A method and system for the construction of a reproductive medicine database

Abstract

The present application relates to the technical field of database construction, and particularly relates to a method and system for constructing a reproductive medicine database. The method comprises the following steps: obtaining a patient biological tissue sample; obtaining reproductive endocrine hormone data, embryo morphology parameter data, immune cell function data and genomic variation data of the patient biological tissue sample; performing time series analysis and processing on hormone fluctuation patterns according to the reproductive endocrine hormone data to obtain endocrine dynamic characteristic data; performing image recognition processing based on light coherence superposition characteristics on the embryo morphology parameter data, and performing embryo development potential evaluation to obtain embryo development potential data; and performing hierarchical clustering processing on patient individual heterogeneity according to the immune cell function data and the genomic variation data to obtain patient immune-heredity typing data. The present application realizes standardization of data acquisition through quantum dot sensing technology and a multi-channel microfluidic chip, and solves the problem of heterogeneity of data sources.

Description

Technical Field

[0001] This invention relates to the field of database construction technology, and in particular to a method and system for constructing a reproductive medicine database. Background Technology

[0002] Reproductive medicine databases are an important component of modern medical informatics, enabling the effective management and analysis of clinical and research data related to reproductive health, including endocrine testing data, embryonic morphology and development data, immune function data, and genomics data. When constructing a reproductive medicine database, users typically include clinicians, researchers, administrators, and patients. Each user group has different needs; for example, doctors need to quickly retrieve patient medical records, researchers need to analyze large amounts of anonymized data, and administrators need statistical reporting functions.

[0003] Traditional methods for constructing reproductive medicine databases often suffer from the following problems: Firstly, the different data collection and recording standards used by various medical institutions make data integration difficult. Secondly, in reproductive medicine, an absolutely accurate correspondence must be maintained between samples (eggs, sperm, embryos) and patient identities, which directly relates to ethical and legal risks. Thirdly, traditional methods relying on manual data entry and barcode labeling carry the potential risk of confusion. Fourthly, reproductive medicine involves multi-level biological data, from genomics and transcriptomics to cell function and organ morphology; integrating this multi-scale, multi-modal data has always been a technical challenge. Traditional methods often employ simple data splicing or feature concatenation, failing to effectively capture cross-scale biological relationships. Summary of the Invention

[0004] Therefore, the present invention needs to provide a method and system for constructing a reproductive medicine database to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a method for constructing a reproductive medicine database includes the following steps:

[0006] Step S1: Obtain biological tissue samples from the patient; obtain reproductive endocrine hormone data, embryonic morphological parameter data, immune cell function data, and genomic variation data from the patient's biological tissue samples;

[0007] Step S2: Perform time-series analysis on hormone fluctuation patterns based on reproductive endocrine hormone data to obtain endocrine dynamic characteristic data; perform image recognition processing based on the coherent superposition characteristics of light on embryonic morphological parameter data, and assess embryonic developmental potential to obtain embryonic developmental potential data.

[0008] Step S3: Based on immune cell function data and genomic variation data, stratified clustering is performed to analyze individual patient heterogeneity, resulting in patient immunogenetic typing data. The stratified clustering process involves... A memristor array with a double-layer oxide structure is realized, with each memristor having 64 adjustable conductance states;

[0009] Step S4: Perform multidimensional feature fusion processing on endocrine dynamic characteristic data, embryonic developmental potential data and patient immune-genetic typing data, and construct a homomorphically encrypted distributed knowledge graph index to obtain the reproductive medicine database structure.

[0010] This invention achieves precise collection of multi-dimensional information by acquiring patient biological tissue samples and extracting reproductive endocrine hormone data, embryonic morphological parameter data, immune cell function data, and genomic variation data, providing a solid foundation for subsequent data analysis and processing. For reproductive endocrine hormone data, time-series analysis is used to analyze hormone fluctuation patterns and obtain dynamic endocrine characteristic data, quantifying individual endocrine change trends and helping to identify potential physiological abnormalities or hormonal imbalances, thereby improving the precise regulation capabilities of assisted reproduction. Simultaneously, image recognition processing of embryonic morphological parameter data is performed based on the coherent superposition characteristics of light, combined with embryonic developmental potential assessment methods, to accurately determine the developmental potential of embryos, making the selection of high-quality embryos more scientific and reducing errors caused by subjective judgment. In terms of immuno-genetic characteristic analysis, through... A memristor array with a double-layer oxide structure is used for hierarchical clustering to achieve in-depth mining of patient immune cell function data and genomic variation data. Each memristor has 64 adjustable conductance states, which can finely adjust the data clustering process, improve the accuracy of individual heterogeneity identification, and thus achieve precise immuno-genetic typing, making personalized treatment strategies more targeted. Based on the acquired endocrine dynamic characteristic data, embryonic developmental potential data, and patient immuno-genetic typing data, multi-dimensional feature fusion technology is used for deep integration, and a homomorphically encrypted distributed knowledge graph index is constructed. This enables reproductive medicine data to be efficiently stored and retrieved while ensuring privacy, improving data security and usability. The overall methodology, from data acquisition, feature extraction, individual heterogeneity analysis to multi-dimensional data fusion, forms a complete analysis process, which not only improves the intelligence of data processing in the field of reproductive medicine, but also provides reliable technical support for precision medicine and personalized assisted reproduction.

[0011] The present invention also provides a system for constructing a reproductive medicine database, for executing the above-described method for constructing a reproductive medicine database, the system comprising:

[0012] The biological data acquisition module is used to acquire biological tissue samples from patients; and to acquire data on reproductive endocrine hormones, embryonic morphological parameters, immune cell function, and genomic variation from these biological tissue samples.

[0013] The potential assessment module is used to perform time-series analysis of hormone fluctuation patterns based on reproductive endocrine hormone data to obtain endocrine dynamic characteristic data; and to perform image recognition processing based on the coherent superposition characteristics of light on embryonic morphological parameter data, and to assess embryonic developmental potential to obtain embryonic developmental potential data.

[0014] The memristor heterogeneity typing module is used to perform hierarchical clustering of individual patient heterogeneity based on immune cell function data and genomic variation data to obtain patient immuno-genetic typing data. The hierarchical clustering process is achieved through... A memristor array with a double-layer oxide structure is realized, with each memristor having 64 adjustable conductance states;

[0015] An encrypted storage module is used to perform multidimensional feature fusion processing on endocrine dynamic characteristic data, embryonic developmental potential data, and patient immune-genetic typing data, and to construct a homomorphically encrypted distributed knowledge graph index to obtain the reproductive medicine database structure.

[0016] This invention utilizes a biological data acquisition module to systematically acquire patient biological tissue samples and precisely extract data on reproductive endocrine hormones, embryonic morphological parameters, immune cell function, and genomic variations, thus comprehensively integrating personalized medical information. Based on this data, the potential assessment module analyzes the temporal fluctuation patterns of reproductive endocrine hormones through time-series analysis, constructing dynamic characteristic data to reveal the regulatory laws of the endocrine system on the reproductive process. Embryonic morphological parameter data is analyzed using high-precision image recognition based on the coherent superposition characteristics of light, achieving a quantitative assessment of embryonic developmental potential and ensuring high stability and reliability of the assessment results. The memristor heterogeneity typing module utilizes... A memristor array with a double-layer oxide structure performs hierarchical clustering of immune cell function data and genomic variation data to identify immuno-genetic heterogeneity among individuals and construct a refined patient typing model. The 64 tunable conductance states within the memristor array enable it to accurately characterize the complex information flow patterns between different immuno-genetic subtypes, improving the resolution and adaptability of the typing. At the data management level, an encrypted storage module performs multi-dimensional feature fusion on endocrine dynamic characteristic data, embryonic developmental potential data, and patient immuno-genetic typing data, allowing for in-depth mining of cross-correlation patterns between data and storing the data using homomorphic encryption to ensure data is not leaked during computation. Furthermore, a knowledge graph index is constructed based on distributed ledger technology, giving the reproductive medicine database structure traceability and tamper-proof capabilities, providing reliable data support for personalized reproductive health assessment. This database not only enhances the systematic nature of reproductive health analysis but also enables precision medicine strategies to be implemented under data security, providing efficient and reliable intelligent auxiliary support for clinical decision-making. Attached Figure Description

[0017] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0018] Figure 1 This is a schematic diagram of the steps in a method for constructing a reproductive medicine database according to the present invention;

[0019] Figure 2 for Figure 1 A detailed flowchart of step S1;

[0020] Figure 3 for Figure 1 A detailed flowchart of step S3. Detailed Implementation

[0021] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0023] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for constructing a reproductive medicine database, the method comprising the following steps:

[0025] Step S1: Obtain biological tissue samples from the patient; obtain reproductive endocrine hormone data, embryonic morphological parameter data, immune cell function data, and genomic variation data from the patient's biological tissue samples;

[0026] In this embodiment, biological tissue samples were first extracted from the patient's peripheral blood, and plasma and mononuclear cells were separated using flow cytometry to obtain data on reproductive endocrine hormones and immune cell function, respectively. Simultaneously, for the acquisition of embryonic morphological parameters, high-resolution optical microscopy was used to acquire real-time images of the fertilized egg and embryonic development process, recording morphological changes every 5 minutes until blastocyst formation. Furthermore, to obtain genomic variation data, DNA was extracted from the patient's biological tissue samples, and whole-exome sequencing was used for genomic analysis to detect key genetic information such as single nucleotide polymorphisms, insertion / deletion variations, and copy number variations. The sequencing depth was set to 100× to ensure data reliability. Finally, all data were stored in a standardized database for subsequent analysis.

[0027] Step S2: Perform time-series analysis on hormone fluctuation patterns based on reproductive endocrine hormone data to obtain endocrine dynamic characteristic data; perform image recognition processing based on the coherent superposition characteristics of light on embryonic morphological parameter data, and assess embryonic developmental potential to obtain embryonic developmental potential data.

[0028] In this embodiment, for the time-series analysis and processing of reproductive endocrine hormone data, the Savitzky-Golay filtering method is first used to smooth the data to remove short-cycle noise signals. Then, a hormone fluctuation prediction model is established based on a Long Short-Term Memory (LSTM) neural network to perform nonlinear time-series modeling of the dynamic changes in hormone levels such as estradiol, progesterone, and luteinizing hormone, and key inflection point information is extracted to identify the endocrine dynamic characteristics of the ovulation cycle. For the processing of embryonic morphological parameter data, based on the coherent superposition characteristics of light, spectral interferometry microscopy is used in conjunction with an improved U-Net deep learning model to automatically identify and quantify parameters such as embryonic cell division, zona pellucida thickness, and blastocoel expansion rate, and the developmental potential of the embryo is assessed based on a multivariate regression model. Finally, the output endocrine dynamic characteristic data and embryonic developmental potential data are used to guide personalized ovulation induction protocols and embryo transfer decisions, respectively.

[0029] Step S3: Based on immune cell function data and genomic variation data, stratified clustering is performed to analyze individual patient heterogeneity, resulting in patient immunogenetic typing data. The stratified clustering process involves... A memristor array with a double-layer oxide structure is realized, with each memristor having 64 adjustable conductance states;

[0030] In this embodiment, immune cell function data are first standardized. Single-cell RNA sequencing technology is used to analyze the subtype composition of immune cells in peripheral blood, and cytokine secretion detection is used to assess the functional status of immune cells, including the cytotoxicity level of T cells and the immunosuppressive capacity of regulatory T cells. Then, combined with genomic variation data, an improved hierarchical clustering algorithm is used to perform hierarchical clustering analysis on the patient's immuno-genetic characteristics. This algorithm... A bilayer oxide memristor array was used, where each memristor node corresponds to a gene-immune trait combination. By adjusting the conductance of the memristors (up to 64 adjustable conductance values), the differences in immune response and genetic background among different patients were simulated. Ultimately, the clustering results classified patients into different immune-genetic subtypes.

[0031] Step S4: Perform multidimensional feature fusion processing on endocrine dynamic characteristic data, embryonic developmental potential data and patient immune-genetic typing data, and construct a homomorphically encrypted distributed knowledge graph index to obtain the reproductive medicine database structure.

[0032] In this embodiment, firstly, multidimensional feature fusion is performed on the acquired endocrine dynamic characteristic data, embryonic developmental potential data, and patient immuno-genetic typing data. A self-attention mechanism based on the Transformer architecture is used to establish global dependencies between different feature dimensions to enhance the interactive information between data. Subsequently, homomorphic encryption technology is used to encrypt the fused data to ensure the security of patient privacy in the distributed computing environment. The encrypted data adopts the CKKS (Cheon-Kim-Kim-Song) homomorphic encryption scheme, which supports addition and multiplication operations without decryption, thereby ensuring the security of data during remote storage and computation. Finally, based on the knowledge graph construction method, the data is organized in RDF (Resource Description Framework) format, and the index is optimized using a graph database to achieve efficient query and retrieval functions, ultimately forming a reproductive medicine database structure that provides a high-quality data foundation for clinical decision support and scientific research analysis.

[0033] This invention achieves precise collection of multi-dimensional information by acquiring patient biological tissue samples and extracting reproductive endocrine hormone data, embryonic morphological parameter data, immune cell function data, and genomic variation data, providing a solid foundation for subsequent data analysis and processing. For reproductive endocrine hormone data, time-series analysis is used to analyze hormone fluctuation patterns and obtain dynamic endocrine characteristic data, quantifying individual endocrine change trends and helping to identify potential physiological abnormalities or hormonal imbalances, thereby improving the precise regulation capabilities of assisted reproduction. Simultaneously, image recognition processing of embryonic morphological parameter data is performed based on the coherent superposition characteristics of light, combined with embryonic developmental potential assessment methods, to accurately determine the developmental potential of embryos, making the selection of high-quality embryos more scientific and reducing errors caused by subjective judgment. In terms of immuno-genetic characteristic analysis, through... A memristor array with a double-layer oxide structure is used for hierarchical clustering to achieve in-depth mining of patient immune cell function data and genomic variation data. Each memristor has 64 adjustable conductance states, which can finely adjust the data clustering process, improve the accuracy of individual heterogeneity identification, and thus achieve precise immuno-genetic typing, making personalized treatment strategies more targeted. Based on the acquired endocrine dynamic characteristic data, embryonic developmental potential data, and patient immuno-genetic typing data, multi-dimensional feature fusion technology is used for deep integration, and a homomorphically encrypted distributed knowledge graph index is constructed. This enables reproductive medicine data to be efficiently stored and retrieved while ensuring privacy, improving data security and usability. The overall methodology, from data acquisition, feature extraction, individual heterogeneity analysis to multi-dimensional data fusion, forms a complete analysis process, which not only improves the intelligence of data processing in the field of reproductive medicine, but also provides reliable technical support for precision medicine and personalized assisted reproduction.

[0034] Preferably, step S1 includes the following steps:

[0035] Step S11: Obtain biological tissue samples from the patient;

[0036] In this embodiment, peripheral blood or endometrial tissue samples were first collected from the patient. Peripheral blood samples were collected using EDTA anticoagulant tubes to ensure the integrity of cellular components, while endometrial tissue samples were obtained via hysteroscopic biopsy. Subsequently, the biological tissue samples were transported to the laboratory at 4°C and pre-processed in a biosafety cabinet. This included washing the blood samples with PBS buffer, separating mononuclear cells using density gradient centrifugation (400g, 15 minutes), or digesting the endometrial tissue with tissue dissociation enzymes (such as 0.25% trypsin) to release a single-cell suspension. Finally, the concentration of the resulting cell suspension was adjusted to approximately 1 × 10⁻⁶ cells per milliliter. 6 To ensure the stability of subsequent experiments, each cell was counted.

[0037] Step S12: The patient's biological tissue sample is introduced into a PDMS microfluidic chip with an integrated CdSe / ZnS core-shell quantum dot sensing array to obtain the dispersed flow data of the biological sample. The PDMS microfluidic chip has a multi-channel structure with a channel width of 30-50 μm. m;

[0038] In this embodiment, a PDMS microfluidic chip with an integrated CdSe / ZnS core-shell quantum dot sensing array was used to detect the flow of biological samples. This chip has a multi-channel structure, with each channel having a width of 30-50 μm to match the flow characteristics of individual cells. First, the microfluidic chip was pre-wetted in a CO2 incubator for 30 minutes to enhance the affinity for the biological fluid. Then, a patient's biological tissue sample was introduced into the microfluidic channel using an infusion pump at a constant flow rate (approximately 1 μL / min), allowing the cells to disperse under the shear force of the fluid and ensuring that the cells did not adhere to the channel walls. The flow state of the cells in the microfluidic channel was monitored in real time using a fluorescence microscopy imaging system, and the spatial distribution data and flow rate change data of the cells were recorded to analyze the dispersion characteristics of different cell types in a microfluidic environment.

[0039] Step S13: Modify the biological sample dispersion flow data with specific antibodies on the quantum dot sensing array, and sort and capture NK cells, T cell subsets, B cells and monocytes in the patient's biological tissue sample to obtain immune cell capture data, wherein the modified specific antibodies include CD56, CD3, CD4, CD8, CD19, CD14 and HLA-DR;

[0040] In this embodiment, a quantum dot sensing array is used to sort and capture target immune cells from flowing biological samples. First, specific antibodies, including CD56 (for recognizing NK cells), CD3 (for recognizing T cells), CD4 (for recognizing helper T cells), CD8 (for recognizing cytotoxic T cells), CD19 (for recognizing B cells), CD14 (for recognizing monocytes), and HLA-DR (for recognizing antigen-presenting cells), are modified onto the surface of the quantum dots. The antibody modification employs a biotin-streptavidin conjugation strategy to ensure stable fixation of the antibodies on the quantum dot surface. Subsequently, the biological tissue sample is passed through the sensing array. After the specific antibodies specifically bind to the corresponding cell surface antigens, unbound cells are eluted using weak fluid shear forces, ultimately retaining the target cells on the array. The captured immune cells are imaged using a fluorescence microscope, and the number of captured immune cells of each type is calculated using automated image analysis software to obtain immune cell capture data.

[0041] Step S14: Perform multispectral excitation on the immune cell capture data in the wavelength range of 525-655nm and record the changes in quantum dot fluorescence resonance energy transfer signals to obtain immune cell function data;

[0042] In this embodiment, to analyze the functional state of captured immune cells, changes in quantum dot fluorescence resonance energy transfer (FRET) signals were recorded using multispectral excitation (525-655 nm wavelength range). First, quantum dot-labeled cells were irradiated sequentially at wavelengths of 525 nm, 575 nm, and 655 nm using a single-cell laminar flow excitation device, and the changes in fluorescence intensity at different excitation wavelengths were measured. Second, based on the FRET mechanism, the fluorescence energy transfer effect caused by conformational changes in surface receptor molecules when immune cells are in an active state was observed. For example, during T cell activation, activation of the CD3 / CD28 signaling pathway causes enhancement or attenuation of specific fluorescence signals. FRET signals were recorded using a high-throughput spectral imaging system, and principal component analysis (PCA) was used to perform dimensionality reduction analysis on the multichannel fluorescence data to distinguish the functional states of different immune cell subsets, ultimately obtaining immune cell functional data.

[0043] Step S15: Obtain reproductive endocrine hormone data, embryonic morphological parameter data, and genomic variation data from the patient's biological tissue samples.

[0044] In this embodiment, reproductive endocrine hormone data were first obtained from the patient's biological tissue samples, and estradiol (ELISA) was detected using ELISA (enzyme-linked immunosorbent assay). ), progesterone ( The concentrations of luteinizing hormone (LH) and follicle-stimulating hormone (FSH) were measured and quantitatively analyzed using an automated chemiluminescence immunoassay analyzer to ensure the sensitivity and accuracy of hormone assays. Secondly, for acquiring embryonic morphological parameters, an embryo monitoring system based on coherent light imaging was used to acquire images of embryonic morphological changes every 5 minutes, recording key parameters such as zona pellucida thickness and blastocoel volume. Finally, to analyze genomic variation data, DNA was extracted from patient tissue samples and whole-exome sequencing was performed using high-throughput sequencing technology (such as Illumina NovaSeq 6000). The sequencing data coverage depth was set to 100×, and mutation screening was performed using bioinformatics algorithms (such as the GATK mutation detection tool) to obtain genomic variation information related to reproductive health. Ultimately, the obtained reproductive endocrine hormone data, embryonic morphological parameter data, and genomic variation data will be used for personalized reproductive medicine decision support.

[0045] This invention acquires patient biological tissue samples and introduces a PDMS microfluidic chip with an integrated CdSe / ZnS core-shell quantum dot sensing array. This enables efficient dispersion and flow of biological samples within a multi-channel structure, ensuring uniformity and stability in sample processing. The 30-50 μm channel width of the multi-channel structure is suitable for the flow behavior characteristics of different cell types, improving the accuracy of cell screening. Furthermore, by modifying the quantum dot sensing array with specific antibodies, NK cells, T cell subsets, B cells, and monocytes in the patient biological tissue samples are sorted and captured, accurately identifying and separating different immune cell types, making subsequent analysis more targeted. The specific antibody markers CD56, CD3, CD4, CD8, CD19, CD14, and HLA-DR cover the major immune cell types, ensuring comprehensive immune function assessment. Further, by using multispectral excitation technology in the 525-655 nm wavelength range, combined with changes in quantum dot fluorescence resonance energy transfer signals, precise detection of immune cell function is achieved, making dynamic monitoring of the immune response state more sensitive and accurate. Furthermore, the system simultaneously acquires reproductive endocrine hormone data, embryonic morphological parameters, and genomic variation data from patient biological tissue samples, enabling comprehensive integration of core biological information related to reproductive medicine and providing multi-dimensional data support for precision medicine analysis. The overall solution leverages microfluidic technology, quantum dot labeling, and multispectral detection to achieve efficient integration of immune cell screening, functional assessment, and reproductive-related data acquisition, providing a precise technical foundation for personalized diagnosis and treatment and the optimization of assisted reproductive strategies.

[0046] Preferably, step S15 includes the following steps:

[0047] Step S151: The follicular fluid in the patient's biological tissue sample is processed by a microfluidic chip electrochemical sensor array to obtain a primary endocrine data stream. The microfluidic chip electrochemical sensor array contains 64 independent electrode units, each of which is surface-modified with specific aptamers for estradiol, progesterone, FSH and LH.

[0048] In this embodiment, follicular fluid samples were collected from the patient, ensuring the sample temperature was maintained at 37°C to minimize loss of bioactivity. The samples were introduced into a microfluidic chip electrochemical sensor array via a PDMS microfluidic channel. This array contains 64 independent electrode units, each with a surface modified with specific aptamers via chemical self-assembly, including aptamers for estradiol, progesterone, follicle-stimulating hormone (FSH), and luteinizing hormone (LH). After the samples entered the channel, an external micro-electric field (voltage range 0.1V-0.5V) was applied to induce the hormone molecules to bind to the aptamers. Differential pulse voltammetry (DPV) was used to measure the current signal changes of each electrode unit, converting them into corresponding primary endocrine data streams, recording real-time concentration changes at nanoampere-level current resolution.

[0049] Step S152: A diamond quantum sensor based on nitrogen-vacancy defects, fabricated using a 28nm process, is used to detect hormone concentrations in the primary endocrine data stream with pT-level precision, thereby obtaining a high-precision endocrine concentration matrix. The detection precision of the diamond quantum sensor reaches [percentage missing]. Molar concentration, sampling frequency 300Hz;

[0050] In this embodiment, the obtained primary endocrine data stream is input into a diamond quantum sensor based on nitrogen-vacancy defects for ultra-high precision hormone concentration measurement. The diamond quantum sensor used is fabricated using a 28nm process and employs optically detected magnetic resonance (ODMR) technology. Microwave excitation and a 532nm laser pump are used to detect the spin resonance signal at the nitrogen-vacancy center, thereby achieving pT-level concentration detection of estradiol, progesterone, FSH, and LH. During the measurement process, the sensor's sampling frequency is set to 300Hz to ensure the capture of rapidly changing hormone secretion events. Furthermore, Fourier transform of the optical readout signal further improves the detection accuracy, achieving a concentration level of 10^(-12) moles. Finally, all measured values ​​constitute a high-precision endocrine concentration matrix, providing accurate hormone concentration information for subsequent analysis.

[0051] Step S153: Extract pulse signals from the high-precision endocrine concentration matrix to obtain hormone secretion pulse feature data;

[0052] In this embodiment, a pulse signal extraction algorithm is applied to the high-precision endocrine concentration matrix to identify the pulse-like changes in hormone concentration. A time-frequency analysis method based on continuous wavelet transform (CWT) is used to scale and normalize hormone concentrations at different time points, and a threshold (e.g., 1.5 times the standard deviation) is set to screen for significant pulse signals. This method can effectively distinguish between basal secretion and pulse secretion patterns, thereby obtaining hormone secretion pulse characteristic data, including important parameters such as the amplitude, frequency, interpeak interval, and deceleration rate of secretion peaks, providing data support for assessing patients' endocrine function.

[0053] Step S154: Combine hormone secretion pulse characteristic data with time series markers to obtain reproductive endocrine hormone data;

[0054] In this embodiment, the acquired hormone secretion pulse characteristic data are combined with time-series markers to construct a reproductive endocrine hormone dataset. The time-series markers are obtained based on factors such as the patient's circadian rhythm, cyclical hormone secretion patterns, and ovulation cycle stage. A time synchronization algorithm is used to automatically align the pulse data, and a Long Short-Term Memory (LSTM) network is employed to model the data over time to predict future hormone fluctuation trends. The results of this step can be used to analyze individualized reproductive endocrine status, providing a reference for subsequent embryo culture and implantation decisions.

[0055] Step S155: The patient's biological tissue sample is continuously monitored through an integrated microfluidic culture chamber and a phase contrast microscopy imaging system to obtain a sequence of dynamic embryonic development images. The phase contrast microscopy imaging system consists of a frequency-tunable optical resonant cavity and a high-speed CMOS image sensor.

[0056] In this embodiment, the patient's embryonic sample is placed in a microfluidic culture chamber for continuous in vitro culture, while an integrated phase-contrast microscopy imaging system is used for real-time monitoring. This system consists of a tunable optical resonant cavity and a high-speed CMOS image sensor, utilizing optical coherence phase-shifting (PCI) technology to record label-free imaging data of embryonic cells. The resonant frequency of the optical resonant cavity is set at 560 THz, enabling single-frame imaging within 1 ms, ensuring that key dynamic changes during embryonic development are fully recorded. Imaging data is continuously acquired at a frame rate of 50 fps, and an automatic image enhancement algorithm is used to improve cell boundary sharpness, thereby obtaining a high-resolution sequence of dynamic embryonic development images.

[0057] Step S156: Optical coherence processing is performed on the dynamic development image sequence of the embryo to obtain a multi-parameter morphological feature set of the embryo. The optical coherence processing is implemented using a photonic interference computing unit containing an integrated microring resonator array of 256×256.

[0058] In this embodiment, optical coherence processing is performed on the acquired embryonic dynamic development image sequence to extract multi-parameter morphological features. Optical coherence processing is implemented using a photonic interferometry computing unit, which includes a 256×256 array of integrated microring resonators capable of demodulating optical field information and extracting key features such as embryonic cell growth rate, volume change, and refractive index distribution. First, the photonic interferometry computing unit performs spatiotemporal coherence analysis on embryonic images at different time points to extract intercellular refractive index change patterns. Then, a deep convolutional neural network (CNN) is used to extract high-dimensional features from the interferogram, constructing a multi-parameter morphological feature set for the embryo, providing data support for subsequent embryo quality assessment.

[0059] Step S157: Perform optical computation processing on the multi-parameter morphological feature set of the embryo to obtain embryo morphological parameter data, which includes cleavage rate, cleavage symmetry, blastocoel formation dynamics and chromosome euploidy prediction value.

[0060] In this embodiment, optical computational processing is performed on the multi-parameter morphological feature set of the embryo to obtain embryonic morphological parameter data. During processing, firstly, a morphological analysis algorithm based on curve fitting is used to calculate the cleavage rate, and the uniformity of cleavage cell division is assessed using a morphological symmetry measure function. Secondly, dynamic image processing technology is used to detect the temporal characteristics of blastocoel formation, including the blastocoel expansion rate and cell mass contraction frequency. Furthermore, a chromosome euploidy prediction model trained using a machine learning algorithm, combined with cell division dynamics and blastocoel morphological characteristics, is used to predict the genetic stability of the embryo, thereby obtaining complete embryonic morphological parameter data and providing a basis for clinical decision-making in assisted reproduction.

[0061] Step S158: Perform reproductive-related variant site analysis on nucleic acids extracted from the patient's biological tissue sample to obtain genomic variant data.

[0062] In this embodiment, nucleic acids were extracted from patient tissue samples, and reproductive-related variant sites were analyzed. Magnetic bead extraction was used to ensure efficient enrichment and purification of DNA and RNA. Next, a next-generation sequencing (NGS) platform was used to sequence target gene regions, with particular attention to genes related to reproductive health, such as FOXL2, AMH, FSHR, and ESR1. Variance analysis employed a bioinformatics pipeline, including base quality filtering, alignment to a reference genome, single nucleotide polymorphism (SNP) and insertion / deletion (INDEL) detection, combined with machine learning methods to predict the functional impact of variants. Finally, genomic variant data was generated to provide data support for the patient's reproductive health assessment.

[0063] This invention utilizes a microfluidic chip electrochemical sensor array to process follicular fluid, enabling precise quantification of key reproductive hormones such as estradiol, progesterone, FSH, and LH through efficient detection by 64 independent electrode units. Specific aptamers modified on the surface of each electrode ensure high sensitivity and specificity, resulting in a stable primary endocrine data stream. Subsequently, a nitrogen-vacancy defect diamond quantum sensor based on 28nm technology is used to achieve pT-level high-precision detection of hormone concentrations, realizing ultrasensitive measurements of 10^(-12) molar concentrations. A high sampling frequency of 300Hz ensures the ability to capture the temporal fluctuations of hormone secretion, making the establishment of a high-precision endocrine concentration matrix more comprehensive and accurate. Based on this, pulse signal extraction technology is used to convert the high-precision data into hormone secretion pulse characteristic data, which, combined with time-series labeling, forms complete reproductive endocrine hormone data, providing comprehensive temporal information support for subsequent analysis. Furthermore, by integrating a microfluidic culture chamber and a phase-contrast microscopy system, continuous monitoring of the dynamic development of embryos is achieved, ensuring the acquisition of data throughout the cleavage and blastocyst development process. The phase-contrast microscopy system, combining a frequency-tunable optical resonant cavity and a high-speed CMOS image sensor, enables high-resolution and high-frame-rate embryonic image acquisition, accurately reproducing details of embryonic growth. The obtained dynamic embryonic development image sequences are subjected to optical coherence processing using a photonic interferometry computing unit with an integrated microring resonator. This extracts a multi-parameter morphological feature set, including cleavage rate, cleavage symmetry, blastocyst cavity formation dynamics, and chromosome euploidy prediction values. Optical computation further enhances the accuracy of embryonic morphological analysis, making it more valuable for reference. Simultaneously, nucleic acid analysis of biological tissue samples is performed to accurately screen for reproductive-related variant sites, thereby obtaining high-quality genomic variation data and ensuring comprehensive information support at the genetic level. The overall solution combines multiple advanced technologies such as electrochemical sensing, quantum precision measurement, microfluidic imaging monitoring, and optical coherence computing to achieve deep integration of reproductive endocrine analysis, embryo morphology assessment, and genetic background analysis, providing accurate and systematic data support for reproductive health research and assisted reproductive medicine.

[0064] Preferably, step S158 includes the following steps:

[0065] Step S1581: Perform single-molecule sequencing based on graphene nanopore array on nucleic acids extracted from the patient's biological tissue sample to obtain DNA transmission signals at single-base resolution;

[0066] In this embodiment, cells from patient biological tissue samples are lysed and DNA is extracted. Target DNA fragments are enriched using magnetic beads, and sequencing libraries are constructed using end modification and adapter ligation. The library is then loaded onto a single-molecule sequencing platform based on a graphene nanopore array. A transmembrane electric field is applied to allow single-stranded DNA to pass through the graphene nanopores one by one, while current changes are monitored in real time. Graphene nanopores have sub-nanometer pore sizes and, combined with surface functional group modifications, enhance the selectivity and stability for DNA molecules. When DNA passes through the nanopores, different base genes have different spatial steric hindrance and charge distributions, generating characteristic current blocking signals as they pass through the channels. The system records these signals using a highly sensitive current measurement module, thereby obtaining DNA passage signals at single-base resolution. This process uses a 100kHz sampling frequency for data acquisition and incorporates a machine learning model to correct for noise interference in real time, ensuring signal accuracy.

[0067] Step S1582: The DNA transmission signal is acquired and processed in parallel using a multi-channel graphene transistor array to obtain the raw genome sequence data. The sensitivity of the multi-channel graphene transistor array is high enough to detect the current change caused by the passage of a single DNA molecule.

[0068] In this embodiment, the acquired DNA is input as a signal to a multi-channel graphene transistor array. Each channel consists of an independent graphene field-effect transistor (GFET), whose surface is modified with specific probes to enhance the response to DNA signals. Through external voltage control, the charge change induced by the passage of DNA molecules modulates the conductivity of the graphene transistors, forming a current signal directly related to the DNA base types. The array employs a 128×128-channel parallel architecture, with each channel achieving a detection limit down to the single-molecule level and exhibiting a current sensitivity at the 1pA level. The acquired signal is amplified by a high-precision, low-noise amplifier, then converted into a digital signal in parallel by a high-speed analog-to-digital converter (ADC), and transmitted in real-time to a central processing unit (CPU) for preliminary data integration, ultimately yielding the raw genome sequence data.

[0069] Step S1583: Perform base accuracy verification on the raw genome reads to obtain high-precision genome sequencing data;

[0070] In this embodiment, the base accuracy of the raw genome sequence data is verified. First, a noise model based on the Markov chain Monte Carlo (MCMC) method is used to assess sequencing errors in the data and remove anomalous signals. Then, a recurrent neural network (RNN) model is constructed using a deep learning algorithm to compare DNA signals collected from multiple channels, and error correction is performed using a statistical probability model. A reference genome is used as a correction template, and the Smith-Waterman alignment algorithm is employed for local alignment to detect base substitution, insertion, and deletion errors. Finally, based on multi-round iterative confidence-weighted processing, high-confidence sequencing fragments are selected to obtain high-precision genome sequencing data with a base accuracy of 99.99%.

[0071] Step S1584: Use the memristor neuromorphic computing unit to analyze the reproductive-related variant sites of the high-precision genome sequencing data to obtain genome variation data, which includes single nucleotide polymorphisms, copy number variations and structural variation information of reproductive-related genes.

[0072] In this embodiment, high-precision genome sequencing data is input into a memristor neuromorphic computing unit. This unit, based on a cross-array memristor (RRAM) architecture, can simulate synaptic weighted computation and features low power consumption and high parallelism. First, a convolutional neural network (CNN) is used to extract features from the sequencing data, converting the DNA sequence into a numerical matrix and performing pattern matching analysis. Then, a long short-term memory (LSTM) network is used to identify variant sites in reproduction-related genes, including single nucleotide polymorphisms (SNPs), copy number variations (CNVs), and structural variations (SVs). The system is trained using known variant data from a genome database and calculates the probability of occurrence for each variant using Bayesian inference. Finally, key gene variants that may affect reproductive function are extracted and classified according to their impact, yielding the patient's genome variant data, which can be used to assess reproductive health risks or develop personalized medical plans.

[0073] This invention utilizes a microfluidic chip electrochemical sensor array to process follicular fluid, enabling precise quantification of key reproductive hormones such as estradiol, progesterone, FSH, and LH through efficient detection by 64 independent electrode units. Specific aptamers modified on the surface of each electrode ensure high sensitivity and specificity, resulting in a stable primary endocrine data stream. Subsequently, a nitrogen-vacancy defect diamond quantum sensor based on 28nm technology is used to achieve pT-level high-precision detection of hormone concentrations, realizing ultrasensitive measurements of 10^(-12) molar concentrations. A high sampling frequency of 300Hz ensures the ability to capture the temporal fluctuations of hormone secretion, making the establishment of a high-precision endocrine concentration matrix more comprehensive and accurate. Based on this, pulse signal extraction technology is used to convert the high-precision data into hormone secretion pulse characteristic data, which, combined with time-series labeling, forms complete reproductive endocrine hormone data, providing comprehensive temporal information support for subsequent analysis. Furthermore, by integrating a microfluidic culture chamber and a phase-contrast microscopy system, continuous monitoring of the dynamic development of embryos is achieved, ensuring the acquisition of data throughout the cleavage and blastocyst development process. The phase-contrast microscopy system, combining a frequency-tunable optical resonant cavity and a high-speed CMOS image sensor, enables high-resolution and high-frame-rate embryonic image acquisition, accurately reproducing details of embryonic growth. The obtained dynamic embryonic development image sequences are subjected to optical coherence processing using a photonic interferometry computing unit with an integrated microring resonator. This extracts a multi-parameter morphological feature set, including cleavage rate, cleavage symmetry, blastocyst cavity formation dynamics, and chromosome euploidy prediction values. Optical computation further enhances the accuracy of embryonic morphological analysis, making it more valuable for reference. Simultaneously, nucleic acid analysis of biological tissue samples is performed to accurately screen for reproductive-related variant sites, thereby obtaining high-quality genomic variation data and ensuring comprehensive information support at the genetic level. The overall solution combines multiple advanced technologies such as electrochemical sensing, quantum precision measurement, microfluidic imaging monitoring, and optical coherence computing to achieve deep integration of reproductive endocrine analysis, embryo morphology assessment, and genetic background analysis, providing accurate and systematic data support for reproductive health research and assisted reproductive medicine.

[0074] Preferably, the time-series analysis of hormone fluctuation patterns in step S2 includes:

[0075] The reproductive endocrine hormone data were standardized by time axis processing. A 400-hour unified time window was constructed with the first day of the menstrual cycle as the reference point to obtain standardized hormone data.

[0076] Standardized hormone data were sampled at the femtosecond level with a temporal resolution of 10 femtoseconds to obtain high temporal resolution hormone fluctuation curves;

[0077] Extracting intrinsic mode function sets from high temporal resolution hormone fluctuation curves;

[0078] Correlation analysis was performed on the intrinsic modal function set to obtain data on synergistic effects among hormones;

[0079] Extracting nonlinear kinetic features from data on synergistic effects among hormones;

[0080] By systematically identifying the nonlinear dynamic characteristics, a kinetic model of hormone regulation is obtained;

[0081] Forward evolution simulation of the hormone regulation kinetic model was performed to obtain dynamic endocrine characteristic data.

[0082] In this embodiment, reproductive endocrine hormone data of subjects were collected over several consecutive months, including luteinizing hormone (LH), follicle-stimulating hormone (FSH), estradiol (E2), and progesterone (P4), and the start time of the menstrual cycle was recorded. First, based on the length of the subjects' menstrual cycles, hormone data from different cycles were converted to the same reference time axis. Using the first day of the menstrual cycle as the time reference point, a unified 400-hour time window was constructed, and cycle normalization was performed using linear time scaling. For missing data points, bicubic interpolation was used to complete the data to ensure the integrity of the time series. Finally, a hormone concentration data matrix represented by a standardized time axis was obtained, providing a unified time reference framework for subsequent analysis. High-precision optical detection technology was used to perform ultra-high resolution time sampling of the standardized hormone data. Using a femtosecond optical pulsed laser combined with Raman scattering spectroscopy analysis, ultrafast dynamic measurements of hormone molecules in biological fluids were performed. First, plasma or urine samples are placed in nanofluidic channels, and multispectral acquisition is performed using an ultrafast time-resolved optical detection system to obtain continuous data points at 10 femtosecond (fs) intervals. Then, the discrete data is reconstructed using time interpolation reconstruction algorithms (such as spline interpolation) to generate high-temporal-resolution hormone fluctuation curves, enabling them to more accurately reflect the instantaneous dynamic changes in hormone secretion. Empirical Mode Decomposition (EMD) is used to decompose the high-temporal-resolution hormone fluctuation curves and extract their intrinsic mode functions (IMFs). First, envelope fitting is applied to separate different time-frequency components of the hormone fluctuation data, and IMFs containing different frequency characteristics are decomposed layer by layer. Then, Hilbert transform is used to calculate the instantaneous frequency and amplitude of each IMF component, constructing a Hilbert time spectrum to analyze the time-varying characteristics of hormone fluctuations. For example, in the process of hormone regulation, different IMF components may correspond to short-period pulse release, diurnal rhythm changes, and longer-term regulatory trends, respectively. This information can be used to reveal the dynamic regulatory mechanisms of reproductive endocrine hormones. An adaptive time-frequency joint analysis method combining wavelet packet transform (WPT) and short-time Fourier transform (STFT) was employed to perform multidimensional correlation analysis on the extracted IMF components. First, time-frequency matrices were constructed for the IMF components of different hormones, and their cross-time spectra were calculated to assess the time-frequency coupling relationships between different hormones. Then, mutual information analysis was used to quantify the degree of synergistic change between hormone pairs, screening out hormone pairs with high temporal correlation and resonance modes. For example, some IMF components of LH and FSH may exhibit high time-frequency coupling before ovulation, reflecting their synergistic regulatory effects. Finally, data on the synergistic effects between hormones were output, including characteristic parameters such as the synchronicity, phase shift, and coupling strength of different hormone pairs. A delayed coordinate reconstruction method was used to construct the phase space trajectory of the hormone system and analyze its nonlinear characteristics. First, Takens' theorem was used to determine the optimal embedding dimension and time delay parameters, mapping the synergistic effect data to a high-dimensional phase space.Then, the chaotic properties of the system were assessed by calculating the maximum Lyapunov exponent, and the fractal structure of the system was measured using the correlation dimension. Furthermore, recurrence plot analysis was applied to extract the periodic and non-periodic components of the hormone system, identifying endocrine regulation patterns at different physiological stages. For example, during ovulation, the hormone system may exhibit low-dimensional chaotic characteristics, while during the luteal phase, it may show more stable quasi-periodic oscillations. A kinetic model of hormone regulation was constructed using the fractional derivative method to describe the long-term memory effect of hormone secretion and metabolism. First, based on experimental data, the fractional order of the hormone kinetic system was estimated using the least squares fitting method, and the state equation of the model was defined using the Caputo fractional differential operator. Then, the particle swarm optimization (PSO) algorithm was used to globally optimize the model parameters to fit the experimental data, and cross-validation was performed using data from multiple groups of subjects. For example, the fractional-order kinetic model can describe the time-delayed feedback regulation process of estrogen on luteinizing hormone and reveal the dynamic response patterns of the hormone system under different physiological states. Ultimately, the output hormone regulation kinetic model can be used to predict hormone level fluctuations and early identification of abnormal states. A numerical simulation framework based on the Lattice Boltzmann Method (LBM) is used to computationally simulate the hormone regulation kinetic model. First, an LBM grid suitable for nonlinear physiological systems is constructed, and a hormone concentration distribution function is defined based on the Boltzmann transport equation. Then, the BGK (Bhatnagar-Gross-Krook) collision operator is used to calculate the diffusion and feedback regulation processes of hormones, simulating the evolutionary behavior of hormones over time. For example, in the simulation, initial concentrations of LH, FSH, E2, and P4 can be set, and periodic perturbations can be applied to simulate changes in physiological rhythms, observing the stability and self-organizing behavior of the hormone system. The simulation results can be used to predict the potential mechanisms of abnormal hormone fluctuations, such as precise intervention programs for polycystic ovary syndrome (PCOS) or luteal insufficiency.

[0083] This invention employs time-axis standardization, using the first day of the menstrual cycle as a baseline to construct a unified 400-hour time window. This allows for comparative analysis of hormone data from different patients within the same timeframe, improving data consistency and comparability. Subsequently, femtosecond-level sampling technology with a temporal resolution of 10 femtoseconds is used to precisely capture minute fluctuations in hormone levels, ensuring high temporal accuracy of the hormone fluctuation curve and thus accurately reflecting the dynamic changes of the reproductive endocrine system. The high-temporal-resolution hormone fluctuation curve is decomposed using the Hilbert-Huang transform to extract intrinsic mode function sets at different time scales. This helps identify periodic signals and short-term oscillation characteristics while preserving nonlinear information in the hormone regulation process. Based on this, correlation analysis of the intrinsic mode function set is performed using adaptive time-frequency joint analysis technology to accurately analyze the synergistic change patterns among hormones and reveal the coupling relationships of various hormones in physiological regulation. Furthermore, nonlinear dynamic features are extracted to characterize the complex feedback mechanisms in the hormone regulation process, and system identification is achieved through fractional differential equation modeling, enabling the hormone regulation kinetic model to accurately describe the regulatory characteristics of the endocrine system. Finally, the lattice Boltzmann method was used to perform forward evolution simulation of the hormone regulation kinetic model, predict the changing trends of hormone levels at different time scales, provide a global perspective on the dynamic regulation of the reproductive endocrine system, and provide accurate data support for the formulation of personalized intervention strategies.

[0084] Preferably, the image recognition processing based on the coherent superposition characteristics of light in step S2 includes:

[0085] Embryo morphological parameter data were imported into a silicon-based photonic interferometer chip with an integrated 128×128 microring resonator array to obtain multi-angle embryo light field information.

[0086] Phase modulation processing is performed on the multi-angle embryo light field information to obtain the characteristic light phase map of the embryo;

[0087] Parallel optical calculations were performed on the characteristic optical phase map of the embryo using the coherence of light to obtain the embryo morphological feature vector;

[0088] Low-loss transmission of embryonic morphological feature vectors yields high-fidelity morphological feature data, with waveguide loss of less than 0.01 dB / cm.

[0089] Nonlinear transformations were performed on high-fidelity morphological feature data to obtain embryonic structure-function correlation features;

[0090] Dimensional optimization of embryonic structure-function correlation features yields optimized morphological feature representations.

[0091] The optimized morphological features were classified using interferometric methods to obtain embryo morphological data.

[0092] In this embodiment, embryonic morphological parameters, including key indicators such as embryonic cell count, cytoplasmic uniformity, fragmentation rate, and zona pellucida thickness, are first acquired and converted into optical signals, which are then input into a silicon-based photonic interferometer chip. This chip employs a 128×128 microring resonator array, where each microring resonator can modulate the interference state of light waves at a specific wavelength, forming multi-angle optical information. To ensure accurate data input, a high-speed optical modulator maps the morphological parameters to the amplitude, phase, and polarization direction of the optical signal, and then imports the data into the chip via fiber optic coupling. During the experiment, the incident light wavelength is set in the near-infrared band at 1550 nm to obtain high optical resolution, while a low-loss silicon waveguide is used to reduce signal attenuation. When the embryonic morphological parameter optical signal enters the microring resonator array, different resonators selectively modulate the light at specific wavelengths, causing the multi-angle embryonic light field information to coherently superimpose within the array. To further extract effective information, a phase modulation technique based on the electro-optic effect is used, adjusting the phase shift of the optical signal by applying an external modulation electric field to form a characteristic phase map of the embryo. Specifically, the modulation electric field strength is controlled within the range of 10V / μm, and the modulation frequency is set to 10GHz to achieve fine control of the optical field distribution inside the microring and ensure high-precision phase adjustment. After acquiring the characteristic optical phase map of the embryo, information is calculated using optical coherence technology. Specifically, a Mach-Zehnder interferometer array is used to interfere and superimpose the coherent optical signals output from different microring resonators, and the morphological feature vector of the embryo is demodulated through the light intensity distribution. During this calculation, the optical path difference of each interference channel is controlled at the subwavelength level, typically within 100nm, to ensure high resolution of the interference fringes. The entire optical calculation process is executed in parallel on the chip, eliminating the need for traditional electronic computing resources, thus achieving ultra-high-speed data processing and ensuring real-time morphological feature extraction. After acquiring the embryo morphological feature vector, it needs to be transmitted over long distances to ensure data integrity and accuracy. To this end, a low-loss optical transmission scheme based on silicon waveguides is adopted, using high-purity silicon material with a loss of less than 0.01dB / cm, and integrating a passive mode-matching structure in the waveguide to reduce scattering loss. In the specific implementation process, the waveguide width was optimized to 450nm, and a bending radius of not less than 5μm was adopted to reduce mode mismatch loss. Simultaneously, to prevent environmental noise interference, the entire transmission channel employed fiber optic encapsulation technology to improve anti-interference capabilities, thereby obtaining high-fidelity embryonic morphological data. After acquiring high-fidelity embryonic morphological feature data, nonlinear transformation methods were used to analyze the structure-function correlation characteristics of the embryo. Specifically, the nonlinear Fourier transform (NFT) method was used to map the spatiotemporal optical field data to the nonlinear spectral domain, thereby extracting the complex nonlinear relationship between embryonic morphology and developmental potential.In the experiment, the computation window for the nonlinear transformation was set to 5 fs to ensure ultra-high resolution data extraction. Simultaneously, Higher-Order Mode Decomposition (HOD) was used to further mine potential biophysical features from embryonic morphological characteristics, such as the coupling effect between cell division rate and metabolic activity. After the nonlinear transformation, the extracted structure-function correlation features typically exhibit high dimensionality, thus requiring optimization and dimensionality reduction. Specifically, Principal Component Analysis (PCA) combined with Manifold Learning (LLE) was used for nonlinear dimensionality reduction. In the experiment, the PCA feature selection threshold was set to 95%, retaining 95% of the information to minimize information loss. Simultaneously, LLE was used to further optimize the feature space, ensuring that morphological features retain their key information in the low-dimensional space. The final output of this process is the optimized morphological feature representation, which can be used for subsequent classification and analysis. Finally, optical interferometry was used to classify the optimized morphological feature representation to identify embryonic characteristics at different developmental stages. Specifically, a classification algorithm based on optical Fourier transform was employed. The light field features were projected into the frequency domain using Fast Fourier Transform (FFT), and a Support Vector Machine (SVM) classifier was used for pattern recognition. During the experiment, the kernel function of the SVM was set to a Gaussian radial basis function (RBF), and the parameter C was set to 10 to improve classification accuracy. Ultimately, the interferometry-based classification method effectively distinguished between high-quality and low-quality embryos and output the final embryo morphology data.

[0093] This invention achieves efficient acquisition of multi-angle light field information by importing embryonic morphological parameter data into a silicon-based photonic interferometer chip integrated with a 128×128 microring resonator array, enabling comprehensive recording of the spatial structural features of the embryo's optical properties in different directions. Phase modulation processing is applied to the multi-angle embryonic light field information to ensure high-resolution reconstruction of the light field data, generating an embryonic characteristic light phase map, providing accurate coherent information for subsequent morphological analysis. Parallel optical calculations are performed on the embryonic characteristic light phase map using the coherence of light, fully leveraging the high-speed parallel computing capabilities of optical processing to obtain efficient and high-precision embryonic morphological feature vectors. Based on this, low-loss optical waveguides are used for high-fidelity data transmission, effectively reducing the loss of morphological information during transmission and ensuring data integrity and accuracy. Nonlinear transformations are applied to the high-fidelity morphological feature data to deeply explore the relationship between the embryo's spatial structural information and biological functional characteristics, revealing the intrinsic correlation between morphology and function during embryonic development. Subsequently, dimensionality optimization techniques are used to reduce data complexity and improve the expressive power of morphological features, making subsequent pattern recognition more efficient. Finally, an interferometric classification method was used to analyze the optimized morphological features, enabling the embryo morphology data to have good distinguishability on the basis of high precision and high resolution, providing accurate data support for embryo developmental potential assessment and personalized reproductive medicine decision-making.

[0094] Preferably, the embryonic developmental potential assessment in step S2 includes:

[0095] Temporal features were extracted from the embryo morphology data to obtain a dynamic feature map of embryonic development;

[0096] Energy efficiency was calculated from the dynamic characteristic map of embryonic development and the energy distribution pattern during cell division, thereby obtaining the metabolic efficiency index;

[0097] By analyzing the dynamic characteristic map of embryonic development, critical turning points in the embryonic development process are identified, the stability and plasticity of the developmental trajectory are determined, and developmental elasticity score data are obtained.

[0098] Quantitative assessment of the dynamic characteristics of embryonic development based on cell arrangement and polarization patterns was performed to obtain morphological and tissue integrity data.

[0099] Based on metabolic efficiency index, developmental elasticity score data and morphological tissue integrity data, a preliminary implantation potential prediction value is obtained by probabilistic inference through photon interferometry.

[0100] The preliminary implantation potential predictions are calibrated based on the complex nonlinear relationship of coherent optical path coding to obtain calibrated implantation probability data.

[0101] Embryonic developmental potential data were obtained by assessing the embryonic developmental potential based on chromosome euploidy estimation, mitochondrial function score, and epigenetic status using calibrated implantation probability data.

[0102] In this embodiment, a high-resolution time-lapse imaging system is first used to record the entire process of embryonic development from fertilization to blastocyst formation, with a sampling interval of 10 minutes to capture key morphological changes. Subsequently, a time-series analysis method based on light field reconstruction is used to extract key temporal features during embryonic development, such as cell division time points, cytoplasmic reorganization dynamics, cell symmetry, and spatial arrangement. By constructing an optical time-series matrix with a temporal resolution of at least 100 ms, short-time Fourier transform (STFT) is used to analyze the frequency domain features during development, and a multi-scale recurrent neural network (MS-RNN) is combined to perform deep modeling of morphological change trends, ultimately forming a dynamic feature map of embryonic development. This map can intuitively show the evolution of embryonic morphological features at different stages over time. After obtaining the dynamic feature map of embryonic development, optical metabolic imaging technology is further used to analyze energy utilization during cell division. The specific implementation method employs two-photon fluorescence lifetime imaging (FLIM) combined with phosphorylation and dephosphorylation characteristic ratio measurements to monitor changes in the fluorescence lifetime of nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD) within embryonic cells, thereby quantifying cellular metabolic activities. In the experiment, excitation wavelengths were set at 750 nm and 920 nm, and the fluorescence lifetime measurement range was set between 100 ps and 10 ns. The energy consumption rate at different stages was calculated using a fluorescence lifetime decay fitting model, and an energy distribution heatmap during cell division was constructed. Finally, a metabolic efficiency index was calculated based on the time-energy integral, which can be used to assess the stability and efficiency of energy utilization during embryonic development. To identify key turning points in embryonic development, a phase space reconstruction method was used to map the embryonic morphological evolution process onto a multidimensional dynamic system, analyzing cell division rhythm, morphological change rate, and developmental rhythm fluctuations. In the specific experiment, a phase space trajectory was constructed using the delayed embedding method, with a time delay τ = 30 min and an embedding dimension d = 5. The fractal dimension and maximum Lyapunov exponent of the developmental trajectory were calculated. If the embryonic developmental trajectory has a high fractal dimension and a Lyapunov exponent below 0.1, it indicates a relatively stable developmental process; conversely, it means that the developmental trajectory has greater plasticity. The developmental elasticity score data calculated by this method can be used to assess the embryo's adaptability to changes in the external environment. The acquisition of morphological and tissue integrity data relied on optical coherence tomography (OCT) combined with polarized light microscopy (PSM) to measure the embryonic cell arrangement structure and polarization direction with high precision. In the specific experiment, the axial resolution of OCT was set to 1 μm, and the lateral resolution to 5 μm. A spectral domain scanning method was used to acquire the hierarchical structure of cell arrangement. Simultaneously, PSM technology assessed cell polarization patterns by measuring the birefringence properties of the cell membrane lipid bilayer.During the experiment, the Cell Packing Density (CPD) and Polarization Deviation Angle (PDA) of the embryos were calculated. Embryos with a CPD greater than 0.85 and a PDA less than 10° were considered to have high morphological and tissue integrity. The resulting morphological and tissue integrity data can be used to assess the stability of embryonic structure and the coordination of intercellular connections. After obtaining the metabolic efficiency index, developmental elasticity score, and morphological and tissue integrity data, multi-factor fusion analysis was performed using photonic interferometry to achieve a preliminary assessment of embryo implantation potential. Specifically, an integrated Mach-Zehnder interferometer (MZI) array was used. High-dimensional quantum state mapping between data was achieved through phase modulation of different interference paths, and the probability distribution was derived using the interference results of coherent light waves. In the experiment, the incident light wavelength was set to 1310 nm, and a silicon-based optical waveguide chip was used for calculation. The correlation between the relative phase shift of the optical interference output and the various characteristic data of the embryo was greater than 0.95, thus ensuring the reliability of the calculation results. Finally, a preliminary implantation potential prediction value was obtained through coherent interferometry calculations. This prediction value can be used to assess the likelihood of successful embryo implantation after implantation. Since the prediction of embryo implantation potential involves complex nonlinear relationships, the preliminary implantation potential prediction value needs to be calibrated. Specifically, a deep learning optimization strategy based on optical path coding was used to construct a Photonic Neural Network (PNN), mapping embryo feature data to a high-dimensional optical computation space, and nonlinear feature learning was achieved by modulating the optical interferometry path. During the experiment, a 16-channel phased-array optical path modulation technique was used, with the phase shift control accuracy of each optical path within 0.01π to achieve high-precision fitting of complex nonlinear relationships. The final calibrated implantation probability data provides more accurate embryo implantation prediction results, reduces the false positive rate, and improves the reliability of clinical decisions. Based on the calibrated implantation probability data, embryonic genomic and metabolic characteristics were further integrated to achieve a comprehensive assessment of embryonic developmental potential. The specific implementation method employed single-cell whole-genome amplification (scWGA) technology to determine embryonic chromosome euploidy and analyzed embryonic karyotype stability using fluorescence in situ hybridization (FISH). Fluorescence resonance energy transfer (FRET) imaging was used to assess mitochondrial oxidative phosphorylation (OXPHOS) function and measure mitochondrial membrane potential (ΔΨm). Simultaneously, single-molecule epigenetic sequencing was used to analyze DNA methylation patterns and histone modification status. In the experiment, the excitation wavelength for FRET measurement was set at 405 nm, and the emission wavelength at 525 nm. Time-correlated fluorescence lifetime measurement was employed to ensure measurement accuracy.Finally, by combining the above data, a photon probability calculation model based on Bayesian inference was used to derive embryo development potential data. This data can provide a precise basis for clinical embryo selection and improve the success rate of in vitro fertilization.

[0103] This invention extracts temporal features from embryonic morphological data to construct a dynamic feature atlas of embryonic development, systematically quantifying morphological changes at different developmental stages and providing precise data support for analyzing growth patterns. Based on this atlas, energy utilization during cell division is calculated, identifying energy allocation patterns and quantifying the metabolic efficiency of the embryo at different developmental stages, generating a metabolic efficiency index to characterize the balance and stability of embryonic energy utilization. Combining the dynamic feature atlas, key turning points in embryonic development are analyzed to assess the stability and plasticity of the developmental trajectory, ultimately obtaining developmental resilience scores to quantify the embryo's adaptability under different environmental influences. Furthermore, a quantitative analysis method based on cell arrangement and polarization patterns is employed to assess the integrity of embryonic morphology and tissues, accurately depicting the developmental coordination of the embryo in spatial structure. Subsequently, the metabolic efficiency index, developmental resilience scores, and morphological integrity data are integrated, and probabilistic inference is performed using photon interferometry to generate preliminary implantation potential prediction values, providing preliminary quantitative indicators for embryo selection. Building upon this foundation, the implantation potential predictions are further calibrated by encoding complex nonlinear relationships through coherent optical paths, making the calculation of embryonic developmental potential more robust and ensuring more reliable prediction results. Finally, by combining multidimensional biological indicators such as chromosome euploidy estimation, mitochondrial function scoring, and epigenetic status, the calibrated implantation probability data is analyzed in depth to comprehensively assess the embryonic developmental potential, providing a high-precision reference for clinical decision-making and making embryo selection and personalized assisted reproductive strategies more scientific and accurate.

[0104] Preferably, step S3 includes the following steps:

[0105] Step S31: Utilize A bilayer oxide structure memristor array performs feature dimensionality reduction processing on immune cell function data and genomic variation data to obtain a high-dimensional feature compression representation, in which each memristor has 64 adjustable conductance states.

[0106] In this embodiment, using A bilayer oxide structure memristor array is used to perform feature dimensionality reduction on immune cell function data and genomic variation data to reduce data dimensionality while retaining key information. Specifically, immune cell function data includes multiple parameters such as cell proliferation rate, cytokine secretion level, and cytotoxicity score, while genomic variation data involves single nucleotide polymorphisms (SNPs), copy number variations (CNVs), and methylation levels. First, the input data is standardized to ensure that the numerical range of the data fits the conductance adjustment range of the memristors. Then, the data matrix is ​​input into the memristor array, where the conductance state of each memristor can be adjusted in 64 discrete levels, representing different feature weights. Pulse programming is used to apply training pulses to the memristors, adjusting their conductance state according to the target dimensionality reduction mapping, ultimately obtaining a high-dimensional feature compression representation. This representation effectively reduces data redundancy and improves computational efficiency.

[0107] Step S32: Perform a nonlinear transformation on the high-dimensional feature compression representation, and apply a pulse sequence of ±0.8V to ±1.2V to regulate the conduction state of the memristor, thereby obtaining the nonlinear feature mapping matrix;

[0108] In this embodiment, the obtained high-dimensional feature compression representation undergoes a nonlinear transformation to enhance the data's discriminative ability. Specifically, a pulse sequence ranging from ±0.8V to ±1.2V is first applied to the memristor array. Variations in pulse amplitude and pulse width control the memristor's conductance, inducing a nonlinear response. During this process, a nonlinear transformation method based on exponential mapping is employed to expand the feature dimension of the input data to a more complex feature space, while simultaneously enhancing the separability between data. Utilizing the programmable nature of the memristor, the mapping results are adaptively adjusted for different data categories to ensure that the nonlinear feature mapping matrix correctly represents the data's distribution characteristics and provides support for subsequent clustering analysis.

[0109] Step S33: Identify the natural clustering structure on the nonlinear feature mapping matrix to obtain preliminary clustering grouping data;

[0110] In this embodiment, natural clustering structures are identified on a nonlinear feature mapping matrix to obtain preliminary clustering information of the data. First, a similarity matrix based on the conduction states of memristors is constructed using spectral clustering, reflecting the similarity relationships between different data points. Then, random perturbation signals are applied to the memristor array to detect the inherent data structure in the feature mapping matrix. Unsupervised learning methods such as k-means or density clustering (DBSCAN) are used to classify the changing trends of the memristor conduction states, thereby generating preliminary clustering data. During this process, the Silhouette coefficient or Davies-Bouldin index can be used to evaluate the clustering effect and ensure the rationality of the clustering structure.

[0111] Step S34: Calculate the feature vector similarity of the preliminary clustering group data based on the memristor network in the memristor array to obtain the immune-genetic subtype data;

[0112] In this embodiment, to further refine the clustering results, immune-genetic subtype data is calculated based on the feature vector similarity of the memristor array. Specifically, representative feature vectors are first extracted from each cluster group, and cosine similarity, Euclidean distance, and dynamic time warping (DTW) similarity between different groups are calculated on the memristor network to measure the cohesion and segregation of data categories. Then, the preliminary cluster group data is adjusted according to the calculation results, so that data points within the same subtype have high similarity, while the similarity between different subtypes is low. Finally, immune-genetic subtype data is generated to provide a basis for subsequent information flow simulation.

[0113] Step S35: Simulate complex information flow based on immune-genetic subtype data to obtain the typing stability index;

[0114] In this embodiment, complex information flow is simulated based on immuno-genetic subtype data to obtain a genotyping stability index. Specifically, firstly, a random walk method based on memristor arrays is used to simulate the dynamic information transfer process between different subtypes. This method, based on the adjustable conductivity characteristics of memristors, maps the data flow path to the topology of the memristor array and calculates the information propagation efficiency and steady-state distribution. Then, a Markov chain steady-state probability calculation method is used to evaluate the stability of different subtypes, calculating the information retention rate within subtypes and the information exchange rate across subtypes, thereby obtaining the genotyping stability index. Subtypes with higher genotyping stability indices typically have stable gene regulatory networks, while subtypes with lower indices may be more susceptible to external environmental factors or genetic drift.

[0115] Step S36: Identify the key driving factors of each subtype in the immuno-genetic subtype data according to the subtyping stability index, and sort the importance of high-dimensional features to obtain subtype feature spectrum data;

[0116] In this embodiment, the key driving factors of each subtype in the immuno-genetic subtype data are identified using the subtyping stability index, and high-dimensional feature importance is ranked to obtain subtype feature spectrum data. First, feature variables with significant differences within different subtypes are extracted, and the contribution of each feature to the classification results is calculated using the SHAP (Shapley Additive Explanations) method based on memristor arrays. Then, the feature contributions are normalized and ranked according to methods such as information gain and mutual information entropy to screen out key driving factors. For the top-ranked feature variables, their distribution patterns across different subtypes are further analyzed to construct subtype feature spectrum data, providing a basis for subsequent subtyping tree construction.

[0117] Step S37: Construct a hierarchical classification tree based on the subtype feature spectrum data and the classification stability index calculation unit to obtain the patient's immune-genetic classification data.

[0118] In this embodiment, a hierarchical classification tree is constructed based on subtype feature spectrum data and a classification stability index calculation unit to obtain patient immuno-genetic typing data. Specifically, firstly, a decision tree algorithm or random forest method is used, with subtype feature spectrum data as input variables and immuno-genetic subtype labels as target variables, to construct an initial classification tree. Then, an optimal pruning algorithm based on memristors is used to remove redundant feature branches, improving the generalization ability of the classification tree. Finally, the confidence of the hierarchical classification tree is calculated by combining the classification stability index, and the classification accuracy is evaluated using cross-validation to ensure the reliability and interpretability of the immuno-genetic typing data.

[0119] This invention adopts A memristor array with a double-layer oxide structure performs feature dimensionality reduction on immune cell function data and genomic variation data, effectively compressing high-dimensional data. Sixty-four tunable conductance states improve the accuracy of feature representation and data storage efficiency. Nonlinear transformation is applied to the dimensionality-reduced high-dimensional features, and a pulse sequence of ±0.8V to ±1.2V is applied to regulate the memristor's conduction state, enabling the feature mapping process to adapt to complex nonlinear relationships and generating a nonlinear feature mapping matrix, enhancing data separability. Based on this matrix, natural clustering structures are identified, and inherent grouping patterns are extracted to obtain preliminary clustering data, improving the adaptive ability of classification. Subsequently, feature vector similarity is calculated based on the memristor network to further optimize clustering accuracy and obtain immune-genetic subtype data, enabling a more precise characterization of individual differences in different immune-genetic traits. The immune-genetic subtype data is used to simulate complex information flow and calculate the subtype stability index, thereby quantifying the robustness and evolutionary trends of different subtypes. Further analysis revealed the key driving factors for each subtype and ranked the importance of high-dimensional features, extracting subtype feature spectrum data to effectively reveal the core characteristics of the immune-genetic pattern. Finally, by comprehensively calculating the subtype feature spectrum data and the subtyping stability index, a hierarchical classification tree was constructed to accurately classify patients' immune-genetic subtypes.

[0120] Preferably, step S4 includes the following steps:

[0121] Step S41: Perform a unified mathematical representation transformation on the endocrine dynamic characteristic data, embryonic developmental potential data, and patient immune-genetic typing data to obtain a multidimensional joint tensor representation;

[0122] In this embodiment, the endocrine dynamics data, embryonic developmental potential data, and patient immuno-genetic typing data are normalized using standardization methods. Specifically, the endocrine dynamics data uses a sliding window method to extract time-series features; the embryonic developmental potential data uses feature vectors constructed based on morphological scores and metabolic dynamic parameters; and the immuno-genetic typing data undergoes dimensionality reduction using principal component analysis. Then, the Kronecker product method is used to perform a unified mathematical representation transformation on the three types of data, constructing a third-order tensor structure. Each tensor dimension corresponds to the time-series features, genetic features, and embryonic features, respectively, ultimately forming a multidimensional joint tensor representation of size N×M×K, where N is the number of samples, M is the comprehensive feature dimension, and K is the time step.

[0123] Step S42: Perform cross-modal feature alignment on the multidimensional joint tensor representation to obtain aligned feature space data;

[0124] In this embodiment, for the multidimensional joint tensor representation, the mutual information maximization method is used to calculate the correlation matrix between features of different modalities, and the orthogonal projection method is used to reduce the dimensionality of low-correlation features, thereby reducing noise interference. Then, a generator-discriminator framework is constructed using a cross-modal adversarial learning network. The generator uses a convolutional neural network (CNN) to learn the common representation of data from different modalities, while the discriminator uses a recurrent neural network (RNN) to determine whether the data comes from the same modality, thereby optimizing the alignment effect of the feature space. Finally, the maximum mean difference (MMD) loss is used to constrain the feature distribution between different modalities, mapping them to the same feature space to obtain aligned feature space data. Its structural dimension is the same as that of the multidimensional joint tensor representation, but the feature distribution between modalities tends to be consistent.

[0125] Step S43: Construct a heterogeneous graph neural network based on silicon photonics chip, and use the heterogeneous graph neural network to perform correlation modeling on the aligned feature space data to obtain multimodal correlation graph structure data. The heterogeneous graph neural network integrates 1024 programmable micro-ring resonators as graph nodes.

[0126] In this embodiment, 1024 programmable microring resonators are integrated on a silicon photonic chip, each resonator corresponding to a graph node for storing single sample features of aligned feature space data. Then, an optical beamsplitter and a tunable filter are used to modulate light intensity, thereby constructing edge weights between different resonators and using them to construct the adjacency matrix of the heterogeneous graph neural network. Next, a graph attention mechanism is used to calculate the information propagation weights between different nodes, combined with photonic matrix multiplication units to implement multi-layer heterogeneous graph convolution operations, extracting nonlinear correlation features between data, and finally outputting multimodal correlation graph structure data, where the embedding of each node represents the cross-modal features of an individual patient.

[0127] Step S44: Extract persistent cohomology features from the multimodal association graph structure data, identify stable patterns across data types, and thus obtain multi-level clinical association pattern data;

[0128] In this embodiment, for multimodal association graph structured data, the high-order topological structure of cross-modal features is first extracted using the persistent cohomology method in Topological Data Analysis (TDA). Specifically, this includes computing the Betti number, persistence barcode, and persistence diagram in the graph to quantify stable patterns among different modalities. Then, a feature clustering model is constructed based on a graph neural network, and the persistent cohomology features are partitioned using a spectral clustering algorithm to identify stable patterns across data types. This results in the construction of multi-level clinical association pattern data, which can be used to predict the clinical risk characteristics of different patients in reproductive medicine.

[0129] Step S45: Perform homomorphic encryption on the multi-level clinical association pattern data to obtain encrypted clinical association data;

[0130] In this embodiment, to ensure the privacy and security of multi-level clinical correlation pattern data, the Paillier homomorphic encryption algorithm is first used to encrypt the data. Specifically, a 1024-bit key is used for public-key encryption, and a random noise perturbation strategy is employed to enhance data privacy protection. Then, a distributed encrypted computing method is used to perform matrix operations on the encrypted data, including basic operations such as addition and multiplication, without decrypting the original data. Finally, encrypted clinical correlation data is obtained, which can be securely shared among different institutions while maintaining data confidentiality.

[0131] Step S46: Construct a knowledge graph index structure based on encrypted clinical correlation data using distributed ledger technology to obtain the initial knowledge graph framework data;

[0132] In this embodiment, a decentralized storage architecture is constructed based on distributed ledger technology. Smart contracts are created within the Hyperledger Fabric framework to implement storage and access control for encrypted clinically relevant data. Then, for different clinical data sources, a knowledge graph construction method based on RDF (Resource Description Framework) is used to convert encrypted data into subject-predicate-object representations, and an index structure is built using the SPARQL query language. Finally, a preliminary knowledge graph framework is formed, which contains relationships between different data entities and supports distributed access and querying.

[0133] Step S47: Deploy a temporal logic reasoning engine on the primary knowledge graph framework and establish a secure access control mechanism to obtain the reproductive medicine database structure.

[0134] In this embodiment, a temporal logic reasoning engine is deployed on a primary knowledge graph framework. First, temporal reasoning rules are defined based on Temporal Description Logic (TDL), and ontology constraint modeling is performed using OWL 2 (Web Ontology Language). Then, combining entity relationships in the knowledge graph, causal relationships between clinical data are automatically derived using inductive reasoning. Finally, a secure access control mechanism is established based on a Zero Trust Architecture, including Role-Based Access Control (RBAC) and Attribute-Based Access Control (ABAC), thereby ensuring data security and traceability. Ultimately, a complete reproductive medicine database structure is obtained, enabling intelligent data storage and analysis.

[0135] This invention provides a unified mathematical representation transformation for endocrine dynamic characteristic data, embryonic developmental potential data, and patient immuno-genetic typing data, enabling data from different sources to be described within the same mathematical framework. Multidimensional joint tensor representation ensures data integrity and correlation. Subsequently, cross-modal feature alignment is performed on the multidimensional joint tensor representation to eliminate modal differences between data, construct a unified feature space, and improve the effectiveness of information fusion. A heterogeneous graph neural network based on silicon photonics chips is used for correlation modeling of the aligned feature space data. This network integrates 1024 programmable microring resonators as graph nodes, enabling efficient modeling of complex nonlinear relationships between data and fully exploring potential interaction patterns between endocrine, embryonic development, and immuno-genetic factors. Persistent homology features are extracted through the multimodal correlation graph structure to identify stable patterns across data types, thereby revealing multi-level clinical correlation information and improving the stability and robustness of medical data analysis. Homomorphic encryption is implemented for multi-level clinical correlation pattern data to ensure data privacy and security during computation, allowing sensitive medical information to be analyzed and computed in an encrypted state without decryption. A knowledge graph index structure based on encrypted clinical data is constructed using distributed ledger technology, making data storage and management more transparent and traceable, and ensuring the integrity and tamper-proof capabilities of the knowledge graph framework. Building upon this initial knowledge graph framework, a temporal logic reasoning engine is deployed, enabling the system to perform temporal reasoning on dynamic medical data. Combined with secure access control mechanisms, this ensures the security and controllability of data queries and reasoning results, ultimately forming a reproductive medicine database structure. This database structure can integrate and manage multi-source medical data, support precision medicine analysis, and provide a solid data foundation for personalized reproductive health assessment, embryo selection, and clinical decision-making.

[0136] The present invention also provides a system for constructing a reproductive medicine database, for executing the above-described method for constructing a reproductive medicine database, the system comprising:

[0137] The biological data acquisition module is used to acquire biological tissue samples from patients; and to acquire data on reproductive endocrine hormones, embryonic morphological parameters, immune cell function, and genomic variation from these biological tissue samples.

[0138] The potential assessment module is used to perform time-series analysis of hormone fluctuation patterns based on reproductive endocrine hormone data to obtain endocrine dynamic characteristic data; and to perform image recognition processing based on the coherent superposition characteristics of light on embryonic morphological parameter data, and to assess embryonic developmental potential to obtain embryonic developmental potential data.

[0139] The memristor heterogeneity typing module is used to perform hierarchical clustering of individual patient heterogeneity based on immune cell function data and genomic variation data to obtain patient immuno-genetic typing data. The hierarchical clustering process is achieved through... A memristor array with a double-layer oxide structure is realized, with each memristor having 64 adjustable conductance states;

[0140] An encrypted storage module is used to perform multidimensional feature fusion processing on endocrine dynamic characteristic data, embryonic developmental potential data, and patient immune-genetic typing data, and to construct a homomorphically encrypted distributed knowledge graph index to obtain the reproductive medicine database structure.

[0141] This invention utilizes a biological data acquisition module to systematically acquire patient biological tissue samples and precisely extract data on reproductive endocrine hormones, embryonic morphological parameters, immune cell function, and genomic variations, thus comprehensively integrating personalized medical information. Based on this data, the potential assessment module analyzes the temporal fluctuation patterns of reproductive endocrine hormones through time-series analysis, constructing dynamic characteristic data to reveal the regulatory laws of the endocrine system on the reproductive process. Embryonic morphological parameter data is analyzed using high-precision image recognition based on the coherent superposition characteristics of light, achieving a quantitative assessment of embryonic developmental potential and ensuring high stability and reliability of the assessment results. The memristor heterogeneity typing module utilizes... A memristor array with a double-layer oxide structure performs hierarchical clustering of immune cell function data and genomic variation data to identify immuno-genetic heterogeneity among individuals and construct a refined patient typing model. The 64 tunable conductance states within the memristor array enable it to accurately characterize the complex information flow patterns between different immuno-genetic subtypes, improving the resolution and adaptability of the typing. At the data management level, an encrypted storage module performs multi-dimensional feature fusion on endocrine dynamic characteristic data, embryonic developmental potential data, and patient immuno-genetic typing data, allowing for in-depth mining of cross-correlation patterns between data and storing the data using homomorphic encryption to ensure data is not leaked during computation. Furthermore, a knowledge graph index is constructed based on distributed ledger technology, giving the reproductive medicine database structure traceability and tamper-proof capabilities, providing reliable data support for personalized reproductive health assessment. This database not only enhances the systematic nature of reproductive health analysis but also enables precision medicine strategies to be implemented under data security, providing efficient and reliable intelligent auxiliary support for clinical decision-making.

[0142] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0143] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for constructing a reproductive medicine database, characterized by, The method comprises the following steps: Step S1: obtaining a patient biological tissue sample; obtaining reproductive endocrine hormone data, embryo morphological parameter data, immune cell function data, and genomic variation data of the patient biological tissue sample, step S1 comprising: Step S11: obtaining a patient biological tissue sample; Step S12: introducing the patient biological tissue sample into the PDMS microfluidic chip of the quantum dot sensing array with integrated CdSe / ZnS core-shell structure, so as to obtain biological sample dispersion flow data, wherein the PDMS microfluidic chip has a multi-channel structure, and the channel width is 30-50 ; Step S13: modifying specific antibodies on the biological sample dispersion flow data on the quantum dot sensing array, and sorting and capturing NK cells, T cell subsets, B cells, and monocytes in the patient biological tissue sample, thereby obtaining immune cell capture data, wherein the modification specific antibodies include CD56, CD3, CD4, CD8, CD19, CD14, and HLA-DR; Step S14: multi-spectral excitation of the immune cell capture data in the wavelength range of 525-655 nm, and recording the quantum dot fluorescence resonance energy transfer signal change, thereby obtaining immune cell function data; Step S15: obtaining reproductive endocrine hormone data, embryo morphological parameter data, and genomic variation data of the patient biological tissue sample, step S15 comprising: Step S151: processing the follicular fluid in the patient biological tissue sample by a microfluidic chip electrochemical sensor array, thereby obtaining primary endocrine data flow, wherein the microfluidic chip electrochemical sensor array contains 64 independent electrode units, and each unit is modified with specific aptamers for estradiol, progesterone, FSH, and LH on the surface; Step S152: The primary endocrine data stream is detected by the nitrogen vacancy defect-based diamond quantum sensor prepared by the 28nm process to obtain a high-precision endocrine concentration matrix, and the detection accuracy of the diamond quantum sensor reaches Molar concentration, sampling frequency is 300Hz; Step S153: extracting pulse signal from the high-precision endocrine concentration matrix, thereby obtaining hormone secretion pulse feature data; Step S154: combining the hormone secretion pulse feature data with time sequence labels to obtain reproductive endocrine hormone data; Step S155: continuously monitoring the patient biological tissue sample by an integrated microfluidic culture cavity and a phase contrast microscopic imaging system, thereby obtaining an embryo dynamic development image sequence, wherein the phase contrast microscopic imaging system is composed of a frequency-adjustable optical resonant cavity and a high-speed CMOS image sensor; Step S156: performing optical coherence processing on the embryo dynamic development image sequence, thereby obtaining an embryo multi-parameter morphological feature set, wherein the optical coherence processing is realized by a photonic interference calculation unit containing an integrated micro-ring resonator with a 256x256 array; Step S157: performing optical calculation processing on the embryo multi-parameter morphological feature set, thereby obtaining embryo morphological parameter data, wherein the embryo morphological parameter data includes cleavage rate, cleavage symmetry, blastocyst cavity formation dynamics, and chromosome euploidy prediction value; Step S158: performing reproductive related variation site analysis on the extracted nucleic acid in the patient biological tissue sample, thereby obtaining genomic variation data; Step S2: performing time series analysis processing on the hormone fluctuation pattern according to the reproductive endocrine hormone data to obtain endocrine dynamic feature data; performing image recognition processing based on the coherent superposition characteristics of light on the embryo morphological parameter data, and performing embryo development potential evaluation to obtain embryo development potential data; Step S3: stratified clustering of patient individual heterogeneity from immune cell function data and genomic variation data, resulting in patient immune-genetic profiling data, wherein stratified clustering is by 32 x 32 Implementation of a two-layer oxide structure memristor array, each memristor having 64 adjustable conductance states, step S3 includes: Step S31: using 32x32 The memristor array with double-layer oxide structure is used for feature dimension reduction processing on immune cell function data and genomic variation data, so as to obtain high-dimensional feature compressed representation, wherein each memristor has 64 adjustable conductance states. Step S32: Nonlinear transformation is performed on the high-dimensional feature compressed representation, and a pulse sequence regulation of ±0.8V to ±1.2V is applied to control the conductive state of the memristor, thereby obtaining a nonlinear feature mapping matrix; Step S33: Natural clustering structure is identified on the nonlinear feature mapping matrix, thereby obtaining preliminary clustering grouping data; Step S34: Feature vector similarity calculation based on the memristor network in the memristor array is performed on the preliminary clustering grouping data, thereby obtaining immune-genetic subtype data; Step S35: Complex information flow is simulated according to the immune-genetic subtype data, thereby obtaining a typing stability index; Step S36: According to the typing stability index, the key driving factors of each subtype in the immune-genetic subtype data are identified, and high-dimensional feature importance is sorted, thereby obtaining subtype feature spectrum data; Step S37: A hierarchical classification tree is constructed according to the subtype feature spectrum data and the typing stability index calculation unit, thereby obtaining patient immune-genetic typing data; Step S4: Multidimensional feature fusion processing is performed on the endocrine dynamic feature data, embryonic development potential data, and patient immune-genetic typing data, and a homomorphic encryption distributed knowledge graph index is constructed, thereby obtaining a reproductive medicine database structure, and step S4 includes: Step S41: Unified mathematical representation conversion is performed on the endocrine dynamic feature data, embryonic development potential data, and patient immune-genetic typing data, thereby obtaining a multidimensional joint tensor representation; Step S42: Cross-modal feature alignment is performed on the multidimensional joint tensor representation, thereby obtaining aligned feature space data; Step S43: A heterogeneous graph neural network based on a silicon photonics chip is constructed, and the aligned feature space data is modeled using the heterogeneous graph neural network, thereby obtaining multimodal association graph structure data, wherein the heterogeneous graph neural network integrates 1024 programmable microring resonators as graph nodes; Step S44: Persistent homology features are extracted on the multimodal association graph structure data, and stable patterns across data types are identified, thereby obtaining multi-level clinical association mode data; Step S45: Homomorphic encryption processing is performed on the multi-level clinical association mode data, thereby obtaining encrypted clinical association data; Step S46: A knowledge graph index structure is constructed based on the encrypted clinical association data through a distributed ledger technology, thereby obtaining primary knowledge graph framework data; Step S47: A temporal logic reasoning engine is deployed on the primary knowledge graph framework, and a secure access control mechanism is established, thereby obtaining a reproductive medicine database structure; Step S158 includes the following steps: Step S1581: Single molecule sequencing based on a graphene nanopore array is performed on the nucleic acid extracted from the patient's biological tissue sample, thereby obtaining single base resolution DNA translocation signals; Step S1582: The DNA translocation signals are collected and processed in parallel through a multi-channel graphene transistor array, thereby obtaining genomic raw read data, wherein the sensitivity of the multi-channel graphene transistor array reaches the detection of current changes caused by a single DNA molecule passing through; Step S1583: Base accuracy verification processing is performed on the genomic raw reads, thereby obtaining high-precision genomic sequencing data; Step S1584: using the memristor neuromorphic computing unit to analyze the high-precision genomic sequencing data for reproductive-related variant sites, thereby obtaining genomic variant data, wherein the genomic variant data includes single nucleotide polymorphism, copy number variation and structural variation information of the reproductive-related genes; including: using a convolutional neural network (CNN) to extract features from the sequencing data, converting the DNA sequence into a numerical matrix, and performing pattern matching analysis; using a long short-term memory (LSTM) network to identify variant sites of the reproductive-related genes, including single nucleotide polymorphism (SNP), copy number variation (CNV) and structural variation (SV); the system uses known variant data in the genomic database for training, and calculates the occurrence probability of each variant through Bayesian inference; finally, the key gene variants that may affect reproductive function are extracted, and classified according to their impact, to obtain the patient's genomic variant data.

2. The method for constructing a reproductive medicine database according to claim 1, wherein, The time series analysis and processing of the hormone fluctuation pattern in step S2 includes: Performing time axis standardization processing on the reproductive endocrine hormone data, constructing a 400-hour unified time window with the first day of the menstrual cycle as the reference point to obtain standardized hormone data; Performing femtosecond-level sampling on the standardized hormone data with a time resolution of 10 femtoseconds to obtain a high-time-resolution hormone fluctuation curve; Extracting an intrinsic mode function set from the high-time-resolution hormone fluctuation curve; Performing correlation analysis on the intrinsic mode function set to obtain hormone interaction effect data; Extracting nonlinear dynamics features from the hormone interaction effect data; Performing system identification on the nonlinear dynamics features to obtain a hormone regulation dynamics model; Performing forward evolution simulation on the hormone regulation dynamics model to obtain endocrine dynamic characteristic data.

3. The method for constructing a reproductive medicine database according to claim 2, wherein, The image recognition processing based on the coherent superposition characteristics of light in step S2 includes: Importing the embryo morphological parameter data into a silicon-based photonic interferometer chip integrated with a 128x128 microring resonator array to obtain multi-angle embryo light field information; Performing phase modulation processing on the multi-angle embryo light field information to obtain an embryo characteristic light phase map; Performing parallel optical calculation on the embryo characteristic light phase map using the coherence of light to obtain an embryo morphological feature vector; Performing low-loss transmission on the embryo morphological feature vector to obtain high-fidelity morphological feature data, wherein the waveguide loss of the low-loss transmission is less than 0.01 dB / cm; Performing nonlinear transformation on the high-fidelity morphological feature data to obtain embryo structure-function correlation features; Performing dimension optimization on the embryo structure-function correlation features to obtain optimized morphological feature representations; Performing classification processing based on interferometric measurement on the optimized morphological feature representations to obtain embryo morphological data.

4. The method for constructing a reproductive medicine database according to claim 1, wherein, The embryo development potential evaluation in step S2 includes: Extracting the time series features in the embryo morphological data to obtain an embryo development dynamic feature map; Performing energy efficiency calculation on the embryo development dynamic feature map, and the energy distribution pattern in the cell division process, thereby obtaining a metabolic efficiency index; Analyzing the critical turning points in the embryo development process according to the embryo development dynamic feature map, identifying the stability and plasticity of the development trajectory, and thereby obtaining development elasticity score data; The morphological integrity data is obtained by quantitatively evaluating the dynamic characteristics of embryonic development based on cell arrangement and polarization mode; The preliminary implantation potential prediction value is obtained by probabilistic reasoning through photon interference according to the metabolic efficiency index, the development elasticity score data and the morphological integrity data; The calibrated implantation probability data is obtained by calibration processing of the preliminary implantation potential prediction value based on complex nonlinear relationship of coherent light path encoding; The embryo development potential data is obtained by embryo development potential evaluation based on chromosome ploidy estimation, mitochondrial function score and epigenetic state for the calibrated implantation probability data.

5. A system for the construction of a reproductive medicine database, characterized in that, The system for constructing a reproductive medicine database comprises: a biological data acquisition module for acquiring a patient biological tissue sample, and obtaining reproductive endocrine hormone data, embryo morphological parameter data, immune cell function data and genomic variation data of the patient biological tissue sample; a potential evaluation module for performing time series analysis processing on a hormone fluctuation mode according to the reproductive endocrine hormone data to obtain endocrine dynamic characteristic data, and performing image recognition processing on the embryo morphological parameter data based on light coherent superposition characteristics, and performing embryo development potential evaluation to obtain embryo development potential data; A memristor heterogeneity profiling module for stratified clustering of patient individual heterogeneity from immune cell function data and genomic variation data to obtain patient immune-genetic profiling data, wherein the stratified clustering is performed by 32x32 Memristor array implementation of a bilayer oxide structure, each memristor having 64 tunable conductance states; an encrypted storage module for performing multi-dimensional feature fusion processing on the endocrine dynamic characteristic data, the embryo development potential data and the patient immune-heritage typing data, and constructing a homomorphic encryption distributed knowledge graph index to obtain a reproductive medicine database structure.

Images