Artificial intelligence prediction method and system for multi-embryo blastocyst culture outcome
By constructing a feature vector based on morphological parameter data in assisted reproductive technology and inputting a classifier for prediction, the problem of predicting individual embryonic blastocyst formation outcomes in population blastocyst culture environment is solved, and high accuracy and versatility are achieved. Reduced data labeling costs.
Patent Information
- Application Number
- CN202411502065.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2025-02-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to accurately predict the blastocyst formation outcome of individual embryos in population blastocyst culture environments, resulting in limited accuracy and universality of prediction models in different laboratory environments.
By collecting morphological parameter data of embryo culture, building feature vectors and inputting them into a preset classifier to predict the culture results of embryo groups. This method does not require independent labeling of individual embryos, and is suitable for a variety of culture environments and is able to handle the prediction of population and individual culture cycles.
It achieves high accuracy and versatility in different laboratory environments, reduces the cost of data labeling, is suitable for a variety of culture environments, and maintains prediction accuracy comparable to existing single embryo probability prediction methods.
Smart Images
Figure CN119446555A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of assisted reproductive technology, and in particular to an artificial intelligence prediction method and system for the outcome of multi-embryo blastocyst culture. Background Art
[0002] In the field of assisted reproductive technology, accurate prediction of blastocyst formation is crucial for optimizing embryo culture strategies for in vitro fertilization-embryo transfer (IVF-ET). If it is predicted that multiple blastocysts may be produced in a cycle, the doctor may choose to extend the in vitro culture to the blastocyst stage; conversely, if the blastocyst formation rate is predicted to be low, early transplantation may be considered to reduce embryo loss. With the continuous advancement of artificial intelligence technology, researchers have gradually applied a variety of machine learning algorithms to the prediction of blastocyst formation and achieved initial results. For example, Liao et al. (2021) developed a model based on convolutional neural networks to predict blastocyst formation using early embryo development videos, with an AUC of up to 0.8. Jiang et al. (2022) proposed a machine learning algorithm based on a tree model, using early embryo morphological parameters (such as cell number, fragment ratio, etc.) to predict blastocyst formation with an AUC of up to 0.85. Although existing models perform well in predictive accuracy, they are mainly limited to predicting whether a single embryo will form a usable blastocyst. Multiple embryos can usually be obtained in one IVF-ET treatment cycle, and blastocyst culture is usually charged uniformly by cycle, that is, the same user is charged the same fee for culturing one or more embryos. Therefore, predicting the outcome of the entire cycle (such as the absence of usable blastocyst formation and the number of usable blastocysts) has important clinical significance and economic value.
[0003] Recently, some studies have begun to try to infer cycle outcomes by integrating the predicted probabilities of multiple embryos. However, this method still faces significant limitations: in a laboratory environment that adopts an embryo group culture strategy (i.e., multiple embryos are cultured in each drop of culture medium), it is difficult to accurately correspond the characteristics of a single embryo to the outcome of blastocyst culture. Existing methods are not suitable for laboratories where individual embryo tracking is difficult, which seriously limits their widespread application in clinical practice. In view of this, there is an urgent need to develop an innovative prediction method to fill the gap between current prediction methods (mainly focusing on individual culture) and clinical practice needs (involving both individual and group culture), thereby improving the accuracy and versatility of prediction models in different laboratory environments. Summary of the invention
[0004] The purpose of the present invention is to propose an artificial intelligence prediction method and system for the outcome of multi-embryo blastocyst culture to solve the problems existing in the above-mentioned prior art.
[0005] To achieve the above object, the present invention provides the following solutions:
[0006] Artificial intelligence prediction methods for the outcomes of multiple-embryo blastocyst culture, including:
[0007] Collect morphological parameter data of embryo culture;
[0008] Based on the morphological parameter data, construct a feature vector;
[0009] The feature vector is input into a preset classifier to predict the culture results of the embryo group.
[0010] Optionally, the morphological parameter data include: cell number, fragment ratio and symmetry of embryos on the second and third days.
[0011] Optionally, the embryo sample feature vector is a two-dimensional vector, wherein the first dimension represents each independent embryo sample, the second dimension represents the characteristic parameters of a single embryo, and each of the embryo sample feature vectors includes several embryo samples.
[0012] Optionally, the feature vector is: [[embryo 1_second day cell number, embryo 1_second day fragment ratio, embryo 1_second day symmetry, embryo 1_third day cell number, embryo 1_third day fragment ratio, embryo 1_third day symmetry], [embryo 2_second day cell number, embryo 2_second day fragment ratio, embryo 2_second day symmetry, embryo 2_third day cell number, embryo 2_third day fragment ratio, embryo 2_third day symmetry], ... [embryo n_second day cell number, embryo n_second day fragment ratio, embryo n_second day symmetry, embryo n_third day cell number, embryo n_third day fragment ratio, embryo n_third day symmetry]].
[0013] Optionally, constructing an embryo sample feature vector based on the morphological parameter data includes:
[0014] Performing K-means clustering on the morphological parameter data to obtain K cluster centers;
[0015] Calculate the distance from each embryo in the embryo sample to the center of each cluster;
[0016] Count the number of instances belonging to each cluster in each embryo sample to form a K-dimensional histogram, which is the embryo sample feature vector;
[0017] The objective function of the K-means clustering is:
[0018]
[0019] Among them, S is a set of k clusters, μ i is the center point of the i-th cluster, and x belongs to cluster S i A data point of ||x-μ i || 2 is the distance from data point x to cluster center μi The goal of K-means clustering is to find the cluster center μ that minimizes J. i .
[0020] Optionally, before inputting the embryo sample feature vector into a preset classifier, the method further includes: training the preset classifier;
[0021] Training a preset classifier includes:
[0022] Construct a data set; wherein the data set includes: a single embryo ID, an embryo sample feature vector, an embryo group ID, and a group label;
[0023] The data set was used to train the preset classifier, and 5-fold cross validation was used to optimize the model parameters during the training process.
[0024] An artificial intelligence prediction system for the outcome of multi-embryo blastocyst culture, the system comprising: a collection module, a construction module, and a prediction module;
[0025] The acquisition module is used to collect morphological parameter data of embryo culture;
[0026] The construction module is used to construct a feature vector based on the morphological parameter data;
[0027] The prediction module is used to input the feature vector into a preset classifier to predict the culture results of the embryo group.
[0028] Optionally, the morphological parameter data include: cell number, fragment ratio and symmetry of embryos on the second and third days.
[0029] Optionally, the embryo sample feature vector is a two-dimensional vector, wherein the first dimension represents each independent embryo sample, the second dimension represents the characteristic parameters of a single embryo, and each of the embryo sample feature vectors includes several embryo samples.
[0030] Optionally, the feature vector is: [[embryo 1_second day cell number, embryo 1_second day fragment ratio, embryo 1_second day symmetry, embryo 1_third day cell number, embryo 1_third day fragment ratio, embryo 1_third day symmetry], [embryo 2_second day cell number, embryo 2_second day fragment ratio, embryo 2_second day symmetry, embryo 2_third day cell number, embryo 2_third day fragment ratio, embryo 2_third day symmetry], ... [embryo n_second day cell number, embryo n_second day fragment ratio, embryo n_second day symmetry, embryo n_third day cell number, embryo n_third day fragment ratio, embryo n_third day symmetry]].
[0031] Optionally, constructing an embryo sample feature vector based on the morphological parameter data in the construction module includes:
[0032] Performing K-means clustering on the morphological parameter data to obtain K cluster centers;
[0033] Calculate the distance from each embryo in the embryo sample to the center of each cluster;
[0034] Count the number of instances belonging to each cluster in each embryo sample to form a K-dimensional histogram, which is the embryo sample feature vector;
[0035] The objective function of the K-means clustering is:
[0036]
[0037] Among them, S is a set of k clusters, μ i is the center point of the i-th cluster, and x belongs to cluster S i A data point of ||x-μ i || 2 is the distance from data point x to cluster center μ i The goal of K-means clustering is to find the cluster center μ that minimizes J. i .
[0038] Optionally, the artificial intelligence prediction system further includes a training module;
[0039] The training module is used to train a preset classifier; it includes:
[0040] Construct a data set; wherein the data set includes: a single embryo ID, an embryo sample feature vector, an embryo group ID, and a group label;
[0041] The data set was used to train the preset classifier, and 5-fold cross validation was used to optimize the model parameters during the training process.
[0042] The beneficial effects of the present invention are:
[0043] Since there is no need to independently label individual embryos, the present invention greatly reduces the time and labor costs associated with data annotation, saving laboratory resource investment, including manpower and equipment.
[0044] The present invention is applicable to a variety of culture environments, and can not only handle group culture, but also be applied to cycle prediction of individual culture. This diversity bridges the gap between current technologies in different culture strategies.
[0045] While reducing the reliance on individual labels, the prediction accuracy of the present invention is comparable to that of existing methods based on single embryo probability prediction, thus ensuring technical reliability. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0047] Figure 1 Schematic diagram of the process flow of the artificial intelligence prediction method for the outcome of multi-embryo blastocyst culture according to an embodiment of the present invention. DETAILED DESCRIPTION
[0048] 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.
[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0050] This embodiment proposes an artificial intelligence prediction method for the outcome of multi-embryo blastocyst culture, including:
[0051] Collect morphological parameter data of embryo culture;
[0052] Based on the morphological parameter data, construct feature vectors;
[0053] The feature vector is input into the preset classifier to predict the culture results of the embryo group.
[0054] Furthermore, the morphological parameter data included: cell number, fragment ratio and symmetry of embryos on the second and third days.
[0055] Furthermore, the embryo sample feature vector is a two-dimensional vector, wherein the first dimension represents a number of embryos obtained by a user in a treatment cycle, and the second dimension represents the characteristic parameters of a single embryo.
[0056] Further, the eigenvector is: [[embryo 1_second day cell number, embryo 1_second day fragment ratio, embryo 1_second day symmetry, embryo 1_third day cell number, embryo 1_third day fragment ratio, embryo 1_third day symmetry], [embryo 2_second day cell number, embryo 2_second day fragment ratio, embryo 2_second day symmetry, embryo 2_third day cell number, embryo 2_third day fragment ratio, embryo 2_third day symmetry], ... [embryo n_second day cell number, embryo n_second day fragment ratio, embryo n_second day symmetry, embryo n_third day cell number, embryo n_third day fragment ratio, embryo n_third day symmetry]].
[0057] like Figure 1 As shown, the specific implementation method of this embodiment is:
[0058] 1. Data collection and preprocessing;
[0059] The data comes from the laboratory of an assisted reproductive center, covering 9,000 embryo samples. The morphological parameter data of embryo culture are collected, including the number of cells, fragment ratio and symmetry of embryos on the second and third days. In order to better analyze and use these data, this embodiment processes them into a two-dimensional vector format. The specific steps are as follows:
[0060] 1.1. Sample combination: Each patient may have multiple embryo samples. This embodiment combines all embryo samples of the same patient for comprehensive analysis.
[0061] 1.2. Construction of two-dimensional vector: The first dimension represents each independent embryo sample. The second dimension represents the characteristic parameters of each sample, including the above six morphological parameters.
[0062] The data of this embodiment contains multiple samples, each of which comes from a different treatment cycle. One sample = a group of embryos (a mother will produce multiple embryos in one treatment cycle) = a package. For example, the data has 2000 patients (sample size), each patient can obtain 1-20 embryos, and each embryo has 6 morphological characteristics. Therefore, the data as a whole is three-dimensional [sample size, number of embryos, parameters of each embryo], and the characteristics of a single sample are two-dimensional [number of embryos, parameters of each embryo].
[0063] 2. Model Construction
[0064] The multi-instance learning model is built using Python programming language and Scikit-learn library. The model applies K-means clustering to transform the indefinite length group data into a fixed length feature vector, and then applies XGBoost classifier for prediction, which can effectively handle complex data sets with only package labels.
[0065] Feature vector generation:
[0066] The K-means clustering algorithm is used to generate histograms, and the original vector of each packet is aggregated into a feature vector of fixed length. The specific process is as follows:
[0067] a) Collect the original vectors of all instance features in all bags, where bags refer to embryo samples.
[0068] b) Apply K-means clustering to these original vectors to obtain K cluster centers (K is set to 100 in the present invention). These K cluster centers constitute the "vocabulary".
[0069] c) For each bag, calculate which cluster each instance in the bag belongs to (by calculating the distance to the center of each cluster). The instances are embryos.
[0070] d) Count the number of instances belonging to each cluster in each package to form a K-dimensional histogram, which is the feature vector of the package.
[0071] The objective function of K-means clustering is:
[0072]
[0073] Among them, S is a set of k clusters, μ i is the center point (centroid) of the i-th cluster, and x is the number of nodes belonging to cluster S i A data point of ||x-μ i || 2 is the distance from data point x to cluster center μ i The goal of K-means clustering is to find the cluster center μ that minimizes J. i .
[0074] Here, K is an adjustable parameter. K cluster centers represent K types of instance features. These K cluster centers can theoretically cover all instance types. Distance can be used as a measure of the similarity between an instance and a cluster center. A smaller distance means that the instance is more similar to the cluster center. By calculating the distance from each instance to all cluster centers, it can be determined which cluster the instance is closest to, that is, which cluster it belongs to. This is one of the core steps of the K-means algorithm. For example, if K is set to 50, K-means can be used to convert a package dimension of [number of embryos, parameters of each embryo] into a feature of [1, 50] (by assigning instances to the nearest cluster center).
[0075] Classifier training and prediction:
[0076] The present invention inputs the generated feature vector into the XGBoost classifier to train the model and predict the training outcomes of the new group. The core principle of the XGBoost model is as follows:
[0077] Prediction Mechanism:
[0078] The model prediction is based on the cumulative output of multiple decision trees. Each decision tree predicts the input features, and the final result is the sum of the prediction values of all trees. It can efficiently handle nonlinear relationships and is suitable for complex group culture outcome prediction.
[0079] Loss function:
[0080] The training loss function is used to evaluate the model prediction accuracy. Depending on the problem type, squared error (regression) or log loss (classification) is used.
[0081] Optimize the process:
[0082] a) Initialization: Use a constant value as the initial prediction.
[0083] b) Iterative optimization: Calculate the gradient statistics of the current model; build a new decision tree to fit the gradient statistics; add the new tree to the model and update the prediction
[0084] c) Repeat the iteration until the termination condition is met
[0085] 3. Experimental Implementation
[0086] The data is divided into a training set (80%) and a test set (20%).
[0087] The dataset includes:
[0088] Individual ID: uniquely identifies each individual
[0089] Individual characteristics: The feature vector describing each individual includes [number of cells on the second day, fragment ratio on the second day, symmetry on the second day, number of cells on the third day, fragment ratio on the third day, symmetry on the third day]
[0090] Group ID: Indicates which group the individual belongs to
[0091] Group label: The culture outcome of the entire group (whether one or more usable blastocysts are formed, and the number of usable blastocysts formed)
[0092] During the training process, 5-fold cross validation was used to optimize the model parameters.
[0093] 4. Results and Analysis
[0094] On the independent test set, the model's binary classification prediction (whether one or more usable blastocysts are formed) AUC reached 0.85, and the mean absolute error (MAE) of regression prediction (the number of usable blastocysts formed) was 0.87, which is comparable to existing methods based on single embryo probability prediction.
[0095] This embodiment aims to solve the key technical problem of group blastocyst culture stage prediction in the current field of assisted reproductive technology. Existing prediction methods mainly focus on the prediction of a single embryo, and only infer the periodic results by integrating the prediction probabilities of multiple embryos. This is limited by the problem that embryo characteristics and labels cannot be matched one-to-one in a group culture environment, and therefore it is difficult to apply to group blastocyst culture programs.
[0096] To this end, the primary goal of this embodiment is to develop a method for predicting blastocyst formation at the group level without individual labels, aiming to more comprehensively evaluate the overall viability of all embryos of the patient. This embodiment intends to integrate the morphological parameters of multiple embryos through an innovative multi-instance learning method to provide a more practical and efficient prediction tool. In summary, the goal of this embodiment is to develop a blastocyst formation prediction method based on multi-instance learning, with the following features:
[0097] 1. Classification or regression prediction of blastocyst formation at the population level, including the presence or absence of available blastocysts and the number of available blastocysts.
[0098] 2. No individual labels are required, and the training strategy is applicable to both individuals and groups.
[0099] 3. Consider the interactions between embryos to improve the accuracy and applicability of predictions.
[0100] By achieving the above objectives, this embodiment is expected to fill the gaps in current technology and bring new breakthroughs and developments to the field of assisted reproductive technology.
[0101] Compared with the conventional technology, this embodiment has achieved significant improvements in the following aspects:
[0102] 1. No need for individual labeling: The model of the present invention does not require independent labeling of each embryo, and is therefore suitable for different individual and group culture strategies.
[0103] 2. Consider the interactions between embryos: Through multi-instance learning models, the complex interactions between embryos can be better captured, thereby improving the accuracy and applicability of predictions.
[0104] The vocabulary-based embedding space paradigm of MIL is adopted, which can adapt to variable input sizes. We define a mapping function M:X→v that maps a bag X to a feature vector v that comprehensively summarizes the features in the bag. This mapping function identifies the instance categories in a given bag X with the help of a vocabulary containing a set of prototypes. Using a specific histogram vocabulary, we can characterize the instance composition of each bag in detail, providing an explicit mapping to highlight the composition of X. These feature vectors can be processed using standard classifiers such as XGBoost.
[0105] In the process of implementing this technology, data collection and processing are required. The conventional morphological parameters of the first three days of embryonic development (including the number of cells, fragment ratio, and symmetry on the second and third days) are selected as input features, and the group outcome (whether or not a usable blastocyst is formed, and the number of usable blastocysts) is used as a label. A prediction model that can process group data is constructed under the framework of multi-instance learning, and the model is trained with the processed data, and its generalization ability is verified on an independent test set.
[0106] Although this embodiment supports other MIL paradigms as alternatives, such as the instance space paradigm (focusing on instance-level distinctions, which may ignore important embryo interactions), or the bag space paradigm (which treats the group as a whole, but may not accurately capture complex intra-group relationships), overall, the embedding space paradigm can map each group into a feature vector, more effectively capture the collective characteristics and interactions between embryos, fully extract and utilize input information, and has unique advantages in processing embryo data.
[0107] Through the above technical scheme, this embodiment effectively fills the shortcomings of current technology in the prediction of group blastocyst culture, provides a new solution for the field of assisted reproductive technology, and shows broad application prospects.
[0108] This embodiment also proposes an artificial intelligence prediction system for the outcome of multi-embryo blastocyst culture, the system comprising:
[0109] Acquisition module, construction module, prediction module;
[0110] A collection module is used to collect morphological parameter data of embryo culture;
[0111] A construction module is used to construct a feature vector based on morphological parameter data;
[0112] The prediction module is used to input the feature vector into a preset classifier to predict the culture results of the embryo group.
[0113] Furthermore, the morphological parameter data included: cell number, fragment ratio and symmetry of embryos on the second and third days.
[0114] Furthermore, the embryo sample feature vector is a two-dimensional vector, wherein the first dimension represents a number of embryos obtained by a user in a treatment cycle, and the second dimension represents the characteristic parameters of a single embryo.
[0115] Further, the eigenvector is: [[embryo 1_second day cell number, embryo 1_second day fragment ratio, embryo 1_second day symmetry, embryo 1_third day cell number, embryo 1_third day fragment ratio, embryo 1_third day symmetry], [embryo 2_second day cell number, embryo 2_second day fragment ratio, embryo 2_second day symmetry, embryo 2_third day cell number, embryo 2_third day fragment ratio, embryo 2_third day symmetry], ... [embryo n_second day cell number, embryo n_second day fragment ratio, embryo n_second day symmetry, embryo n_third day cell number, embryo n_third day fragment ratio, embryo n_third day symmetry]].
[0116] Furthermore, in the construction module, based on the morphological parameter data, the construction of the embryo sample feature vector includes:
[0117] K-means clustering is performed on the morphological parameter data to obtain K cluster centers;
[0118] Calculate the distance from each instance in the bag to the center of each cluster;
[0119] Count the number of instances belonging to each cluster in each package to form a K-dimensional histogram, which is the embryo sample feature vector of the package;
[0120] The objective function of K-means clustering is:
[0121]
[0122] Among them, S is a set of k clusters, μ i is the center point (centroid) of the i-th cluster, and x is the number of nodes belonging to cluster S i A data point of ||x-μ i || 2 is the distance from data point x to cluster center μ i The goal of K-means clustering is to find the cluster center μ that minimizes J. i .
[0123] Furthermore, the artificial intelligence prediction system also includes a training module;
[0124] The training module is used to train the preset classifier; it includes:
[0125] Construct a data set; the data set includes: single embryo ID, embryo sample feature vector, embryo group ID, and group label;
[0126] The preset classifier is trained using the data set, and 5-fold cross validation is used to optimize the model parameters during the training process.
[0127] This example introduces an innovative method based on multi-instance learning, focusing on improving the prediction of the outcome of group blastocyst culture, and achieves the following significant technical effects:
[0128] 1. Reduce data annotation costs:
[0129] Since there is no need to independently label individual embryos, this embodiment greatly reduces the time and labor costs associated with data labeling, saving laboratory resource investment, including manpower and equipment.
[0130] 2. Enhance model adaptability:
[0131] This technology is applicable to a variety of culture environments and can not only handle group culture, but also be applied to cycle prediction of individual culture. This diversity bridges the gap between current technologies in different culture strategies.
[0132] 3. Maintain high forecast accuracy:
[0133] While reducing the reliance on individual labels, the prediction accuracy of the present invention is comparable to that of existing methods based on single embryo probability prediction, thus ensuring technical reliability.
[0134] 4. Improve economic benefits:
[0135] Accurately predicting the blastocyst formation rate helps identify cycles without available blastocyst formation, avoids unnecessary blastocyst culture, and saves treatment costs for patients. In addition, the model of the present invention can be applied to group embryo culture laboratories, reducing the demand for human resources, improving work efficiency and increasing business processing volume, thereby improving economic returns.
[0136] 5. Improve social benefits:
[0137] Accurate prediction of blastocyst formation rate can help assist clinical decision-making on whether to perform blastocyst culture, which is beneficial to improving embryo utilization, increasing the success rate of IVF-ET treatment, shortening treatment time, improving patient satisfaction and trust, and at the same time helping to increase the overall population birth rate.
[0138] Overall, this implementation not only provides an efficient method for predicting blastocyst culture, but also brings significant scientific, economic and social benefits to the field of assisted reproductive technology. The results of practical application experiments have proved that this technology can effectively solve the shortcomings of existing methods and provide solid support for clinical practice.
[0139] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.
Claims
1. An artificial intelligence prediction method for the outcome of multiple embryo blastocyst culture, characterized in that: include: Morphological parameters of embryo culture were collected; Based on the morphological parameters, constructing an embryo sample feature vector; The embryo sample feature vector is input into a preset classifier to predict the culture results of the embryo group.
2. The artificial intelligence prediction method for the outcome of multiple embryo blastocyst culture according to claim 1, characterized in that: The morphological parameters included: cell number, fragmentation ratio and symmetry of day 2 and day 3 embryos.
3. The artificial intelligence prediction method for the outcome of multiple embryo blastocyst culture according to claim 2, characterized in that: The embryo sample feature vector is a two-dimensional vector, wherein the first dimension represents a number of embryos obtained by a user in a treatment cycle, and the second dimension represents the feature parameters of a single embryo.
4. The artificial intelligence prediction method for the outcome of multiple embryo blastocyst culture according to claim 1, characterized in that: Based on the morphological parameter data, constructing an embryo sample feature vector includes: Performing K-means clustering on the morphological parameter data to obtain K cluster centers; Calculate the distance from each embryo in the embryo sample to the center of each cluster; Count the number of instances belonging to each cluster in each embryo sample to form a K-dimensional histogram, which is the embryo sample feature vector; The objective function of the K-means clustering is: Among them, S is a set of k clusters, μ i is the center point of the i-th cluster, and x belongs to cluster S i A data point of ||x-μ i || 2 is the distance from data point x to cluster center μ i The goal of K-means clustering is to find the cluster center μ that minimizes J. i .
5. The artificial intelligence prediction method for the outcome of multiple embryo blastocyst culture according to claim 1, characterized in that: Before inputting the embryo sample feature vector into the preset classifier, the method further includes: training the preset classifier; Training a preset classifier includes: Construct a data set; wherein the data set includes: a single embryo ID, an embryo sample feature vector, an embryo group ID, and a group label; The data set was used to train the preset classifier, and 5-fold cross validation was used to optimize the model parameters during the training process.
6. An artificial intelligence prediction system for the outcome of multi-embryo blastocyst culture, characterized in that: An artificial intelligence prediction method for the outcome of multi-embryo blastocyst culture as described in any one of claims 1 to 5, the system comprising: a collection module, a construction module, and a prediction module; The acquisition module is used to collect morphological parameter data of embryo culture; The construction module is used to construct a feature vector based on the morphological parameter data; The prediction module is used to input the feature vector into a preset classifier to predict the culture results of the embryo group.
7. The artificial intelligence prediction system for multiple embryo blastocyst culture outcomes according to claim 6, characterized in that: The morphological parameter data include: cell number, fragment ratio and symmetry of embryos on the second and third days.
8. The artificial intelligence prediction system for multiple embryo blastocyst culture outcomes according to claim 6, characterized in that: The embryo sample feature vector is a two-dimensional vector, wherein the first dimension represents each independent embryo sample, the second dimension represents the characteristic parameters of a single embryo, and each embryo sample feature vector includes a number of embryo samples.
9. The artificial intelligence prediction system for multiple embryo blastocyst culture outcomes according to claim 6, characterized in that: In the construction module, constructing an embryo sample feature vector based on the morphological parameter data includes: Performing K-means clustering on the morphological parameter data to obtain K cluster centers; Calculate the distance from each embryo in the embryo sample to the center of each cluster; Count the number of instances belonging to each cluster in each embryo sample to form a K-dimensional histogram, which is the embryo sample feature vector; The objective function of the K-means clustering is: Among them, S is a set of k clusters, μ i is the center point of the i-th cluster, and x belongs to cluster S i A data point of ||x-μ i || 2 is the distance from data point x to cluster center μ i The goal of K-means clustering is to find the cluster center μ that minimizes J. i .
10. The artificial intelligence prediction system for multiple embryo blastocyst culture outcomes according to claim 6, characterized in that: The artificial intelligence prediction system also includes a training module; The training module is used to train a preset classifier; it includes: Construct a data set; wherein the data set includes: a single embryo ID, an embryo sample feature vector, an embryo group ID, and a group label; The data set was used to train the preset classifier, and 5-fold cross validation was used to optimize the model parameters during the training process.