Multi-parameter based mesenchymal stem cell culture monitoring system

Abstract

The application provides a multi-parameter-based mesenchymal stem cell culture monitoring system and relates to the technical field of data processing, and comprises: a unit for generating a first control instruction when the metabolic-morphological abnormality correlation strength in the joint fuzzy state matrix is determined to reach a preset correlation threshold value through a fuzzy rule base; generating a second control instruction when an abnormal state flag indicates isolated metabolic abnormality; a unit for driving the buffer injection unit and the nutrient supplement pump in cooperation in response to the first control instruction; and a unit for dynamically adjusting the stirring rate controller, the temperature feedback loop and the gas proportional valve opening of the bioreactor in response to the second control instruction. The application improves the control efficiency and the cell culture stability.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a multi-parameter-based mesenchymal stem cell culture monitoring system. Background Technology

[0002] Large-scale culture of mesenchymal stem cells (MSCs) is a core component of regenerative medicine, and its quality directly impacts clinical efficacy. Current culture monitoring systems primarily rely on single-dimensional parameter threshold alarms (such as pH, dissolved oxygen, or microscopic image analysis), and some have the following limitations:

[0003] For example, traditional systems cannot dynamically correlate metabolic abnormalities (such as lactate accumulation) with cell morphological changes (such as pseudopodia contraction), resulting in delayed early warnings; the nonlinear increase in cell metabolic rate during the later stages of culture can easily lead to false alarms when fixed thresholds are used (such as misjudging normal metabolic fluctuations in high-density culture as abnormal); and regulatory instructions targeting local abnormalities (such as single temperature drift) may exacerbate the overall microenvironment imbalance (such as shear damage caused by accelerated stirring).

[0004] Therefore, although some studies have attempted multi-parameter fusion analysis in the existing technology, it is still not possible to achieve graded and precise intervention, resulting in a certain batch of cell activity reduction due to remedial delays. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a multi-parameter-based mesenchymal stem cell culture monitoring system, which improves control efficiency and cell culture stability.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] Firstly, a multi-parameter-based mesenchymal stem cell culture monitoring system includes:

[0008] The acquisition module is used to acquire real-time metabolic dynamic parameter sets of the culture medium and periodic morphological parameter sets of the cell population;

[0009] The generation module is used to train and generate a dynamic metabolic threshold library based on historical data of the metabolic dynamic parameter set, and at the same time, to construct a fuzzy morphological benchmark library based on historical data of the morphological parameter set.

[0010] The analysis module is used to compare the deviation between the real-time dynamic metabolic parameter set and the dynamic metabolic threshold library to generate the membership degree of metabolic abnormality; the real-time morphological parameter set is mapped to the preset feature space through the mirror projection algorithm, and the morphological abnormality confidence is calculated in the space based on the membership function of the fuzzy morphological benchmark library; the membership degree of metabolic abnormality and the morphological abnormality confidence are fused to generate a joint fuzzy state matrix.

[0011] The decision module is used to generate a first-level control instruction when the correlation strength between metabolic and morphological abnormalities in the joint fuzzy state matrix reaches a preset correlation threshold as determined by the fuzzy rule base; and to generate a second-level control instruction when the abnormal state flag indicates an isolated metabolic abnormality.

[0012] The output module is used to respond to primary control commands and collaboratively drive the buffer injection unit and nutrient supplementation pump; and to respond to secondary control commands and dynamically adjust the stirring rate controller, temperature feedback loop and gas proportional valve opening of the bioreactor.

[0013] Furthermore, the real-time metabolic dynamic parameter set of the culture medium and the periodic morphological parameter set of the cell population are obtained, including:

[0014] Raw time-series data streams of culture medium were acquired using a biosensor array; a sliding window processor was used to perform segmented feature extraction on the raw time-series data streams to generate a primary metabolic feature set containing a mean vector, a standard deviation vector, and a parameter change rate sequence; time dimension alignment and feature concatenation were performed on the primary metabolic feature set to generate a structured metabolic dynamic parameter set;

[0015] High-resolution image sets of cell populations are periodically captured using a microscopic imaging unit; layered processing is then performed on the high-resolution image sets.

[0016] Edge segmentation is performed at the pixel level to generate a cell contour dataset. Based on the cell contour dataset, geometric topological features are extracted at the single cell level to generate a single cell feature set. At the population level, population morphology indicators are generated based on the spatial distribution relationship of the single cell feature set.

[0017] Single-cell feature sets and population morphological indicators are input into a multi-scale feature fusion engine, and cross-scale associated features are generated through spatial distribution topology modeling.

[0018] A structured morphological parameter set is constructed based on cross-scale correlation features and group morphological indicators.

[0019] Furthermore, a dynamic metabolic threshold library is generated by training based on historical data from the metabolic dynamic parameter set, and a fuzzy morphological benchmark library is constructed based on historical data from the morphological parameter set, including:

[0020] The system retrieves the dynamic metabolic parameter sets for each culture cycle from the historical metabolic database; performs timestamp alignment on the parameter sets to generate time-series-related metabolic feature tensors; constructs a dynamic probability distribution model based on the metabolic feature tensors, and generates dynamic confidence interval sequences by extracting boundary values ​​under pre-set confidence levels.

[0021] The dynamic confidence interval sequences were integrated into a structured dynamic metabolic threshold library according to the culture stage;

[0022] Extract morphological parameter sets across batches from a historical morphological database; perform time-aware incremental clustering on single-cell feature sets:

[0023] The feature subsets are divided according to the culture time window. Density peak clustering is performed on each subset to generate local morphological benchmarks. The local benchmarks are then fused with time decay weights to generate a global typical cell morphology benchmark vector.

[0024] Dynamic evolution modeling of group morphology indicators:

[0025] Based on the timestamp index, the evolution trajectory of population indicators is constructed, and a dynamic benchmark interval set is generated by fitting a Gaussian mixture model in segments according to the cultivation stage.

[0026] The typical cell morphology baseline vector and the dynamic baseline interval set are input into the rule generator, the adaptive membership function set for morphological parameter anomaly determination is defined, and the membership function set and the corresponding baseline data are associated and stored to generate a structured fuzzy morphological baseline library.

[0027] Furthermore, the joint fuzzy state matrix includes:

[0028] Obtain the real-time dynamic metabolic parameter set and call the threshold vector corresponding to the culture stage in the dynamic metabolic threshold library;

[0029] Perform component confidence interval deviation calculations on the mean vector, standard deviation vector, and parameter change rate sequence of the parameter set to generate the original deviation set;

[0030] Based on the original deviation set, an abnormal weight coefficient set is generated according to the sensitivity of the parameters during the cultivation stage;

[0031] By merging the original deviation set and the abnormality weight coefficient set, a scalar metabolic abnormality membership degree is output.

[0032] Obtain single-cell feature sets and population morphological indices from a real-time morphological parameter set;

[0033] Call upon the typical cell morphology reference vectors and dynamic reference interval sets from the fuzzy morphology reference library;

[0034] Project the single-cell feature set onto the feature space spanned by the reference vector; calculate the Mahalanobis distance between the cell population distribution and the reference vector in the projection space to generate the single-cell deviation vector; perform dynamic interval cross-boundary detection on the population morphology index to generate a population cross-boundary marker set;

[0035] Based on the single-cell deviation vector and the population out-of-bounds marker set, a scalar morphological anomaly confidence level is obtained;

[0036] The membership degree of metabolic abnormality and the confidence degree of morphological abnormality are combined into a two-dimensional abnormal state vector. Based on the historical abnormal pattern library, an inter-dimensional correlation weight matrix is ​​generated. Tensor multiplication is performed on the abnormal state vector and the correlation weight matrix, and the output operation result is used as the joint fuzzy state matrix.

[0037] Furthermore, the membership degree of metabolic abnormalities and the confidence degree of morphological abnormalities are combined into a two-dimensional abnormal state vector; an inter-dimensional correlation weight matrix is ​​generated based on a historical abnormal pattern library, and a tensor multiplication operation is performed on the abnormal state vector and the correlation weight matrix. The output result is used as a joint fuzzy state matrix, including:

[0038] Metabolic abnormality membership degree is used as the first dimension value, and morphological abnormality confidence degree is used as the second dimension value;

[0039] Generate a two-dimensional abnormal state vector by combining elements in dimensional order;

[0040] When the value of the first dimension exceeds the metabolic dominance threshold, the metabolic abnormality dominance factor is activated; when the value of the second dimension exceeds the morphological dominance threshold, the morphological abnormality dominance factor is activated; a weight allocation instruction is generated based on the type of activated dominance factor.

[0041] The system responds to weight allocation instructions by invoking the historical anomaly pattern library; it then uses the correlation characteristics of similar dominant factors within the historical anomaly pattern library as a basis for further analysis.

[0042] Based on metabolic and morphological abnormalities, an adaptive weight matrix is ​​generated that includes self-effect weights and cross-dimensional correlation weights.

[0043] Perform tensor convolution operation on the two-dimensional abnormal state vector and the adaptive weight matrix, and output the result as the joint fuzzy state matrix.

[0044] Furthermore, when the correlation strength between metabolic and morphological abnormalities in the joint fuzzy state matrix reaches a preset correlation threshold as determined by the fuzzy rule base, a first-level control instruction is generated; when the abnormal state flag indicates an isolated metabolic abnormality, a second-level control instruction is generated, including:

[0045] Extract the combined intensity of metabolic abnormalities and the combined intensity of morphological abnormalities from the joint fuzzy state matrix;

[0046] An abnormal intensity level identifier set is generated based on preset intensity grading rules;

[0047] Input the abnormal intensity level identifier set into the fuzzy rule base. When both metabolic and morphological abnormality intensities reach the high-level threshold, activate the metabolic-morphological strong correlation flag. When the metabolic abnormality intensity reaches the high level and the morphological abnormality intensity is lower than the decoupling threshold, activate the isolated metabolic abnormality flag.

[0048] In response to metabolic-morphology strongly correlated markers, a primary control instruction generation pathway is triggered.

[0049] The response to isolated metabolic abnormality flags triggers the generation pathway of secondary control instructions.

[0050] Furthermore, in response to primary control commands, it coordinates the driving of the buffer injection unit and nutrient supplementation pump; in response to secondary control commands, it dynamically adjusts the bioreactor's stirring rate controller, temperature feedback loop, and gas proportional valve opening, including:

[0051] In response to the first-level control command, extract the comprehensive intensity of metabolic abnormality and the comprehensive intensity of morphological abnormality from the joint fuzzy state matrix;

[0052] The buffer regulation model is used to process the overall intensity of metabolic abnormalities and generate a buffer pulse injection parameter set; the intensity of metabolic and morphological abnormalities is fused to generate a nutrient supplementation flow rate regulation curve; and the buffer injection unit and nutrient supplementation pump are coordinated to perform linked operations.

[0053] In response to the secondary control command, the parameter change rate sequence and metabolic abnormality membership degree are invoked from the real-time metabolic dynamic parameter set;

[0054] Perform multi-parameter coupling analysis using a dynamic control strategy engine:

[0055] An adaptive correction for stirring rate is generated based on the parameter change rate sequence; a temperature feedback compensation signal is generated based on the membership degree of metabolic anomalies; and a proportional valve opening increment sequence is generated through a gas valve control model.

[0056] Synchronous adjustment of stirring rate controller, temperature feedback loop and gas proportional valve.

[0057] In a second aspect, a computing device includes:

[0058] One or more processors;

[0059] A storage device for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the system.

[0060] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the system.

[0061] The above-described solution of the present invention has at least the following beneficial effects:

[0062] By acquiring real-time dynamic metabolic parameters of the culture medium and periodic morphological parameters of the cell population, the limitations of traditional offline sampling and monitoring are overcome, enabling the immediate capture of dynamic changes in cell metabolism and morphology. A dynamic metabolic threshold library built based on historical data can dynamically adjust the judgment criteria according to the cell growth stage, while a fuzzy morphological benchmark library quantifies the morphological characteristics of different cycles, solving the problem of misjudgment caused by static thresholds and subjective human judgment, and improving the accuracy of cell state monitoring.

[0063] By automating parameter acquisition, analysis, and control command execution, manual sampling, observation, and operation are reduced, which not only lowers labor costs but also reduces external environmental interference with the culture system, lowers the risk of contamination, and helps ensure the safety of mesenchymal stem cell culture. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the process of a multi-parameter-based mesenchymal stem cell culture monitoring system provided in an embodiment of the present invention.

[0065] Figure 2 This invention combines the membership degree of metabolic abnormalities and the confidence degree of morphological abnormalities into a two-dimensional abnormal state vector, generates an inter-dimensional correlation weight matrix based on a historical abnormal pattern library, performs tensor multiplication on the abnormal state vector and the correlation weight matrix, and uses the output operation result as a flowchart of the joint fuzzy state matrix. Detailed Implementation

[0066] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0067] like Figure 1 As shown, embodiments of the present invention propose a multi-parameter-based mesenchymal stem cell culture monitoring system, comprising:

[0068] The acquisition module is used to acquire real-time metabolic dynamic parameter sets of the culture medium and periodic morphological parameter sets of the cell population;

[0069] The generation module is used to train and generate a dynamic metabolic threshold library based on historical data of the metabolic dynamic parameter set, and at the same time, to construct a fuzzy morphological benchmark library based on historical data of the morphological parameter set.

[0070] The analysis module is used to compare the deviation between the real-time dynamic metabolic parameter set and the dynamic metabolic threshold library to generate the membership degree of metabolic abnormality; the real-time morphological parameter set is mapped to the preset feature space through the mirror projection algorithm, and the morphological abnormality confidence is calculated in the space based on the membership function of the fuzzy morphological benchmark library; the membership degree of metabolic abnormality and the morphological abnormality confidence are fused to generate a joint fuzzy state matrix.

[0071] The decision module is used to generate a first-level control instruction when the correlation strength between metabolic and morphological abnormalities in the joint fuzzy state matrix reaches a preset correlation threshold as determined by the fuzzy rule base; and to generate a second-level control instruction when the abnormal state flag indicates an isolated metabolic abnormality.

[0072] The output module is used to respond to primary control commands and collaboratively drive the buffer injection unit and nutrient supplementation pump; and to respond to secondary control commands and dynamically adjust the stirring rate controller, temperature feedback loop and gas proportional valve opening of the bioreactor.

[0073] In this embodiment of the invention, by acquiring the dynamic metabolic parameter set of the culture medium and the periodic morphological parameter set of the cell population in real time, the limitations of traditional offline sampling and monitoring due to lag are overcome, enabling the immediate capture of dynamic changes in cell metabolism and morphology. The dynamic metabolic threshold library constructed based on historical data can dynamically adjust the judgment criteria according to the cell growth stage, while the fuzzy morphological benchmark library quantifies the morphological characteristics of different cycles, solving the problem of misjudgment caused by static thresholds and subjective human judgment, and improving the accuracy of cell state monitoring.

[0074] By automating parameter acquisition, analysis, and control command execution, manual sampling, observation, and operation are reduced, which not only lowers labor costs but also reduces external environmental interference with the culture system, lowers the risk of contamination, and helps ensure the safety of mesenchymal stem cell culture.

[0075] In a preferred embodiment of the present invention, obtaining the real-time metabolic dynamic parameter set of the culture medium and the periodic morphological parameter set of the cell population includes:

[0076] Raw time-series data streams of culture medium were acquired using a biosensor array; a sliding window processor was used to perform segmented feature extraction on the raw time-series data streams to generate a primary metabolic feature set containing a mean vector, a standard deviation vector, and a parameter change rate sequence; time dimension alignment and feature concatenation were performed on the primary metabolic feature set to generate a structured metabolic dynamic parameter set;

[0077] High-resolution image sets of cell populations are periodically captured using a microscopic imaging unit; layered processing is then performed on the high-resolution image sets.

[0078] Edge segmentation is performed at the pixel level to generate a cell contour dataset. Based on the cell contour dataset, geometric topological features are extracted at the single cell level to generate a single cell feature set. At the population level, population morphology indicators are generated based on the spatial distribution relationship of the single cell feature set.

[0079] Single-cell feature sets and population morphological indicators are input into a multi-scale feature fusion engine, and cross-scale associated features are generated through spatial distribution topology modeling.

[0080] A structured morphological parameter set is constructed based on cross-scale correlation features and group morphological indicators.

[0081] In this embodiment of the invention, a biosensor array (monitoring key metabolic indicators such as pH, dissolved oxygen concentration, glucose content, and lactic acid concentration in the culture medium) continuously collects data at set time intervals (e.g., once per second) to form a raw data stream that changes continuously over time (each indicator corresponds to a time-series curve).

[0082] The parameters of the sliding window processor are adjusted according to the metabolic activity level: Logarithmic growth phase: the window length is set to 5 minutes (covering 600 data points within 300 seconds), and the sliding step size is 1 minute (moving 60 seconds each time) to ensure dense capture of changes; Non-logarithmic growth phase: the window length is set to 10 minutes (covering 300 data points within 600 seconds), and the sliding step size is 2 minutes to reduce redundant calculations.

[0083] For a certain type of indicator data within a single window (such as pH value within 5 minutes), the average value within that window is obtained by summing the values ​​of all data points and dividing by the number of data points. For example, if there are 600 pH value data points within a 5-minute window, the sum is 4260, and the average value is 7.1. The window average values ​​of all indicators are integrated into a vector (such as [pH average 7.1, dissolved oxygen average 5.3 mg / L, glucose average 4.8 mmol / L, lactate average 1.1 mmol / L]).

[0084] Calculate the deviation of a certain index data within a single window from the window mean, square all deviations, take the average, and then take the square root to obtain the standard deviation, which reflects the magnitude of data fluctuation. If the pH value fluctuates little within the window, the standard deviation may be 0.06; if dissolved oxygen changes drastically due to rapid cell consumption, the standard deviation may reach 0.5 mg / L.

[0085] Take the average value of the same index from two adjacent windows, subtract the average value of the previous window from the average value of the next window, and then divide by the time interval between the two windows (i.e., the sliding step size) to obtain the rate of change of the index. For example, if the average glucose value in the previous window is 5.0 mmol / L and the average value in the next window is 4.7 mmol / L, and the step size is 1 minute, then the rate of change is (4.7-5.0)÷1=-0.3 mmol / (L·min). Calculate the rate of change of adjacent windows continuously to form a rate sequence over time.

[0086] Due to differences in response delays among different sensors (e.g., pH sensor delay is approximately 0.3 seconds, glucose sensor delay is approximately 0.5 seconds), the characteristic sequences of each indicator are calibrated based on the unified system clock: if the window characteristics of pH have been recorded at a certain moment (e.g., at the 1200th second), but the characteristics of glucose have not yet been generated, then the approximate value at the 1200th second is estimated based on the glucose characteristic values ​​at the 1199th and 1201st seconds, ensuring that the four types of indicators all have corresponding mean, standard deviation, and rate of change data at the same time point.

[0087] The characteristic data at each time point are assembled according to a fixed structure: first, the means of four types of indicators (pH, dissolved oxygen, glucose, lactic acid) are arranged; then, the standard deviations of the four types of indicators are arranged; and finally, the rates of change of the four types of indicators are arranged.

[0088] A feature combination containing 12 data items (e.g., [7.1, 5.3, 4.8, 1.1, 0.06, 0.5, 0.2, 0.1, -0.02, -0.05, -0.3, 0.08]) is formed. The feature combinations of all time points are arranged in chronological order to form a structured set of metabolic dynamic parameters. Each item can fully reflect the metabolic state of the culture medium at a certain moment (average level, fluctuation range and trend of change).

[0089] High-resolution images of cell populations are captured at fixed intervals (e.g., once per hour) using a microscopic imaging unit, forming a time-series image set.

[0090] Edge segmentation is performed on each image. By distinguishing the gray-level differences and texture features between cells and the background (such as threshold segmentation and edge detection algorithms), the contour boundaries of each cell are identified and extracted, generating a cell contour dataset containing information such as contour pixel coordinates and boundary continuity.

[0091] Geometric and topological features of individual cells are extracted based on the contour dataset, including geometric features (area, perimeter, equivalent diameter, roundness, aspect ratio) and topological features (relative position of cell nucleus and cell contour, degree of cell membrane folding), etc. Each cell corresponds to a set of features, which are summarized into a single-cell feature set.

[0092] Analyze the spatial distribution relationship of single-cell feature sets and calculate population morphology indicators, including: cell density (number of cells per unit area), spatial distribution uniformity (such as average distance between cells, local clustering coefficient), population heterogeneity of size / shape (such as standard deviation of cell size, proportion of cells with specific shapes), and spatial orientation (such as the dominant direction of cell arrangement).

[0093] The single-cell feature set and population morphology indicators are input into the multi-scale feature fusion processor. By analyzing "how single-cell features affect population distribution" (such as whether large cells tend to cluster in high-density areas, and whether cells of specific shapes maintain a fixed distance from surrounding cells), a spatial distribution topology model (such as the association rules between local cell features and population density) is constructed, and cross-scale association features (such as "population density gradient of large cell clusters" and "spatial distribution entropy of highly rounded cells") are generated.

[0094] Cross-scale correlation features and population morphology indicators are integrated in chronological order (corresponding to the time points of periodic image acquisition) to form a multi-dimensional morphological feature matrix corresponding to each time point (such as the average area of ​​a single cell, population density, and distribution entropy of large cell aggregation areas at a certain moment), which is a structured morphological parameter set.

[0095] By reflecting the metabolic state of the culture medium in real time through multi-dimensional features (mean, fluctuation, rate of change), it can not only show the overall trend but also capture subtle fluctuations, providing precise quantitative evidence for cell metabolic regulation. Hierarchical analysis from pixel to single cell to population, combined with cross-scale correlation, comprehensively reflects the individual differences in cell morphology and the spatial organization patterns of the population, avoiding the limitations of single-scale analysis. The structured parameter set unifies the time dimension, allowing direct correlation between metabolic environment and cell morphological changes at the same point in time.

[0096] In a preferred embodiment of the present invention, a dynamic metabolic threshold library is generated by training based on historical data of a metabolic dynamic parameter set, and a fuzzy morphological benchmark library is constructed based on historical data of a morphological parameter set, including:

[0097] The system retrieves the dynamic metabolic parameter sets for each culture cycle from the historical metabolic database; performs timestamp alignment on the parameter sets to generate time-series-related metabolic feature tensors; constructs a dynamic probability distribution model based on the metabolic feature tensors, and generates dynamic confidence interval sequences by extracting boundary values ​​under pre-set confidence levels.

[0098] The dynamic confidence interval sequences were integrated into a structured dynamic metabolic threshold library according to the culture stage;

[0099] Extract morphological parameter sets across batches from a historical morphological database; perform time-aware incremental clustering on single-cell feature sets:

[0100] The feature subsets are divided according to the culture time window. Density peak clustering is performed on each subset to generate local morphological benchmarks. The local benchmarks are then fused with time decay weights to generate a global typical cell morphology benchmark vector.

[0101] Dynamic evolution modeling of group morphology indicators:

[0102] Based on the timestamp index, the evolution trajectory of population indicators is constructed, and a dynamic benchmark interval set is generated by fitting a Gaussian mixture model in segments according to the cultivation stage.

[0103] The typical cell morphology baseline vector and the dynamic baseline interval set are input into the rule generator, the adaptive membership function set for morphological parameter anomaly determination is defined, and the membership function set and the corresponding baseline data are associated and stored to generate a structured fuzzy morphological baseline library.

[0104] In this embodiment of the invention, a structured set of dynamic metabolic parameters from multiple culture cycles is extracted from a historical metabolic database. The absolute time difference is eliminated by relative time conversion (with a unified start time of "0 time"), and a time-series-related metabolic feature tensor (with the dimension of "culture batch × relative time point × metabolic feature dimension") is generated.

[0105] A dynamic probability distribution model is constructed based on metabolic feature tensors. For each relative time point, the numerical distribution of metabolic features (such as mean pH and glucose change rate) of all historical batches is statistically analyzed. Using a pre-set confidence level of 90% to 95% (95% for strong regularity parameters such as glucose concentration, and 90% for weak regularity parameters such as local pH fluctuations), boundary values ​​at the corresponding confidence levels are extracted. For example, at a confidence level of 95%, the upper and lower limits are calculated as "mean ± 1.96 × standard deviation" through the fitted normal distribution, and at a confidence level of 90%, it is "mean ± 1.64 × standard deviation", forming the normal interval for a single time point. Arranged in relative time order, a dynamic confidence interval sequence is generated.

[0106] Constructing a dynamic probability distribution model:

[0107] Based on the metabolic feature tensor, the model uses "a single relative time point + a single metabolic feature" as the smallest analytical unit, such as "pH standard deviation at the 2nd hour of culture," "dissolved oxygen change rate at the 6th hour of culture," and "glucose concentration at the 10th hour of culture." Each unit corresponds to an independent analytical object, ensuring the model's specificity. For each analytical unit (e.g., "lactate concentration at the 4th hour"), corresponding data from all historical normal culture batches are selected from the metabolic feature tensor. Assuming there are 80 historical normal batches, the lactate concentration values ​​recorded at "the 4th hour of culture" for these 80 batches are extracted, forming a dedicated dataset (80 values ​​in total). Only data from "the same relative time point," "the same metabolic feature," and "normal batches" are retained, excluding data from abnormal batches or other time points to ensure that the data only reflects normal fluctuations. The specific distribution of this dataset is observed: for example, whether the values ​​are concentrated around a certain central value (e.g., mostly between 5.2 and 5). (Between 0.6mM) Does it exhibit a symmetrical pattern of "dense in the middle and sparse at both ends," or is there any skewness (e.g., most values ​​are low, a few are high but still within the normal range)? Based on the distribution characteristics of the data, choose the corresponding method to quantify the fluctuation pattern: If the data shows a symmetrical and concentrated distribution (e.g., most routine indicators of metabolic characteristics), use the fitting logic of "central trend + dispersion": use the central value in the dataset (e.g., the core value around which most values ​​are located) to reflect the common range, and use the dispersion magnitude of the data (e.g., the average distance of most values ​​from the central value) to reflect the magnitude of the fluctuation; If the data distribution is more complex (e.g., there are multiple fluctuation centers or asymmetrical distribution), use the fitting logic of "density clustering": identify the numerical intervals with high frequency in the data (e.g., two dense intervals of 3.1-3.3mM and 3.5-3.7mM), use the position and coverage of the interval to describe normal fluctuations, and record the probability of the occurrence of values ​​outside the interval (e.g., only 1% of normal data fall outside the interval).

[0108] Based on the above fitting logic, a unique distribution model is generated for each analysis unit:

[0109] Regarding the "lactate concentration at the 4th hour," if the logic of central tendency + dispersion is used, the model will clearly state: "The common central value of lactate concentration at this time point is 5.4 mM, and most normal data (such as 90%) will fluctuate between 5.0-5.8 mM. The further away from this range, the lower the probability of occurrence." If the logic of density clustering is used, the model will clearly state: "The lactate concentration at this time point is mainly concentrated in 3.2-3.4 mM (accounting for 60% of normal data) and 3.6-3.8 mM (accounting for 30% of normal data). Values ​​within these two intervals are all normal, and only 10% of normal data fall outside the intervals." The specific distribution models of all analysis units are integrated according to the dimension of "relative time point + metabolic characteristics" to form the final "dynamic probability distribution model."

[0110] The model as a whole is a structured set that includes the normal fluctuation patterns of all metabolic characteristics at all time points throughout the culture process. For example, from the first hour to the 48th hour of culture, there are corresponding distribution models for the average pH, dissolved oxygen, and glucose concentration characteristics at each hour, clearly recording the normal range, probability of occurrence, and fluctuation pattern of the characteristic at that time point.

[0111] The dynamic confidence interval sequence is divided according to the cultivation stage (adaptation period, logarithmic growth period, etc.). Each stage integrates the confidence intervals of metabolic characteristics at all time points within that stage (the value range fluctuates with the dispersion of historical data, such as the 95% confidence interval of glucose concentration in the logarithmic growth period may be 3.4~6.6mM), forming a structured storage of "stage-time point-feature-threshold upper and lower limits".

[0112] Multiple batches of structured morphological parameter sets were extracted from historical morphological databases, including single-cell features, population indices, and cross-scale correlation features. The data were divided into continuous time windows according to culture time (e.g., one window every 4 hours), and the single-cell features within each window constituted a feature subset.

[0113] Density peak clustering is performed on the feature subset of each window, and the cluster center is the point with the highest density in the feature space. This generates a local morphological benchmark (e.g., "typical cell area 20μm" within a certain window). 2 (Circularity 0.8”).

[0114] A time-decay weighted fusion of local benchmarks is adopted, with individual weights ranging from 0 to 1, and the sum of the weights of all time windows being 1. Priority is given to assigning higher weights to recent time windows (e.g., if the current window is t, the weight of window t-1 is 0.8, window t-2 is 0.15, and window t-3 and earlier is 0.05), or exponential decay is adopted (the decay coefficient τ is usually taken as 3 to 10 time windows to ensure that the weight drops to less than 5% of the initial value after an interval of 3 windows). A global typical cell morphology benchmark vector covering the entire culture process is generated by weighted averaging.

[0115] Historical batch data of various population indicators (cell density, distribution evenness, etc.) are integrated according to timestamps to form a trajectory set that changes over time. The data is then segmented by culture stage, and a Gaussian mixture model is used to fit the indicator distribution. 90%–95% confidence intervals (95% for strong regularity indicators such as cell density, and 90% for weak regularity indicators such as distribution evenness) are extracted as the dynamic baseline interval for that stage (e.g., the 95% confidence interval for cell density in the stable phase might be 1.2 × 10⁻⁶). 6 ~2.5×10 6 (cells / mL).

[0116] The global typical benchmark vector and the dynamic benchmark interval set are input into the rule generator. An adaptive membership function set is defined, with a membership degree range of 0 to 1, where 1 indicates that the parameter "fully conforms to the normal benchmark", 0 indicates "completely deviates from the normal benchmark", and values ​​between 0 and 1 represent "transitional states" (e.g., a membership degree of 0.3 indicates "slight deviation from normal"). The membership function boundaries are dynamically adjusted according to the benchmark (e.g., the typical cell area is 50 μm). 2 At that time, the "normal" membership degree was 40~60μm. 2 The range gradually decreases from 1 to 0.1 within the interval. The membership function set (0~1 value range) and the benchmark data are associated and stored to form a structured fuzzy morphological benchmark library.

[0117] The dynamic metabolic threshold library dynamically adjusts the normal range according to the culture stage and time. The fuzzy morphological benchmark library captures the stage-specificity of morphological features through time-aware clustering and dynamic modeling. Both can adapt to the dynamic changes in the cell culture process and avoid misjudgments caused by fixed thresholds / benchmarks. The metabolic threshold is determined based on the probability distribution of historical data. The morphological benchmark handles the continuity and fuzziness of morphological features (such as the case of "approaching abnormality but not exceeding the absolute threshold") through fuzzy membership functions, improving the ability to identify early abnormalities and minor abnormalities. The benchmark library integrates historical patterns across batches and can serve as the "gold standard" for the culture process. It can be used to compare the deviation between real-time parameters and historical normal patterns, providing a quantitative basis for adjusting culture conditions (such as supplementing nutrients and optimizing environmental parameters) and reducing batch-to-batch differences.

[0118] In a preferred embodiment of the present invention, the joint fuzzy state matrix includes:

[0119] Obtain the real-time dynamic metabolic parameter set and call the threshold vector corresponding to the culture stage in the dynamic metabolic threshold library;

[0120] Perform component confidence interval deviation calculations on the mean vector, standard deviation vector, and parameter change rate sequence of the parameter set to generate the original deviation set;

[0121] Based on the original deviation set, an abnormal weight coefficient set is generated according to the sensitivity of the parameters during the cultivation stage;

[0122] By merging the original deviation set and the abnormality weight coefficient set, a scalar metabolic abnormality membership degree is output.

[0123] Obtain single-cell feature sets and population morphological indices from a real-time morphological parameter set;

[0124] Call upon the typical cell morphology reference vectors and dynamic reference interval sets from the fuzzy morphology reference library;

[0125] Project the single-cell feature set onto the feature space spanned by the reference vector; calculate the Mahalanobis distance between the cell population distribution and the reference vector in the projection space to generate the single-cell deviation vector; perform dynamic interval cross-boundary detection on the population morphology index to generate a population cross-boundary marker set;

[0126] Based on the single-cell deviation vector and the population out-of-bounds marker set, a scalar morphological anomaly confidence level is obtained;

[0127] The membership degree of metabolic abnormality and the confidence degree of morphological abnormality are combined into a two-dimensional abnormal state vector. Based on the historical abnormal pattern library, an inter-dimensional correlation weight matrix is ​​generated. Tensor multiplication is performed on the abnormal state vector and the correlation weight matrix, and the output operation result is used as the joint fuzzy state matrix.

[0128] In this embodiment of the invention, a set of real-time metabolic dynamic parameters (including mean vector, standard deviation vector, and parameter change rate sequence) at the current culture time is obtained. Based on the current culture stage (such as logarithmic growth phase), the threshold vector of the corresponding stage (i.e., the normal confidence interval of each metabolic parameter in this stage, such as glucose concentration 3.4~6.6mM and pH 6.8~7.2) is called from the dynamic metabolic threshold library.

[0129] For each parameter in the metabolic parameter set, a "confidence interval deviation" calculation is performed. If the parameter value is within the threshold interval, the deviation is 0. If it exceeds the interval (e.g., glucose concentration 7.0 mM, higher than the upper limit 6.6 mM), the deviation degree is calculated (e.g., the ratio of the excess to the interval width, 7.0-6.6=0.4 mM, interval width 6.6-3.4=3.2 mM, deviation degree is 0.4 / 3.2=0.125). The deviation degrees of all parameters are summarized to generate an original deviation degree set (e.g., including glucose deviation degree 0.125, pH deviation degree 0, dissolved oxygen deviation degree 0.05, etc.).

[0130] Based on the physiological sensitivity of each parameter at the current culture stage (e.g., the effect of glucose concentration on cell proliferation during the logarithmic growth phase is much greater than that of local pH fluctuations), an abnormal weight coefficient is assigned to each parameter (value range 0~1, the higher the sensitivity, the greater the weight, e.g., glucose weight 0.8, pH weight 0.2), forming an abnormal weight coefficient set.

[0131] The original deviation set and the abnormal weight coefficient set are weighted and fused (e.g., the deviation of each parameter is multiplied by the corresponding weight and then summed) to obtain the scalar metabolic abnormality membership degree (value range 0~1, 0 indicates no abnormality, 1 indicates severe abnormality, such as a calculation result of 0.15 indicating mild metabolic abnormality).

[0132] Obtain the real-time morphological parameter set at the current culture time (including single-cell feature sets such as area and roundness, and population morphological indices such as cell density and distribution uniformity), and call the typical cell morphological benchmark vector for the corresponding stage from the fuzzy morphological benchmark library (e.g., typical cell area of ​​50 μm for this stage). 2 Circularity 0.8) and population dynamics baseline interval set (e.g., cell density 1.2 × 10⁻⁶) 6 ~2.5×10 6 (cells / mL).

[0133] The real-time single-cell feature set (area, roundness, etc. of each cell) is projected onto the feature space spanned by the reference vector of typical cell morphology (i.e., aligned with the dimension of the reference vector to ensure feature comparability). Within the projection space, the Mahalanobis distance between the feature distribution of the cell population and the reference vector is calculated (the distance considering the correlation between features, such as the relationship between area and roundness), and the deviation degree of each cell is obtained. These are then summarized into a single-cell deviation vector (element value range 0~1, the larger the value, the farther away from the reference).

[0134] Perform "dynamic range out-of-bounds detection" on population morphology indicators. If the indicator is within the dynamic baseline range (e.g., cell density 2.0 × 10⁻⁶), the indicator will be considered valid. 6 cells / mL, in the range of 1.2~2.5×10 6 If the value is within the specified interval, the flag is 0; if it exceeds the specified interval (e.g., 3.0 × 10), the flag is 0. 6 If the cell / mL value exceeds the limit, a value between 0 and 1 is assigned based on the degree of excess (e.g., exceeding the upper limit by 0.5 × 10⁻⁶). 6 Interval width 1.3×10 6 (with a flag value of 0.5 / 1.3≈0.38), a set of out-of-bounds flags for the group is generated.

[0135] The average value of the single-cell deviation vector (reflecting the overall deviation of the population) is combined with the weighted sum of the population boundary crossing marker set (the comprehensive impact of population indicator boundary crossing) to obtain the scalar morphological abnormality confidence score (value range 0~1, such as a calculation result of 0.2 indicating mild morphological abnormality).

[0136] The membership degree of metabolic abnormality (e.g., 0.15) and the confidence degree of morphological abnormality (e.g., 0.2) are combined into a two-dimensional abnormal state vector (e.g., [0.15, 0.2]), with the vector elements corresponding to the degree of metabolic and morphological abnormalities, respectively.

[0137] Extract the interdimensional correlation weight matrix from the historical anomaly pattern library. This matrix is ​​generated based on the correlation patterns between metabolic anomalies and morphological anomalies in historical data (e.g., if the probability of "metabolic anomalies accompanied by morphological anomalies" is high in the past, then the corresponding element in the matrix has a high weight). The matrix element values ​​range from 0 to 1 (e.g., the weight matrix is ​​[[0.9, 0.3], [0.3, 0.8]], which means that the weight of metabolic anomaly on its own state is 0.9, and the correlation weight on morphological state is 0.3).

[0138] Perform tensor multiplication (i.e., element-wise weighted combination of vector and matrix) on the two-dimensional abnormal state vector and the associated weight matrix. The output is a joint fuzzy state matrix (such as a 2×2 matrix, where the element values ​​reflect the comprehensive abnormal states of "metabolism-metabolism", "metabolism-morphology", "morphology-metabolism", and "morphology-morphology", with a value range of 0~1).

[0139] By assessing anomalies in both metabolic and morphological dimensions, the limitations of misjudgment based on a single indicator (such as metabolism alone or morphology alone) are avoided, providing a more comprehensive reflection of the true state of cells. The correlation weight matrix incorporates historical anomaly patterns, quantifying the intrinsic relationship between metabolic and morphological anomalies (e.g., insufficient metabolic substrates are often accompanied by cell shrinkage), enhancing the logical consistency of anomaly detection. The degree of anomaly is described using continuous values ​​from 0 to 1, rather than a simple "normal / abnormal" dichotomy, providing a more nuanced quantitative basis for early warning and precise control of the cell culture process.

[0140] In a preferred embodiment of the present invention, the membership degree of metabolic abnormality and the confidence degree of morphological abnormality are combined into a two-dimensional abnormal state vector; an inter-dimensional correlation weight matrix is ​​generated based on a historical abnormal pattern library; tensor multiplication is performed on the abnormal state vector and the correlation weight matrix; and the output operation result is used as a joint fuzzy state matrix, including:

[0141] Metabolic abnormality membership degree is used as the first dimension value, and morphological abnormality confidence degree is used as the second dimension value;

[0142] Generate a two-dimensional abnormal state vector by combining elements in dimensional order;

[0143] When the value of the first dimension exceeds the metabolic dominance threshold, the metabolic abnormality dominance factor is activated; when the value of the second dimension exceeds the morphological dominance threshold, the morphological abnormality dominance factor is activated; a weight allocation instruction is generated based on the type of activated dominance factor.

[0144] The system responds to weight allocation instructions by invoking the historical anomaly pattern library; it then uses the correlation characteristics of similar dominant factors within the historical anomaly pattern library as a basis for further analysis.

[0145] Based on metabolic and morphological abnormalities, an adaptive weight matrix is ​​generated that includes self-effect weights and cross-dimensional correlation weights.

[0146] Perform tensor convolution operation on the two-dimensional abnormal state vector and the adaptive weight matrix, and output the result as the joint fuzzy state matrix.

[0147] In this embodiment of the invention, the previously generated scalar metabolic abnormality membership degree (value range 0~1, such as 0.6) is used as the first dimension value of the two-dimensional vector (representing the degree of metabolic abnormality); the scalar morphological abnormality confidence degree (value range 0~1, such as 0.3) is used as the second dimension value (representing the degree of morphological abnormality).

[0148] Vector combination: The values ​​of the two dimensions are combined into a two-dimensional abnormal state vector (e.g., [0.6, 0.3]) in the order of "metabolic dimension first, morphological dimension second". The vector directly reflects the current level of metabolic and morphological abnormality.

[0149] The system presets a metabolic dominance threshold (e.g., 0.5, which can be adjusted based on historical data) and a morphology dominance threshold (e.g., 0.5). If the first dimension value of the two-dimensional vector (metabolic abnormality membership degree) exceeds the metabolic dominance threshold (e.g., 0.6 > 0.5), the "metabolic abnormality dominance factor" is activated; if the second dimension value (morphological abnormality confidence degree) exceeds the morphology dominance threshold (e.g., 0.6 > 0.5), the "morphological abnormality dominance factor" is activated; if both exceed the thresholds, both dominance factors are activated simultaneously.

[0150] Instructions are generated based on the type of dominant factor activated. For example, when only the metabolic dominant factor is activated, the instruction is "enhance the weight of the metabolic dimension itself and adjust the cross-dimensional correlation weight of metabolism to morphology"; when only the morphology dominant factor is activated, the instruction is "enhance the weight of the morphology dimension itself and adjust the cross-dimensional correlation weight of morphology to metabolism"; when both factors are activated, the instruction is "balance the weights of both and strengthen the bidirectional cross-dimensional correlation weight".

[0151] In response to weight allocation instructions, the system extracts correlation features of similar dominant factors from historical abnormal patterns. For example, it calls historical data on "metabolic dominant abnormalities" to obtain the correlation between metabolic abnormalities and morphological abnormalities under this pattern (such as the correlation strength between the degree of metabolic abnormality and cell morphological shrinkage when there is insufficient metabolic substrate); and it calls historical data on "morphological dominant abnormalities" to obtain the influence of morphological abnormalities (such as cell aggregation) on the metabolic environment (such as aggregation leading to local accumulation of metabolic waste).

[0152] An adaptive weight matrix (2×2 matrix) is generated based on historical correlation features, including self-influence weights: the influence weight of the metabolic dimension on itself (e.g., 0.8 when metabolism is dominant) and the influence weight of the morphological dimension on itself (e.g., 0.4 when metabolism is dominant); and cross-dimensional correlation weights: the influence weight of the metabolic dimension on the morphological dimension (e.g., 0.3 when metabolism is dominant, reflecting the degree of influence of metabolic abnormalities on morphology) and the influence weight of the morphological dimension on the metabolic dimension (e.g., 0.2 when metabolism is dominant, reflecting the reverse influence of morphology on metabolism); the weight values ​​are dynamically adjusted according to historical patterns (value range 0~1, with higher self-weights for dominant dimensions and related cross-dimensional weights).

[0153] Tensor convolution is performed on a two-dimensional abnormal state vector (e.g., [0.6, 0.3]) and an adaptive weight matrix. This involves weighting and fusing the two dimensions of the vector through matrix elements, preserving the abnormal information of each dimension while incorporating the inter-dimensional correlations (e.g., metabolic abnormalities influence the assessment results of morphological abnormalities through cross-dimensional weights). The output is a joint fuzzy state matrix (2×2 matrix, element value range 0~1), where each element represents a combined abnormal state of "metabolism-metabolism", "metabolism-morphology", "morphology-metabolism", and "morphology-morphology".

[0154] By activating dominant factors, weight allocation is tilted towards the more significant abnormal dimensions (e.g., when metabolic abnormalities are more severe, metabolic-related weights are prioritized), avoiding secondary abnormalities from interfering with core judgments. An adaptive weight matrix is ​​generated based on a historical abnormality pattern library, quantifying the potential correlation between metabolic and morphological abnormalities (e.g., metabolic disorders often lead to morphological changes) into weights, making abnormality assessment more closely aligned with the actual laws of biological processes. The joint fuzzy state matrix not only reflects single-dimensional abnormalities but also comprehensively characterizes complex abnormal patterns of cell states through the quantification of "inter-dimensional interactions," providing more specific targets for precise intervention in the culture process (e.g., adjusting nutrient supply and culture environment for "metabolic-morphological" correlation abnormalities).

[0155] In a preferred embodiment of the present invention, when the correlation strength between metabolic and morphological abnormalities in the joint fuzzy state matrix reaches a preset correlation threshold as determined by the fuzzy rule base, a first-level control instruction is generated; when the abnormal state flag indicates an isolated metabolic abnormality, a second-level control instruction is generated, including:

[0156] Extract the combined intensity of metabolic abnormalities and the combined intensity of morphological abnormalities from the joint fuzzy state matrix;

[0157] An abnormal intensity level identifier set is generated based on preset intensity grading rules;

[0158] Input the abnormal intensity level identifier set into the fuzzy rule base. When both metabolic and morphological abnormality intensities reach the high-level threshold, activate the metabolic-morphological strong correlation flag. When the metabolic abnormality intensity reaches the high level and the morphological abnormality intensity is lower than the decoupling threshold, activate the isolated metabolic abnormality flag.

[0159] In response to metabolic-morphology strongly correlated markers, a primary control instruction generation pathway is triggered.

[0160] The response to isolated metabolic abnormality flags triggers the generation pathway of secondary control instructions.

[0161] In this embodiment of the invention, the comprehensive intensity of metabolic abnormalities and the comprehensive intensity of morphological abnormalities are extracted from the joint fuzzy state matrix. The comprehensive intensity of metabolic abnormalities integrates the values ​​of the "metabolism-metabolism" and "morphology-metabolism" elements in the matrix (reflecting the sum of metabolic abnormalities themselves and the correlation between morphology and metabolism), while the comprehensive intensity of morphological abnormalities integrates the values ​​of the "morphology-morphology" and "metabolism-morphology" elements (reflecting the sum of morphological abnormalities themselves and the correlation between metabolism and morphology). Both have a value range of 0 to 1.

[0162] Based on preset intensity grading rules (e.g., 0~0.3 is "low level", 0.3~0.7 is "medium level", and 0.7~1 is "high level"), the overall intensity of metabolism and morphology is graded. For example, a metabolic overall intensity of 0.8 is graded as "high level" and a morphological overall intensity of 0.2 is graded as "low level", generating an abnormal intensity grade identifier set containing two dimensions (e.g., [high level, low level]).

[0163] Input the set of abnormal intensity level identifiers into the fuzzy rule base and perform the following judgment: when the comprehensive intensity of metabolic abnormality reaches "high level" (e.g., ≥0.7) and the comprehensive intensity of morphological abnormality also reaches "high level" (e.g., ≥0.7), it is determined that there is a strong correlation between the two abnormalities, and the "metabolic-morphological strong correlation flag bit" is activated.

[0164] When the overall intensity of metabolic abnormalities reaches "high level" (e.g., ≥0.7), but the overall intensity of morphological abnormalities is lower than the preset decoupling threshold (e.g., ≤0.3, i.e. "low level"), metabolic abnormalities are determined to be independent of morphological abnormalities, and the "isolated metabolic abnormality marker" is activated.

[0165] Responding to the “strongly correlated metabolism-morphology marker”: triggers the primary control instruction generation pathway to generate comprehensive regulatory instructions for abnormalities in the synergistic relationship between metabolism and morphology (such as simultaneously adjusting nutrient supply and culture environment parameters to alleviate the abnormal linkage between the two); responding to the “isolated metabolic abnormality marker”: triggers the secondary control instruction generation pathway to generate precise regulatory instructions that target only metabolic abnormalities (such as supplementing metabolic substrates separately to avoid unnecessary intervention in morphologically normal cells).

[0166] By distinguishing between strongly correlated and isolated anomalies, ineffective interventions can be reduced, and disturbances to the culture system can be minimized. For metabolic-morphology strongly correlated anomalies, primary instructions can be rapidly generated to prevent the spread of correlated anomalies (e.g., metabolic disorders leading to morphological deterioration further exacerbate metabolic abnormalities), achieving early containment. Secondary instructions target isolated metabolic anomalies, regulating only necessary steps, saving regulatory resources (e.g., eliminating the need to adjust morphology-related equipment parameters), and improving the regulatory efficiency and stability of the cell culture process.

[0167] In a preferred embodiment of the present invention, in response to a primary control command, the buffer injection unit and the nutrient supplementation pump are driven in a coordinated manner; in response to a secondary control command, the stirring rate controller, the temperature feedback loop, and the opening degree of the gas proportional valve of the bioreactor are dynamically adjusted, including:

[0168] In response to the first-level control command, extract the comprehensive intensity of metabolic abnormality and the comprehensive intensity of morphological abnormality from the joint fuzzy state matrix;

[0169] The buffer regulation model is used to process the overall intensity of metabolic abnormalities and generate a buffer pulse injection parameter set; the intensity of metabolic and morphological abnormalities is fused to generate a nutrient supplementation flow rate regulation curve; and the buffer injection unit and nutrient supplementation pump are coordinated to perform linked operations.

[0170] In response to the secondary control command, the parameter change rate sequence and metabolic abnormality membership degree are invoked from the real-time metabolic dynamic parameter set;

[0171] Perform multi-parameter coupling analysis using a dynamic control strategy engine:

[0172] An adaptive correction for stirring rate is generated based on the parameter change rate sequence; a temperature feedback compensation signal is generated based on the membership degree of metabolic anomalies; and a proportional valve opening increment sequence is generated through a gas valve control model.

[0173] Synchronous adjustment of stirring rate controller, temperature feedback loop and gas proportional valve.

[0174] In this embodiment of the invention, in response to a first-level control command, the comprehensive intensity of metabolic abnormalities (e.g., 0.8) and the comprehensive intensity of morphological abnormalities (e.g., 0.7) are extracted from the joint fuzzy state matrix. Both reflect the severity of metabolic and morphological correlation abnormalities.

[0175] The overall intensity of metabolic abnormalities is input into the buffer regulation model (which is built based on the correlation between metabolic abnormalities and buffer demand in historical data) to generate a set of buffer pulse injection parameters, including single injection volume (e.g., 50 mL for an abnormality intensity of 0.8), injection interval (e.g., once every 10 minutes), and duration (e.g., 30 minutes), in order to quickly alleviate metabolic environmental imbalances (e.g., pH abnormalities).

[0176] The combined intensity of metabolic abnormalities and morphological abnormalities (e.g., weighted at a 7:3 ratio) are input into the nutrient supplementation model to generate a nutrient supplementation flow rate regulation curve. The curve dynamically adjusts the flow rate over time (e.g., an initial flow rate of 2 mL / min, which gradually decreases to 0.5 mL / min as the abnormality intensity decreases), while also taking into account the metabolic substrate requirements and the cell morphology's ability to absorb nutrients (e.g., avoiding excessive nutrient intake that could lead to toxicity accumulation when there are morphological abnormalities).

[0177] The buffer pulse injection parameter set is sent to the buffer injection unit, and the nutrient supplementation flow rate control curve is sent to the nutrient supplementation pump. The two units are controlled to perform operations synchronously in time (such as adjusting the nutrient flow rate synchronously during buffer injection to avoid superimposed disturbances to the culture medium environment).

[0178] In response to secondary control commands, the system calls upon the parameter change rate sequence (e.g., glucose concentration decreases by 0.6 mM per hour, dissolved oxygen decreases by 5% per hour) and metabolic abnormality membership degree (e.g., 0.6) from the real-time metabolic dynamic parameter set.

[0179] Adaptive stirring rate correction: Based on the parameter change rate sequence, the uniformity of the culture medium is judged (e.g., large fluctuations in glucose concentration indicate local concentration unevenness), and a stirring rate correction is generated (e.g., the current rate is 200 rpm, and it is corrected to 220 rpm to enhance mixing); Temperature feedback compensation signal: Based on the membership degree of metabolic abnormalities, the temperature adjustment range is determined (e.g., a membership degree of 0.6 corresponds to a compensation of +0.5℃ to alleviate abnormal metabolic enzyme activity), and a compensation signal for the temperature feedback loop is generated; Gas proportional valve opening increment sequence: Through the gas valve control model (corresponding to the dissolved oxygen change rate and gas ratio requirements), the opening increment of the oxygen / carbon dioxide proportional valve is generated (e.g., if the dissolved oxygen decreases rapidly, the oxygen valve opening increases by 5%, while the carbon dioxide valve remains unchanged).

[0180] The stirring rate correction is sent to the stirring rate controller, the temperature compensation signal is input to the temperature feedback loop, and the opening increment sequence is sent to the gas proportional valve. The operating parameters of the three devices are adjusted synchronously to specifically improve metabolic abnormalities (avoiding interference with cells of normal morphology caused by adjusting morphology-related parameters).

[0181] Primary control commands, through the coordinated regulation of buffer and nutrients, synchronously improve abnormalities in the relationship between the metabolic environment and cell morphology (such as cell shrinkage caused by insufficient metabolic substrates), avoiding secondary imbalances caused by single regulation. Secondary control commands only regulate reactor parameters directly related to metabolism (stirring, temperature, gas), without involving morphology-related interventions, reducing disturbance to normally shaped cells and improving regulation efficiency. Regulatory values ​​are dynamically generated based on real-time metabolic parameter change rates and the intensity of abnormalities, rather than fixed values, making regulation more closely aligned with the current culture state (e.g., automatically increasing the regulatory amplitude when metabolic abnormalities worsen), thus improving the stability and controllability of the cell culture process.

[0182] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0183] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0184] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-parameter-based mesenchymal stem cell culture monitoring system, characterized in that, include: The acquisition module is used to acquire real-time metabolic dynamic parameter sets of the culture medium and periodic morphological parameter sets of the cell population. This includes: acquiring raw time-series data streams of the culture medium via a biosensor array; performing segmented feature extraction on the raw time-series data stream using a sliding window processor to generate a primary metabolic feature set containing a mean vector, standard deviation vector, and parameter change rate sequence; performing time dimension alignment and feature stitching on the primary metabolic feature set to generate a structured metabolic dynamic parameter set; periodically capturing high-resolution image sets of the cell population using a microscopic imaging unit; performing layered processing on the high-resolution image sets: performing edge segmentation at the pixel level to generate a cell contour dataset; extracting geometric topological features at the single-cell level based on the cell contour dataset to generate a single-cell feature set; generating population morphological indicators at the population level based on the spatial distribution relationship of the single-cell feature sets; inputting the single-cell feature sets and population morphological indicators into a multi-scale feature fusion processor to generate cross-scale associated features through spatial distribution topology modeling; and constructing a structured morphological parameter set based on the cross-scale associated features and population morphological indicators. The generation module is used to train and generate a dynamic metabolic threshold library based on historical data of metabolic dynamic parameter sets, and to construct a fuzzy morphological benchmark library based on historical data of morphological parameter sets. This includes: calling the metabolic dynamic parameter sets for each culture cycle from the historical metabolic database; performing timestamp alignment on the parameter sets to generate time-series correlated metabolic feature tensors; constructing a dynamic probability distribution model based on the metabolic feature tensors, and generating dynamic confidence interval sequences by extracting boundary values ​​under pre-set confidence levels; integrating the dynamic confidence interval sequences into a structured dynamic metabolic threshold library according to culture stages; extracting cross-batch morphological parameter sets from the historical morphological database; and processing single-cell data. The feature set performs time-aware incremental clustering: the feature subsets are divided according to the culture time window, and density peak clustering is performed on each subset to generate local morphological benchmarks. The local benchmarks are then fused with time decay weights to generate a global typical cell morphology benchmark vector. The population morphology index is dynamically modeled: the population index evolution trajectory is constructed based on the timestamp index, and a Gaussian mixture model is fitted in segments according to the culture stage to generate a dynamic benchmark interval set. The typical cell morphology benchmark vector and the dynamic benchmark interval set are input into the rule generator to define an adaptive membership function set for morphological parameter anomaly judgment. The membership function set and the corresponding benchmark data are associated and stored to generate a structured fuzzy morphological benchmark library. The analysis module is used to compare the deviation between the real-time dynamic metabolic parameter set and the dynamic metabolic threshold library to generate metabolic abnormality membership degrees; to map the real-time morphological parameter set to a preset feature space through a mirror projection algorithm, and to calculate the morphological abnormality confidence degree based on the membership function of the fuzzy morphological benchmark library in this space; to fuse the metabolic abnormality membership degree and the morphological abnormality confidence degree to generate a joint fuzzy state matrix; the joint fuzzy state matrix includes: obtaining the real-time dynamic metabolic parameter set and calling the threshold vector corresponding to the culture stage in the dynamic metabolic threshold library; performing component confidence interval deviation calculation on the mean vector, standard deviation vector and parameter change rate sequence of the parameter set to generate an original deviation set; generating an abnormality weight coefficient set based on the sensitivity of the parameters in the culture stage according to the original deviation set; and fusing the original deviation set and the abnormality weight coefficient set to output a standard deviation set. The algorithm performs the following steps:

1. Quantitative metabolic abnormality membership degree; 2. Obtain single-cell feature sets and population morphological indicators from a real-time morphological parameter set; 3. Call typical cell morphological benchmark vectors and dynamic benchmark interval sets from a fuzzy morphological benchmark library; 4. Project the single-cell feature sets onto the feature space spanned by the benchmark vectors; 5. Calculate the Mahalanobis distance between the cell population distribution and the benchmark vectors within the projection space to generate a single-cell deviation vector; 6. Perform dynamic interval boundary crossing detection on the population morphological indicators to generate a population boundary crossing marker set; 7. Obtain scalar morphological abnormality confidence based on the single-cell deviation vector and the population boundary crossing marker set; 8. Combine the metabolic abnormality membership degree and the morphological abnormality confidence into a two-dimensional abnormality state vector; 9. Generate an inter-dimensional correlation weight matrix based on a historical abnormality pattern library; 10. Perform tensor multiplication on the abnormality state vector and the correlation weight matrix, and output the result as a joint fuzzy state matrix. The decision module is used to generate a first-level control instruction when the correlation strength between metabolic and morphological abnormalities in the joint fuzzy state matrix reaches a preset correlation threshold as determined by the fuzzy rule base; and to generate a second-level control instruction when the abnormal state flag indicates an isolated metabolic abnormality. The output module is used to respond to primary control commands and collaboratively drive the buffer injection unit and nutrient supplementation pump; and to respond to secondary control commands and dynamically adjust the stirring rate controller, temperature feedback loop and gas proportional valve opening of the bioreactor.

2. The multi-parameter-based mesenchymal stem cell culture monitoring system according to claim 1, characterized in that, The membership degree of metabolic abnormalities and the confidence degree of morphological abnormalities are combined into a two-dimensional abnormal state vector. Based on a historical abnormal pattern library, an inter-dimensional correlation weight matrix is ​​generated. Tensor multiplication is performed on the abnormal state vector and the correlation weight matrix, and the output result is used as a joint fuzzy state matrix, including: Metabolic abnormality membership degree is used as the first dimension value, and morphological abnormality confidence degree is used as the second dimension value; Generate a two-dimensional abnormal state vector by combining elements in dimensional order; When the value of the first dimension exceeds the metabolic dominance threshold, the metabolic abnormality dominance factor is activated; when the value of the second dimension exceeds the morphological dominance threshold, the morphological abnormality dominance factor is activated; a weight allocation instruction is generated based on the type of activated dominance factor. The system responds to weight allocation instructions by invoking the historical anomaly pattern library; it then uses the correlation characteristics of similar dominant factors within the historical anomaly pattern library as a basis for further analysis. Based on metabolic and morphological abnormalities, an adaptive weight matrix is ​​generated that includes self-effect weights and cross-dimensional correlation weights. Perform tensor convolution operation on the two-dimensional abnormal state vector and the adaptive weight matrix, and output the result as the joint fuzzy state matrix.

3. The multi-parameter-based mesenchymal stem cell culture monitoring system according to claim 2, characterized in that, When the correlation strength between metabolic and morphological abnormalities in the joint fuzzy state matrix reaches a preset correlation threshold as determined by the fuzzy rule base, a first-level control command is generated. When the abnormal status flag indicates an isolated metabolic abnormality, a secondary control instruction is generated, including: Extract the combined intensity of metabolic abnormalities and the combined intensity of morphological abnormalities from the joint fuzzy state matrix; An abnormal intensity level identifier set is generated based on preset intensity grading rules; Input the abnormal intensity level identifier set into the fuzzy rule base. When both metabolic and morphological abnormality intensities reach the high-level threshold, activate the metabolic-morphological strong correlation flag. When the metabolic abnormality intensity reaches the high level and the morphological abnormality intensity is lower than the decoupling threshold, activate the isolated metabolic abnormality flag. In response to metabolic-morphology strongly correlated markers, a primary control instruction generation pathway is triggered. The response to isolated metabolic abnormality flags triggers the generation pathway of secondary control instructions.

4. The multi-parameter-based mesenchymal stem cell culture monitoring system according to claim 3, characterized in that, In response to primary control commands, it coordinates the driving of the buffer injection unit and nutrient supplementation pump; in response to secondary control commands, it dynamically adjusts the bioreactor's stirring rate controller, temperature feedback loop, and gas proportional valve opening, including: In response to the first-level control command, extract the comprehensive intensity of metabolic abnormality and the comprehensive intensity of morphological abnormality from the joint fuzzy state matrix; The buffer regulation model is used to process the overall intensity of metabolic abnormalities and generate a buffer pulse injection parameter set; the intensity of metabolic and morphological abnormalities is fused to generate a nutrient supplementation flow rate regulation curve; and the buffer injection unit and nutrient supplementation pump are coordinated to perform linked operations. In response to the secondary control command, the parameter change rate sequence and metabolic abnormality membership degree are invoked from the real-time metabolic dynamic parameter set; Perform multi-parameter coupling analysis using a dynamic control strategy engine: An adaptive correction for stirring rate is generated based on the parameter change rate sequence; a temperature feedback compensation signal is generated based on the membership degree of metabolic anomalies; and a proportional valve opening increment sequence is generated through a gas valve control model. Synchronous adjustment of stirring rate controller, temperature feedback loop and gas proportional valve.

5. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the system as described in any one of claims 1 to 4.

Images