Success Rate-Oriented Prediction Method, System and Product for Superovulation Medication Regimen

Abstract

The present invention belongs to the field of medical artificial intelligence, and provides a prediction method, system and product for superovulation drug use regimens oriented to success rates. The technical solution is to obtain the monitoring index data after the treatment of the patient's superovulation drug use regimen; according to the obtained monitoring index data after the treatment of the patient's superovulation drug use regimen, form the success rate evaluation value of each drug use regimen; convert the single classification label of the drug use regimen into a multi-classification label, and then combine the success rate evaluation value of each drug use regimen to weight the multi-classification label to obtain a multi-objective regression label; train the regression model based on the multi-objective regression label, and based on the trained regression model, combine the index data of the patient to be predicted for prediction, and convert the regression value of the prediction result to obtain the superovulation drug use regimen recommendation result. The present invention uses the multi-objective regression label for model training and prediction to avoid the problem of poor single label consistency, and the model can predict multiple drug use regimens for doctors to make decisions.

Description

Technical Field

[0001] The present invention belongs to the field of medical artificial intelligence, and particularly relates to a method, system and product for predicting a superovulation drug use plan oriented to success rate. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention, and do not necessarily constitute prior art.

[0003] "Superovulation drug use plan treatment" is a key step in assisted reproductive technology. This treatment method stimulates the ovaries with drugs to produce multiple eggs in one cycle, thereby increasing the chance of successful pregnancy. Although this technology is widely used clinically, there is currently no recognized gold standard to guide the determination of superovulation drug use plans. Doctors usually formulate drug use plans based on their own experience and the wishes of patients.

[0004] In clinical practice, the quality of the drug use plan directly affects the pregnancy outcome and the family economic cost of patients. Using machine learning models to train and learn on the basis of big data to predict and optimize the drug use plan will greatly improve the scientificity and accuracy of doctors' decisions. This technology can not only assist clinicians in formulating more effective drug use plans, but also significantly improve the treatment effect, reduce the economic burden of patients, and improve the overall quality of reproductive medical services. Therefore, by introducing machine learning models to predict and optimize the superovulation drug use plan, a new data-driven decision-making tool will be provided for doctors, enhancing the individuation and precision of treatment, thereby improving the success rate of assisted reproductive technology and patient satisfaction.

[0005] Currently, the prediction of superovulation drug use plans is mainly based on single-label multi-class machine learning methods or deep learning methods. For single-label classification, current machine learning or deep learning methods are relatively mature; however, in actual clinical practice, it is not appropriate to use single-label prediction for superovulation drug use plans; there is no authoritative gold standard for determining the superovulation drug use plan in assisted reproduction, so doctors' formulation of superovulation drug use plans for patients is often subjective. If the actual drug use plan name in the sample data is directly used as the sample label for model training, the problem of poor label consistency will inevitably be encountered. At the same time, similar patients may adopt different drug use plans, which leads to high heterogeneity of sample labels. This situation makes the data very challenging for any model, and directly training the model usually cannot achieve ideal prediction results. Summary of the Invention

[0006] To solve at least one of the technical problems existing in the above-mentioned background art, the present invention provides a prediction method, system and product for superovulation drug use plan oriented to success rate, which converts classification labels into multi-objective regression labels, uses multi-objective regression labels for model training and prediction, avoids the problem of poor single-label consistency, and the model can predict multiple drug use plans for doctors to make decisions, which better meets the actual clinical needs.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The first aspect of the present invention provides a prediction method for superovulation drug use plan oriented to success rate, including the following steps:

[0009] Obtain the monitoring index data of the patient after the superovulation drug use plan treatment;

[0010] According to the obtained monitoring index data of the patient after the superovulation drug use plan treatment, form the success rate evaluation value of each drug use plan;

[0011] Convert the single classification label of the drug use plan into a multi-classification label, and then combine the success rate evaluation value of each drug use plan to weight the multi-classification label to obtain a multi-objective regression label;

[0012] Train the drug use plan prediction model based on the multi-objective regression label, and predict based on the trained drug use plan prediction model combined with the index data of the patient to be predicted, and convert the regression value of the prediction result to obtain the recommended result of the superovulation drug use plan.

[0013] Further, the success rate evaluation value of each drug use plan is calculated according to the 2PN evaluation value, the high-quality embryo evaluation value and the oocyte retrieval number evaluation value;

[0014] Or, the success rate evaluation value of each drug use plan is calculated according to the 2PN evaluation value, the available embryo evaluation value and the oocyte retrieval number evaluation value. Further, the conversion of the single classification label of the drug use plan into a multi-classification label, and then combining the success rate evaluation value of each drug use plan to weight the multi-classification label to obtain a multi-objective regression label includes:

[0015] Expand a single classification label into multiple target variables, and assign weights to each target variable so that each target variable reflects the success rate of the actual drug use plan; among them, set the target value of the drug use plan that has not been actually treated to 0.

[0016] Further, use one-hot encoding to convert the single classification label of the drug use plan into a multi-classification label.

[0017] Further, the method further includes, after obtaining the multi-objective regression labels, performing upsampling using the Synthetic Minority Over-sampling Technique (SMOTE) to obtain an upsampling result, and then, based on the upsampling result, using random under-sampling to reduce the majority-class samples.

[0018] Further, performing upsampling using the Synthetic Minority Over-sampling Technique (SMOTE) to obtain an upsampling result includes:

[0019] Dividing the monitoring index data after the patient's superovulation drug use plan treatment into input features X , classification labels y , multi-objective regression labels y-score ;

[0020] Combining the input features X with the multi-objective regression labels y-score , inputting the combined result to the position of the feature X , and during the upsampling process, only interpolating X at the position of the feature X , while y-score the part remains unchanged.

[0021] Further, when converting the regression value of the prediction result to obtain the superovulation drug use plan recommendation result, the obtained multi-objective regression values are sorted in descending order, and during the recommendation, truncated value comparison is performed according to the sorting to obtain the main superovulation drug use plan and the secondary superovulation drug use plan.

[0022] Further, the drug use plan prediction model uses a multi-output regressor MultiOutputRegressor as a wrapper, and the regression model uses XGBRegressor.

[0023] Further, key features are screened based on the feature data of the patient to be predicted. The screening basis is to screen out the features with a p-value < the set threshold through analysis of variance and p-value test, including: age, weight, BMI, basal estradiol hormone, basal luteinizing hormone, basal follicle-stimulating hormone, basal progesterone, basal prolactin, basal testosterone, basal thyroid-stimulating hormone, anti-Müllerian hormone, menstrual cycle, whether having polycystic ovary syndrome, whether having endometriosis, whether having pelvic tubal disease, whether having primary infertility, ovarian response, antral follicle count, and infertility duration.

[0024] The second aspect of the present invention provides a superovulation drug use plan prediction system for success rate, including:

[0025] A data acquisition module, which is used to acquire the monitoring index data after the patient's superovulation drug use plan treatment;

[0026] A medication plan evaluation module, which is used to form a success rate evaluation value for each medication plan according to the monitoring index data obtained after the patient's superovulation medication plan is treated;

[0027] A label conversion module, which is used to convert the single-category label of the medication plan into a multi-category label, and then assign weights to the multi-category label in combination with the success rate evaluation value of each medication plan to obtain a multi-objective regression label;

[0028] A plan prediction module, which is used to train a regression model based on the multi-objective regression label, and make a prediction based on the trained regression model in combination with the index data of the patient to be predicted, and convert the regression value of the prediction result to obtain a recommended result of the superovulation medication plan.

[0029] The third aspect of the present invention provides a program product.

[0030] A program product, which is a computer program product, includes a computer program, and when the computer program is executed by a processor, it implements the steps in the above-mentioned superovulation medication plan prediction method oriented to success rate.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1. Aiming at the problem that the prior art will encounter poor consistency of medication plan labels when dealing with superovulation medication plan prediction modeling, even with the support of big data, the model is difficult to meet clinical needs. The present invention converts the single-label classification problem into a multi-objective regression problem through sample processing, and then assigns weights to the multi-category label in combination with the success rate evaluation value of each medication plan to obtain a multi-objective regression label; converts the regression value of the prediction result to obtain a recommended result of the superovulation medication plan, realizes the evaluation of the medication plan before patient ovulation induction, and enables the superovulation medication plan prediction model to better meet clinical needs.

[0033] 2. Aiming at the problem that the distribution of medication plan sample data is usually unbalanced and affects the prediction performance, the present invention uses improved SMOTE for lossless upsampling during sample sampling to ensure the integrity of the multi-objective regression label during upsampling.

[0034] The advantages of the additional aspects of the present invention will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of the present invention. Description of the Drawings

[0035] The specification drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0036] Figure 1It is a flowchart of a prediction method for a superovulation drug use plan oriented to success rate provided by an embodiment of the present invention;

[0037] Figure 2 It is a schematic diagram of the conversion of multi-objective regression sample labels provided by an embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the SMOTE lossless sampling process provided by an embodiment of the present invention;

[0039] Figure 4 It is a schematic diagram of the conversion process of the model output label provided by an embodiment of the present invention;

[0040] Figure 5 It is a comparison chart of whether to sample and train the multi-classification results provided by an embodiment of the present invention; among them, (a) is the comparison chart of the multi-classification results of non-sampled training, and (b) is the comparison chart of the multi-classification results of sampled training;

[0041] Figure 6 A comparison chart of the recommended results of the key index model provided by an embodiment of the present invention; among them, (a) is the comparison chart of the recommended results of the egg retrieval number index, (b) is the comparison chart of the recommended results of the 2PN number, (c) is the comparison chart of the recommended results of the available embryo number, and (d) is the comparison chart of the recommended results of the high-quality embryo number;

[0042] Figure 7 A comparison chart of the recommended results of the clinical outcome model provided by an embodiment of the present invention;

[0043] Figure 8 A comparison chart of the recommended results of the abortion outcome model provided by an embodiment of the present invention;

[0044] Figure 9 A comparison chart of the recommended results of the externally validated abortion outcome model provided by an embodiment of the present invention;

[0045] Figure 10 A comparison chart of the recommended results of the biochemical pregnancy abortion rate and the clinical pregnancy abortion rate of the externally validated clinical outcome model provided by an embodiment of the invention;

[0046] Figure 11 A comparison chart of the recommended results of the ovarian hyperstimulation incidence rate of the externally validated clinical outcome model provided by an embodiment of the invention. Detailed implementation manners

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed description is illustrative and is intended to provide further description of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0050] Aiming at the problems existing in the use of single-label prediction for the superovulation drug regimen mentioned in the background technology, the present invention adopts more complex and meticulous data preprocessing and feature engineering technologies to ensure that the model can accurately capture the key patterns and relationships in the data, and proposes a prediction modeling of the ovulation induction drug regimen oriented to the success rate. By using the comprehensive evaluation value of the treatment success rate of the ovulation induction drug regimen to weight the multi-objective regression label, it mainly includes the following two aspects:

[0051] First, through sample processing, the classification label is converted into a multi-objective regression label, and an improved SMOTE is used for lossless upsampling during sample sampling to ensure the integrity of the multi-objective regression label during upsampling;

[0052] Second, using the multi-objective regression label for model training and prediction can avoid the problem of poor single-label consistency, and the model can predict multiple drug regimens for doctors to make decisions, which better meets the actual clinical needs.

[0053] Embodiment 1

[0054] This embodiment provides a method for predicting a superovulation drug regimen oriented to the success rate, including the following steps:

[0055] Step 1: Obtain the monitoring index data after the patient's superovulation drug regimen treatment;

[0056] After the patient's drug regimen treatment, there will be corresponding monitoring data records, such as indicators like 2PN rate, good embryo rate, number of retrieved oocytes, etc. These information can objectively evaluate the success of the drug treatment outcome.

[0057] In this embodiment, after the patient's drug regimen treatment, there will be corresponding monitoring data records, and the monitoring indicators include indicators such as 2PN rate, good embryo rate, number of retrieved oocytes, number of antral follicles, number of good embryos, and number of 2PN embryos.

[0058] Step 2: According to the obtained monitoring index data after the patient's superovulation drug regimen treatment, form the success rate evaluation value of each drug regimen;

[0059] In this embodiment, the success rate evaluation value of each drug regimen is calculated based on the 2PN evaluation value, good embryo evaluation value, and number of retrieved oocytes evaluation value;

[0060] Among them, the success rate scoring formula for 2PN is:

[0061] effect_score_a = Two_pn_rate, where Two_pn_rate is the 2PN rate;

[0062] The excellent embryo scoring formula is:

[0063] effect_score_b = Yp_rate × yp_num_rate,

[0064] Let yp_num_rate = 0,

[0065] If yp_num > 11, then yp_num = 11;

[0066] If yp_num > 0, then yp_num_rate = yp_num / 11;

[0067] Among them, Yp_rate is the excellent embryo rate, yp_num is the number of excellent embryos, and yp_num_rate is the excellent embryo number ratio;

[0068] The oocyte retrieval number scoring formula is:

[0069] effect_score_c = (qulgs / (qulgs + BFN)) × qulgs_advantage,

[0070] If qulgs > 100, then qulgs = 100,

[0071] If qulgs <= 15, then qulgs_advantage = qulgs / 15,

[0072] Otherwise, qulgs_advantage = (100 - qulgs) / (100 - 15),

[0073] Among them, qulgs is the number of oocytes retrieved, and BFN is the number of antral follicles.

[0074] The formula for calculating the comprehensive score value is:

[0075] yp_effect_score = effect_score_a × k[0] + effect_score_b × k[1] + effect_score_c × k[2]. In this embodiment, k = [0.3, 1.5, 0.5]. In this embodiment, the success rate evaluation value of each medication regimen can also be calculated based on the 2PN evaluation value, the available embryo evaluation value, and the oocyte retrieval number evaluation value;

[0076] Among them, the success rate scoring formula for 2PN is:

[0077] effect_score_a = Two_pn_rate, where Two_pn_rate is the 2PN rate;

[0078] The scoring formula for available embryos is:

[0079] usable_effect_score_b = Usable_rate × usable_num_rate,

[0080] Let usable_num _rate = 0,

[0081] If usable_num > 11, then usable_num = 11;

[0082] If usable_num > 0, then usable_num_rate = usable_num / 11;

[0083] Among them, Usable_rate is the available embryo rate, usable_num is the number of available embryos, and usable_num_rate is the ratio of the number of available embryos;

[0084] The scoring formula for the number of retrieved eggs is:

[0085] effect_score_c = (qulgs / (qulgs + BFN)) × qulgs_advantage,

[0086] If qulgs > 100, then qulgs = 100,

[0087] If qulgs <= 15, then qulgs_advantage = qulgs / 15,

[0088] Otherwise qulgs_advantage = (100 - qulgs) / (100 - 15),

[0089] Among them, qulgs is the number of retrieved eggs, and BFN is the number of antral follicles.

[0090] The calculation formula for the comprehensive score value is:

[0091] usable_effect_score = effect_score_a × k[0] + usable_effect_score_b × k[1] + effect_score_c × k[2]. In this embodiment, k = [0.3, 1.5, 0.5].

[0092] The success rate scoring formula yp_effect_score or usable_effect_score in this embodiment, both of these scoring formulas are applicable to model training; mainly considering indicators such as the 2PN rate of patients, the excellent embryo rate or the available embryo rate, and the number of retrieved eggs, the scoring result tends to be that the larger these indicator values are, the larger the scoring value is. However, when calculating the scoring value of the retrieved eggs, according to relevant literature, the number of retrieved eggs is not the larger the better. The optimal egg retrieval coefficient is used as a constraint. In this embodiment, the optimal egg retrieval value is set to 15;

[0093] Step 3: Convert the single-category label of the medication plan list in the original sample into a multi-category label, and then after weighting the multi-category label in combination with the success rate evaluation value, obtain the multi-objective regression label.

[0094] As Figure 2 shown, this processing method expands the single classification label into multiple target variables and assigns weights to these target variables, so that each target variable can reflect the success rate of the actual medication plan. The target variable with a value of 0 represents the success rate of other medication plans. Since there are only clinical actual labels in the sample data, setting the target value of the un-actually-treated medication plan to 0 is a simple processing method. The feasibility of this method is based on the following points:

[0095] 1. Make full use of existing labels: In actual clinical practice, each sample has only one actual medication plan label. By converting this label into multiple target variables and assigning corresponding success rate scores to the actual treatment plan, the existing label information can be fully utilized.

[0096] 2. Simplify the processing: For other un-actually-used medication plans, set the target value to 0, which simplifies the data processing process. This processing method avoids complex assumptions and inferences while maintaining the integrity of the data.

[0097] 3. Multi-objective regression model: This method allows the use of a multi-objective regression model, enabling the model to simultaneously predict the success rate scores of multiple medication plans. In this way, multiple medication plan options can be provided for each patient and sorted according to the scores.

[0098] 4. Reflect the actual success rate: By assigning weights, each target variable can reflect the success rate of the actual medication plan, thereby improving the prediction accuracy and practicality of the model.

[0099] 5. Flexibility and scalability: This processing method has high flexibility, can adapt to different clinical scenarios and requirements, and can be continuously optimized as more data accumulates and the model is improved.

[0100] The distribution of sample data of medication regimens is usually unbalanced. Statistical findings show that the number of samples for short regimens or ultra-long regimens only accounts for 1 / 10 of those for long regimens and antagonist regimens. This uneven class distribution can cause the algorithm model to tend to predict the majority class because this can maximize the overall accuracy rate, but this will result in lower recall rates and precision rates for the minority class. Due to the scarcity of minority class samples, the model may have difficulty learning the features of the minority class during the training process, thereby affecting the prediction performance.

[0101] For the processing of this unevenly distributed data, mixed sampling is performed at the data level. First, the Synthetic Minority Over-sampling Technique (SMOTE) is used for oversampling, and then random undersampling is used to reduce the number of majority class samples. SMOTE is a commonly used data augmentation technique for dealing with minority class samples in imbalanced datasets. It increases the number of minority class samples by generating synthetic samples. The main steps are as follows:

[0102] Select a minority class sample: Randomly select a sample from the minority class sample set.

[0103] Determine k the k number of nearest neighbors: Find the

[0104] number of nearest neighbor minority class samples for this sample. k Randomly select a neighbor: Randomly select a neighbor from the

[0105] number of nearest neighbor samples.

[0106] SMOTE oversampling mainly targets classification datasets, that is, the labels (Labels) in the dataset are 1-dimensional class labels. However, in multi-objective regression tasks, the labels need to be converted into multi-dimensional vectors. Therefore, during the oversampling process, the labels need to be losslessly converted to ensure the integrity and consistency of the data.

[0107] The specific steps are as follows:

[0108] Divide the dataset into input features X , classification labels y , and multi-objective regression labels y-score ;

[0109] Merge X horizontally with y-score . As shown in Figure 3 for the SMOTE lossless sampling process, assume X there are 6 cases, y-scoreThere are 4 columns. After merging, the improved SMOTE algorithm is input (passing in X the feature positions features_to_interpolate), and interpolation is only performed at the positions of features_to_interpolate during the upsampling process, X while the y-score part remains unchanged.

[0110] The upsampled sample data is re-split into X features and y-score multi-objective regression labels, which are respectively used for the training and prediction of the subsequent medication plan prediction model.

[0111] Step 4: Train the medication plan prediction model based on the multi-objective regression labels, and predict based on the trained medication plan prediction model combined with the feature data of the patient to be predicted. Convert the regression value of the prediction result to obtain the recommended result of the superovulation medication plan;

[0112] After model prediction, label conversion is performed. By sorting the multi-objective regression values in descending order, during recommendation, cut-off value comparison is performed in order, and the main plan or secondary plan is recommended for the doctor to choose, as Figure 4 shown.

[0113] In this embodiment, the basic feature data of the patient before the superovulation medication plan treatment, such as: basic information, basic hormones, etc. Key indicators for prediction are obtained through preprocessing and screening of the basic features. The basis for screening is: through analysis of variance and p-value test, features with p-value < the set threshold are screened. In this embodiment, the set threshold is taken as 0.05.

[0114] Finally, 19 features are selected, as shown in Table 1:

[0115] Table 1 Basic feature data of the patient to be predicted

[0116]

[0117] In this embodiment, the medication plan prediction model uses the MultiOutputRegressor as a wrapper, and the regression model uses the XGBRegressor.

[0118] MultiOutputRegressor (multi-output regressor): A meta-estimator for handling multi-objective regression problems, which makes predictions by fitting a separate regression model for each objective.

[0119] XGBRegressor: A regression model based on gradient boosting, belonging to the XGBoost library. The XGBRegressor improves the prediction accuracy by gradually adding weak learners and is widely used in various regression tasks.

[0120] Through the above solution, the problem of poor consistency of medication plan labels encountered in the prior art when dealing with the prediction modeling of superovulation medication plans is solved. Even with the support of big data, the model is difficult to meet the clinical needs. The present invention converts the single-label classification problem into a multi-objective regression problem through sample processing, enabling the superovulation medication plan prediction model to better meet the clinical needs.

[0121] Experimental verification

[0122] By sampling to balance the sample data, the model can also predict minority-class samples during training without making the prediction results severely biased towards majority-class samples; as Figure 5 shown, Figure 5 in (a) is the multi-classification result without sampling training, Figure 5 in (b) is the comparison chart of the multi-classification results after sampling training; it can be seen from the actual experimental results that for the minority classes: long protocol (long), super-long protocol (superlong), the prediction results after sampling have improved. It is verified that for the problem that the distribution of the sample data of the medication plan is usually unbalanced and affects the prediction performance, the present invention uses improved SMOTE for lossless upsampling during sample sampling to ensure the integrity of the multi-objective regression labels during upsampling.

[0123] Internal model verification:

[0124] 6115 patients from 2020 to 2023 were used as the test set, and the model was used to retrospectively analyze their medication plans.

[0125] The results show that:

[0126] 1. Analysis of ovulation induction outcomes;

[0127] As Figure 6 shown, the key indicators of the recommended plan using this model are respectively shown in Figure 6 in (a)- Figure 6 in (d), including the number of retrieved oocytes, the number of 2PNs, the number of available embryos, and the number of high-quality embryos, are all significantly higher than those of the applicable plan and the non-recommended plan (p < 0.001). And the number of retrieved oocytes, the number of 2PNs, and the number of high-quality embryos of the applicable plan are all higher than those of the non-recommended plan (p < 0.05),

[0128] 2. Analysis of clinical outcomes;

[0129] Figure 7 is the comparison chart of the recommended results of the clinical outcome model, asFigure 7 As shown, it can be seen that for patients using the recommended protocol, key indicators such as the HCG positive rate, clinical pregnancy rate, continued pregnancy rate, and live birth rate are significantly higher than those of the applicable protocol and the non-recommended protocol. (p < 0.05). The HCG positive rate, clinical pregnancy rate, continued pregnancy rate, live birth rate, etc. of the applicable protocol are also higher than those of the non-recommended protocol, although there is no statistical difference.

[0130] 3. Analysis of abortion outcomes;

[0131] Figure 8 It is a comparison chart of the recommended results of the abortion outcome model. As Figure 8 shown, the biochemical pregnancy abortion rate and clinical pregnancy abortion rate of patients using the recommended protocol are both lower than those of the applicable protocol and the non-recommended protocol.

[0132] External validation of the model:

[0133] The data of 9,255 patients in 2019 was used as external data for model validation.

[0134] The results show:

[0135] 1. Analysis of clinical outcomes;

[0136] Figure 9 It is a comparison chart of the recommended results of the external validation abortion outcome model. As Figure 9 shown, for patients using the recommended protocol, key indicators such as the HCG positive rate, clinical pregnancy rate, continued pregnancy rate, live birth rate, and cumulative live birth rate are significantly higher than those of the applicable protocol and the non-recommended protocol. (p < 0.001). The cumulative live birth rate of the applicable protocol is significantly higher than that of the non-recommended protocol (p = 0.04). Its clinical pregnancy rate, continued pregnancy rate, live birth rate, etc. are higher than those of the non-recommended protocol, although there is no statistical difference.

[0137] 2. Analysis of abortion outcomes;

[0138] Figure 10 It is a comparison chart of the recommended results of the biochemical pregnancy abortion rate and the clinical pregnancy abortion rate. The biochemical pregnancy abortion rate and clinical pregnancy abortion rate of patients using the recommended protocol are significantly lower than those of the non-recommended protocol (p < 0.05). Figure 11 It is a comparison chart of the recommended results of the incidence of ovarian hyperstimulation. There is no statistical difference in the incidence of ovarian hyperstimulation among the three.

[0139] Through the above experimental verification, the superovulation drug use protocol prediction model can better meet the clinical needs.

[0140] Example Two

[0141] This example provides a superovulation drug use protocol prediction system for success rate, including:

[0142] A data acquisition module for acquiring monitoring index data of a patient after treatment with a superovulation drug use plan;

[0143] A drug use plan evaluation module for forming a success rate evaluation value of each drug use plan according to the acquired monitoring index data of a patient after treatment with a superovulation drug use plan;

[0144] A label conversion module for converting a single-category label of a drug use plan into a multi-category label, and then obtaining a multi-objective regression label after weighting the multi-category label by combining the success rate evaluation value of each drug use plan;

[0145] A plan prediction module for training a regression model based on the multi-objective regression label, predicting based on the trained regression model combined with the index data of a patient to be predicted, and converting the regression value of the prediction result to obtain a recommended result of a superovulation drug use plan.

[0146] Embodiment III

[0147] This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in a superovulation drug use plan prediction method for success rate as described above are implemented.

[0148] Embodiment IV

[0149] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in a superovulation drug use plan prediction method for success rate as described above are implemented.

[0150] Embodiment V

[0151] This embodiment provides a program product, which is a computer program product, including a computer program. When the computer program is executed by a processor, the steps in a superovulation drug use plan prediction method for success rate as described above are implemented.

[0152] The above are only preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A prediction method for a superovulation drug use plan oriented to success rate, characterized in that, Including the following steps: Obtain the monitoring index data after the patient's superovulation drug treatment plan; According to the obtained monitoring index data after the patient's superovulation drug treatment plan, form the success rate evaluation value of each drug treatment plan; Convert the single-classification label of the drug treatment plan into a multi-classification label, and then combine the success rate evaluation value of each drug treatment plan to weight the multi-classification label to obtain a multi-objective regression label, including: converting a single classification label into multiple target variables and assigning weights to each target variable so that each target variable reflects the success rate of the actual drug treatment plan; where the target value of the drug treatment plan that has not been actually treated is set to 0; According to the multi-objective regression label, the success rate scores of multiple drug treatment plans can be predicted simultaneously, providing multiple drug treatment plan options for each patient to adapt to different clinical scenarios and requirements; Train the drug treatment plan prediction model based on the multi-objective regression label, and predict based on the trained drug treatment plan prediction model combined with the characteristic data of the patient to be predicted, and convert the regression value of the prediction result to obtain the superovulation drug treatment plan recommendation result; Among them, when converting the regression value of the prediction result to obtain the superovulation drug treatment plan recommendation result, sort the obtained multi-objective regression values in descending order, and perform cut-off value comparison in order during recommendation to obtain the main superovulation drug treatment plan and the secondary superovulation drug treatment plan; Among them, the success rate evaluation value of each drug treatment plan is calculated according to the 2PN evaluation value, the good embryo evaluation value, and the number of retrieved oocytes evaluation value; Or, the success rate evaluation value of each drug treatment plan is calculated according to the 2PN evaluation value, the available embryo evaluation value, and the number of retrieved oocytes evaluation value; Among them, the drug treatment plan prediction model uses the MultiOutputRegressor as a wrapper, and the regression model uses XGBRegressor.

2. The method for predicting a superovulation drug use plan oriented to success rate according to claim 1, wherein, The method further includes, after obtaining the multi-objective regression label, performing upsampling using the synthetic minority over-sampling technique to obtain an upsampling result, and then based on the upsampling result, using random under-sampling to reduce the majority class samples.

3. The method for predicting a superovulation drug use plan oriented to success rate according to claim 2, wherein, The performing upsampling using the synthetic minority over-sampling technique to obtain an upsampling result includes: The monitoring index data after the treatment of the patient's superovulation drug use plan are divided into input features X , classification labels y , multi-objective regression labels y-score ; Combine the input features X with the multi-objective regression labels y-score and input the combined result to the position of the feature X During the upsampling process, interpolation is only performed at the position of the feature X for X while y-score remains unchanged as it is 4. The method for predicting a superovulation drug use plan oriented to success rate according to claim 1, wherein, Screen out key features based on the characteristic data of the patient to be predicted. The screening basis is to screen out features with p-value < the set threshold through analysis of variance and p-value test, including: age, weight, BMI, basal estradiol hormone, basal luteinizing hormone, basal follicle-stimulating hormone, basal progesterone, basal prolactin, basal testosterone, basal thyroid-stimulating hormone, anti-Müllerian hormone, menstrual cycle, whether polycystic, whether endometriosis, whether pelvic tubal disease, whether primary infertility, ovarian response, antral follicle count, and infertility duration.

5. A prediction system for superovulation drug use regimens oriented towards success rate, characterized in that, Using the method for predicting a superovulation drug treatment plan for success rate according to any one of claims 1-4, including: A data acquisition module, which is used to obtain the monitoring index data after the patient's superovulation drug treatment plan; A drug treatment plan evaluation module, which is used to form the success rate evaluation value of each drug treatment plan according to the obtained monitoring index data after the patient's superovulation drug treatment plan; A label conversion module, which is used to convert the classification labels of the medication plan list into multi-classification labels, and then obtain multi-objective regression labels after weighting the multi-classification labels by combining the success rate evaluation values of each medication plan; A scheme prediction module, which is used to train a regression model based on the multi-objective regression labels, make predictions based on the trained regression model combined with the index data of the patient to be predicted, and convert the regression value of the prediction result to obtain a recommended result of the superovulation medication plan.

6. A program product, which is a computer program product and includes a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps in a success rate-oriented superovulation medication plan prediction method according to any one of claims 1-4.

Images