Automatic embryo morphology evaluation system based on artificial intelligence image recognition

Abstract

The invention provides an automatic embryo morphology evaluation system based on artificial intelligence image recognition, and relates to the technical field of biomedical engineering. The invention discloses an artificial intelligence image recognition-based embryo morphology automatic evaluation system. Comprising an AI system, a data acquisition and preprocessing module, a multi-modal data fusion module, a model training and optimization module, a user interface and interaction module, a security and privacy protection module, a model interpretation module, a clinical verification and evaluation module, a system integration and deployment module, a knowledge base and decision support module and a real-time monitoring and early warning module. The system comprises an individual treatment scheme module, a multi-language support and localization module, a user training and support module, a data backup and recovery module and a performance monitoring and optimizing module. Through a multi-modal data fusion module, data from different sources, embryo images, gene data, patient medical history and hormone levels are integrated, data enhancement is performed by using diversity, and the generalization ability of the model is improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical engineering technology, and in particular to an automated embryo morphology evaluation system based on artificial intelligence image recognition. Background Art

[0002] In assisted reproductive technology, the quality and developmental potential of the embryo are key factors in determining the success of in vitro fertilization. Traditional embryo selection mainly relies on morphological evaluation by embryologists, that is, observing the appearance characteristics of the embryo through a microscope, such as the number of cells, symmetry and degree of fragmentation. Artificial intelligence systems rely on large amounts of data for training and verification. Assisted reproductive technology has accumulated a large amount of embryo image data and clinical data, providing rich resources for the training of AI models. These technologies provide rich visual information and provide a basis for the training and evaluation of AI models.

[0003] There are still some problems in the use of existing automated embryo morphology assessment systems based on artificial intelligence image recognition. The performance of AI models is highly dependent on the quality and diversity of training data. The data set is limited and cannot cover all possible embryo morphologies and developmental conditions. Data labeling is subjective, and different embryologists may have different labels for the same embryo. Deep learning models are usually regarded as "black boxes" and it is difficult to explain their decision-making process. This leads to insufficient trust in the AI ​​system among clinicians, affecting its application in clinical practice. Therefore, those skilled in the art provide an automated embryo morphology assessment system based on artificial intelligence image recognition to solve the problems raised in the above background technology. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In response to the shortcomings of the prior art, the present invention provides an automated embryo morphology assessment system based on artificial intelligence image recognition, which solves the problems that the performance of the AI ​​model is highly dependent on the quality and diversity of the training data, the data set is limited in size and cannot cover all possible embryo morphologies and developmental conditions, data labeling is subjective, and different embryologists may have different labels for the same embryo, and deep learning models are usually viewed as "black boxes" and their decision-making process is difficult to explain, which leads to insufficient trust in the AI ​​system among clinicians and affects its application in clinical practice.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an automated embryo morphology evaluation system based on artificial intelligence image recognition, comprising:

[0008] The data acquisition and preprocessing module obtains embryo images and related information from different types of equipment such as microscopes, B-ultrasound and gene sequencers, removes noise data, duplicate data and incomplete data, ensures data quality, and converts data from different sources into a unified format and standard for subsequent processing;

[0009] The multimodal data fusion module integrates image data, genetic data and data from different sources of patient medical history, extracts features from data of different modalities, and performs feature fusion to improve the prediction ability of the model, establish an efficient data storage system, and support the storage and rapid retrieval of large-scale data;

[0010] Model training and optimization module: select deep learning models for training, predict embryo quality and developmental potential, optimize model hyperparameters through grid search, random search and Bayesian optimization to improve model performance, and evaluate the model using validation and test sets to ensure the generalization ability and stability of the model;

[0011] The user interface and interaction module provides a simple and intuitive user interface to facilitate clinicians to use the AI ​​system for embryo evaluation. It provides real-time feedback and suggestions based on the prediction results of the AI ​​model to help doctors make more accurate decisions and support interactive operations.

[0012] Security and privacy protection module: encrypts sensitive data to prevent data leakage, implements strict access control measures to ensure that only authorized personnel can access patient data, and ensures that the system complies with relevant laws, regulations and ethical standards;

[0013] Model interpretability module, which provides explanations for model predictions to help clinicians understand the basis for AI decisions. It develops visualization tools to intuitively display the model's decision-making process, such as using heat maps to show which areas in embryo images have the greatest impact on the prediction results. It also uses Grad-CAM interpretable AI technology to provide explanations for the model's internal mechanisms.

[0014] The clinical validation and evaluation module works with medical institutions to design and implement large-scale, randomized controlled clinical trials to verify the effectiveness of the AI ​​system in real clinical settings, collect data from clinical trials, and conduct statistical analysis to evaluate the effectiveness of the AI ​​system in improving pregnancy rates and reducing miscarriage rates, and conduct long-term follow-up studies to evaluate the long-term effects and safety of the AI ​​system.

[0015] The system integration and deployment module integrates various modules to form a complete system, establishes CI / CD processes, realizes the automated construction, testing and deployment of the system, monitors the system in real time, and promptly discovers and resolves problems in system operation;

[0016] The knowledge base and decision support module establishes a database containing embryological knowledge, clinical guidelines, research literature, and historical cases to provide reference for the AI ​​model. Based on the prediction results of the AI ​​model and the information in the knowledge base, it provides detailed decision support and suggestions for clinicians. The knowledge base is updated regularly to ensure that it contains the latest research results and clinical guidelines.

[0017] The real-time monitoring and early warning module monitors the embryonic development process in real time and detects abnormalities in a timely manner. When an abnormality is detected, the early warning mechanism is automatically triggered to remind clinicians to intervene. Combined with historical data, it analyzes the trends and laws of embryonic development to improve the accuracy of early warning.

[0018] The personalized treatment plan module provides personalized embryo evaluation results based on the individual differences in age, genetic background and medical history of the patient. It combines the prediction results of the AI ​​model with the individual characteristics of the patient to recommend personalized treatment plans, predict the possible efficacy of different treatment plans, and help doctors choose the best plan;

[0019] Multi-language support and localization module, providing a multi-language user interface to facilitate users in different countries and regions. It adjusts system content and functions according to the laws, regulations, cultural habits and clinical guidelines of different regions, integrates automatic translation tools, and supports users to switch between different languages.

[0020] User training and support module, which provides online training courses to help users become familiar with system functions and operating procedures, establish detailed help documents and FAQs, provide instant help, provide customer support services, and answer questions users encounter during use;

[0021] Data backup and recovery module: regularly back up system data to ensure data security and integrity, establish a disaster recovery mechanism, quickly restore data when system failure or data loss occurs, and archive historical data for easy subsequent query and analysis;

[0022] Performance monitoring and optimization module: monitors system performance indicators in real time, analyzes system performance bottlenecks, makes optimization suggestions, optimizes the use of system resources, and improves the overall performance of the system;

[0023] The data acquisition and preprocessing module uses a high-resolution microscope, a phase contrast microscope and a fluorescence microscope to obtain image data of the embryo, including different developmental stages, fertilized eggs, cleavage stages and blastocyst stages, and obtains the patient's clinical data, age, hormone levels and medical history, gene sequencing results and video data of embryo development in a time-difference incubator;

[0024] The multimodal data fusion module aligns image data, gene data and clinical data from different sources in time and space to ensure data consistency, uses timestamps and spatial coordinates for data alignment, converts data in different formats into a unified format to facilitate subsequent processing, stores the integrated data in a database to support fast retrieval and access, uses a deep learning model to extract feature vectors of images, uses gene expression analysis technology to extract feature vectors of genes, fuses feature vectors of different modalities to form a comprehensive feature vector, and uses a feature selection algorithm to improve the prediction ability of the model.

[0025] Preferably, the data acquisition and preprocessing module obtains an image sequence of the embryo through a microscope and a camera, obtains continuous video data of embryo development using time-difference culture technology, extracts the patient's clinical data from the hospital's electronic medical record system, obtains genetic data from a gene sequencing device or database, applies Gaussian filtering to remove image noise, adjusts brightness and contrast, and applies histogram equalization and other methods to enhance image quality, adjusts all images to the same size, resolution and format, removes blurred, occluded and other poor quality images, as well as incomplete or erroneous data records, and has experienced embryologists annotate the image data, and the annotation content includes the quality grade of the embryo, the development stage and whether there are abnormalities.

[0026] Preferably, the model training module selects a suitable machine learning or deep learning model for training, trains the model using a training set, adjusts model parameters to minimize the loss function, and evaluates the model using a validation set and a test set to ensure the generalization ability and stability of the model.

[0027] Preferably, for the gradient-based interpretation method, the model interpretability module needs to perform forward propagation and gradient calculation of the model, input the input embryo image data into the trained deep learning model, and obtain the output embryo quality score of the model, and its algorithm formula is as follows:

[0028] y=f(x;θ)

[0029] Among them, y is the output of the model, x is the input data, and θ is the model parameter;

[0030] Calculate the gradient of the model output relative to the input data to determine which input features have the greatest impact on the output. The algorithm formula is as follows:

[0031]

[0032] According to the selected explanatory method, the corresponding explanation results are generated, and the gradient of each feature map relative to the model output is calculated. The algorithm formula is as follows:

[0033]

[0034] in, is the gradient weight of category c for the kth feature map, y c is the output of category c, is the activation value of the kth feature map at position (i, j), and Z is the normalization factor;

[0035] Calculate the weighted feature map, multiply the gradient weight by the feature map to obtain the weighted feature map. The algorithm formula is as follows:

[0036]

[0037] in, is the Grad-CAM heatmap of category c, and ReLU is the linear rectification function.

[0038] Preferably, the input data is locally perturbed to generate multiple perturbation samples, model prediction is performed on the perturbed samples to obtain prediction results, a linear regression model is used to fit the relationship between the perturbation samples and the prediction results, the local behavior of the model is explained, and an explanation of the model prediction is generated based on the weights of the linear regression model. The contribution of each feature to the model output is calculated, and the Shapley value is used for attribution. The explanation results are visualized to intuitively show which areas have the greatest impact on the model prediction results, and a report containing the explanation results is automatically generated for reference by clinicians.

[0039] Preferably, the clinical verification and evaluation module verifies the effect of the AI ​​system in improving the accuracy of embryo selection, improving the pregnancy rate and reducing the miscarriage rate, the age range and embryo quality of patients receiving in vitro fertilization treatment, patients with certain genetic diseases or other medical conditions that affect pregnancy, and randomly divides patients who meet the inclusion criteria into an experimental group and a control group, using the AI ​​system for embryo selection and a control group, using traditional methods for embryo selection, submitting the clinical trial plan to the ethics committee, and only after approval can the trial be carried out, the patient's age, BMI, cause of infertility, previous medical history and treatment plan, embryo image data, developmental stage, morphological characteristics, AI system evaluation results and embryologist evaluation results, pregnancy outcome data collection data content, pregnancy rate, miscarriage rate and live birth rate, time point within a certain period of time after embryo transfer, follow-up at 12 weeks, 24 weeks and 36 weeks, collect pregnancy results, perform descriptive statistical analysis on baseline data, embryo data and pregnancy outcome data, mean, standard deviation, frequency and percentage data.

[0040] Preferably, based on the differences in pregnancy rate, miscarriage rate, live birth rate, etc. between the comparative test group and the control group, the Kaplan-Meier curve and the Cox proportional hazard model are used to analyze the duration of pregnancy and the risk of miscarriage after embryo transfer, and the sensitivity recall rate and 1-specific false positive rate curves at different thresholds are plotted to evaluate the overall performance of the model. The larger the AUC, the better the model performance. Combined with the statistical analysis results, the significance and potential impact of the AI ​​system in clinical applications are explained, and the safety of the AI ​​system in practical applications is evaluated, including the risk of misdiagnosis and missed diagnosis. The report writing content includes research background, methods, results, discussion and conclusions.

[0041] (III) Beneficial effects

[0042] The present invention provides an automated embryo morphology assessment system based on artificial intelligence image recognition. It has the following beneficial effects:

[0043] 1. In the present invention, through the multimodal data fusion module, data from different sources, embryo images, genetic data, patient medical history and hormone levels are integrated to form a comprehensive data set. The diversity of multimodal data is used to enhance data and improve the generalization ability of the model. By combining genetic data and image data, the developmental potential and health status of the embryo can be more comprehensively evaluated. Data from different sources are standardized to ensure the consistency of data formats and standards, which is convenient for subsequent fusion and analysis.

[0044] 2. In the present invention, Grad-CAM, SHAP and LIME interpretable AI technologies are used to provide explanations for model predictions. By calculating gradient weights, heat maps are generated to intuitively show which areas in the embryo image have the greatest impact on the prediction results, calculate the Shapley value of each feature, quantify the contribution of each feature to the model output, perform local perturbations on the input data, train linear models, and explain the local behavior of the model. In addition to the prediction results of the AI ​​model, detailed explanatory information is also provided in the embryo assessment report, which morphological features have the greatest impact on the prediction results, to help clinicians make more informed decisions.

[0045] 3. In the present invention, by cooperating with medical institutions, large-scale, randomized controlled clinical trials are carried out to verify the effectiveness of the AI ​​system in a real clinical environment, and long-term follow-up studies are conducted to evaluate the long-term effects and safety of the AI ​​system, including duration of pregnancy, miscarriage rate, live birth rate and infant health status. The follow-up data are statistically analyzed to evaluate the long-term impact of the AI ​​system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0047] 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.

[0048] Embodiment 1:

[0049] like Figure 1 As shown, an embodiment of the present invention provides an automated embryo morphology assessment system based on artificial intelligence image recognition, comprising:

[0050] The data acquisition and preprocessing module obtains embryo images and related information from different types of equipment such as microscopes, B-ultrasound and gene sequencers, removes noise data, duplicate data and incomplete data, ensures data quality, and converts data from different sources into a unified format and standard for subsequent processing;

[0051] The multimodal data fusion module integrates image data, genetic data and data from different sources of patient medical history, extracts features from data of different modalities, and performs feature fusion to improve the prediction ability of the model, establish an efficient data storage system, and support the storage and rapid retrieval of large-scale data;

[0052] Model training and optimization module: select deep learning models for training, predict embryo quality and developmental potential, optimize model hyperparameters through grid search, random search and Bayesian optimization to improve model performance, and evaluate the model using validation and test sets to ensure the generalization ability and stability of the model;

[0053] The user interface and interaction module provides a simple and intuitive user interface to facilitate clinicians to use the AI ​​system for embryo evaluation. It provides real-time feedback and suggestions based on the prediction results of the AI ​​model to help doctors make more accurate decisions and support interactive operations.

[0054] Security and privacy protection module: encrypts sensitive data to prevent data leakage, implements strict access control measures to ensure that only authorized personnel can access patient data, and ensures that the system complies with relevant laws, regulations and ethical standards;

[0055] Model interpretability module, which provides explanations for model predictions to help clinicians understand the basis for AI decisions. It develops visualization tools to intuitively display the model's decision-making process, such as using heat maps to show which areas in embryo images have the greatest impact on the prediction results. It also uses Grad-CAM interpretable AI technology to provide explanations for the model's internal mechanisms.

[0056] The clinical validation and evaluation module works with medical institutions to design and implement large-scale, randomized controlled clinical trials to verify the effectiveness of the AI ​​system in real clinical settings, collect data from clinical trials, and conduct statistical analysis to evaluate the effectiveness of the AI ​​system in improving pregnancy rates and reducing miscarriage rates, and conduct long-term follow-up studies to evaluate the long-term effects and safety of the AI ​​system.

[0057] The system integration and deployment module integrates various modules to form a complete system, establishes CI / CD processes, realizes the automated construction, testing and deployment of the system, monitors the system in real time, and promptly discovers and resolves problems in system operation;

[0058] The knowledge base and decision support module establishes a database containing embryological knowledge, clinical guidelines, research literature, and historical cases to provide reference for the AI ​​model. Based on the prediction results of the AI ​​model and the information in the knowledge base, it provides detailed decision support and suggestions for clinicians. The knowledge base is updated regularly to ensure that it contains the latest research results and clinical guidelines.

[0059] The real-time monitoring and early warning module monitors the embryonic development process in real time and detects abnormalities in a timely manner. When an abnormality is detected, the early warning mechanism is automatically triggered to remind clinicians to intervene. Combined with historical data, it analyzes the trends and laws of embryonic development to improve the accuracy of early warning.

[0060] The personalized treatment plan module provides personalized embryo evaluation results based on the individual differences in age, genetic background and medical history of the patient. It combines the prediction results of the AI ​​model with the individual characteristics of the patient to recommend personalized treatment plans, predict the possible efficacy of different treatment plans, and help doctors choose the best plan;

[0061] Multi-language support and localization module, providing a multi-language user interface to facilitate users in different countries and regions. It adjusts system content and functions according to the laws, regulations, cultural habits and clinical guidelines of different regions, integrates automatic translation tools, and supports users to switch between different languages.

[0062] User training and support module, which provides online training courses to help users become familiar with system functions and operating procedures, establish detailed help documents and FAQs, provide instant help, provide customer support services, and answer questions users encounter during use;

[0063] Data backup and recovery module: regularly back up system data to ensure data security and integrity, establish a disaster recovery mechanism, quickly restore data when system failure or data loss occurs, and archive historical data for easy subsequent query and analysis;

[0064] Performance monitoring and optimization module: monitors system performance indicators in real time, analyzes system performance bottlenecks, makes optimization suggestions, optimizes the use of system resources, and improves the overall performance of the system;

[0065] The data acquisition and preprocessing module uses high-resolution microscopes, phase contrast microscopes, and fluorescence microscopes to obtain image data of embryos, including different developmental stages, fertilized eggs, cleavage stages, and blastocyst stages, as well as the patient's clinical data, age, hormone levels, and medical history, gene sequencing results, and video data of embryo development in a time-difference incubator;

[0066] The multimodal data fusion module aligns image data, genetic data and clinical data from different sources in time and space to ensure data consistency, uses timestamps and spatial coordinates for data alignment, converts data in different formats into a unified format to facilitate subsequent processing, stores the integrated data in a database to support fast retrieval and access, uses deep learning models to extract feature vectors of images, uses gene expression analysis technology to extract feature vectors of genes, fuses feature vectors of different modalities to form a comprehensive feature vector, and uses feature selection algorithms to improve the predictive ability of the model.

[0067] The data acquisition and preprocessing module obtains embryo image sequences through microscopes and cameras, uses time-difference culture technology to obtain continuous video data of embryo development, extracts patients' clinical data from the hospital's electronic medical record system, obtains genetic data from gene sequencing equipment or databases, applies Gaussian filtering to remove image noise, adjusts brightness and contrast, and applies histogram equalization and other methods to enhance image quality. All images are adjusted to the same size, resolution and format, and blurred, occluded and other poor quality images, as well as incomplete or erroneous data records are removed. Experienced embryologists annotate the image data, and the annotation content includes the quality grade of the embryo, the development stage and whether there are abnormalities. Through the multimodal data fusion module, data from different sources, including embryo images, genetic data, patient medical history and hormone levels, are integrated to form a comprehensive data set. The diversity of multimodal data is used to enhance data and improve the generalization ability of the model. Combining genetic data and image data can more comprehensively evaluate the developmental potential and health status of the embryo. Data from different sources are standardized to ensure the consistency of data formats and standards, which is convenient for subsequent fusion and analysis.

[0068] The model training module selects a suitable machine learning or deep learning model for training, trains the model using a training set, adjusts model parameters to minimize the loss function, and evaluates the model using a validation set and a test set to ensure the generalization ability and stability of the model.

[0069] Model interpretability module For the gradient-based interpretation method, it is necessary to perform forward propagation and gradient calculation of the model, input the input embryo image data into the trained deep learning model, and obtain the output embryo quality score of the model. The algorithm formula is as follows:

[0070] y=f(x;θ)

[0071] Among them, y is the output of the model, x is the input data, and θ is the model parameter;

[0072] Calculate the gradient of the model output relative to the input data to determine which input features have the greatest impact on the output. The algorithm formula is as follows:

[0073]

[0074] According to the selected explanatory method, the corresponding explanation results are generated, and the gradient of each feature map relative to the model output is calculated. The algorithm formula is as follows:

[0075]

[0076] in, is the gradient weight of category c for the kth feature map, y c is the output of category c, is the activation value of the kth feature map at position (i, j), and Z is the normalization factor;

[0077] Calculate the weighted feature map, multiply the gradient weight by the feature map to obtain the weighted feature map. The algorithm formula is as follows:

[0078]

[0079] in, is the Grad-CAM heatmap of category c, and ReLU is the linear rectification function.

[0080] The input data is locally perturbed to generate multiple perturbed samples. The perturbed samples are used for model prediction to obtain the prediction results. The linear regression model is used to fit the relationship between the perturbed samples and the prediction results. The local behavior of the model is explained. According to the weights of the linear regression model, an explanation of the model prediction is generated. The contribution of each feature to the model output is calculated. The Shapley value is used for attribution. The explanation results are visualized to intuitively show which areas have the greatest impact on the model prediction results. A report containing the explanation results is automatically generated for reference by clinicians. Grad-CAM, SHAP and LIME interpretable AI technologies are used to provide explanations for model predictions. By calculating the gradient weights, a heat map is generated to intuitively show which areas in the embryo image have the greatest impact on the prediction results. The Shapley value of each feature is calculated to quantify the contribution of each feature to the model output. The input data is locally perturbed, a linear model is trained, and the local behavior of the model is explained. In addition to the prediction results of the AI ​​model, detailed explanation information is also provided in the embryo assessment report, which morphological features have the greatest impact on the prediction results to help clinicians make more informed decisions.

[0081] The clinical validation and evaluation module verifies the effectiveness of the AI ​​system in improving the accuracy of embryo selection, increasing pregnancy rate and reducing miscarriage rate, the age range and embryo quality of patients undergoing in vitro fertilization treatment, patients with certain genetic diseases or other medical conditions that affect pregnancy, and patients who meet the inclusion criteria are randomly divided into an experimental group and a control group, using the AI ​​system for embryo selection and a control group, using traditional methods for embryo selection, submitting the clinical trial plan to the ethics committee, and only after approval can the trial be carried out, the patient's age, BMI, cause of infertility, previous medical history and treatment plan, embryo image data, developmental stage, morphological characteristics, AI system evaluation results and embryologist evaluation results, pregnancy outcome data collection data content, pregnancy rate, miscarriage rate and live birth rate, time point within a certain period of time after embryo transfer, follow-up at 12 weeks, 24 weeks and 36 weeks, collect pregnancy results, perform descriptive statistical analysis on baseline data, embryo data and pregnancy outcome data, mean, standard deviation, frequency and percentage data.

[0082] According to the differences in pregnancy rate, miscarriage rate, live birth rate, etc. between the experimental group and the control group, the Kaplan-Meier curve and Cox proportional hazard model were used to analyze the duration of pregnancy and the risk of miscarriage after embryo transfer, and the sensitivity recall rate and 1-specific false positive rate curves at different thresholds were plotted to evaluate the overall performance of the model. The larger the AUC, the better the model performance. Combined with the statistical analysis results, the significance and potential impact of the AI ​​system in clinical applications were explained, and the safety of the AI ​​system in practical applications was evaluated, including the risk of misdiagnosis and missed diagnosis. The report writing content includes research background, methods, results, discussion and conclusion. Through cooperation with medical institutions, large-scale, randomized controlled clinical trials were carried out to verify the effectiveness of the AI ​​system in a real clinical environment, and long-term follow-up studies were conducted to evaluate the long-term effect and safety of the AI ​​system, including the duration of pregnancy, miscarriage rate, live birth rate and infant health status. The follow-up data were statistically analyzed to evaluate the long-term impact of the AI ​​system.

[0083] 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. Embryo morphology automated assessment system based on artificial intelligence image recognition, characterized by: include: AI system, data acquisition and preprocessing module, multimodal data fusion module, model training and optimization module, user interface and interaction module, security and privacy protection module, model interpretability module, clinical verification and evaluation module, system integration and deployment module, knowledge base and decision support module, real-time monitoring and early warning module, personalized treatment plan module, multilingual support and localization module, user training and support module, data backup and recovery module and performance monitoring and optimization module; The data acquisition and preprocessing module uses a high-resolution microscope, a phase contrast microscope and a fluorescence microscope to obtain image data of the embryo, including different developmental stages, fertilized eggs, cleavage stages and blastocyst stages, and obtains the patient's clinical data, age, hormone levels and medical history, gene sequencing results and video data of embryo development in a time-difference incubator; The multimodal data fusion module aligns image data, gene data and clinical data from different sources in time and space to ensure data consistency, uses timestamps and spatial coordinates for data alignment, converts data in different formats into a unified format to facilitate subsequent processing, stores the integrated data in a database to support fast retrieval and access, uses a deep learning model to extract feature vectors of images, uses gene expression analysis technology to extract feature vectors of genes, fuses feature vectors of different modalities to form a comprehensive feature vector, and uses a feature selection algorithm to improve the prediction ability of the model.

2. The embryo morphology automated assessment system based on artificial intelligence image recognition according to claim 1, characterized in that: The data acquisition and preprocessing module obtains an image sequence of the embryo through a microscope and a camera, obtains continuous video data of embryo development using time-difference culture technology, extracts the patient's clinical data from the hospital's electronic medical record system, obtains genetic data from a genetic sequencing device or a database, applies Gaussian filtering to remove image noise, adjusts brightness and contrast, and applies histogram equalization and other methods to enhance image quality, adjusts all images to the same size, resolution and format, removes blurred, occluded and other poor quality images, as well as incomplete or erroneous data records, and has experienced embryologists annotate the image data, and the annotation content includes the quality grade of the embryo, the development stage and whether there are abnormalities.

3. The embryo morphology automated assessment system based on artificial intelligence image recognition according to claim 1, characterized in that: The model training module selects a suitable machine learning or deep learning model for training, trains the model using a training set, adjusts model parameters to minimize the loss function, and evaluates the model using a validation set and a test set to ensure the generalization ability and stability of the model.

4. The embryo morphology automated assessment system based on artificial intelligence image recognition according to claim 1, characterized in that: For the gradient-based interpretation method, the model interpretation module needs to perform forward propagation and gradient calculation of the model, input the input embryo image data into the trained deep learning model, and obtain the output embryo quality score of the model. The algorithm formula is as follows: y=f(x;θ) Among them, y is the output of the model, x is the input data, and θ is the model parameter; Calculate the gradient of the model output relative to the input data to determine which input features have the greatest impact on the output. The algorithm formula is as follows: According to the selected explanatory method, the corresponding explanation results are generated, and the gradient of each feature map relative to the model output is calculated. The algorithm formula is as follows: in, is the gradient weight of category c for the kth feature map, y c is the output of category c, is the activation value of the kth feature map at position (i, j), and Z is the normalization factor; Calculate the weighted feature map, multiply the gradient weight by the feature map to obtain the weighted feature map. The algorithm formula is as follows: in, is the Grad-CAM heatmap of category c, and ReLU is the linear rectification function.

5. The embryo morphology automated assessment system based on artificial intelligence image recognition according to claim 4, characterized in that: The input data is locally perturbed to generate multiple perturbed samples. Model prediction is performed on the perturbed samples to obtain the prediction results. The linear regression model is used to fit the relationship between the perturbed samples and the prediction results. The local behavior of the model is explained. According to the weights of the linear regression model, an explanation of the model prediction is generated. The contribution of each feature to the model output is calculated, and the Shapley value is used for attribution. The explanation results are visualized to intuitively show which areas have the greatest impact on the model prediction results. A report containing the explanation results is automatically generated for reference by clinicians.

6. The embryo morphology automated assessment system based on artificial intelligence image recognition according to claim 1, characterized in that: The clinical validation and evaluation module verifies the effect of the AI ​​system in improving the accuracy of embryo selection, increasing pregnancy rate and reducing miscarriage rate, the age range and embryo quality of patients receiving in vitro fertilization treatment, patients with certain genetic diseases or other medical conditions that affect pregnancy, and randomly divides patients who meet the inclusion criteria into an experimental group, using the AI ​​system for embryo selection and a control group, using traditional methods for embryo selection, submitting the clinical trial plan to the ethics committee, and only after approval can the trial be carried out, the patient's age, BMI, cause of infertility, previous medical history and treatment plan, embryo image data, developmental stage, morphological characteristics, AI system evaluation results and embryologist evaluation results, pregnancy outcome data collection data content, pregnancy rate, miscarriage rate and live birth rate, time point within a certain period of time after embryo transfer, follow-up at 12 weeks, 24 weeks and 36 weeks, collect pregnancy results, perform descriptive statistical analysis on baseline data, embryo data and pregnancy outcome data, mean, standard deviation, frequency and percentage data.

7. The embryo morphology automated assessment system based on artificial intelligence image recognition according to claim 6, characterized in that: According to the differences in pregnancy rate, miscarriage rate, live birth rate, etc. between the comparative test group and the control group, the Kaplan-Meier curve and Cox proportional hazard model were used to analyze the duration of pregnancy and the risk of miscarriage after embryo transfer, and the sensitivity recall rate and 1-specific false positive rate curves at different thresholds were plotted to evaluate the overall performance of the model. The larger the AUC, the better the model performance. Combined with the statistical analysis results, the significance and potential impact of the AI ​​system in clinical applications were explained, and the safety of the AI ​​system in practical applications was evaluated, including the risk of misdiagnosis and missed diagnosis. The report writing content includes research background, methods, results, discussion and conclusion.

Images