnancy outcome model after surgery for deep endometriosis as a tool for predicting pregnancy outcomes, compared with the traditional list chart that relies on expert experience or reflects the influence of a single indicator, it can integrate multi-dimensional clinical data information, avoid human subjective judgment, improve prediction accuracy, and at the same time improve prediction efficiency.

[0115] like Figure 3 As shown, based on the above-mentioned embodiment of the method for constructing a pregnancy outcome model after deep endometriosis surgery, an embodiment of the present application provides a device for constructing a pregnancy outcome model after deep endometriosis surgery, including: a first data acquisition module 301, a first prediction model acquisition module 302, a second data acquisition module 303 and a deep endometriosis surgery pregnancy outcome model acquisition module 304.

[0116] Furthermore, in some embodiments of the present application, the first data acquisition module 301 is used to collect retrospectively collected data of a first target population before a first time point, and divide the retrospectively collected data into a first training set and a first validation set; wherein, the first target population is patients diagnosed with deep endometriosis and infertility, who have undergone surgery and have a pregnancy outcome; the first prediction model acquisition module 302 is used to train several different types of preset prediction models through the first training set, and then screen the first prediction model from each of the trained preset prediction models through the first validation set; wherein, the preset prediction models include: Cox regression model, random forest model and LASSO regression model; the second data acquisition module 303 is used to collect prospectively collected data of a second target population after the first time point, and divide the prospectively collected data into a second training set and a second validation set; wherein, the second target population is patients who have been diagnosed with deep endometriosis and infertility, have undergone surgery and are waiting for postoperative follow-up; the deep endometriosis postoperative pregnancy outcome model acquisition module 304 is used to determine whether the first prediction model needs to be calibrated based on the second validation set; if so, fine-tune the first prediction model based on the second training set to obtain the deep endometriosis postoperative pregnancy outcome model; if not, use the first prediction model as the deep endometriosis postoperative pregnancy outcome model.

[0117] Furthermore, in some embodiments of the present application, the first prediction model is selected from the preset prediction models that have been trained through the first verification set, including: drawing the first ROC curve corresponding to each of the preset prediction models through the first verification set, and calculating the first area under the curve corresponding to each of the first ROC curves; taking the preset prediction model corresponding to the largest area under the first curve among the preset prediction models as the first prediction model; verifying the first prediction model according to preset performance indicators, and when the first prediction model fails to pass the verification, fine-tuning the first prediction model according to the first training set.

[0118] Furthermore, in some embodiments of the present application, determining whether the first prediction model needs to be calibrated based on the second validation set includes: drawing a second ROC curve corresponding to the first prediction model through the second validation set, and calculating the area under the second curve corresponding to the second ROC curve; when the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the corresponding first curve, the first prediction model needs to be calibrated.

[0119] Furthermore, in some embodiments of the present application, fine-tuning the first prediction model based on the second training set includes: drawing a calibration curve corresponding to the first prediction model through the second validation set, and calculating the calibration slope of the calibration curve; determining whether the calibration slope is abnormal; if it is abnormal, fine-tuning the first prediction model by adding a nonlinear term and combining it with the second training set; if it is not abnormal, fine-tuning the first prediction model by optimizing feature engineering and combining it with the second training set; verifying the fine-tuned first prediction model according to the preset performance indicators, and when the fine-tuned first prediction model fails the verification, retraining the first prediction model based on the first training set and the second training set.

[0120] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present application, which can realize the method for constructing a postoperative pregnancy outcome model for deep endometriosis provided by any of the above-mentioned method embodiments of the present application.

[0121] In summary, it can be seen that compared with the prior art, the apparatus for constructing a postoperative pregnancy outcome model for deep endometriosis provided in the embodiment of the present application has the following beneficial effects: in response to the problem of the lack of high-quality prediction models, the present application simultaneously uses multiple statistical methods (such as Cox proportional hazard regression, random forest, and LASSO regression) to construct a prediction model, and screens out the optimal first prediction model from it, significantly improving the scientificity and reliability of model selection and overcoming the potential limitations of a single model; in addition, the present application not only uses retrospectively collected data for internal training and preliminary verification, but more importantly, through prospective cohort studies, and collects prospective collected data for external verification and correction of the first prediction model, by adopting a comprehensive evaluation system (including discrimination AUC / C-index, calibration curve / Brier score, clinical practicality DCA decision curve analysis, and Kaplan-Meier risk stratification) to comprehensively evaluate the predictive performance, calibration, and clinical value of the model, thereby ensuring the robustness and clinical applicability of the model, and also solving the problem of the lack of high-quality postoperative pregnancy outcome prediction models for patients with DIE and infertility. This model can improve the accuracy of the prediction of the probability of pregnancy after surgery for patients with deep DIE, thereby improving the accuracy of risk stratification for patients.

[0122] like Figure 4 As shown, based on the above embodiment of the method for predicting pregnancy outcome after deep endometriosis surgery, an embodiment of the present application provides a device for predicting pregnancy outcome after deep endometriosis surgery, comprising: a third data acquisition module 401 and a pregnancy outcome acquisition module 402 after deep endometriosis surgery;

[0123] Furthermore, in some embodiments of the present application, the third data acquisition module 401 is used to collect clinical data of the target to be predicted; the deep endometriosis postoperative pregnancy outcome acquisition module 402 is used to input the clinical data into the deep endometriosis postoperative pregnancy outcome model to obtain the deep endometriosis postoperative pregnancy outcome; wherein, the deep endometriosis postoperative pregnancy outcome model is constructed and acquired through the deep endometriosis postoperative pregnancy outcome model construction device described in any one of the embodiments of the present application.

[0124] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present application, and it can implement the method for predicting pregnancy outcomes after surgery for deep endometriosis provided by any of the above-mentioned method embodiments of the present application.

[0125] In summary, it can be seen that compared with the existing technology, the device for predicting pregnancy outcomes after surgery for deep endometriosis provided in the embodiment of the present application has the following beneficial effects: by constructing a deep endometriosis postoperative pregnancy outcome model as a tool for predicting pregnancy outcomes, compared with the traditional list chart that relies on expert experience or reflects the influence of a single indicator, it can integrate multi-dimensional clinical data information, avoid human subjective judgment, improve prediction accuracy, and at the same time improve prediction efficiency.

[0126] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided herein, the connection relationship between the modules indicates that there is a communication connection between them, which may be implemented as one or more communication buses or signal lines. Those skilled in the art may understand and implement the present invention without inventive effort.

[0127] Based on the above-mentioned embodiments of the method for constructing a model of pregnancy outcome after deep endometriosis surgery / the method for predicting pregnancy outcome after deep endometriosis surgery, another embodiment of the present application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for constructing a model of pregnancy outcome after deep endometriosis surgery / the method for predicting pregnancy outcome after deep endometriosis surgery of any embodiment of the present application is implemented.

[0128] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more module elements may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0129] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0130] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0131] On the basis of the above-mentioned method embodiments, another embodiment of the present application provides a computer-readable storage medium, comprising a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the method for constructing a model for pregnancy outcome after surgery for deep endometriosis / the method for predicting pregnancy outcome after surgery for deep endometriosis as described in any of the above-mentioned method embodiments of the present application.

[0132] Wherein, the module / unit integrated in the device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program, when executed by the processor, can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device that can carry the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium, etc.

Claims

1. A method for constructing a pregnancy outcome model after deep endometriosis surgery, characterized in that: include: Before the first time point, retrospective data of a first target population is collected and divided into a first training set and a first validation set; wherein the first target population comprises patients diagnosed with deep endometriosis and infertility, who have undergone surgery and have achieved pregnancy outcomes; After training several different types of preset prediction models using the first training set, a first prediction model is selected from the trained preset prediction models using the first validation set; wherein the preset prediction models include: a Cox regression model, a random forest model, and a LASSO regression model; After the first time point, prospectively collected data from a second target population is collected and divided into a second training set and a second validation set; wherein the second target population comprises patients diagnosed with deep endometriosis and infertility, who have undergone surgery and are awaiting postoperative follow-up; Determine whether the first prediction model needs to be calibrated based on the second validation set; if necessary, fine-tune the first prediction model based on the second training set to obtain a deep endometriosis postoperative pregnancy outcome model; if not, use the first prediction model as the deep endometriosis postoperative pregnancy outcome model.

2. The method for constructing a pregnancy outcome model after deep endometriosis surgery according to claim 1, characterized in that: The selecting the first prediction model from the preset prediction models using the first validation set includes: Using the first validation set, plotting first ROC curves corresponding to the respective preset prediction models, and calculating first areas under the curves corresponding to the respective first ROC curves; The preset prediction model corresponding to the largest area under the first curve among the preset prediction models is used as the first prediction model; The first prediction model is verified according to a preset performance indicator, and when the first prediction model fails the verification, the first prediction model is fine-tuned according to the first training set.

3. The method for constructing a pregnancy outcome model after deep endometriosis surgery according to claim 2, wherein: The determining whether the first prediction model needs to be calibrated according to the second validation set includes: Using the second validation set, plotting a second ROC curve corresponding to the first prediction model, and calculating a second area under the curve corresponding to the second ROC curve; When the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the first curve corresponding to the first prediction model, the first prediction model needs to be calibrated.

4. The method for constructing a pregnancy outcome model after deep endometriosis surgery according to claim 2, wherein: The fine-tuning of the first prediction model according to the second training set includes: Using the second validation set, plotting a calibration curve corresponding to the first prediction model, and calculating a calibration slope of the calibration curve; determining whether the calibration slope is abnormal; If it is abnormal, then fine-tune the first prediction model by adding nonlinear terms and combining with the second training set; if it is not abnormal, then fine-tune the first prediction model by optimizing feature engineering and combining with the second training set; The fine-tuned first prediction model is verified according to the preset performance indicator. When the fine-tuned first prediction model fails the verification, the first prediction model is retrained according to the first training set and the second training set.

5. A method for predicting pregnancy outcome after surgery for deep endometriosis, characterized in that: include: Collect clinical data of the target to be predicted; The clinical data are input into a deep endometriosis postoperative pregnancy outcome model to obtain a deep endometriosis postoperative pregnancy outcome model; wherein, the deep endometriosis postoperative pregnancy outcome model is constructed and obtained by the deep endometriosis postoperative pregnancy outcome model construction method described in any one of claims 1 to 4.

6. A device for constructing a pregnancy outcome model after deep endometriosis surgery, characterized in that: include: a first data acquisition module, a first prediction model acquisition module, a second data acquisition module, and a deep endometriosis postoperative pregnancy outcome model acquisition module; The first data collection module is configured to collect retrospectively collected data of a first target population before a first time point, and divide the retrospectively collected data into a first training set and a first validation set; wherein the first target population is patients diagnosed with deep endometriosis and infertility, who have undergone surgery and have achieved pregnancy outcomes; The first prediction model acquisition module is configured to train several different types of preset prediction models using the first training set, and then select a first prediction model from the trained preset prediction models using the first validation set; wherein the preset prediction models include: a Cox regression model, a random forest model, and a LASSO regression model; The second data collection module is configured to collect prospectively collected data of a second target population after the first time point, and divide the prospectively collected data into a second training set and a second validation set; wherein the second target population comprises patients diagnosed with deep endometriosis and infertility, who have undergone surgery and are awaiting postoperative follow-up; The deep endometriosis postoperative pregnancy outcome model acquisition module is used to determine whether the first prediction model needs to be calibrated based on the second validation set; if necessary, fine-tune the first prediction model based on the second training set to obtain the deep endometriosis postoperative pregnancy outcome model; if not, use the first prediction model as the deep endometriosis postoperative pregnancy outcome model.

7. The apparatus for constructing a pregnancy outcome model after deep endometriosis surgery according to claim 6, characterized in that: The selecting a first prediction model from the trained preset prediction models using the first validation set includes: Using the first validation set, plotting first ROC curves corresponding to the respective preset prediction models, and calculating first areas under the curves corresponding to the respective first ROC curves; The preset prediction model corresponding to the largest area under the first curve among the preset prediction models is used as the first prediction model; The first prediction model is verified according to a preset performance indicator, and when the first prediction model fails the verification, the first prediction model is fine-tuned according to the first training set.

8. The apparatus for constructing a pregnancy outcome model after deep endometriosis surgery according to claim 7, characterized in that: The determining whether the first prediction model needs to be calibrated according to the second validation set includes: Using the second validation set, plotting a second ROC curve corresponding to the first prediction model, and calculating a second area under the curve corresponding to the second ROC curve; When the area under the second curve corresponding to the first prediction model decreases by a preset area compared to the area under the first curve corresponding to the first prediction model, the first prediction model needs to be calibrated.

9. The apparatus for constructing a pregnancy outcome model after deep endometriosis surgery according to claim 7, wherein: The fine-tuning of the first prediction model according to the second training set includes: Using the second validation set, plotting a calibration curve corresponding to the first prediction model, and calculating a calibration slope of the calibration curve; determining whether the calibration slope is abnormal; If it is abnormal, then fine-tune the first prediction model by adding nonlinear terms and combining with the second training set; if it is not abnormal, then fine-tune the first prediction model by optimizing feature engineering and combining with the second training set; The fine-tuned first prediction model is verified according to the preset performance indicator. When the fine-tuned first prediction model fails the verification, the first prediction model is retrained according to the first training set and the second training set.

10. A device for predicting pregnancy outcome after surgery for deep endometriosis, characterized in that: include: The third data collection module and the pregnancy outcome acquisition module after deep endometriosis surgery; Wherein, the third data acquisition module is used to collect clinical data of the target to be predicted; The deep endometriosis postoperative pregnancy outcome acquisition module is used to input the clinical data into the deep endometriosis postoperative pregnancy outcome model to obtain the deep endometriosis postoperative pregnancy outcome; wherein, the deep endometriosis postoperative pregnancy outcome model is constructed and acquired through the deep endometriosis postoperative pregnancy outcome model construction device described in any one of claims 6 to 9.

Images