A personalized dynamic embryo culture system based on AI feedback

Abstract

This invention discloses an AI-based personalized dynamic embryo culture system, belonging to the field of embryo culture technology. The system adopts a modular architecture, including a core control module, at least one independent culture chamber module, an AI analysis module, and a dynamic regulation module. By employing microfluidic chip technology and an integrated microsensor array, this invention achieves high-precision real-time monitoring of embryo morphology images and physiological parameters of the culture medium. Combined with multimodal data fusion analysis using the AI ​​analysis module, it significantly improves the accuracy and reliability of embryo developmental potential assessment, thereby contributing to increased rates of high-quality embryos and clinical pregnancy success. Furthermore, this invention, through a built-in personalized protocol generation unit, can dynamically generate and execute personalized culture protocols based on the patient's age, ovarian function indicators, and basal hormone levels, while utilizing the dynamic regulation module to simulate the uterine diurnal rhythm for temperature fluctuation control.

Description

Technical Field

[0001] This invention relates to the field of embryo culture technology, and more specifically, to a personalized dynamic embryo culture system based on AI feedback. Background Technology

[0002] In vitro fertilization (IVF) and embryo transfer (IVT) technology has become an important means of treating infertility. The success rate of this technology largely depends on the quality and stability of embryo culture in vitro. Currently, most mainstream embryo culture systems use static culture methods, that is, culturing embryos in a culture medium with fixed components in an incubator with constant temperature, humidity, and gas concentration. While such systems can provide a relatively stable basic culture environment, they cannot simulate the dynamic and personalized physiological conditions experienced by embryos during in vivo development, such as maternal circadian rhythms, real-time clearance of metabolic waste, and adjustments to the culture medium composition tailored to different patient characteristics. Furthermore, traditional embryo assessment mainly relies on embryologists' morphological observations under a microscope at specific time points. The assessment results are easily influenced by subjective experience and cannot achieve continuous and quantitative monitoring of the entire embryonic development process.

[0003] Existing technologies have the following main drawbacks: First, the culture environment is singular and static, making it impossible to dynamically and personally regulate the embryo based on its actual developmental state, resulting in the embryo's developmental potential not being fully stimulated. Second, embryo assessment relies on manual, static observation, lacking high-precision, continuous, and objective means of assessing developmental potential, which affects the accuracy and consistency of high-quality embryo selection. Third, there is a lack of real-time monitoring and feedback regulation capabilities for the culture medium microenvironment, and the accumulation of metabolic waste may have toxic effects on the embryo, and it is difficult to accurately adapt the culture medium composition to different patient characteristics (such as age, ovarian function, etc.). Finally, the system's scalability and intelligence level are limited, making it difficult to integrate multimodal data for fusion analysis and achieve rapid, closed-loop regulation and feedback, while also facing challenges in data security and cross-center collaboration.

[0004] Based on this, the present invention designs an AI-based personalized dynamic embryo culture system to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based personalized dynamic embryo culture system to solve the problems mentioned in the background art.

[0006] A personalized dynamic embryo culture system based on AI feedback. The system adopts a modular architecture, including a core control module, at least one independent culture chamber module, an AI analysis module, and a dynamic regulation module. The independent culture chamber module is constructed using microfluidic chip technology and integrates a group of microsensors to monitor the morphological images of the embryo and the physiological parameters of the culture medium. The AI ​​analysis module is built on a lightweight Transformer architecture and is used to fuse the morphological images and physiological parameters for multimodal analysis, assess the embryonic developmental potential and generate regulatory instructions. The dynamic control module receives the control command and uses the microfluidic unit to periodically adjust the culture conditions in the culture chamber and directionally remove metabolic waste. The system also includes an edge computing unit for local processing of AI analysis tasks to achieve second-level feedback; The core control module supports data interface with the hospital's reproductive medicine management system and has a built-in customized unit that generates personalized culture programs based on individual patient clinical characteristics.

[0007] Preferably, the AI ​​analysis module is trained and optimized using multi-center desensitized data through federated learning technology, and integrates a manual correction feedback mechanism, allowing embryologists to review the AI ​​evaluation results and correct the parameters. The dynamic regulation module incorporates an embryonic development elasticity coefficient into its regulation algorithm, allowing the embryonic development rhythm to fluctuate within a preset range, and can activate a personalized regulation mode that monitors but does not intervene when a rare developmental pattern is identified. The microsensor group includes a calibration-free, long-life pH electrode and a stable gas sensor for continuous monitoring of the culture environment; The independent culture chamber module features a fully enclosed physical isolation design and integrates a microbial contamination sensor and a one-button disinfection unit combining ultraviolet light and high-temperature steam to achieve contamination control and rapid maintenance.

[0008] Preferably, the system adopts a combination of core modules and optional expansion modules. The basic version provides single-compartment culture and basic AI monitoring functions, while the advanced version supports multi-compartment parallel culture and advanced metabolic analysis functions through expansion modules. The microfluidic unit enables the recycling of micro-volume culture medium and is compatible with standardized culture media from different brands. The customization unit can dynamically optimize the concentration of specific components in the culture medium based on the patient's age, ovarian function, and hormone level data. The system's user interface is a visual touch interface with built-in standardized process guidance and integrates remote diagnostic functions.

[0009] Preferably, the specific process of the dynamic control module performing the periodic adjustment is as follows: based on the algorithm simulating the circadian rhythm in the uterus, the culture temperature is fluctuated and controlled within a 24-hour cycle, wherein the first temperature is maintained during the daytime stage and the second temperature, which differs from the first temperature by 0.1-0.3℃, is automatically switched during the nighttime stage. The targeted removal of metabolic waste is achieved by the microsensor group monitoring the concentration of lactic acid and / or ammonia in the culture medium in real time. When the concentration exceeds a preset threshold, a microfluidic flushing procedure is initiated to selectively remove metabolic waste while preserving the nutrient microenvironment around the embryo.

[0010] Preferably, the logic for the AI ​​analysis module to generate control instructions further includes: setting a dynamic embryonic development elasticity coefficient for each embryo, which defines the allowable time deviation range of the embryo at key developmental time points; When the actual developmental rhythm of the embryo is within the allowable time deviation range, the dynamic control module maintains the current culture conditions unchanged. Only when the developmental deviation of the embryo exceeds the range defined by the elasticity coefficient will the AI ​​analysis module generate interventional control instructions.

[0011] Preferably, the microfluidic unit realizes the recycling of micro-volume culture medium, and its specific configuration is: a micro-circulation pipeline connected to the independent culture chamber module, which integrates a selective filter membrane for separating metabolic waste and a micro-quantitative pump for replenishing nutrients. The system controls the consumption of culture medium for a single embryo between 0.1 mL and 0.15 mL through the micro-circulation pipeline.

[0012] Preferably, the AI ​​analysis module of the system further includes a personalized solution generation unit, which is configured to perform the following operations: acquire the patient's clinical individual characteristic data, including age, ovarian function indicators and basal hormone levels; associate the clinical individual characteristic data with the initial state data of the embryo; run the AI ​​sub-model; dynamically generate a personalized culture medium composition adjustment instruction based on the associated data; wherein the adjustment instruction includes at least an optimized setting of the concentration of antioxidant components or the proportion of insulin-like growth factor in the culture medium; and send the culture medium composition adjustment instruction to the dynamic control module to control the microfluidic unit to adjust the culture medium composition of the specified culture chamber in real time.

[0013] Preferably, the embryonic developmental elasticity coefficient is dynamically calculated and generated by the AI ​​analysis module for each embryo. The calculation process includes: analyzing the historical developmental time series data of the embryo and comparing it with the standard developmental curve; extracting specific factors from the patient's clinical individual characteristic data, including the patient's age and fertilization method; and calculating the personalized time tolerance window of the embryo at each subsequent key developmental point through a pre-trained regression model based on the comparison results of the historical developmental time series data and the patient's specific factors. This time tolerance window is defined as the embryonic developmental elasticity coefficient.

[0014] Preferably, the microsensor group further includes a metabolite concentration detection unit, which is communicatively connected to the microfluidic unit to form a closed-loop feedback subsystem. The metabolite concentration detection unit is configured to monitor the concentration of at least one metabolic waste marker in the culture medium in real time. The microfluidic unit is configured to, in response to the monitoring result of the metabolite concentration detection unit, perform at least one of the following operations: initiate the microfluidic flushing procedure, or adjust the flow rate of the culture medium to change the local metabolite diffusion efficiency, or activate an adsorbent module integrated in the circulation pipeline to selectively adsorb metabolic waste.

[0015] Compared with the prior art, the advantages of this invention are: 1. This invention achieves high-precision real-time monitoring of embryo morphology images and culture medium physiological parameters by employing microfluidic chip technology and integrated micro-sensor groups. Combined with an AI analysis module based on a lightweight Transformer architecture for multimodal data fusion analysis, it significantly improves the accuracy and reliability of embryo development potential assessment, thereby helping to increase the rate of high-quality embryos and clinical pregnancy success rate.

[0016] 2. This invention, through its built-in personalized protocol generation unit, can dynamically generate and execute personalized culture protocols based on the patient's age, ovarian function indicators, and baseline hormone levels, among other clinical characteristics. Simultaneously, it utilizes a dynamic regulation module to simulate the uterine diurnal rhythm for temperature fluctuation control, and achieves targeted removal of metabolic waste and precise maintenance of the nutrient microenvironment, providing each embryo with a highly personalized dynamic culture environment that closely mimics the physiological state in vivo.

[0017] 3. By adopting a modular architecture and edge computing units, this invention achieves localized processing of AI analysis tasks and second-level feedback control, significantly improving the system's response speed and stability. The microfluidic circulation system controls the consumption of a single embryo culture medium to 0.1-0.15 mL, significantly reducing consumable costs. Furthermore, through a fully enclosed design, contamination sensors, and a combined disinfection unit, a complete contamination prevention and control system is constructed, ensuring the safety and reliability of the system's long-term operation. Attached Figure Description

[0018] Figure 1 This is a system architecture diagram of an AI-based personalized dynamic embryo culture system proposed in this invention. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 A personalized dynamic embryo culture system based on AI feedback is proposed. The system adopts a modular architecture, including a core control module, at least one independent culture chamber module, an AI analysis module, and a dynamic regulation module. The core control module adopts an embedded Linux system and is equipped with ROS (Robot Operating System) middleware to realize asynchronous communication and task scheduling between modules.

[0021] The independent culture chamber module is constructed using microfluidic chip technology and integrates a set of micro-sensors to monitor the morphological images of the embryos and the physiological parameters of the culture medium. The microfluidic chip of the independent culture chamber module is made of PDMS (polydimethylsiloxane) material, which has biocompatibility and gas permeability. The chip size is 20mm×20mm, and each chamber has a volume of 5μL.

[0022] The AI ​​analysis module is built on a lightweight Transformer architecture and is used to fuse morphological images and physiological parameters for multimodal analysis, assess embryonic developmental potential and generate regulatory instructions. As a further refinement of this embodiment, the multimodal fusion process of the AI ​​analysis module is implemented using a lightweight Transformer network. Specifically, the morphological image feature vectors of the embryo are... Physiological parameter eigenvectors of the culture medium The features are fused to generate a joint feature vector. The core attention fusion formula is as follows: ; in, For querying the matrix, by Obtained through linear transformation; For the key-value matrix, by Obtained through linear transformation; This is a scaling factor. This mechanism adaptively assigns weights to features of different modalities, thereby improving the accuracy of developmental potential assessment.

[0023] The dynamic control module receives control commands and uses the microfluidic unit to periodically adjust the culture conditions in the culture chamber and directionally remove metabolic waste. As a further refinement of this embodiment, the dynamic control module performs periodic temperature adjustments based on an algorithm that simulates the circadian rhythm within the uterus. This algorithm uses the following periodic function to achieve fluctuating control of the culture temperature: ; in, This is the base temperature during the day; The diurnal temperature range is 0.1–0.3℃. Current time (hours); This is a phase offset used to align with the patient's individualized sleep schedule. In this way, the culture environment is closer to the physiological state.

[0024] Furthermore, the targeted removal of metabolic waste employs a dynamic threshold triggering mechanism. The threshold for metabolic waste removal... It is not a fixed value, but is dynamically adjusted according to the cultivation stage: ; in, Basic threshold; For training time; These are empirical parameters used to simulate the adaptive changes in embryonic metabolism over time. The concentrations of lactic acid or ammonia are monitored in real-time by the microsensor array. At that time, the system automatically triggers the microfluidic flushing procedure.

[0025] The system also includes an edge computing unit for localized processing of AI analysis tasks to achieve second-level feedback; the edge computing unit uses a Jetson Nano module and is equipped with the TensorRT acceleration library to achieve real-time inference of Transformer models with a response time of ≤500ms.

[0026] The core control module supports data integration with the hospital's reproductive medicine management system and has a built-in customized unit that generates personalized culture programs based on individual patient clinical characteristics.

[0027] The AI ​​analysis module utilizes federated learning technology to train and optimize using multi-center anonymized data, and integrates a manual correction feedback mechanism, allowing embryologists to review and correct the AI ​​assessment results. The federated learning training process is as follows: each round of training involves local updates of model parameters by each reproductive center, with only encrypted gradients uploaded to the aggregation server to ensure data privacy. The manual correction feedback mechanism specifically includes an expert review button on the system interface, allowing embryologists to adjust the developmental score generated by the AI ​​by 1-5 levels. The system records the corrections and uses them for subsequent incremental learning.

[0028] The dynamic regulation module's regulation algorithm incorporates an embryonic development elasticity coefficient, allowing the embryonic development rhythm to fluctuate within a preset range, and can activate a personalized regulation mode that only monitors and does not intervene when rare developmental patterns are identified. As a further refinement of this embodiment, the elasticity coefficient of embryonic development The data is dynamically calculated and generated for each embryo by the AI ​​analysis module. The calculation process is based on the deviation between historical developmental timeline data and standard developmental curves, combined with individual patient characteristics, and is achieved through the following regression model: ; in, This represents the average deviation of the embryonic development timeline from the standard curve. Normalized values ​​for patient age; Encode the fertilization method (e.g., IVF=0, ICSI=1); , , These are the model parameters obtained through training on multi-center data. These coefficients... Personalized time tolerance windows for embryos at subsequent key developmental stages were defined. The miniature sensor suite includes a calibration-free, long-life pH electrode and a stable gas sensor for continuous monitoring of the culture environment; The independent culture chamber module features a fully enclosed, physically isolated design and integrates a microbial contamination sensor and a one-button disinfection unit combining ultraviolet light and high-temperature steam to achieve contamination control and rapid maintenance. The microbial contamination sensor uses impedance spectroscopy to detect bacterial proliferation, with a detection limit of 10³ CFU / mL. Once this limit is exceeded, a combined ultraviolet light (wavelength 254 nm) and high-temperature steam (121 °C) disinfection program is triggered, with a disinfection cycle of 15 minutes.

[0029] The system adopts a combination of core modules and optional expansion modules. The basic version provides single-compartment culture and basic AI monitoring functions, while the advanced version supports multi-compartment parallel culture and advanced metabolic analysis functions through expansion modules. The expansion modules are connected to the core module via USB-C interface and support hot-swapping. The advanced version adds a metabolite mass spectrometry analysis unit, which can detect small molecule metabolites such as pyruvate and glutamate in real time.

[0030] The microfluidic unit enables the recycling of trace amounts of culture medium and is compatible with standardized culture media from different brands. As a further refinement of this embodiment, the microfluidic unit follows the following material balance model when recycling trace amounts of culture medium to optimize rinsing and nutrient replenishment strategies: ; in, This refers to the concentration of metabolic waste. For flow rate; This refers to the volume of the culture medium. Input concentration; The overall clearance rate is determined by the selective filtration membrane and adsorbent module integrated within the tubing. Using this model, the system can precisely control the culture medium consumption of a single embryo between 0.1 mL and 0.15 mL.

[0031] The mechanism by which the microfluidic unit adapts to different brands of culture medium is as follows: the system has a built-in database of culture medium components. After the user selects a brand, the system automatically adjusts the flow rate of the micropump and the pore size of the filter membrane to ensure compatibility.

[0032] The customization unit can dynamically optimize the concentration of specific components in the culture medium based on the patient's age, ovarian function, and hormone levels. As a further refinement of this embodiment, when the customization unit generates a personalized culture medium component adjustment instruction, it first calculates an adjustment factor. : ; in, To anti-Müllerian duct hormone levels; The level of follicle-stimulating hormone (FSH); This is the normalized value for age; The weight coefficients are obtained through training with historical data, and the system is based on... The value is used to dynamically adjust the concentration ratio of antioxidant components (such as glutathione) or insulin-like growth factor (IGF) in the culture medium.

[0033] The system's user interface is a visual touch interface with built-in standardized process guidance, and it also integrates remote diagnostic functions.

[0034] The specific process of the dynamic control module performing periodic adjustments is as follows: based on the algorithm that simulates the circadian rhythm in the uterus, the culture temperature is fluctuated and controlled within a 24-hour cycle. During the daytime phase, the first temperature is maintained, and during the nighttime phase, it automatically switches to a second temperature that differs from the first temperature by 0.1-0.3℃. In temperature control algorithms, phase offset The metabolic waste removal threshold can be personalized based on the patient's sleep cycle questionnaire results, within a range of ±2 hours. empirical parameters and The values ​​were set to 0.2 mM and 12 h, respectively, to simulate the metabolic adaptation of the embryo from the early stage to the blastocyst stage.

[0035] The targeted removal of metabolic waste is achieved by a microsensor array that monitors the concentration of lactic acid and / or ammonia in the culture medium in real time. When the concentration exceeds a preset threshold, a microfluidic flushing procedure is initiated to selectively remove metabolic waste while preserving the nutrient microenvironment around the embryo.

[0036] The logic for generating regulatory instructions by the AI ​​analysis module further includes: setting a dynamic embryonic development elasticity coefficient for each embryo, which defines the allowable time deviation range of the embryo at key developmental time points; When the actual developmental rhythm of the embryo is within the allowable time deviation range, the dynamic regulation module maintains the current culture conditions unchanged. Only when the developmental deviation of the embryo exceeds the range defined by the elasticity coefficient will the AI ​​analysis module generate interventional regulation instructions.

[0037] Embryonic developmental elasticity coefficient The unit is hours, and its value range is usually [0.5, 3.0] hours. For example, if a certain embryo's... =1.5 hours, which means that it can tolerate a time deviation of ±1.5 hours at key stages such as gastrulation.

[0038] The microfluidic unit enables the recycling of micro-volume culture medium. Specifically, it consists of a micro-circulation pipeline connected to an independent culture chamber module. This pipeline integrates a selective filtration membrane for separating metabolic waste and a micro-quantitative pump for replenishing nutrients. Through this micro-circulation pipeline, the system controls the consumption of culture medium for a single embryo to be between 0.1 mL and 0.15 mL.

[0039] The microcirculation tubing has an inner diameter of 100μm and adopts a multi-layer composite membrane structure. The innermost layer is a phospholipid double-layer biomimetic coating to reduce embryo adhesion. The selective filtration membrane has a molecular weight cutoff of 500Da, which can effectively intercept lactic acid (molecular weight 90.08) and ammonia (molecular weight 17), but allows nutrients such as glucose to pass through.

[0040] The system's AI analysis module also includes a personalized solution generation unit, which is configured to perform the following operations: acquire the patient's clinical individual characteristic data, including age, ovarian function indicators, and baseline hormone levels; associate the clinical individual characteristic data with the initial state data of the embryo; run the AI ​​sub-model; dynamically generate a personalized culture medium composition adjustment instruction based on the associated data; wherein the adjustment instruction includes at least the optimized setting of the concentration of antioxidant components or the proportion of insulin-like growth factor in the culture medium; and send the culture medium composition adjustment instruction to the dynamic control module to control the microfluidic unit to adjust the culture medium composition of the specified culture chamber in real time.

[0041] The AI ​​sub-model of the personalized treatment plan generation unit uses the XGBoost algorithm. Input features include: patient age, AMH, FSH, AFC (antral follicle count), fertilization method, and initial embryo morphology score. The output is a component adjustment vector. , The system controls a micro metering pump to inject the corresponding concentrate based on this vector.

[0042] The embryonic developmental elasticity coefficient is dynamically calculated and generated for each embryo by the AI ​​analysis module. The calculation process includes: analyzing the historical developmental time series data of the embryo and comparing it with the standard developmental curve; extracting specific factors from the patient's clinical individual characteristic data, including the patient's age and fertilization method; and calculating the personalized time tolerance window of the embryo at each subsequent key developmental point through a pre-trained regression model based on the comparison results of the historical developmental time series data and the patient's specific factors. This time tolerance window is defined as the embryonic developmental elasticity coefficient.

[0043] The regression model training data came from more than 10,000 embryo development timeline records. The model's prediction error (MAE) for elasticity coefficients on the test set was 0.3 hours. Key developmental points included: pronucleus disappearance, 2-cell stage, 4-cell stage, morula stage, and blastocyst formation stage.

[0044] The microsensor assembly also includes a metabolite concentration detection unit, which is communicatively connected to the microfluidic unit to form a closed-loop feedback subsystem. The metabolite concentration detection unit is configured to monitor the concentration of at least one metabolic waste marker in the culture medium in real time. The microfluidic unit is configured to, in response to the monitoring result of the metabolite concentration detection unit, perform at least one of the following operations: initiate a microfluidic flushing procedure, or adjust the flow rate of the culture medium to change the local metabolite diffusion efficiency, or activate an adsorbent module integrated in the circulation pipeline to selectively adsorb metabolic waste.

[0045] The metabolite concentration detection unit uses a microelectrode array combined with cyclic voltammetry to simultaneously detect three metabolites: lactic acid, ammonia, and pyruvate, with a detection frequency of once per minute. The adsorbent module is an activated carbon-molecular sieve composite filter element located in the bypass of the circulation pipeline, which can be started and stopped by a solenoid valve without interrupting the culture.

[0046] To ensure the reliable operation of the system in a clinical environment and the security of sensitive patient information, this invention constructs a multi-layered security protection system: At the hardware and physical levels, each independent culture chamber module adopts a fully enclosed isolation design. Its built-in storage chip temporarily encrypts and caches embryo images and physiological parameters. The data is decrypted only before processing, effectively preventing malicious interception at the edge.

[0047] At the data transmission layer, all communication links between the edge computing unit and the core control module, as well as between the core control module and the hospital reproductive medicine management system, adopt the TLS 1.3-based security protocol and AES-256 encryption algorithm to ensure the confidentiality and integrity of data during transmission and to resist network eavesdropping and man-in-the-middle attacks.

[0048] At the data access and control layer, the system integrates a role-based access control module. The system strictly defines different roles, including embryologists, laboratory administrators, system maintenance engineers, and remote diagnosticians, and assigns differentiated data access and operation permissions to each role. All operations on the system, including reviewing AI evaluation results, adjusting culture parameters, and initiating sterilization procedures, are recorded in real time and generate tamper-proof audit logs, ensuring the traceability of all operations.

[0049] At the privacy-preserving computation and model training layer, the federated learning technique employed by the AI ​​analysis module has been further enhanced. Before the local model parameters from each party are uploaded to the aggregation server, the system uses differential privacy technology to inject rigorously mathematically proven, privacy-budget-compliant parameters into the gradient information. Gaussian noise.

[0050] This mechanism can fundamentally prevent the reverse engineering of original embryo data or patient characteristics through model updates without significantly affecting the model's convergence accuracy. It achieves a secure collaborative model where data remains within the domain and knowledge can be shared, thus meeting the regulatory requirements for medical data protection.

[0051] The working principle of this invention is as follows: Under the aforementioned security and privacy framework, a micro-sensor array integrated within the microfluidic culture chamber first collects real-time morphological images of the embryo and various physiological parameters of the culture medium. Subsequently, an AI analysis module based on a lightweight Transformer architecture fuses and analyzes this multimodal data, assesses the embryo's developmental potential, and generates regulatory instructions. The dynamic control module then, according to these instructions, uses the microfluidic unit to periodically adjust the temperature fluctuations of the culture environment, directionally remove metabolic waste, and dynamically adjust the composition of the culture medium. The entire process incorporates an embryonic developmental elasticity coefficient to achieve tolerance management and relies on an edge computing unit to achieve second-level feedback. Ultimately, through continuous data closure and AI optimization, the system provides each embryo with a highly simulated and customized dynamic culture environment that mimics the physiological state in vivo, thereby significantly improving the quality and success rate of embryonic development.

[0052] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A personalized dynamic embryo culture system based on AI feedback, characterized in that, The system adopts a modular architecture, including a core control module, at least one independent culture chamber module, an AI analysis module, and a dynamic control module; The independent culture chamber module is constructed using microfluidic chip technology and integrates a group of microsensors to monitor the morphological images of the embryo and the physiological parameters of the culture medium. The AI ​​analysis module is built on a lightweight Transformer architecture and is used to fuse the morphological images and physiological parameters for multimodal analysis, assess the embryonic developmental potential and generate regulatory instructions. The dynamic control module receives the control command and uses the microfluidic unit to periodically adjust the culture conditions in the culture chamber and directionally remove metabolic waste. The system also includes an edge computing unit for local processing of AI analysis tasks to achieve second-level feedback; The core control module supports data interface with the hospital's reproductive medicine management system and has a built-in customized unit that generates personalized culture programs based on individual patient clinical characteristics.

2. The personalized dynamic embryo culture system based on AI feedback according to claim 1, characterized in that, The AI ​​analysis module is trained and optimized using multi-center desensitized data through federated learning technology, and integrates a manual correction feedback mechanism, allowing embryologists to review the AI ​​evaluation results and correct the parameters. The dynamic regulation module incorporates an embryonic development elasticity coefficient into its regulation algorithm, allowing the embryonic development rhythm to fluctuate within a preset range, and can activate a personalized regulation mode that monitors but does not intervene when a rare developmental pattern is identified. The microsensor group includes a calibration-free, long-life pH electrode and a stable gas sensor for continuous monitoring of the culture environment; The independent culture chamber module features a fully enclosed physical isolation design and integrates a microbial contamination sensor and a one-button disinfection unit combining ultraviolet light and high-temperature steam to achieve contamination control and rapid maintenance.

3. The personalized dynamic embryo culture system based on AI feedback according to claim 1, characterized in that, The system adopts a combination of core modules and optional expansion modules. The basic version provides single-compartment culture and basic AI monitoring functions, while the advanced version supports multi-compartment parallel culture and advanced metabolic analysis functions through expansion modules. The microfluidic unit enables the recycling of micro-volume culture medium and is compatible with standardized culture media from different brands. The customization unit can dynamically optimize the concentration of specific components in the culture medium based on the patient's age, ovarian function, and hormone level data. The system's user interface is a visual touch interface with built-in standardized process guidance and integrates remote diagnostic functions.

4. The personalized dynamic embryo culture system based on AI feedback according to claim 1, characterized in that, The specific process of the dynamic control module performing the periodic adjustment is as follows: based on the algorithm that simulates the circadian rhythm in the uterus, the culture temperature is fluctuated and controlled within a 24-hour cycle, wherein the first temperature is maintained during the daytime stage and the second temperature, which differs from the first temperature by 0.1-0.3℃, is automatically switched at nighttime. The targeted removal of metabolic waste is achieved by the microsensor group monitoring the concentration of lactic acid and / or ammonia in the culture medium in real time. When the concentration exceeds a preset threshold, a microfluidic flushing procedure is initiated to selectively remove metabolic waste while preserving the nutrient microenvironment around the embryo.

5. The personalized dynamic embryo culture system based on AI feedback according to claim 1, characterized in that, The logic for generating control instructions by the AI ​​analysis module further includes: setting a dynamic embryonic development elasticity coefficient for each embryo, which defines the allowable time deviation range of the embryo at key developmental time points; When the actual developmental rhythm of the embryo is within the allowable time deviation range, the dynamic control module maintains the current culture conditions unchanged. Only when the developmental deviation of the embryo exceeds the range defined by the elasticity coefficient will the AI ​​analysis module generate interventional control instructions.

6. The personalized dynamic embryo culture system based on AI feedback according to claim 1, characterized in that, The microfluidic unit enables the recycling of micro-volume culture medium. Specifically, it consists of a micro-circulation pipeline connected to the independent culture chamber module. This pipeline integrates a selective filtration membrane for separating metabolic waste and a micro-quantitative pump for replenishing nutrients. Through this micro-circulation pipeline, the system controls the culture medium consumption of a single embryo to between 0.1 mL and 0.15 mL.

7. The personalized dynamic embryo culture system based on AI feedback according to claim 1, characterized in that, The AI ​​analysis module of the system also includes a personalized solution generation unit, which is configured to perform the following operations: acquire the patient's clinical individual characteristic data, including age, ovarian function indicators, and baseline hormone levels; associate the clinical individual characteristic data with the initial state data of the embryo; run an AI sub-model; dynamically generate a personalized culture medium composition adjustment instruction based on the associated data; wherein the adjustment instruction includes at least an optimized setting of the concentration of antioxidant components or the proportion of insulin-like growth factor in the culture medium; and send the culture medium composition adjustment instruction to the dynamic control module to control the microfluidic unit to adjust the culture medium composition of the specified culture chamber in real time.

8. The personalized dynamic embryo culture system based on AI feedback according to claim 5, characterized in that, The embryonic developmental elasticity coefficient is dynamically calculated and generated by the AI ​​analysis module for each embryo. The calculation process includes: analyzing the historical developmental time series data of the embryo and comparing it with the standard developmental curve; extracting specific factors from the patient's clinical individual characteristic data, including the patient's age and fertilization method; and calculating the personalized time tolerance window of the embryo at each subsequent key developmental point through a pre-trained regression model based on the comparison results of the historical developmental time series data and the patient's specific factors. This time tolerance window is defined as the embryonic developmental elasticity coefficient.

9. The personalized dynamic embryo culture system based on AI feedback according to claim 4, characterized in that, The microsensor assembly also includes a metabolite concentration detection unit, which is communicatively connected to the microfluidic unit to form a closed-loop feedback subsystem. The metabolite concentration detection unit is configured to monitor the concentration of at least one metabolic waste marker in the culture medium in real time. The microfluidic unit is configured to, in response to the monitoring result of the metabolite concentration detection unit, perform at least one of the following operations: initiate the microfluidic flushing procedure, or adjust the flow rate of the culture medium to change the local metabolite diffusion efficiency, or activate an adsorbent module integrated in the circulation pipeline to selectively adsorb metabolic waste.

Images