Machine learning-based auxiliary miscarriage prevention decision-making method for whole pregnancy period after frozen embryo transplantation

Abstract

The invention discloses a machine learning-based auxiliary miscarriage prevention decision-making method for a whole pregnancy period after frozen embryo transplantation. The method comprises the following steps: obtaining pregnancy clinical data to construct a feature data set, obtaining a feature data set after initial screening, training and testing an artificial neural network model by using the feature data set after initial screening, performing feature extraction on pregnancy clinical data to be decided by using the artificial neural network model after testing qualification, and determining the pregnancy clinical data to be decided. And according to the pregnancy prediction characteristics, predicting a pregnancy outcome used for assisting a fetus protection decision. An artificial neural network model of the auxiliary miscarriage prevention decision-making method can effectively capture embryo quality, endometrial receptivity and complex nonlinear interaction among beta-hCG through a multi-layer perceptron architecture, so that a clinician is assisted and guided to preferentially select single embryo transplantation, and in addition, the early warning ability for biochemical pregnancy and early abortion is improved, and the early warning ability is improved. A clinical intervention window can be advanced by 5-7 days, an invalid transplantation period is reduced, and the economic cost of a patient is saved in a single period.

Description

Technical Field

[0001] The present invention relates to a method for assisting in decision-making about fetal preservation, and in particular to a method for assisting in decision-making about fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning. Background Art

[0002] Assisted reproductive technology (ART) is the most important and effective means of treating infertility, and its clinical effectiveness has attracted much attention. The clinical pregnancy rate and live birth rate of ART in my country have reached 30% and 28%, respectively. Pregnancy outcomes include non-pregnancy, miscarriage, biochemical pregnancy, single birth, and multiple births. Pregnancy outcomes in infertile populations are affected by multiple factors, including female age, ovarian function, endometrial receptivity, basal hormone levels, male semen quality, and embryonic developmental potential. The repeated transplantation process not only wastes reproductive resources but also brings heavy health impacts and financial pressures to patients' families. Decision-making models based on clinical experience are susceptible to cognitive biases and validity limitations, which may weaken the objectivity and reproducibility of clinical decisions.

[0003] The financial and emotional pressures of long-term infertility and repeated pregnancy failures lead up to 60% of patients to abandon recommended treatments before achieving clinical pregnancy. Against this backdrop, scientifically and accurately predicting pregnancy outcomes, assisting clinical decision-making, and optimizing ART treatment plans have become a key focus in reproductive medicine. Notably, with the rapid development and continuous advancement of artificial intelligence (AI) technology, machine learning is increasingly being applied in clinical research. Artificial neural networks (ANNs), a subfield of machine learning, are information processing systems designed to mimic the structure and function of physiological neural networks. They are based on algorithms modeled after the brain's architecture. ANNs excel in processing large amounts of information simultaneously, primarily by repeatedly presenting a series of ordered patterns, resulting in clearer and more concise results. In medical statistics, ANNs also offer unique advantages over regression equations in nonlinear fitting. ANNs' exceptional ability to recognize complex data patterns and iterate has provided new technological avenues for medical research, and their application in clinical practice is growing. Despite the rapid development of AI in ART, pregnancy outcome prediction for infertile patients still relies on traditional statistical methods, requiring further advancement in intelligence. Currently, research in China remains in its exploratory stages, with no mature models currently in clinical application. Summary of the Invention

[0004] The purpose of the invention is to provide a method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning, using ANN technology to construct an artificial neural network model, and by analyzing the full-process factors of frozen embryo transfer, using accurate predictive analysis to select fetal preservation decisions during the entire pregnancy cycle of patients after frozen embryo transfer, thereby helping to optimize clinical treatment plans and improve the success rate of patient treatment.

[0005] Technical solution: The machine learning-based method for assisting pregnancy preservation decision-making throughout the entire pregnancy cycle after frozen embryo transfer described in the present invention includes the following steps:

[0006] Step 1: Obtain pregnancy clinical data according to the data inclusion criteria and construct a feature dataset using the pregnancy clinical data;

[0007] Step 2, performing an initial screening of the pregnancy clinical data in the feature data set to obtain a feature data set after the initial screening;

[0008] Step 3: constructing an artificial neural network model for extracting pregnancy characteristics after frozen embryo transfer, and using the feature dataset after initial screening to train and test the artificial neural network model;

[0009] Step 4: Obtain the clinical data of the pregnancy to be decided, and use the artificial neural network model that has passed the test to extract features of the clinical data of the pregnancy to be decided to obtain pregnancy prediction features;

[0010] Step 5: Predict the pregnancy outcome based on the pregnancy prediction characteristics to assist in the decision-making of pregnancy preservation.

[0011] Furthermore, in step 1, the specific steps for obtaining pregnancy clinical data according to the data collection standards are as follows:

[0012] Step 1.1: Obtain retrospective cohort data of autologous frozen embryo transfer cycles as the original pregnancy data;

[0013] Step 1.2: The data inclusion criteria were set as all ART cycles that completed embryo transfer and were followed up to β-hCG testing;

[0014] Step 1.3: retain the original pregnancy data that meet the data inclusion criteria as the clinical pregnancy data.

[0015] Furthermore, in step 2, the specific steps for initial screening of the pregnancy clinical data in the feature dataset are as follows:

[0016] Step 2.1: Data exclusion criteria were set as data on cycle loss, data on data with no blood β-HCG test results after assisted pregnancy, data on data lost to follow-up, and / or data on pregnancies that did not reach the end of delivery;

[0017] Step 2.2: perform an initial exclusion screening on the pregnancy clinical data in the feature data set using the data exclusion criteria to obtain the feature data set after the initial screening.

[0018] Furthermore, in step 3, the constructed artificial neural network model includes a batch normalization layer, a first fully connected layer, an activation function layer, a second fully connected layer, and a classification output layer connected in sequence;

[0019] The batch normalization layer is used to normalize the input pregnancy clinical data; the first fully connected layer and the second fully connected layer are used to adjust the weights and biases; the activation function layer is used to implement nonlinear fitting; and the classification output layer is used to convert the output of the second fully connected layer into a probability distribution.

[0020] Furthermore, in step 3, the batch normalization layer process is expressed as:

[0021]

[0022] In the formula, x represents the input value; x norm Represents the normalized value of the input value x, scaled within a specific range; x mean and They represent the mean and variance of x in the dataset, and ∈ is a numerically stable protection quantity. These values ​​are used to scale x to a standard range and are usually in the interval [0,1].

[0023] Furthermore, in step 3, the classification output layer includes the Softmax function of multiple classifications connected in sequence, which is expressed as:

[0024]

[0025] Where y i represents the i-th class, P(y i ) represents the predicted probability of the i-th class, that is, the input belongs to the i-th class y i The possibility of z i represents the i-th class y i The original value of , that is, the original output score of the model before activation or expansion, For conversion z i is a positive value, and a larger value indicates a greater impact on the output probability. It is the sum of the exponentials of the original values, normalizing the probabilities of all classes to ensure that the sum of the predicted probabilities is equal to 1.

[0026] Furthermore, in step 4, the pregnancy prediction characteristics obtained are patient factor characteristics, embryo factor characteristics, endometrial factor characteristics, laboratory test index characteristics, or treatment plan factor characteristics;

[0027] Patient factors include infertility factors or primary / secondary infertility; embryo factors include embryo morphology or the number of transferred embryos; endometrial factors include endometrial morphology on the day of transfer or on the day of transformation; laboratory test index characteristics include serum β-hCG; and treatment regimen factors include the treatment regimen of the current cycle.

[0028] Furthermore, infertility factors include multiple factors on the male side, multiple factors on the female side, factors on both sides, and unknown causes; embryo morphology includes the grading morphology of cleavage-stage embryos and the staging morphology of blastocysts; endometrial morphology includes type A endometrium morphology, type B endometrium morphology, and type C endometrium morphology; serum β-hCG is the β-HCG value on the 13th / 14th day after transplantation; the treatment options for the current cycle include ovulation stimulation cycle, ovulation stimulation cycle LE, ovulation stimulation cycle LE+HMG, ovulation stimulation cycle TMX, ovulation stimulation cycle TMX+HMG, hormone replacement cycle GnRH-a+HRT, hormone replacement cycle HRT, natural cycle, and natural cycle E+P.

[0029] Furthermore, multiple male factors include oligoasthenoteratozoospermia, azoospermia and other male factors; multiple female factors include female ovulation disorders, female pelvic and uterine factors and other female factors; cleavage stage embryo morphology includes grade I, grade II, grade III and grade IV; grade I morphology is that the embryo blastomeres are equal in size, regular in shape, with uniform and clear cytoplasm and no fragments or less than 10% fragments; grade II morphology is that the embryo blastomeres are unequal in size, irregular in shape and with fragments between 10 and 25%; grade III morphology is that the embryo blastomeres are uneven in size, irregular in shape and with fragments between 25 and 50%; grade IV morphology is that the embryo blastomeres are severely uneven in size and with fragments greater than 50%; blastocyst stage morphology includes stage 1 morphology, stage 2 morphology, stage 3 morphology, stage 4 morphology, stage 5 morphology and and stage 6 morphology; stage 1 morphology is an early chambered blastocyst and the blastocyst cavity is less than 1 / 2 of the total volume of the embryo; stage 2 morphology is the blastocyst cavity volume is greater than or equal to 1 / 2 of the total volume of the embryo; stage 3 morphology is a fully expanded blastocyst and the blastocyst cavity completely occupies the total volume of the embryo; stage 4 morphology is an expanded blastocyst, the blastocyst cavity is completely filled with the embryo, the total volume of the embryo increases and the zona pellucida becomes thinner; stage 5 morphology is a hatching blastocyst and part of the blastocyst escapes from the zona pellucida; stage 6 morphology is a hatched blastocyst and the blastocyst completely escapes from the zona pellucida; type A endometrium morphology is three-line, with the outer layer and the center being strong echo lines, and the outer layer and the midline of the uterine cavity being a low echo area or dark area; type B endometrium morphology is uniform medium-intensity echo, and the strong echo midline of the uterine cavity is intermittent and unclear; type C endometrium morphology is homogeneous strong echo, and there is no uterine cavity midline echo.

[0030] Furthermore, in step 5, the pregnancy outcomes include non-pregnancy, early miscarriage, and ongoing pregnancy, and auxiliary pregnancy preservation decisions include preparing a new cycle plan, strengthening pregnancy preservation treatment, and routine pregnancy preservation.

[0031] Compared with the prior art, the present invention has the following advantages: (1) Compared with traditional linear models (such as logistic regression), the artificial neural network model of the present invention can effectively capture the complex nonlinear interactions between embryo quality, endometrial receptivity and β-hCG through the multi-layer perceptron architecture. In the independent validation set, the macro AUC of the artificial neural network model for pregnancy outcome classification reached 0.93 (95% accuracy). CI: 0.90–0.96), especially for live birth outcomes, the specificity is increased to 91.5%, reducing the risk of misjudgment; (2) The artificial neural network model of the present invention accurately predicts the probability of live birth, thereby assisting and guiding clinicians to give priority to single embryo transfer. In addition, the early warning ability of biochemical pregnancy (prediction AUC = 0.89) and early miscarriage (AUC = 0.77) can advance the clinical intervention window by 5-7 days, reduce ineffective transplant cycles, and save patients economic costs per cycle; (3) The artificial neural network model of the present invention breaks through the limitations of traditional single indicators and integrates 31 clinical features, including embryo morphological grade, β-hCG, endometrial morphology and infertility factors. The artificial neural network model automatically extracts key interactive features (such as the synergistic effect of β-hCG and embryo quality) to avoid information omission from artificial feature engineering; (4) The artificial neural network model of the present invention solves the complex nonlinear prediction problem in assisted reproduction, not only providing a high-precision decision-making tool for clinicians, but also significantly reducing the waste of medical resources and the burden on patients, providing an innovative solution for improving the fertility of the Chinese population. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the decision-making method of the present invention;

[0033] Figure 2 This is the artificial neural network model diagram of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the embodiments.

[0035] like Figure 1 As shown, the machine learning-based method for assisting pregnancy preservation decision-making during the entire pregnancy cycle after frozen embryo transfer disclosed in the present invention includes the following steps:

[0036] Step 1: Obtain pregnancy clinical data according to the data inclusion criteria and construct a feature dataset using the pregnancy clinical data;

[0037] Step 2, performing an initial screening of the pregnancy clinical data in the feature data set to obtain a feature data set after the initial screening;

[0038] Step 3: constructing an artificial neural network model for extracting pregnancy characteristics after frozen embryo transfer, and using the feature dataset after initial screening to train and test the artificial neural network model;

[0039] Step 4: Obtain the clinical data of the pregnancy to be decided, and use the artificial neural network model that has passed the test to extract features of the clinical data of the pregnancy to be decided to obtain pregnancy prediction features;

[0040] Step 5: Predict the pregnancy outcome based on the pregnancy prediction characteristics to assist in the decision-making of pregnancy preservation.

[0041] Furthermore, in step 1, the specific steps for obtaining pregnancy clinical data according to the data collection standards are as follows:

[0042] Step 1.1, obtaining retrospective cohort data of autologous frozen embryo transfer cycles as the original pregnancy data. In this embodiment, the original pregnancy data were screened from a retrospective cohort data of 4806 autologous frozen embryo transfer (FET) cycles performed between 2015 and 2024.

[0043] Step 1.2: The data inclusion criteria were set as all ART cycles that completed embryo transfer and were followed up to β-hCG testing;

[0044] Step 1.3: retain the original pregnancy data that meet the data inclusion criteria as the clinical pregnancy data.

[0045] Furthermore, in step 2, the specific steps for initial screening of the pregnancy clinical data in the feature dataset are as follows:

[0046] Step 2.1: Data exclusion criteria were set as data on cycle loss, data on data with no blood β-HCG test results after assisted pregnancy, data on data lost to follow-up, and / or data on pregnancies that did not reach the end of delivery;

[0047] Step 2.2: perform an initial exclusion screening on the pregnancy clinical data in the feature data set using the data exclusion criteria to obtain the feature data set after the initial screening.

[0048] Furthermore, in step 3, the constructed artificial neural network model includes a batch normalization layer, a first fully connected layer, an activation function layer, a second fully connected layer, and a classification output layer connected in sequence, such as Figure 2 As shown;

[0049] The batch normalization layer is used to normalize the input pregnancy clinical data, thereby enhancing the training of deep neural networks, improving learning stability and accelerating training speed;

[0050] The first and second fully connected layers are used to adjust the network weights and biases to learn the relationship between input features and classification outputs;

[0051] The activation function layer is used to provide nonlinear fitting capabilities and realize nonlinear fitting. The ReLU activation function is defined as f(x)=max(0,x);

[0052] The classification output layer is used to convert the output of the connection layer into a probability distribution, where the classification with the highest probability is selected as the final classification output.

[0053] Furthermore, a multi-layer perceptron architecture was constructed by the first fully connected layer, the activation function layer, and the second fully connected layer, which contains three layers of hidden units (256-128-64), and can effectively capture the complex nonlinear interactions between embryo quality, endometrial receptivity, and β-hCG.

[0054] Furthermore, in step 3, the batch normalization layer process is expressed as:

[0055]

[0056] In the formula, x represents the input value; x norm Represents the normalized value of the input value x, scaled within a specific range; x mean and They represent the mean and variation of x in the data set respectively, ∈ is a numerical stability protection quantity, and the value of ∈ is relatively small to prevent division by zero. These values ​​are used to scale x to the standard range and are usually in the interval [0,1].

[0057] Furthermore, in step 3, the classification output layer includes the Softmax function of multiple classifications connected in sequence, which is expressed as:

[0058]

[0059] Where y i represents the i-th class, P(y i ) represents the predicted probability of the i-th class, that is, the input belongs to the i-th class y i The possibility of z i represents the i-th class y i The original value of , that is, the original output score of the model before activation or expansion, For conversion z i is a positive value, and a larger value indicates a greater impact on the output probability. It is the sum of the exponentials of the original values, normalizing the probabilities of all classes to ensure that the sum of the predicted probabilities is equal to 1.

[0060] Furthermore, in step 3, the first fully connected layer and the second fully connected layer are both linear layers, and the linear layer is expressed as:

[0061] y=Wx+b

[0062] This is a weight matrix where x represents the input features and y represents the classification output. It defines the linear relationship between the input and output and is a bias vector that shifts the output, allowing for better model fit. W and b are both learnable parameters that are updated during training.

[0063] Furthermore, when training and testing the artificial neural network model, the feature data set is divided into a training set and a test set, with the ratio of the training set to the test set being 7:3; when testing the artificial neural network model, the pregnancy clinical data in the test set are used to continuously analyze and correct the artificial neural network model, thereby obtaining an ART pregnancy outcome prediction model for feature extraction of pregnancy clinical data.

[0064] Furthermore, after the artificial neural network model evaluation training and testing were completed, the artificial neural network model was evaluated using an independent evaluation data set. The evaluation criteria included accuracy and AUC. In the independent evaluation data set, the macro AUC of the artificial neural network model for pregnancy outcome classification reached 0.93 (95% CI: 0.90–0.96), especially the specificity for live birth outcomes was increased to 91.5%, reducing the risk of misjudgment.

[0065] Furthermore, in step 4, the pregnancy prediction characteristics obtained are patient factor characteristics, embryo factor characteristics, endometrial factor characteristics, laboratory test index characteristics, or treatment plan factor characteristics;

[0066] Patient factors include infertility factors or primary / secondary infertility; embryo factors include embryo morphology or the number of transferred embryos; endometrial factors include endometrial morphology on the day of transfer or on the day of transformation; laboratory test index characteristics include serum β-hCG; and treatment regimen factors include the treatment regimen of the current cycle.

[0067] By setting the pregnancy prediction characteristics as patient factor characteristics, embryo factor characteristics, endometrial factor characteristics, laboratory test indicator characteristics or treatment plan factor characteristics, a comprehensive assessment of pregnancy can be made to ensure the accuracy and reliability of the prediction.

[0068] Furthermore, infertility factors include multiple factors on the male side, multiple factors on the female side, factors on both sides, and unknown causes; embryo morphology includes the grading morphology of cleavage-stage embryos and the staging morphology of blastocysts; endometrial morphology includes type A endometrium morphology, type B endometrium morphology, and type C endometrium morphology; serum β-hCG is the β-HCG value on the 13th / 14th day after transplantation; the treatment options for the current cycle include ovulation stimulation cycle, ovulation stimulation cycle LE, ovulation stimulation cycle LE+HMG, ovulation stimulation cycle TMX, ovulation stimulation cycle TMX+HMG, hormone replacement cycle GnRH-a+HRT, hormone replacement cycle HRT, natural cycle, and natural cycle E+P.

[0069] By further limiting infertility factors, embryo morphology, endometrial morphology, serum β-hCG, and treatment regimen of the current cycle, the importance of related influencing features can be further clarified, ensuring the accuracy and reliability of the prediction.

[0070] Furthermore, multiple male factors include oligoasthenoteratozoospermia, azoospermia and other male factors; multiple female factors include female ovulation disorders, female pelvic and uterine factors and other female factors; cleavage stage embryo morphology includes grade I, grade II, grade III and grade IV; grade I morphology is that the embryo blastomeres are equal in size, regular in shape, with uniform and clear cytoplasm and no fragments or less than 10% fragments; grade II morphology is that the embryo blastomeres are unequal in size, irregular in shape and with fragments between 10 and 25%; grade III morphology is that the embryo blastomeres are uneven in size, irregular in shape and with fragments between 25 and 50%; grade IV morphology is that the embryo blastomeres are severely uneven in size and with fragments greater than 50%; blastocyst stage morphology includes stage 1 morphology, stage 2 morphology, stage 3 morphology, stage 4 morphology, stage 5 morphology and and stage 6 morphology; stage 1 morphology is an early chambered blastocyst and the blastocyst cavity is less than 1 / 2 of the total volume of the embryo; stage 2 morphology is the blastocyst cavity volume is greater than or equal to 1 / 2 of the total volume of the embryo; stage 3 morphology is a fully expanded blastocyst and the blastocyst cavity completely occupies the total volume of the embryo; stage 4 morphology is an expanded blastocyst, the blastocyst cavity is completely filled with the embryo, the total volume of the embryo increases and the zona pellucida becomes thinner; stage 5 morphology is a hatching blastocyst and part of the blastocyst escapes from the zona pellucida; stage 6 morphology is a hatched blastocyst and the blastocyst completely escapes from the zona pellucida; type A endometrium morphology is three-line, with the outer layer and the center being strong echo lines, and the outer layer and the midline of the uterine cavity being a low echo area or dark area; type B endometrium morphology is uniform medium-intensity echo, and the strong echo midline of the uterine cavity is intermittent and unclear; type C endometrium morphology is homogeneous strong echo, and there is no uterine cavity midline echo.

[0071] By further limiting the male's multiple factors, female's multiple factors, cleavage-stage embryo grade morphology, blastocyst stage morphology, type A endometrium morphology, type B endometrium morphology, and type C endometrium morphology, the importance of related influencing features can be further clarified to ensure the accuracy and reliability of the prediction.

[0072] Furthermore, in step 5, the pregnancy outcomes include non-pregnancy, early miscarriage, and ongoing pregnancy, and auxiliary pregnancy preservation decisions include preparing a new cycle plan, strengthening pregnancy preservation treatment, and routine pregnancy preservation.

[0073] Furthermore, when the pregnancy outcome is no pregnancy, an auxiliary decision can be made to cancel the embryo transfer and prepare a new cycle plan; when the pregnancy outcome is early miscarriage, an auxiliary decision can be made to strengthen the tocolytic treatment; when the pregnancy outcome is ongoing pregnancy, an auxiliary decision can be made to perform conventional tocolytic treatment.

[0074] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A machine learning-based method for assisting pregnancy preservation decision-making throughout the entire pregnancy cycle after frozen embryo transfer, characterized in that: The steps include: Step 1: Obtain pregnancy clinical data according to the data inclusion criteria and construct a feature dataset using the pregnancy clinical data; Step 2, performing an initial screening of the pregnancy clinical data in the feature data set to obtain a feature data set after the initial screening; Step 3: constructing an artificial neural network model for extracting pregnancy characteristics after frozen embryo transfer, and using the feature dataset after initial screening to train and test the artificial neural network model; Step 4: Obtain the clinical data of the pregnancy to be decided, and use the artificial neural network model that has passed the test to extract features of the clinical data of the pregnancy to be decided to obtain pregnancy prediction features; Step 5: Predict the pregnancy outcome based on the pregnancy prediction characteristics to assist in the decision-making of pregnancy preservation.

2. The machine learning-based method for assisting pregnancy preservation decision-making throughout the entire pregnancy cycle after frozen embryo transfer according to claim 1, characterized in that: In step 1, the specific steps for obtaining pregnancy clinical data according to the data collection standards are as follows: Step 1.1: Obtain retrospective cohort data of autologous frozen embryo transfer cycles as the original pregnancy data; Step 1.2: The data inclusion criteria were set as all ART cycles that completed embryo transfer and were followed up to β-hCG testing; Step 1.3: retain the original pregnancy data that meet the data inclusion criteria as the clinical pregnancy data.

3. The machine learning-based method for assisting pregnancy preservation decision-making throughout the entire pregnancy cycle after frozen embryo transfer according to claim 1, characterized in that: In step 2, the specific steps for initial screening of the pregnancy clinical data in the feature dataset are as follows: Step 2.1: Data exclusion criteria were set as data on cycle loss, data on data with no blood β-HCG test results after assisted pregnancy, data on data lost to follow-up, and / or data on pregnancies that did not reach the end of delivery; Step 2.2: perform an initial exclusion screening on the pregnancy clinical data in the feature data set using the data exclusion criteria to obtain the feature data set after the initial screening.

4. The method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning according to claim 1, characterized in that: In step 3, the constructed artificial neural network model includes a batch normalization layer, a first fully connected layer, an activation function layer, a second fully connected layer, and a classification output layer connected in sequence; The batch normalization layer is used to normalize the input pregnancy clinical data; the first fully connected layer and the second fully connected layer are used to adjust the weights and biases; the activation function layer is used to implement nonlinear fitting; and the classification output layer is used to convert the output of the second fully connected layer into a probability distribution.

5. The method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning according to claim 4, characterized in that: In step 3, the batch normalization layer process is expressed as: In the formula, x represents the input value; x norm Represents the normalized value of the input value x, scaled within a specific range; x mean and They represent the mean and variance of x in the dataset, and ∈ is a numerically stable protection quantity. These values ​​are used to scale x to a standard range and are usually in the interval [0,1].

6. The method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning according to claim 4, characterized in that: In step 3, the classification output layer includes the Softmax function of multiple classifications connected in sequence, which is expressed as: Where y i represents the i-th class, P(y i ) represents the predicted probability of the i-th class, that is, the input belongs to the i-th class y i The possibility of z i represents the i-th class y i The original value of , that is, the original output score of the model before activation or expansion, For conversion z i is a positive value, and a larger value indicates a greater impact on the output probability. It is the sum of the exponentials of the original values, normalizing the probabilities of all classes to ensure that the sum of the predicted probabilities is equal to 1.

7. The method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning according to claim 1, characterized in that: In step 4, the pregnancy prediction characteristics obtained are patient factor characteristics, embryo factor characteristics, endometrial factor characteristics, laboratory test index characteristics, or treatment plan factor characteristics; Patient factors include infertility factors or primary / secondary infertility; embryo factors include embryo morphology or the number of transferred embryos; endometrial factors include endometrial morphology on the day of transfer or on the day of transformation; laboratory test index characteristics include serum β-hCG; and treatment regimen factors include the treatment regimen of the current cycle.

8. The method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning according to claim 7, characterized in that: Infertility factors include multiple factors on the male side, multiple factors on the female side, factors on both sides, and unknown causes; embryo morphology includes the graded morphology of cleavage-stage embryos and the staged morphology of blastocysts; endometrial morphology includes type A endometrium morphology, type B endometrium morphology, and type C endometrium morphology; serum β-hCG is the β-HCG value on the 13th / 14th day after transplantation; treatment options for the current cycle include ovulation stimulation cycle, ovulation stimulation cycle LE, ovulation stimulation cycle LE+HMG, ovulation stimulation cycle TMX, ovulation stimulation cycle TMX+HMG, hormone replacement cycle GnRH-a+HRT, hormone replacement cycle HRT, natural cycle, and natural cycle E+P.

9. The method for assisting in decision-making on fetal preservation during the entire pregnancy cycle after frozen embryo transfer based on machine learning according to claim 8, characterized in that: Multiple male factors include oligoasthenoteratozoospermia, azoospermia, and other male factors; multiple female factors include ovulatory disorders, pelvic and uterine factors, and other female factors; cleavage-stage embryo morphology is graded into three categories: grade I, grade II, grade III, and grade IV; grade I embryos have uniformly sized blastomeres, regular morphology, uniform and clear cytoplasm, and no or less than 10% fragmentation. Grade II morphology refers to unequal embryonic blastomeres, irregular morphology, and fragmentation between 10 and 25%; Grade III morphology refers to unequal embryonic blastomeres, irregular morphology, and fragmentation between 25 and 50%; Grade IV morphology refers to severely uneven embryonic blastomeres and fragmentation greater than 50%; blastocyst stage morphology includes stage 1 morphology, stage 2 morphology, stage 3 morphology, stage 4 morphology, stage 5 morphology, and stage 6 morphology; stage 1 morphology refers to an early blastocyst with a cavity and the blastocyst cavity is less than 1 / 2 of the total volume of the embryo; stage 2 morphology refers to a blastocyst cavity volume greater than or equal to 1 / 2 of the total volume of the embryo; stage 3 morphology refers to a fully expanded blastocyst. The morphology of stage A is three-line, with the outer layer and the center being strong echo lines, and the area between the outer layer and the midline of the uterine cavity being a low echo area or dark area; the morphology of type B endometrium is uniform with medium intensity echo, and the midline of the strong echo in the uterine cavity is intermittent and unclear; the morphology of type C endometrium is homogeneous with strong echo, and there is no midline echo in the uterine cavity.

10. The machine learning-based method for assisting pregnancy preservation decision-making throughout the entire pregnancy cycle after frozen embryo transfer according to claim 1, characterized in that: In step 5, pregnancy outcomes include non-pregnancy, early miscarriage, and ongoing pregnancy, and auxiliary pregnancy preservation decisions include preparing a new cycle plan, strengthening pregnancy preservation treatment, and routine pregnancy preservation.

Images