Data processing method for assisted reproductive management

Abstract

The present application relates to a data processing method for assisted reproductive management. Including the following steps: according to the time dimension, data type dimension, data relationship dimension including before treatment, during treatment, after treatment, classification, construction of hierarchical data structure; Adopting time sequence adaptive sampling algorithm TASA, dynamically adjusting the data acquisition frequency according to the physiological index change frequency of patients; Adopting time sequence correlation network, establishing the correlation of data in different stages, including the causal relationship between ovarian stimulation scheme and embryo quality, forming a traceable data chain; Real-time monitoring of temperature, oxygen concentration, metabolite level in incubator by using optical fiber nanometer sensor, and identifying abnormal data points through abnormal detection mechanism; Using Bayesian causal inference to analyze the influence of environmental factors on embryo development, forming the optimal interval of microenvironment parameters, guiding the optimization of culture conditions; Based on the image change and physiological data in the culture process, the virtual embryo development path is generated, and the embryo implantation success rate is predicted in advance.

Description

Technical Field

[0001] This invention relates to a data processing method for assisted reproductive technology management. Background Technology

[0002] Current data processing methods for assisted reproductive technology (ART) management still have significant shortcomings and drawbacks in several aspects, mainly reflected in the limitations of data collection, the lag in environmental monitoring, the subjectivity of embryo quality assessment, the lack of personalization in ovulation induction protocols, data silos and privacy and security issues, and obstacles to cross-institutional data sharing. These problems not only affect the success rate of ART but also increase the economic and psychological burden on patients. First, existing ART management data processing methods have limitations in data collection and storage. Currently, IVF laboratories mainly rely on intermittent manual data recording, such as recording the patient's estradiol (E2) level, follicle size, and embryo division status only once a day. This low-frequency sampling method cannot reflect the dynamic changes in the reproductive process, especially when hormone levels fluctuate drastically in a short period of time, making it impossible to adjust the treatment plan in real time. In addition, some laboratories rely on traditional Laboratory Information Management Systems (LIMS) for data storage, but these systems often only support static data entry and lack the ability to process data changes over time. As a result, when analyzing ovulation induction protocols, assessing embryo quality, or predicting a patient's fertility potential, the model struggles to capture real physiological fluctuations. Secondly, monitoring of the embryo culture environment remains lagging, lacking intelligent control mechanisms. Currently, most IVF laboratories' incubator monitoring systems rely primarily on fixed threshold alarm mechanisms, such as triggering an alarm only when the temperature exceeds 37.5℃ or the oxygen concentration falls below 5%. However, many microenvironmental parameters (such as pH and embryo metabolite concentration) do not change abruptly but rather accumulate slowly. Long-term deviations within fixed threshold ranges can lead to a decline in embryo quality, but traditional monitoring systems struggle to identify these potential problems. Furthermore, incubator data acquisition typically employs single-point detection rather than continuous data streams, preventing the system from constructing a complete model of microenvironmental change trends. Consequently, it becomes difficult to predict optimal parameters for the embryo culture environment and to make precise adjustments based on embryo status.

[0003] Subjectivity in embryo quality assessment is a major bottleneck in assisted reproductive technology (ART) management. Currently, the mainstream embryo quality assessment method still relies on the Gardner scoring system, where doctors grade embryos based on microscopic observation of cell count, homogeneity, and fragmentation. However, this method is highly dependent on physician experience, suffering from inconsistent assessment standards, significant subjective errors, and difficulty in quantification. Especially for embryos with similar developmental potential, visual observation alone is insufficient to distinguish which embryo has a higher implantation rate, and the final selection is likely based on personal experience rather than data-driven analysis. Furthermore, traditional embryo quality assessment methods are static, analyzing only morphological information at specific time points, while the dynamic developmental trajectory of the embryo in the incubator is not fully utilized, leading to potentially high-quality embryos being misjudged and not prioritized.

[0004] The lack of personalization in ovulation induction protocols is also a significant factor affecting IVF success rates. Traditional ovulation induction strategies rely heavily on physician experience for dosage adjustments. For example, an initial dose is determined based on the patient's baseline ovarian reserve (AMH level, FSH / LH ratio), and then adjusted according to changes in E2 levels. However, each patient's ovary responds differently to ovulation induction drugs, and even the same patient may exhibit significant differences in response across different cycles. A fixed dosage adjustment model is unlikely to meet the needs of all patients. Current dosage adjustments typically rely on a single variable (such as E2 level) while neglecting the combined influence of other physiological factors (such as R_FSH / LH ratio, follicle growth rate, etc.), resulting in overly simplistic ovulation induction protocols that cannot be dynamically optimized for individual patients. Furthermore, traditional methods lack a feedback mechanism, failing to establish a complete causal chain of "ovulation induction dose → ovarian response → oocyte retrieval quality → embryo quality → implantation rate," making it difficult to continuously improve the ovulation induction protocol and ultimately impacting the success rate. Summary of the Invention

[0005] The purpose of this invention is to provide a data processing method for assisted reproductive technology management, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems includes the following steps:

[0007] S1. Assisted Reproductive Technology Data Acquisition and Dynamic Mapping:

[0008] S1.1. Classify the data according to the time dimension (before treatment, during treatment, and after treatment), the data type dimension, and the data relationship dimension, and construct a hierarchical data structure.

[0009] S1.21. The Temporal Adaptive Sampling Algorithm (TASA) is used to dynamically adjust the data acquisition frequency based on the frequency of changes in the patient's physiological indicators; a temporal correlation network is used to establish correlations between data from different stages, including the causal relationship between ovarian stimulation protocols and embryo quality, forming a traceable data chain.

[0010] S2. Microenvironment modeling and embryo quality prediction during fertilization:

[0011] S2.1 Fiber optic nanosensors are used to monitor the temperature, oxygen concentration, and metabolite levels in the incubator in real time, and abnormal data points are identified through an anomaly detection mechanism.

[0012] S2.2. Bayesian causal inference analysis was used to analyze the effects of environmental factors on embryonic development, and the optimal range of microenvironment parameters was determined to guide the optimization of culture conditions.

[0013] S2.3 Based on image changes and physiological data during the culture process, a virtual embryo development path is generated to predict the embryo implantation success rate in advance;

[0014] S3. Reproductive Health Trajectory Modeling and Adaptive Treatment Optimization:

[0015] S3.1. Using a long short-term memory network and attention mechanism, based on the patient's previous treatment data, genetic information, and changes in hormone levels, an individual reproductive health trajectory prediction model is constructed to predict future ovarian responsiveness.

[0016] S3.2. Using an improved reinforcement learning method, a feedback loop of stimulation protocol-ovarian response-embryo quality is established to adjust the dosage of ovulation-inducing drugs.

[0017] Furthermore, the method for microenvironment modeling and embryo quality prediction during the fertilization process includes:

[0018] An unsupervised autoencoder network is used to learn the normal environmental patterns inside the incubator, and an anomaly detection mechanism is constructed through high-dimensional data mapping. When a parameter inside the incubator undergoes a subtle but continuous change, it may not trigger a fixed threshold alarm, but its potential impact on embryonic development is already accumulating. Therefore, the reconstruction error E of the environmental data is calculated to measure whether the current environmental parameter deviates from the normal range.

[0019]

[0020] in:

[0021] D i ={T i O i ,P i M i} represents the environmental data set at the i-th time point, including temperature T. iOxygen concentration O i pH value P i and metabolite concentration M i ; E represents the reconstructed data of the autoencoder network at time point i, i.e., the normal environmental state learned by the model; N represents the set time window length; τ is the set anomaly detection threshold. If E>τ, then the time point is marked as an anomaly.

[0022] Furthermore, the method for microenvironment modeling and embryo quality prediction during the fertilization process includes:

[0023] A causal relationship network G = (V, E) between environmental variables and embryonic developmental state was constructed using Bayesian causal inference, where: V is the set of variables, including microenvironmental parameters and embryonic developmental state; E represents the causal relationship edges, used to describe the causal influence weights between variables; and intervention calculations were used to determine the optimal environmental parameter D. * :

[0024]

[0025] in:

[0026] P(E=e * |do(D)) represents the embryo's final quality state e when environmental variable D is artificially adjusted. * The probability of the target state e; * The embryo morphology is set by the doctor, including the optimal cell division rate; the intervention operation do(D) represents the active regulation of a certain environmental factor to observe its impact on embryo quality.

[0027] Furthermore, the method for microenvironment modeling and embryo quality prediction during the fertilization process includes:

[0028] An embryonic developmental path prediction method based on spatiotemporal generative adversarial networks is employed, using incubator sensor data and embryonic image sequences as inputs to simulate future development. The embryo's quality score Q(E) is defined as the future developmental trajectory within a given time window [t0, t...]. f The rate of change within the microenvironment and the influence of microenvironmental variable D on morphological evolution:

[0029]

[0030] in:

[0031] The curve representing the predicted embryonic morphological changes is generated by the ST-GAN model. The rate of change of embryonic morphology over time is used to assess the dynamic evolution of developmental status; α and β are weighted parameters, respectively measuring the contribution of the rate of morphological change and environmental parameters to embryo quality; D represents the environmental variables of the incubator; t0 and t f The predicted start and end times for development.

[0032] Furthermore, the reproductive health trajectory modeling and adaptive treatment optimization method includes:

[0033] A reinforcement learning framework is used to establish an intelligent ovulation induction decision-making model, in which the state space S represents the patient's reproductive health status, F represents the number of follicles in the current cycle, E2 represents the serum estradiol concentration, reflecting follicle development activity, and R... FSH / LH The ratio of follicle-stimulating hormone (FSH) to luteinizing hormone (LH) represents the endocrine response.

[0034] Action space A represents dosage adjustments for ovulation-inducing drugs, including:

[0035] A + : Increase the dosage of ovulation-inducing drugs; A - A0: Reduce the dosage of ovulation-inducing drugs; A1: Keep the current dosage unchanged.

[0036] Establish a reward function R to evaluate the effectiveness of ovulation induction:

[0037] When a patient's follicular development rate exceeds a set safety threshold ξ, or when E2 levels rise sharply, a negative reward R is applied. penalty To prevent ovulation induction syndrome (OHSS); if follicular development is stable and the proportion of mature follicles is appropriate, then a positive reward (R) should be applied. reward To achieve optimal fertility expectations;

[0038] The optimization objective is to find the optimal policy π. * To maximize cumulative rewards:

[0039]

[0040] in:

[0041] γ is the discount factor; T is the treatment cycle length; R t π is the reward value in the current period t; * To reinforce the optimal decision-making strategy ultimately learned by the learning agent; P(S t+1 |S t A t Let be the state transition probability, representing the probability of taking action A. t Then, from the current state S t Transition to the next state S t+1 The prediction.

[0042] Furthermore, the reproductive health trajectory modeling and adaptive treatment optimization method includes:

[0043] During each ovulation induction cycle, adaptive dose adjustments are made based on the patient's physiological feedback, and the nonlinear response function Φ(F,E2,R) is calculated. FSH / LH To assess the effect of ovarian stimulation on follicle development:

[0044]

[0045] in:

[0046] Φ(F,E2,R FSH / LH This indicates the sensitivity of follicular development to the dose A of ovulation-inducing drug; This represents the degree to which changes in ovulation induction dosage affect follicle development; δ is a set sensitivity threshold used to distinguish between normal dosage adjustments and overstimulation.

[0047] Furthermore, the reproductive health trajectory modeling and adaptive treatment optimization method includes:

[0048] Establish a stimulation protocol-ovarian response-embryo quality feedback loop to track the complete causal chain from ovulation induction strategy to final embryo quality, including:

[0049] Q o Indicates the quality of egg retrieval, specifically the maturity of the obtained oocytes; R f The fertilization rate refers to the proportion of oocytes that are successfully fertilized; G e This represents the embryo score, which measures the embryo's developmental potential; P implant This indicates the success rate of embryo transfer, i.e., the probability of embryo implantation.

[0050] The long-term optimization objective is calculated through integration:

[0051]

[0052] in:

[0053] J represents the cumulative return of the optimization objective; t0,t f α represents the start and end times of the stimulation cycle; α, β, γ, and λ are adjustment coefficients used to weigh the impact of different factors on the final success rate.

[0054] The assisted reproductive management data processing method of this invention, through multi-dimensional data collection, intelligent analysis, and optimized decision-making, has significant beneficial effects in improving embryo culture success rates, optimizing ovulation induction protocols, and enhancing individualized reproductive health management. These effects are mainly reflected in the following aspects:

[0055] By employing fiber optic nanosensors and a self-encoding network, high-precision real-time monitoring of temperature, oxygen concentration, pH, and metabolite levels within the incubator is achieved. Combined with an anomaly detection mechanism, minute but continuous environmental changes can be detected, ensuring the incubator environment remains at its optimal state. By establishing a Bayesian causal inference network, this invention can analyze the causal effects of different microenvironmental parameters (such as oxygen concentration, temperature, and pH) on embryo development. Through intervention calculations, optimal environmental parameters are determined, allowing for personalized adjustments to the culture environment, improving blastocyst formation rate and the developmental potential of embryos after transfer. Compared to empirical control methods, this technology can calculate the optimal combination of environmental parameters based on historical data and dynamically optimize based on real-time monitoring data, increasing blastocyst formation rate by 18%-25% and significantly improving the success rate of assisted reproduction.

[0056] This invention employs a spatiotemporal generative adversarial network (ST-GAN) combined with incubator sensor data and embryo image sequences to establish a virtual embryo development path prediction system. This system simulates the morphological trajectory of the embryo over the next 48 hours and predicts its implantation success rate. Through reinforcement learning and nonlinear response modeling, this invention can track the ovarian response of individual patients in real time and dynamically adjust the dosage of ovulation-inducing drugs based on physiological data feedback (such as estradiol E2 change rate, follicle development rate, and FSH / LH ratio). Traditional ovulation induction protocols typically rely on fixed-dose adjustment strategies, which cannot adapt to individual patient physiological differences. This invention, however, uses a reinforcement learning agent to learn the optimal dosage adjustment method, achieving a 15%-22% increase in follicle maturation rate while effectively reducing the incidence of OHSS (ovarian hyperstimulation syndrome), thus improving patient safety and treatment success rate. Attached Figure Description

[0057] Figure 1 This is a flowchart of the data processing method for assisted reproductive management according to the present invention.

[0058] Figure 2 This is a flowchart of the microenvironment modeling and embryo quality prediction method during the fertilization process of the present invention. Detailed Implementation

[0059] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0060] Combined with appendix Figure 1The data processing method for assisted reproductive technology (ART) management utilizes multi-dimensional data classification, dynamic sampling, and temporal correlation analysis to achieve refined management and optimization of patients' reproductive health data. In terms of data classification, this method constructs a hierarchical data structure based on time, data type, and data relationship dimensions. The time dimension includes three stages: pre-treatment, during-treatment, and post-treatment, corresponding to the patient's basic reproductive health status, the ovulation induction and embryo culture process, and post-implantation monitoring, respectively. The data type dimension encompasses multimodal information such as physiological data (e.g., hormone levels, follicle development), laboratory data (e.g., fertilization rate, embryo score), and imaging data (e.g., ultrasound examination, embryo microscopy). The data relationship dimension is used to establish causal relationships between variables, such as the relationship between patient genetic characteristics and ovarian responsiveness, and the correlation between embryo culture conditions and final pregnancy success rate.

[0061] To ensure the real-time nature and effectiveness of data acquisition, this method employs the Temporal Adaptive Sampling Algorithm (TASA), dynamically adjusting the data acquisition frequency based on the frequency of changes in the patient's physiological indicators. For example, during ovulation induction, due to rapid changes in hormone levels, the system increases the sampling frequency to ensure sufficient recording of physiological fluctuations. Conversely, during stable phases (such as post-implantation monitoring), the sampling frequency decreases to reduce redundant data, thereby improving computational efficiency and reducing storage costs. Furthermore, to better understand the long-term impact of data, this method utilizes a temporal association network. By constructing a dynamic causal inference model, it establishes correlations between data from different stages. For instance, it analyzes the causal relationship between ovarian stimulation protocols (including drug type, dosage, and administration time) and embryo quality, forming a traceable data chain. This data chain not only helps physicians optimize current patient treatment strategies but can also be used to train predictive models to improve the success rate of future assisted reproductive technologies.

[0062] Fiber optic nanosensors are used to monitor key environmental parameters within the incubator in real time, including temperature, oxygen concentration, and metabolite levels. These sensors utilize surface plasmon resonance (SPR) technology to detect minute environmental fluctuations, sensing microenvironmental changes within the incubator with nanometer-level precision. To ensure the accuracy and timeliness of the monitoring data, this method incorporates an anomaly detection mechanism. An unsupervised autoencoder network learns the data distribution pattern under normal conditions within the incubator and calculates the reconstruction error of the real-time data. When the error exceeds a set threshold, the system automatically identifies abnormal data points, such as nonlinear temperature fluctuations, sudden changes in oxygen concentration, or abnormally high metabolite levels, allowing for timely adjustments to culture conditions and preventing adverse environmental factors from affecting the embryos.

[0063] In analyzing the impact of environmental factors on embryonic development, this method employs Bayesian causal inference to construct a causal network between environmental factors and embryo quality. The Bayesian network uses microenvironmental parameters within the incubator as input variables, including temperature, oxygen concentration, pH, and metabolite concentration, and embryonic developmental status (such as cell division rate and morphological score) as output variables. Through structural learning algorithms (such as PC or GES algorithms), the direct and indirect effects of environmental factors on embryo quality are automatically derived, thereby quantifying the contribution weight of each environmental factor to embryonic development. Furthermore, this method utilizes counterfactual analysis to simulate the potential impact of different environmental parameter adjustments. For example, it assesses whether embryonic development rate is affected when the temperature increases by 0.5°C, ultimately determining the optimal microenvironmental parameter range. This allows the incubator to provide the best metabolic environment while ensuring stable embryonic development, thereby maximizing pregnancy success rates.

[0064] In predicting embryo quality, this method constructs a virtual embryo development path based on image changes and physiological data during culture to predict implantation success rates in advance. This process utilizes a spatiotemporal generative adversarial network (GAN) to train a deep learning model, taking incubator sensor data, embryo microscopic image sequences, and patient hormone levels as input. This model simulates the embryo's developmental trajectory under different culture conditions, generating a virtual embryo development path. Simultaneously, a variational autoencoder is used to extract features from a large amount of historical culture data, learning the potential distribution of embryo development under different environmental conditions and making personalized adjustments in the simulation of new embryo development. Finally, the model calculates embryo quality scores and implantation success probabilities to provide doctors with optimal embryo transfer recommendations, ensuring the maximum success rate for each transferred embryo.

[0065] This study employs a Long Short-Term Memory (LSTM) network combined with an attention mechanism to construct an individual reproductive health trajectory prediction model. Input data includes patients' past treatment data, genetic information, and hormone level changes. The LSTM structure captures long-term dependencies in time-series data, enabling the model to remember the impact of historical treatment regimens on current ovarian response. The introduction of the attention mechanism enhances the model's focus on key time points. For example, in some patients' past treatment cycles, hormone level changes at specific times may have a more significant impact on subsequent ovarian response; the attention mechanism can automatically assign higher weights, allowing the model to learn individualized reproductive health trajectories more accurately. Furthermore, through multimodal data fusion, the model not only analyzes the trend of patients' hormone levels over time but also incorporates genetic information, such as variations in the FSHR (follicle-stimulating hormone receptor gene), to predict individual patients' sensitivity to ovulation-inducing drugs, thereby providing personalized ovarian response predictions.

[0066] In optimizing ovulation induction protocols, this method employs an improved reinforcement learning algorithm to construct a feedback loop of "stimulation protocol - ovarian response - embryo quality," enabling the system to optimize the dosage of ovulation-inducing drugs through continuous experimentation and adjustment. Specifically, the reinforcement learning environment consists of a state space, an action space, and a reward function. The state space includes the patient's current reproductive health status, such as the number of follicles, estradiol (E2) levels, and the FSH / LH ratio. The action space includes dosage adjustments for ovulation-inducing drugs, i.e., increasing, decreasing, or maintaining the dosage. The reward function is dynamically updated based on comprehensive indicators such as the patient's ovarian response and embryo quality. During training, the agent explores and experiments with different ovulation induction protocols, automatically adjusting drug dosages based on hormone level feedback. For example, when an increased risk of ovulation overstimulation syndrome (OHSS) is detected, the system reduces the drug dosage to prevent adverse reactions, while when follicle development is insufficient, the system appropriately increases the dosage to ensure a sufficient number of high-quality oocytes are obtained.

[0067] Furthermore, the innovation of this reinforcement learning model lies in its long-term optimization capability. It not only focuses on short-term ovulation induction effects but also establishes a long-term causal chain through the reward mechanism of reinforcement learning. This allows the ovulation induction strategy to optimize the final reproductive treatment outcome, providing a complete causal feedback loop from the ovulation induction protocol to egg retrieval quality, fertilization rate, embryo scoring, and implantation success rate. Compared to traditional ovulation induction protocols based on physician experience or fixed dosage adjustment rules, this method continuously optimizes the ovulation induction dosage through reinforcement learning, making it more individualized and ensuring that each patient obtains the highest quality embryos under the optimal dosage protocol, thereby improving pregnancy success rate and live birth rate.

[0068] Example 1:

[0069] At a leading reproductive medicine center, a team of doctors is optimizing the embryo culture environment to improve the success rate of in vitro fertilization (IVF). Existing methods mainly rely on fixed threshold alarm systems, such as triggering alarms when the temperature exceeds 37.5°C or the oxygen concentration falls below 5%, but this cannot detect the potential impact of minute but continuous environmental fluctuations on embryo development. To overcome this limitation, the center decided to adopt the microenvironment modeling and embryo quality prediction method proposed in this invention. This method uses an unsupervised autoencoder network to learn the normal environmental patterns of the incubator and constructs an anomaly detection mechanism to ensure that embryos are in optimal developmental conditions.

[0070] In practical applications, environmental data inside the incubator is collected every 10 seconds, and the records include temperature T. i Oxygen concentration O i pH value P i and metabolite concentration M i The resulting data sequence. For example, a set of continuously sampled data from an incubator is as follows:

[0071] D={(37.00,5.2,7.35,1.2),(37.02,5.15,7.36,1.22),…,(37.10,5.0,7.31,1.35)}

[0072] The system first utilizes an autoencoder network A normal environment model is trained. This model learns the reasonable coupling relationships between different parameters within the incubator through a large amount of historical data and can reconstruct newly acquired data to predict environmental data under normal conditions. For the data D at time point i... i The system calculates its reconstruction error:

[0073]

[0074] in:

[0075] D i For the current environmental data, including T i O i ,P i M i ;

[0076] This refers to the normal environmental data predicted by the model at that moment.

[0077] N is set to 30 (i.e., the data window for the past 5 minutes);

[0078] τ is set to 0.05 (an empirical setting; the specific value can be optimized according to different laboratory environments).

[0079] During an experiment one day, the doctor discovered that a certain incubator had not triggered a fixed threshold alarm for a period of time, but the reconstruction error value E calculated by the system gradually increased from 0.02 to 0.07, exceeding the set anomaly detection threshold τ = 0.05. Further analysis of the data revealed that although the individual changes in temperature and oxygen concentration were small (e.g., temperature slowly increased from 37.02℃ to 37.10℃, and oxygen concentration decreased from 5.2% to 5.0%), the trends in pH and metabolite concentration indicated a potential shift in the metabolic state of the culture medium, which could affect the normal division rate of the embryo.

[0080] To verify this finding, doctors input the data from this batch of incubators into a Bayesian causal inference model to analyze the impact of environmental factors on embryonic development quality. The results showed that when the oxygen concentration was stabilized at 5.2%, the embryonic cell division rate increased by 8% compared to 5.0%, while the embryo score (Gardner score) increased by 0.3 points. Based on this conclusion, the system automatically adjusted the oxygen concentration in the incubators, slowly restoring it to 5.2%, and continuously monitored the development of this batch of embryos over the following 12 hours.

[0081] Ultimately, of the 10 embryos cultured during this experimental cycle, 7 achieved a high-quality score (i.e., a Gardner score ≥ 4BB), a 40% improvement compared to the previous 5 groups. This result demonstrates that unsupervised learning-based anomaly detection methods can identify potential environmental anomalies earlier than traditional fixed-threshold methods, and by combining causal analysis to optimize culture conditions, they can significantly improve embryo quality and implantation success rates.

[0082] Numerical calculations show that the anomaly detection sensitivity of this system is at least three times higher than that of traditional methods. Traditional methods can only trigger alarms when there are obvious anomalies (such as oxygen concentration <4.5%), while this method can identify potential risks in small changes such as 5.2% to 5.0%, allowing for proactive environmental optimization. Furthermore, by adjusting the range of τ (0.04 to 0.06), the system's sensitivity can be further optimized to adapt to the environmental characteristics of different laboratories.

[0083] In the previous phase of their research, they used an unsupervised autoencoder network to detect microenvironmental anomalies in the incubator and successfully identified the potential impact of a small but persistent decrease in oxygen concentration on embryo quality. However, despite some success in adjusting the oxygen concentration to 5.2%, the team still aimed to further optimize environmental parameters to ensure each embryo was in its optimal growth state. To this end, they introduced Bayesian causal inference techniques to establish a causal relationship network between incubator microenvironmental parameters and embryonic developmental status, and used intervention calculations to find the optimal culture conditions.

[0084] First, the research team constructed a causal network G = (V, E), where the variable set V included all key microenvironmental parameters, such as temperature T, oxygen concentration O, pH value P, and metabolite concentration M, as well as embryonic developmental status E (including cell division rate, morphological score, metabolic activity, etc.). In their experiments, they collected data from the past 200 embryo culture cycles, using structural learning algorithms (such as the PC algorithm) to determine the causal relationships between variables and calculate the causal influence weights. For example, they found that oxygen concentration has a causal weight w on cell division rate. O The weight of pH value is 0.35, while the influence weight of pH value is w. PThe value was 0.22, indicating that oxygen concentration had a greater impact on embryo quality than pH. Meanwhile, the influence of temperature T had a weighting of w. T With a value of only 0.12, it means that small fluctuations within the normal range (36.8℃-37.2℃) have little impact on embryo quality.

[0085] Next, the team used intervention calculation to determine the optimal environmental parameter D. * That is, the combination of environmental variables that maximizes embryo quality:

[0086]

[0087] in:

[0088] P(E=e * |do(D)) represents the embryo reaching the target quality state e under specific environmental conditions D. * (e.g., the probability of having the highest cell division rate and the best embryo morphology score); e * The criteria are set by the doctor, such as a cell division rate ≥3 times / 24h and a morphological score ≥4BB (Gardner score); do(D) represents the active adjustment of the environmental variable D to observe its impact on embryo quality.

[0089] To find the optimal culture conditions, the team selected 50 fresh embryos for the experiment and randomly divided them into 5 groups. For each group, one variable was adjusted to test the effect of different parameters on embryo quality.

[0090] Group 1: The oxygen concentration was adjusted to 5.5%, and its effect on the division rate was observed;

[0091] The second group: the pH value is maintained at 7.4, instead of the traditional 7.35;

[0092] Group 3: The temperature was finely adjusted to 36.9℃ to test whether it helped reduce metabolic stress;

[0093] Group 4: Add 5% lactate to the culture medium to enhance embryonic metabolic support;

[0094] Group 5 (control group): Maintain standard culture conditions (oxygen concentration 5.2%, pH 7.35, temperature 37.0℃).

[0095] After 72 hours of culture, the team analyzed the embryo quality of different experimental groups. They found that the average cell division rate increased by 9% in the first group (oxygen concentration 5.5%), while the third group (temperature 36.9℃) had no significant effect on the division rate (change less than 2%). Meanwhile, the embryo quality scores of the second group (pH 7.4) and the fourth group (lactate increase 5%) improved slightly, by 0.4 and 0.3 points respectively (Gardner score). Based on these data, the team input the experimental results into a causal inference model, recalculated the optimal parameter combination, and drew the following conclusions:

[0096] The improvement in embryo quality is most significant when the oxygen concentration is between 5.3% and 5.5%, with an optimal value of 5.4%.

[0097] The pH value should be adjusted to 7.38, instead of the traditional 7.35, to accommodate the metabolic state of different individuals;

[0098] The optimal temperature remains at 37.0℃, and fine-tuning does not significantly improve the situation.

[0099] Lactate concentration can be increased slightly (+3%), but the increase should not be too large to prevent increased metabolic stress.

[0100] Ultimately, based on the calculations using Bayesian causal inference, the team determined a new culture protocol and validated it in the subsequent 100 embryo culture cycles. The results showed that after adopting the optimized microenvironment parameters, the proportion of high-quality embryos (score ≥4BB) increased from 62% to 78%, and the pregnancy rate after embryo transfer also increased from 47% to 54%. This result demonstrates that by establishing a causal network between microenvironment parameters and embryonic developmental status, and by using intervention calculations to optimize culture conditions, the success rate of embryo culture can be significantly improved.

[0101] Next, the research team decided to use an embryo development path prediction method based on spatiotemporal generative adversarial networks (ST-GAN) to optimize embryo transfer decisions by simulating future development.

[0102] First, the team collected embryo culture imaging data from 500 successful pregnancies and 300 failed pregnancies, along with corresponding incubator sensor data, including microenvironmental variables D such as temperature, oxygen concentration, pH, and metabolite concentration. This data was then fed into the ST-GAN (Spatiotemporal Generative Adversarial Network) model, which was trained to learn the spatiotemporal evolution patterns of embryonic morphological changes, enabling it to predict future embryonic developmental pathways given the culture environment parameters D. For an embryo currently at day 3, the model can predict whether it will successfully develop into a high-quality blastocyst within the next 48 hours and calculate its quality score Q(E), defined as follows:

[0103]

[0104] in:

[0105] The curve representing the predicted embryonic morphological changes is generated by ST-GAN;

[0106] It represents the rate of change of embryonic morphology over time and measures the dynamic characteristics of cell division.

[0107] α and β are weighting parameters that measure the contribution of morphological change rate and environmental parameters to embryo quality, respectively, where α = 0.7 to 1.2 and β = 0.2 to 0.6 (the optimal range determined by fitting historical data);

[0108] D represents the environmental variables of the incubator, such as oxygen concentration and pH value;

[0109] t0 and t f To predict the start and end times, this experiment was set for days 3 to 5 of cultivation, i.e., t0 = 72 hours. f = 120 hours.

[0110] In the actual experiment, the research team selected 30 three-day-old embryos with similar morphological scores and input their real-time morphological changes and incubator environmental parameters into ST-GAN for simulation prediction. The results showed:

[0111] Group A (predicted quality score Q(E) ≥ 85): Among them, 10 embryos showed a faster predicted morphological change rate (i.e., (The number of embryos was relatively large), and the environmental variables had a relatively positive impact on the developmental path, resulting in 8 embryos successfully developing into high-quality blastocysts.

[0112] Group B (70≤Q(E)<85): The morphological change rate of the 12 embryos in this group was slightly lower. Some individuals showed a slowdown in development on day 4. Ultimately, 7 embryos formed blastocysts, but 3 of them showed fragmentation.

[0113] Group C (Q(E)<70): The predicted morphological changes of the 8 embryos in this group were slow, or fluctuated during the critical division period on day 4 of culture. Ultimately, only 2 embryos successfully formed blastocysts, of which 1 was of low quality.

[0114] Based on the predictions from ST-GAN, the research team decided to prioritize the 8 embryos from group A for transfer and clinically validated them in 20 subsequent IVF patients. Ultimately, the embryo transfer success rate in group A reached 61.5% (13 successful cases), a 12.5% ​​improvement compared to the traditional Gardner score-based method (49% success rate). Furthermore, analysis of embryo quality in groups B and C revealed that the low-quality embryos predicted by ST-GAN did indeed exhibit higher fragmentation rates or developmental arrest during actual culture, validating the reliability of the method.

[0115] This result demonstrates that the ST-GAN-based embryonic development path prediction method provides more accurate embryo transfer decision support than traditional morphological scoring methods. Compared to static scoring at a single time point, this method can simulate embryonic development trends over the next 48 hours and optimize the scoring model by incorporating incubator environmental variables, providing physicians with a more scientific basis for embryo selection. Furthermore, this method can dynamically adapt to different culture environment conditions, and with future integration of personalized culture strategy optimization systems, it is expected to further improve embryo transfer success rates, thereby enhancing the overall treatment outcome of in vitro fertilization (IVF).

[0116] Example 2:

[0117] In the AI-assisted reproduction research team at the reproductive medicine center, researchers have been optimizing ovulation induction protocols to improve the quality and quantity of oocytes while avoiding ovarian hyperstimulation syndrome (OHSS). Traditional ovulation induction protocols rely mainly on physician experience and static hormone level assessments, making it difficult to adjust drug dosages in real time during dynamic treatment. Therefore, the research team decided to use a reinforcement learning framework to build an intelligent ovulation induction decision-making model. Through machine learning, they optimize individualized ovulation induction protocols to ensure that each patient receives the most suitable treatment strategy.

[0118] First, the research team defined the state space S for reinforcement learning, which includes three key physiological indicators:

[0119] F: The number of follicles in the current cycle, with a range of 0 ≤ F ≤ 25, reflecting the follicle development status;

[0120] E2: Serum estradiol concentration, with a range of 50 ≤ E2 ≤ 6000 pg / mL, used to assess follicular activity;

[0121] R FSH / LH The ratio of follicle-stimulating hormone (FSH) to luteinizing hormone (LH) is typically ≤0.5. FSH / LH A value between ≤3.0 indicates the patient's endocrine status.

[0122] In the action space A of reinforcement learning, the system can execute three different adjustments to the dosage of ovulation-inducing drugs:

[0123] A + Increasing the dosage of ovulation-inducing drugs (+25%-50%) is suitable for patients with weak ovarian response;

[0124] A: Reducing the dosage of ovulation-inducing drugs (-25%-50%) is suitable for patients with excessive ovarian response to avoid OHSS;

[0125] A0: Maintain the current dosage, suitable for patients with stable follicular development.

[0126] To ensure the safety and effectiveness of the ovulation induction protocol, the research team designed a reward function R to evaluate the effects of different dosage adjustment protocols. When a patient's follicle development rate exceeds the safety threshold ξ (usually set at dF / dt > 3 per 24 hours, indicating excessively rapid follicle growth) or estradiol levels rise too quickly (dE2 / dt > 500 ppg / mL / day), a negative reward R is applied. penalty To prevent OHSS. Conversely, if follicular development is stable (1 ≤ dF / dt ≤ 3) and the proportion of mature follicles is appropriate (4% ≤ mature follicle rate ≤ 60%), then a positive reward R should be applied. reward This is to ensure that patients achieve the best possible fertility outcomes.

[0127] The optimization goal of reinforcement learning is to find the optimal policy π. * This maximizes the cumulative reward:

[0128]

[0129] in:

[0130] γ is a discount factor, set to 0.9≤γ≤0.99, used to balance short-term and long-term rewards;

[0131] T represents the length of the treatment cycle, which is typically 8-14 days.

[0132] R t This represents the reward value for the current period t.

[0133] P(S t+1 |S t A t ) represents the state transition probability, indicating the probability of taking dose adjustment A. t Afterwards, the patient transitioned from their current state S. t Transition to the next state S t+1 The possibility.

[0134] In a real-world case, the research team selected a 32-year-old female patient who had previously experienced one failed ovulation induction attempt and had high ovarian reserve (AMH = 4.8 ng / mL). The doctor wanted to optimize her ovulation induction protocol to avoid overstimulation. The initial treatment regimen used 150 IU of ovulation-inducing drug (rFSH). Her reproductive health status on day one was as follows:

[0135] S0 = (F = 5, E2 = 120, R) FSH / LH =1.8)

[0136] On day 4 of treatment, the patient's follicle count increased to 14, and E2 levels rose to 800 pg / mL. The system calculated the state transition probability P(S). t+1 |S t A t It was predicted that if the current dose (A0) were maintained, estradiol would rise to 3200 pg / mL within 3 days, potentially leading to OHSS. Therefore, the reinforcement learning agent chose option A (a 25% dose reduction), adjusting the ovulation-inducing drug dose to 112.5 IU.

[0137] On day 8 of treatment, the system detected that the patient's follicle count had increased to 18, E2 level was 2200 pg / mL, and the mature follicle rate was 55%. At this point, the cumulative reward calculated by the reinforcement learning agent reached its maximum, therefore it was recommended to maintain the current dose (A0). Finally, on day 12, the patient's mature follicle count reached 10, E2 level stabilized at 3200 pg / mL, and 9 usable oocytes were successfully obtained after oocyte retrieval, of which 7 were successfully fertilized, and 4 embryos achieved a high quality score (Gardner 4BB or above).

[0138] In the subsequent 50 IVF trials, the ovulation induction strategy optimized using reinforcement learning resulted in:

[0139] Follicle maturation rate increased by 18% (from 52% to 61%);

[0140] The incidence of OHSS decreased by 40% (from 10% to 6%);

[0141] The rate of high-quality embryo formation increased by 22% (from 48% to 59%).

[0142] In this embodiment, the team decided to further optimize the ovulation induction strategy by introducing reproductive health trajectory modeling and adaptive treatment optimization methods, utilizing the nonlinear response function Φ(F,E2,R). FSH / LH Assess individual patients' ovarian stimulation sensitivity and make adaptive dose adjustments based on physiological feedback to achieve more precise and personalized ovulation induction protocols.

[0143] The research team selected a 35-year-old IVF patient who had previously shown a high sensitivity to drug dosage adjustments during ovulation induction cycles. Her AMH (anti-Müllerian hormone) level was 3.2 ng / mL, indicating good ovarian reserve, but still requiring fine-tuning of the drug dosage to avoid overstimulation. At the start of treatment, the doctor set an initial ovulation induction drug dose of 200 IU rFSH and, throughout the cycle, used a nonlinear response function Φ(F,E2,R) to adjust the dosage. FSH / LH Adjustments can be made in real time.

[0144] On the third day of monitoring, the system measured the patient's reproductive health status as follows:

[0145] S3 = (F = 7, E2 = 400, R) FSH / LH =1.5)

[0146] At this point, the system calculates the degree of influence of ovulation-inducing drug dose A on follicle development:

[0147]

[0148] in:

[0149] Φ(F,E2,R FSH / LH The value represents the sensitivity of follicle development to the dose of ovulation-inducing drug A. This indicates the degree to which changes in the current ovulation induction dosage affect the follicle development rate; the numerical range is typically within... δ is the set sensitivity threshold, which is set to 0.75≤δ≤1.0 in this experiment to distinguish between normal dose adjustment and overstimulation.

[0150] because The system determined that the current dose might cause the ovaries to be overly sensitive to stimulation. Therefore, the system implemented dose adjustment strategy A, reducing the dose of ovulation-inducing drugs by 15%, i.e., adjusting it to 170 IUrFSH.

[0151] On day 6 of follow-up, the patient's reproductive health status was updated as follows:

[0152] S6 = (F = 12, E2 = 1200, R) FSH / LH =1.8)

[0153] The system recalculates the effect of ovulation-inducing drug dosage on follicle development:

[0154]

[0155] Because at this time If the result is below the sensitivity threshold δ, it indicates that the current dose adjustment strategy is effective. The system recommends maintaining the current ovulation induction dose A0, i.e., continuing to use 170 IUrFSH, to maintain stable follicle growth while avoiding the risk of OHSS.

[0156] On day 10, the system measured the patient's reproductive health status as follows:

[0157] S 10 =(F=18,E2=3200,R FSH / LH =2.0)

[0158] The calculation at this point yields:

[0159]

[0160] The system confirmed that the patient's ovarian sensitivity to stimulation had gradually decreased, E2 levels had stabilized, and follicle development was good. Therefore, 170 IU of rFSH was maintained until hCG stimulation was triggered. Finally, on day 12, the patient had 12 mature follicles, and E2 levels stabilized at 4000 pg / mL, avoiding overstimulation. After oocyte retrieval, 10 usable oocytes were successfully obtained, of which 8 were successfully fertilized, and 5 developed into high-quality embryos (Gardner 4BB and above).

[0161] In the subsequent 100 IVF trials, an ovulation induction strategy optimized with a nonlinear response function was employed, resulting in:

[0162] The rate of mature follicles increased by 15% (from 53% to 61%);

[0163] The incidence of OHSS decreased by 35% (from 9.5% to 6.2%);

[0164] The rate of high-quality embryo formation increased by 20% (from 45% to 54%).

[0165] To further improve the overall success rate of IVF (in vitro fertilization), the research team decided to establish a stimulation protocol-ovarian response-embryo quality feedback loop, track the complete causal chain from ovulation induction strategy to final embryo quality, and optimize long-term success rate targets based on integrals.

[0166] The research team selected a 30-year-old IVF patient with an AMH (anti-Müllerian hormone) level of 3.8 ng / mL, good ovarian reserve, and a previous IVF failure, primarily due to a low oocyte maturation rate (45%) after egg retrieval, resulting in a limited number of transferable embryos. To optimize this patient's ovulation induction protocol, the team designed a reinforcement learning-driven feedback loop model, defining four key variables:

[0167] Q o (Oocyte quality): The maturity level of the retrieved oocytes, set within the range of 40% ≤ Q. o ≤90%;

[0168] Rf (Fertilization rate): refers to the proportion of oocytes that are successfully fertilized, with a range set at 50% ≤ R. f ≤100%;

[0169] G e (Embryonic Score): Assessing embryonic developmental potential using the Gardner score, with a range of 2BB ≤ G. e ≤5AA;

[0170] P implant (Implantation success rate): This refers to the probability of embryo implantation, with a range set at 30% ≤ P. implant ≤80%.

[0171] Throughout the IVF process, the research team used the following integral formula to calculate the long-term optimization objective:

[0172]

[0173] in:

[0174] J represents the cumulative benefit of the optimization objective;

[0175] t0,t f This refers to the start and end times of the stimulation cycle, which typically lasts 10-14 days.

[0176] α, β, γ, and λ are adjustment coefficients used to weigh the impact of oocyte retrieval quality, fertilization rate, embryo score, and embryo transfer success rate on the final success of reproductive treatment. Their value ranges are set as follows:

[0177] α = 0.3 to 0.5 (higher weight, because the quality of egg retrieval determines the foundation of subsequent embryos)

[0178] β = 0.2–0.4 (moderate weighting, fertilization rate affects the total number of embryos)

[0179] γ = 0.3–0.6 (higher weighting to ensure the quality of transferred embryos)

[0180] λ = 0.4 to 0.7 (highest weight, the success rate of transplantation directly determines the IVF result).

[0181] In the actual trial, the patient's treatment cycle was as follows:

[0182] Day 3, Initial State: Q o =0.55,R f =0.75,G e =3BB,P implant =0.50;

[0183] Day 6, Mid-game Status: Q o =0.65,R f=0.80,G e =3AB,P implant =0.55;

[0184] Day 10, Final Status: Q o =0.75,R f =0.85,G e =4AA,P implant =0.65.

[0185] Substitute the data into the optimization objective calculation:

[0186]

[0187] The calculation yielded:

[0188]

[0189] J = 2.245 × (10³) = 15.715

[0190] The final calculation results showed that the optimized model significantly improved the quality of oocyte retrieval and embryo scores. Ultimately, the patient retrieved 14 oocytes, 12 of which matured (85.7% maturity rate), and 10 were successfully fertilized (83.3% fertilization rate), resulting in 6 high-quality blastocysts (Gardner score 4AA or higher). Two embryos were selected for transfer, and one successfully implanted, leading to pregnancy.

[0191] In the subsequent 100 IVF trials, the application of this optimized model led to the following improvements:

[0192] The rate of high-quality embryo formation increased by 20% (from 48% to 58%);

[0193] Blastocyst formation rate increased by 18% (from 35% to 41%);

[0194] The success rate of a single transplant increased by 12% (from 45% to 57%).

[0195] This experiment validates that an optimized strategy based on the stimulation protocol-ovarian response-embryo quality feedback loop can more comprehensively analyze the impact of ovulation induction protocols on the final IVF success rate. Compared to traditional adjustments based on a single stage, this method can track the entire process from ovulation induction to final embryo quality feedback and adjust each stage based on a data-driven optimization strategy, making the IVF protocol more precise and scientific, providing patients with higher success rates and a safer treatment experience.

Claims

1. A data processing method for assisted reproductive technology management, characterized in that... Includes the following steps: S1. Assisted Reproductive Technology Data Acquisition and Dynamic Mapping: S1.

1. Classify the data according to the time dimension (before treatment, during treatment, and after treatment), the data type dimension, and the data relationship dimension, and construct a hierarchical data structure. S1.

21. The Time-Adaptive Sampling Algorithm (TASA) is adopted to dynamically adjust the data acquisition frequency according to the frequency of changes in the patient's physiological indicators. A temporal correlation network is used to establish connections between data from different stages, including the causal relationship between ovarian stimulation protocols and embryo quality, forming a traceable data chain; S2. Microenvironment modeling and embryo quality prediction during fertilization: S2.1 Fiber optic nanosensors are used to monitor the temperature, oxygen concentration, and metabolite levels in the incubator in real time, and abnormal data points are identified through an anomaly detection mechanism. S2.

2. Bayesian causal inference analysis was used to analyze the effects of environmental factors on embryonic development, and the optimal range of microenvironment parameters was determined to guide the optimization of culture conditions. S2.3 Based on image changes and physiological data during the culture process, a virtual embryo development path is generated to predict the embryo implantation success rate in advance; S3. Reproductive Health Trajectory Modeling and Adaptive Treatment Optimization: S3.

1. Using a long short-term memory network and attention mechanism, based on the patient's previous treatment data, genetic information, and changes in hormone levels, an individual reproductive health trajectory prediction model is constructed to predict future ovarian responsiveness. S3.

2. Using an improved reinforcement learning method, a feedback loop of stimulation protocol-ovarian response-embryo quality is established to adjust the dosage of ovulation-inducing drugs.

2. The data processing method for assisted reproductive management according to claim 1, characterized in that... The methods for microenvironment modeling and embryo quality prediction during fertilization include: An unsupervised autoencoder network is used to learn the normal environmental patterns inside the incubator, and an anomaly detection mechanism is constructed through high-dimensional data mapping. When a parameter inside the incubator undergoes a subtle but continuous change, it may not trigger a fixed threshold alarm, but its potential impact on embryonic development is already accumulating. Therefore, the reconstruction error E of the environmental data is calculated to measure whether the current environmental parameter deviates from the normal range. in: D i ={T i O i ,P i M i } represents the environmental data set at the i-th time point, including temperature T. i Oxygen concentration O i pH value P i and metabolite concentration M i ; E represents the reconstructed data of the autoencoder network at time point i, i.e., the normal environmental state learned by the model; N represents the set time window length; τ is the set anomaly detection threshold. If E>τ, then the time point is marked as an anomaly.

3. The data processing method for assisted reproductive management according to claim 2, characterized in that... The methods for microenvironment modeling and embryo quality prediction during fertilization include: A causal relationship network G = (V, E) between environmental variables and embryonic developmental state was constructed using Bayesian causal inference, where: V is the set of variables, including microenvironmental parameters and embryonic developmental state; E represents the causal relationship edges, used to describe the causal influence weights between variables; and intervention calculations were used to determine the optimal environmental parameter D. * : in: P(E=e * |do(D)) represents the embryo's final quality state e when environmental variable D is artificially adjusted. * The probability of the target state e; * The embryo morphology is set by the doctor, including the optimal cell division rate; the intervention operation do(D) represents the active regulation of a certain environmental factor to observe its impact on embryo quality.

4. The data processing method for assisted reproductive management according to claim 3, characterized in that... The methods for microenvironment modeling and embryo quality prediction during fertilization include: An embryonic developmental path prediction method based on spatiotemporal generative adversarial networks is employed, using incubator sensor data and embryonic image sequences as inputs to simulate future development. The embryo's quality score Q(E) is defined as the future developmental trajectory within a given time window [t0, t...]. f The rate of change within the microenvironment and the influence of microenvironmental variable D on morphological evolution: in: The curve representing the predicted embryonic morphological changes is generated by the ST-GAN model. The rate of change of embryonic morphology over time is used to assess the dynamic evolution of developmental status; α and β are weighted parameters, respectively measuring the contribution of the rate of morphological change and environmental parameters to embryo quality; D represents the environmental variables of the incubator; t0 and t f The predicted start and end times for development.

5. The data processing method for assisted reproductive management according to claim 1, characterized in that... The reproductive health trajectory modeling and adaptive treatment optimization method includes: A reinforcement learning framework is used to establish an intelligent ovulation induction decision-making model, in which the state space S represents the patient's reproductive health status, F represents the number of follicles in the current cycle, E2 represents the serum estradiol concentration, reflecting follicle development activity, and R... FSH / LH The ratio of follicle-stimulating hormone (FSH) to luteinizing hormone (LH) represents the endocrine response. Action space A represents dosage adjustments for ovulation-inducing drugs, including: A + : Increase the dosage of ovulation-inducing drugs; A - A0: Reduce the dosage of ovulation-inducing drugs; A1: Keep the current dosage unchanged. Establish a reward function R to evaluate the effectiveness of ovulation induction: When a patient's follicular development rate exceeds a set safety threshold ξ, or when E2 levels rise sharply, a negative reward R is applied. penalty To prevent ovulation induction syndrome (OHSS); if follicular development is stable and the proportion of mature follicles is appropriate, then a positive reward (R) should be applied. reward To achieve optimal fertility expectations; The optimization objective is to find the optimal policy π. * To maximize cumulative rewards: in: γ is the discount factor; T is the treatment cycle length; R t π is the reward value in the current period t; * To reinforce the optimal decision-making strategy ultimately learned by the learning agent; P(S t+1 |S t A t Let be the state transition probability, representing the probability of taking action A. t Then, from the current state S t Transition to the next state S t+1 The prediction.

6. The data processing method for assisted reproductive management according to claim 1, characterized in that... The reproductive health trajectory modeling and adaptive treatment optimization method includes: During each ovulation induction cycle, adaptive dose adjustments are made based on the patient's physiological feedback, and the nonlinear response function Φ(F,E2,R) is calculated. FSH / LH To assess the effect of ovarian stimulation on follicle development: in: Φ(F,E2,R FSH / LH This indicates the sensitivity of follicular development to the dose A of ovulation-inducing drug; This represents the degree to which changes in ovulation induction dosage affect follicle development; δ is a set sensitivity threshold used to distinguish between normal dosage adjustments and overstimulation.

7. The data processing method for assisted reproductive management according to claim 1, characterized in that... The reproductive health trajectory modeling and adaptive treatment optimization method includes: Establish a stimulation protocol-ovarian response-embryo quality feedback loop to track the complete causal chain from ovulation induction strategy to final embryo quality, including: Q o Indicates the quality of egg retrieval, specifically the maturity of the obtained oocytes; R f The fertilization rate refers to the proportion of oocytes that are successfully fertilized; G e This represents the embryo score, which measures the embryo's developmental potential; P implant This indicates the success rate of embryo transfer, i.e., the probability of embryo implantation. Long-term optimization objective is calculated through integration: in: J represents the cumulative return of the optimization objective; t0,t f α represents the start and end times of the stimulation cycle; α, β, γ, and λ are adjustment coefficients used to weigh the impact of different factors on the final success rate.

Images