A method and system for analyzing assisted reproductive information

Abstract

The present invention relates to the field of assisted reproductive technology, and discloses an assisted reproductive information analysis method and system, which realizes the systematization and standardization of the decision-making process by constructing an assisted reproductive decision-making model based on the Markov decision process. The model defines a state space including patient genetic data and clinical data, as well as an action space such as adjusting drug dosage and selecting treatment plans, and designs a reward function based on pregnancy rate and live birth rate. The deep Q network algorithm is used for model training, and the scientificity and accuracy of decision-making are improved through iterative learning and parameter tuning. The model performance is evaluated by the A / B test method, and compared with the doctor's routine plan to verify the effectiveness of the model. Finally, based on the trained model, the patient's personal data is analyzed to provide doctors with personalized treatment plan recommendations.

Description

Technical Field

[0001] The present invention relates to the field of assisted reproductive technology, and specifically to an assisted reproductive information analysis method and system. Background Art

[0002] In the field of modern medicine, assisted reproductive technology (ART) has become an important means to solve the problem of infertility, bringing hope to many couples who are eager to have offspring. With the advancement of science and technology, ART technology itself has achieved remarkable achievements. However, in its application process, especially in the decision-making process, it still faces a series of challenges, which limit the full realization of the potential of technology.

[0003] At present, the decision-making process of assisted reproductive technology mainly relies on the personal experience and clinical guidelines of doctors. Although this decision-making model can guide treatment to a certain extent, its limitations cannot be ignored. Specifically, the experience and knowledge level of doctors vary from individual to individual, which directly leads to the diversity and uncertainty of treatment options. For example, for the same infertile couple, different doctors may propose completely different treatment plans based on their respective experiences, which not only affects the coherence and consistency of treatment, but also may reduce the success rate of treatment.

[0004] Moreover, doctors often face multiple challenges in the decision-making process, such as time pressure, incomplete information, and complex conditions. Within a limited time, doctors need to comprehensively consider multiple factors such as the patient's medical history, physical examination results, genetic information, etc. to develop the most suitable treatment plan for the patient. However, due to the complexity and multidimensionality of information, it is often difficult for doctors to fully and accurately grasp all relevant information, which to a certain extent limits the scientificity and accuracy of decision-making. Summary of the invention

[0005] The purpose of the present invention is to provide an assisted reproductive information analysis method and system to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an assisted reproductive information analysis method, the method comprising:

[0007] S1: Construct an assisted reproductive decision-making model and model the assisted reproductive decision-making process as a Markov decision process. The specific definitions include:

[0008] State space: includes vector representation of the patient's genetic and clinical data;

[0009] Action space: including decisions on adjusting drug dosage, choosing between different treatment options, and determining the timing of embryo transfer;

[0010] Reward: A reward function designed based on pregnancy rate and live birth rate indicators to evaluate action results;

[0011] Policy: The rule for choosing an action in a given state, optimized through learning to maximize the cumulative reward;

[0012] S2: Design a reward function. Based on the goal of assisted reproduction, design a reward mechanism that includes successful pregnancy and childbirth, improved embryo quality, stable hormone levels, and penalties for treatment failure;

[0013] S3: Train the assisted reproductive decision model, select the deep Q network algorithm, perform iterative training on actual data, use the ε-greedy strategy to balance the model behavior, and tune the learning rate and discount factor hyperparameters;

[0014] S4: Evaluate the assisted reproductive decision-making model. Use the A / B testing method to compare the model recommendation plan with the doctor's routine plan in terms of pregnancy rate and live birth rate. Design specific test cases to evaluate the model performance and make continuous improvements based on the evaluation results.

[0015] S5: Based on the trained assisted reproductive decision-making model, analyze the patient's personal data and provide doctors with personalized treatment plan recommendations.

[0016] Preferably, the specific definition of the reward function is:

[0017] Reward for successful pregnancy and childbirth: If the patient successfully becomes pregnant and gives birth, a large positive reward will be given, recorded as ;

[0018] Reward for improved embryo quality: If the embryo quality improves after treatment, a positive reward will be given according to the degree of improvement, recorded as , the reward is proportional to the improvement in embryo quality;

[0019] Hormone level stability reward: If the patient's hormone level remains stable during treatment, a small positive reward is given, recorded as ;

[0020] Treatment failure penalty: If the treatment fails, including failure to get pregnant or embryo transfer failure, a negative reward is given, recorded as ;

[0021] The reward function R can be expressed as: .

[0022] Preferably, the steps for implementing model training are:

[0023] Step 1: In the deep Q network algorithm, initialize the Q value function Q(s,a) to estimate the expected return of taking action a in state s;

[0024] Step 2: In each training iteration, for a given patient state s, the Q-values ​​of all possible actions a are calculated using the Q-network, expressed as: ;

[0025] Step 3: Select an action based on the Q value. Use the ε-greedy strategy to randomly select an action with probability ε and select the action with the largest Q value with probability 1-ε.

[0026] Step 4: Execute the selected action a, observe the new state reward, and update the Q network using the Q value calculation formula;

[0027] Step 5: Repeat the above steps until the predetermined number of training rounds is reached.

[0028] Preferably, during the model training process, for each round of iteration, the mean square error (MSE) is used as the performance evaluation indicator of the current round of iteration to monitor the training progress and performance changes of the model. The specific method is as follows:

[0029] Obtain the model's prediction results on pregnancy rate and live birth rate for the current training sample, and obtain the corresponding actual results;

[0030] Calculate the difference between the predicted result and the actual result; sum the squares of the differences of all samples and divide by the number of samples to get the mean square error, which is calculated as:

[0031]

[0032] Where n is the number of data points, is the model prediction value, is the true value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

[0033] Preferably, genetic data and clinical data are collected as training data for the assisted reproduction decision-making model; the genetic data include genome sequences and gene expression profiles, which are used to analyze the impact of genetic factors on assisted reproduction; the clinical data include the patient's age, body mass index, medical history records, ultrasound examination results and hormone level tests, which are used to reflect the patient's physiological state.

[0034] Preferably, before training the assisted reproductive decision model, the collected data is preprocessed, including:

[0035] Clean the collected data and remove incomplete, erroneous or outlier data records;

[0036] Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data;

[0037] Perform one-hot encoding conversion on categorical data.

[0038] Preferably, the method for evaluating the performance of the assisted reproductive decision model is:

[0039] Collect a preset number of patient data as a test set, which is independent of the training set data;

[0040] Design an A / B test plan, where group A is treated with the doctor's conventional plan, and group B is treated with the model-recommended plan, and the patients in groups A and B maintain similar distributions in key characteristics;

[0041] Performing tests in a simulated environment, recording the treatment process and outcomes for both groups of patients;

[0042] Evaluation indicators, including pregnancy rate, live birth rate, and number of treatment cycles, were designed to quantify the performance differences between the model-recommended regimen and the physician's conventional regimen. The chi-square test method was used to calculate and compare the statistical differences between Group A and Group B in various evaluation indicators.

[0043] Preferably, an assisted reproductive information analysis system comprises:

[0044] The data collection module is used to collect the patient's genetic data, clinical data, and diagnostic records during treatment;

[0045] Data preprocessing module, which cleans, encodes and standardizes the collected data;

[0046] The model definition module builds the assisted reproduction decision model according to the definition of the Markov decision process, including the state space, action space, reward function and strategy elements;

[0047] In the model training module, the deep Q network algorithm is used to perform iterative training on actual data, the ε-greedy strategy is used to balance the model behavior, and the learning rate and discount factor hyperparameters are tuned;

[0048] The A / B testing module designs an A / B testing plan to compare the model recommendation plan with the doctor's conventional plan in terms of pregnancy rate and live birth rate to evaluate the performance of the model;

[0049] The user interface module designs an interface for user interaction to facilitate doctors to enter patient data, view analysis results and obtain personalized recommendations.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention models the assisted reproductive decision-making process as a Markov decision process. By defining the state space, action space, rewards and strategies, the decision-making process is systematized and standardized. This helps doctors to comprehensively and accurately consider the patient's genetic data and clinical data when facing complex diseases, thereby formulating more scientific and accurate treatment plans.

[0052] The method provided by the present invention can analyze the patient's personal data, including genetic information and clinical data, and provide doctors with personalized treatment plan recommendations, fully considering the individual differences of patients, making the treatment plan more in line with the patient's actual situation and improving the personalization level of the treatment effect.

[0053] By designing a reward mechanism that includes successful pregnancy and delivery, improved embryo quality, stable hormone levels, and penalties for treatment failure, the present invention can guide the model to continuously optimize the treatment plan during the training process, helping doctors make more informed decisions in selecting drug dosages, treatment plans, and the timing of embryo transfer, thereby improving the success rate of treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A step diagram of an assisted reproductive information analysis method according to the present invention;

[0055] Figure 2 Schematic diagram of reward function design scheme;

[0056] Figure 3 Flowchart for training the assisted reproduction decision-making model. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] See also Figure 1-3The present invention provides a technical solution: a method for analyzing assisted reproductive information, the steps of the method comprising:

[0059] S1. Constructing the assisted reproductive decision model: First, the assisted reproductive decision process is modeled as a Markov decision process (MDP). The specific definition is as follows:

[0060] State space: includes vector representations of the patient’s genetic data and clinical data. These data are converted into a format that the model can understand through preprocessing and feature extraction, and serve as state information for model input.

[0061] Action space: This includes decisions such as adjusting drug dosage, choosing between different treatment options, and deciding when to transfer an embryo. These actions are what the model can do in a given state to optimize the outcome of the treatment.

[0062] Reward: A reward function designed based on the pregnancy rate and live birth rate indicators is used to evaluate the results of actions. When the actions taken by the model lead to pregnancy or live birth, positive rewards will be obtained; conversely, if the treatment fails or is ineffective, it will be punished.

[0063] Policy: The rule for choosing an action in a given state. Through learning optimization, the model can gradually learn to choose the best action in different states to maximize the cumulative reward.

[0064] S2. Design reward function: Based on the goal of assisted reproduction, design a reward mechanism that includes successful pregnancy and childbirth, improved embryo quality, stable hormone levels, and penalties for treatment failure. Specifically, the reward function can be set as:

[0065] Successful pregnancy and childbirth: Give a higher positive reward to encourage the model to pursue this goal.

[0066] Improved embryo quality: Certain positive rewards will be given to reflect the impact of embryo quality on treatment outcomes.

[0067] Hormone level stability: Give moderate positive rewards to maintain the balance of hormone levels in the patient's body.

[0068] Treatment failure penalty: negative rewards (i.e., penalties) are given for cases where treatment fails or has poor effects to encourage the model to avoid these adverse outcomes.

[0069] S3. Training the assisted reproductive decision model: Use the deep Q network (DQN) algorithm for model training. Perform iterative training on actual data and continuously adjust the model parameters so that the model gradually learns to choose the optimal action under a given state. The specific steps include:

[0070] Initialize model parameters and hyperparameters (such as learning rate, discount factor, etc.).

[0071] The ε-greedy strategy is used to balance the model behavior. That is, in the early stage of training, the model randomly selects actions with a certain probability to explore the effects of different actions; as the training progresses, the model gradually tends to select the known optimal action.

[0072] Tune hyperparameters such as learning rate and discount factor to improve model training efficiency and performance.

[0073] S4. Evaluate the assisted reproductive decision-making model: Use the A / B test method to compare the performance of the model-recommended plan and the doctor's routine plan in terms of pregnancy rate and live birth rate. Specifically, design an A / B test plan, in which Group A is treated with the doctor's routine plan and Group B is treated with the model-recommended plan. Ensure that patients in Group A and Group B maintain similar distributions in key features to ensure the fairness of the test. Then, collect the treatment process and outcome data of the two groups of patients and perform statistical analysis to evaluate the performance of the model.

[0074] Based on the evaluation results, the model is continuously improved. The performance of the model in the test is analyzed to find out the existing problems and deficiencies, and then the model structure is adjusted, the reward function is improved, and the training data is increased to improve the performance and adaptability of the model.

[0075] S5. Provide personalized treatment plan recommendations: Based on the trained assisted reproductive decision-making model, analyze the patient's personal data and provide doctors with personalized treatment plan recommendations. Specifically, the patient's genetic data and clinical data are input into the model, and the model selects the optimal action based on the current state and outputs the corresponding treatment plan recommendations. Doctors can refer to these recommendations and formulate more appropriate treatment plans based on the patient's actual situation and their own experience.

[0076] The present invention will be further described below in conjunction with Examples 1 to 3: Embodiment 1:

[0077] The reward function is used to comprehensively evaluate various factors in the treatment process according to the goals of assisted reproductive treatment and provide clear guidance for the model. The specific definition of the reward function is:

[0078] Reward for successful pregnancy and delivery: If the patient successfully becomes pregnant and delivers, a large positive reward will be given.

[0079] For example, you can set = 100. This reward value was chosen to ensure that the model places a high priority on the goal of successful pregnancy and childbirth during training.

[0080] Reward for embryo quality improvement: Embryo quality is one of the important factors for treatment success. In order to encourage the model to choose actions that can improve embryo quality, rewards are given according to the extent of embryo quality improvement. For example, the reward for embryo quality improvement can be set as = k × ΔQ, where k is a constant (such as k = 5) and ΔQ is the percentage of embryo quality improvement. In this way, when the embryo quality is significantly improved, the model will receive a corresponding positive reward.

[0081] Hormone level stability reward: To encourage the model to consider the stability of hormone levels when choosing actions, a small positive reward can be set. For example, when hormone levels remain stable, = 10. This reward value was chosen so that the model can take into account the stability of hormone levels when optimizing the strategy.

[0082] Treatment failure penalty: If the treatment fails, such as no pregnancy, embryo transfer failure, etc., a negative reward is given. For example, =−50. The purpose of this negative reward is to prevent the model from choosing actions that may lead to treatment failure during training.

[0083] Taking all the above factors into consideration, the reward function R can be expressed as follows in each step of the update:

[0084] . Embodiment 2:

[0085] The Deep Q Network (DQN) algorithm is a reinforcement learning algorithm that combines deep learning and Q learning. It uses a neural network to approximate the Q-value function, so that it can effectively learn strategies in high-dimensional state space and action space. Collect genetic data and clinical data as training data for the assisted reproduction decision-making model; the genetic data includes genome sequences and gene expression profiles, which are used to analyze the impact of genetic factors on assisted reproduction; the clinical data includes the patient's age, body mass index, medical history records, ultrasound examination results, and hormone level tests, which are used to reflect the patient's physiological state. Before training the assisted reproduction decision-making model, preprocess the collected data, including:

[0086] The collected data are cleaned to remove incomplete, erroneous or outlier data records.

[0087] Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data.

[0088] Perform one-hot encoding conversion on categorical data.

[0089] In the specific implementation, a neural network, called a Q network, is first constructed to estimate the expected return of taking a specific action in a given state. The input of this neural network is the patient's state vector, including age, body mass index, medical history, current treatment plan, and embryo quality information. The output is the Q value estimate corresponding to each possible action.

[0090] In each training iteration, the following steps are performed:

[0091] A patient's state s is randomly selected from the dataset as the current state.

[0092] Use the Q network to calculate the Q values ​​of all possible actions in the current state s. For example, if the action space includes adjusting drug dosage, choosing different treatment options, and deciding the timing of embryo transfer, then the Q network will output the Q values ​​corresponding to these three actions.

[0093] Select actions according to the ε-greedy strategy. Set a small probability ε (such as 0.1), randomly select an action with a probability of ε, and select the action with the largest Q value with a probability of 1-ε. This is done to maintain a certain degree of exploratory nature in the early stages of training to prevent the model from converging to a local optimal solution too early.

[0094] Perform the selected action and observe the new state s' and reward r obtained. The reward r is calculated based on the reward function, which takes into account factors such as successful pregnancy and delivery, improved embryo quality, stable hormone levels, and treatment failure penalties.

[0095] The Q-value update formula is used to update the parameters of the Q network. The current state s, the selected action a, the new state s' and the reward r are used as inputs, and the weights of the Q network are adjusted through the back-propagation algorithm to minimize the error between the predicted Q value and the actual Q value.

[0096] Repeat the above steps until the predetermined number of training rounds is reached or other stopping conditions are met. During the model training process, for each round of iteration, the mean square error (MSE) is used as the performance evaluation indicator of the current round of iteration to monitor the training progress and performance changes of the model. The specific method is as follows:

[0097] Obtain the model's prediction results on pregnancy rate and live birth rate for the current training sample, and obtain the corresponding actual results;

[0098] Calculate the difference between the predicted result and the actual result; sum the squares of the differences of all samples and divide by the number of samples to get the mean square error, which is calculated as:

[0099]

[0100] Where n is the number of data points, is the model prediction value, is the true value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

[0101] Let's take a specific example to illustrate the above process: Assume that the current patient's state is s = [32 years old, BMI 22, medical history: none, current treatment plan: plan A, embryo quality: medium], and the action space is A = {adjust drug dosage, select plan B, decide on embryo transplant}. First, use the Q network to calculate the Q values ​​of these three actions, assuming that the obtained Q values ​​are [0.6, 0.8, 0.5] respectively. Then, according to the ε-greedy strategy, the action with the largest Q value is selected with a probability of 0.9, that is, plan B is selected, and other actions are randomly selected with a probability of 0.1. After executing the selected action, the new state s' and reward r are observed, for example, s' = [32 years old, BMI 22, medical history: none, current treatment plan: plan B, embryo quality: good], r = +1 (indicating improved embryo quality). Finally, the parameters of the Q network are updated using the Q value update formula to estimate the Q value more accurately in the next iteration.

[0102] Through a large number of iterative training, the Q network will gradually learn the rules for selecting the best action in a given state to maximize the cumulative reward. After the training is completed, the trained Q network can be used in actual assisted reproductive decision-making to provide patients with personalized treatment plan recommendations. Embodiment 3:

[0103] In order to implement the method for evaluating the performance of the assisted reproductive decision model, the present invention adopts the following specific implementation methods:

[0104] Data collection: Randomly select a certain number of patient data from patients who have received treatment as the test set. Ensure that the test set data is independent of the data set used to train the model to avoid evaluation bias caused by data overlap.

[0105] A / B test design: The patients participating in the test are randomly divided into Group A and Group B. Group A is treated with the doctor's routine treatment plan, and Group B is treated with the plan recommended by the assisted reproductive decision model. To ensure the fairness and accuracy of the test, patients in Group A and Group B need to be matched on key features, such as age, cause of infertility, previous treatment history, etc., to maintain a similar distribution of these features between the two groups of patients.

[0106] Simulation environment setting: In the simulation environment, according to the designed A / B test plan, the two groups of patients are given corresponding treatment plans. The treatment process and results of the two groups of patients are recorded in detail, including but not limited to the number of treatment cycles, pregnancy status, live birth status, etc.

[0107] Evaluation index design: According to the characteristics and goals of assisted reproductive treatment, evaluation indicators such as pregnancy rate, live birth rate and number of treatment cycles were designed to quantify the performance difference between the model-recommended scheme and the doctor's conventional scheme. The chi-square test method was used to perform statistical analysis on the data of group A and group B on various evaluation indicators, and calculate and compare whether the differences between the two groups were statistically significant.

[0108] Result analysis: According to the results of the chi-square test, the performance differences between the model-recommended plan and the doctor's routine plan in various evaluation indicators were analyzed to determine whether the model is better than the doctor's routine plan.

[0109] The following is a specific implementation example:

[0110] First, data preparation was performed. Data from 1,000 patients were collected from the assisted reproductive center of a hospital as a test set. These data cover the patient's age, body mass index, medical history (including causes of infertility, previous treatment history, etc.), current treatment plan (such as drug type, dosage, etc.), embryo quality information (such as embryo rating, quantity, etc.) and actual treatment results (pregnancy rate and live birth rate). To ensure the representativeness and fairness of the data, patients were selected by random sampling, and the test set and training set data were ensured to maintain similar distributions in key features.

[0111] Design an A / B test plan. Randomly divide these 1,000 patients into Group A and Group B, with 500 patients in each group. Group A continues to use the doctor's routine treatment plan, while Group B uses the model-recommended plan. To ensure the fairness of the test, key factors such as the patient's age, body mass index, and medical history were considered when grouping to ensure that the two groups of patients were balanced in these characteristics.

[0112] Then, the test is performed. In the actual clinical environment, two groups of patients are treated according to the set plan, and the treatment process and results are recorded. To ensure the effectiveness and safety of the test, strict ethical standards and medical procedures are followed.

[0113] In the performance evaluation phase, a number of evaluation indicators were designed, including pregnancy rate, live birth rate, number of treatment cycles, etc. Through statistical analysis, the performance of Group A and Group B on various evaluation indicators was calculated and compared. For example, it was found that the pregnancy rate and live birth rate of Group B were higher than those of Group A, and the number of treatment cycles was shortened. This shows that the model recommendation plan is better than the doctor's conventional plan in some aspects.

[0114] Finally, the evaluation results were compiled into a detailed report and submitted to the medical team and model developers. Based on these feedbacks, the model was iteratively optimized, including adjusting model parameters, improving reward functions, etc. Then, the evaluation process of step S4 was repeated to verify whether the performance of the optimized model was improved.

[0115] The present invention relates to an assisted reproduction information analysis system, the system comprising:

[0116] Data collection module: This module is mainly responsible for collecting the patient's genetic data, clinical data and diagnostic records during the treatment process. Genetic data includes but is not limited to gene sequences, gene mutation information, etc.; clinical data includes the patient's age, gender, weight, height, history of underlying diseases, etc.; diagnostic records during treatment detail the patient's various examination results, treatment plans and reactions during assisted reproductive treatment. These data are collected through electronic medical record systems, genetic testing equipment and other channels, and stored in the system's database, providing a basis for subsequent data analysis and model training.

[0117] Data preprocessing module: The data preprocessing module cleans, encodes and standardizes the collected data. The cleaning process includes removing duplicate data, filling missing values, correcting erroneous data, etc. The encoding process converts non-numeric data into numerical data for subsequent calculations; standardization ensures that the data is compared on the same scale to improve the accuracy and robustness of the model. The preprocessed data will be used as input for model training.

[0118] Model definition module: Based on the definition of Markov decision process, this module constructs an assisted reproductive decision model. The model includes state space, action space, reward function and strategy elements. The state space describes the various possible states of the patient during the treatment process, such as hormone levels, embryo quality, etc.; the action space defines the various treatment measures that doctors can take, such as drug adjustments, surgical operations, etc.; the reward function gives corresponding rewards or penalties based on the treatment results to guide the model to learn the optimal strategy; the strategy elements determine the actions that the model should take in a given state.

[0119] Model training module: The deep Q network algorithm is used to perform iterative training on actual data. The deep Q network is a method that combines deep learning and reinforcement learning, and can handle problems with high-dimensional state and action space. During the training process, the ε-greedy strategy is used to balance the model behavior, that is, the current optimal action is selected under a certain probability, and the action is randomly selected under the remaining probabilities to explore new possible strategies. At the same time, the learning rate and discount factor hyperparameters are tuned to improve the convergence speed and performance of the model.

[0120] A / B test module: In order to evaluate the performance of the model, an A / B test scheme is designed. The patients are randomly divided into two groups, one group receives the treatment recommended by the model, and the other group receives the doctor's conventional plan. The performance of the two groups of patients in terms of pregnancy rate and live birth rate indicators is compared to evaluate the pros and cons of the model. The results of the A / B test will provide an important basis for further optimization and clinical application of the model.

[0121] User interface module: Design an interface for user interaction to facilitate doctors to input patient data, view analysis results, and obtain personalized recommendations. The interface should be concise and easy to operate. Doctors can input the patient's genetic data, clinical data and other information through the interface, and the system will automatically perform preprocessing and model analysis, and output personalized treatment plan recommendations. Doctors can also view the patient's historical treatment records and analysis results to better understand the patient's condition and treatment progress.

[0122] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0123] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing assisted reproductive information, characterized in that: The method comprises: S1: Construct an assisted reproductive decision-making model and model the assisted reproductive decision-making process as a Markov decision process. The specific definitions include: State space: includes vector representation of the patient's genetic and clinical data; Action space: including decisions on adjusting drug dosage, choosing between different treatment options, and determining the timing of embryo transfer; Reward: A reward function designed based on pregnancy rate and live birth rate indicators to evaluate action results; Policy: The rule for choosing an action in a given state, optimized through learning to maximize the cumulative reward; S2: Design a reward function. Based on the goal of assisted reproduction, design a reward mechanism that includes successful pregnancy and childbirth, improved embryo quality, stable hormone levels, and penalties for treatment failure; S3: Train the assisted reproductive decision model, select the deep Q network algorithm, perform iterative training on actual data, use the ε-greedy strategy to balance the model behavior, and tune the learning rate and discount factor hyperparameters; S4: Evaluate the assisted reproductive decision-making model, compare the model recommendation plan with the doctor's routine plan in terms of pregnancy rate and live birth rate through A / B testing, design specific test cases to evaluate model performance, and make continuous improvements based on the evaluation results; S5: Based on the trained assisted reproductive decision-making model, analyze the patient's personal data and provide doctors with personalized treatment plan recommendations; The specific definition of the reward function is: Reward for successful pregnancy and childbirth: If the patient successfully becomes pregnant and gives birth, a large positive reward will be given, recorded as ; Embryo quality improvement reward: If the embryo quality improves after treatment, a positive reward will be given according to the degree of improvement, recorded as , the reward is proportional to the improvement in embryo quality; Hormone level stability reward: If the patient's hormone level remains stable during treatment, a small positive reward is given, recorded as ; Treatment failure penalty: If the treatment fails, including failure to get pregnant or embryo transfer failure, a negative reward is given, recorded as ; The reward function R can be expressed as: ; Among them, the implementation steps of model training are: Step 1: In the deep Q network algorithm, initialize the Q value function Q(s,a) to estimate the expected return of taking action a in state s; Step 2: In each training iteration, for a given patient state s, the Q-values ​​of all possible actions a are calculated using the Q-network, expressed as: ; Step 3: Select an action based on the Q value. Use the ε-greedy strategy to randomly select an action with probability ε and select the action with the largest Q value with probability 1-ε. Step 4: Execute the selected action a, observe the new state reward, and update the Q network using the Q value calculation formula; Step 5: Repeat the above steps until the predetermined number of training rounds is reached; Genetic data and clinical data are collected as training data for the assisted reproduction decision-making model; the genetic data include genome sequences and gene expression profiles, which are used to analyze the impact of genetic factors on assisted reproduction; the clinical data include the patient's age, body mass index, medical history records, ultrasound examination results and hormone level tests, which are used to reflect the patient's physiological state.

2. The method for analyzing assisted reproductive information according to claim 1, characterized in that: During the model training process, for each round of iteration, the mean square error (MSE) is used as the performance evaluation indicator of the current round of iteration to monitor the training progress and performance changes of the model. The specific method is as follows: Obtain the model's prediction results on pregnancy rate and live birth rate for the current training sample, and obtain the corresponding actual results; Calculate the difference between the predicted result and the actual result; The square sum of the differences of all samples is calculated and divided by the number of samples to obtain the mean square error, which is calculated as: ； Where n is the number of data points, is the model prediction value, is the true value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

3. The method for analyzing assisted reproductive information according to claim 2, characterized in that: Before training the assisted reproductive decision-making model, the collected data is preprocessed, including: Clean the collected data and remove incomplete, erroneous or outlier data records; Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data; Perform one-hot encoding conversion on categorical data.

4. The method for analyzing assisted reproductive information according to claim 1, characterized in that: Methods for evaluating the performance of assisted reproduction decision models are: Collect a preset number of patient data as a test set, which is independent of the training set data; Design an A / B test plan, where group A is treated with the doctor's conventional plan, and group B is treated with the model-recommended plan, and the patients in groups A and B maintain similar distributions in key characteristics; Performing tests in a simulated environment, recording the treatment process and outcomes for both groups of patients; Design evaluation indicators, including pregnancy rate, live birth rate, and number of treatment cycles, to quantify the performance difference between the model-recommended regimen and the physician's conventional regimen; The chi-square test method was used to calculate and compare the statistical differences between group A and group B in various evaluation indicators.

5. An assisted reproductive information analysis system, used to perform any one of the analysis methods of claims 1-4, characterized in that: The system comprises: The data collection module is used to collect the patient's genetic data, clinical data, and diagnostic records during treatment; Data preprocessing module, which cleans, encodes and standardizes the collected data; The model definition module builds the assisted reproduction decision model according to the definition of the Markov decision process, including the state space, action space, reward function and strategy elements; In the model training module, the deep Q network algorithm is used to perform iterative training on actual data, the ε-greedy strategy is used to balance the model behavior, and the learning rate and discount factor hyperparameters are tuned; The A / B testing module designs an A / B testing plan to compare the model recommendation plan with the doctor's conventional plan in terms of pregnancy rate and live birth rate to evaluate the performance of the model; The user interface module designs an interface for user interaction to facilitate doctors to enter patient data, view analysis results and obtain personalized recommendations.

Images