Intelligent quality evaluation system and method for whole process of cell preparation

Abstract

The invention relates to the technical field of cell preparation quality evaluation, in particular to a cell preparation whole-process quality intelligent evaluation system and a method thereof, and the system comprises a data acquisition module, a preprocessing module, a mining module, a modeling module, a report automatic generation module and a data storage module. The modeling module is used for constructing a cell product index evaluation system and a product traceability model and comprises a cell culture state evaluation model and an environment monitoring analysis module, and the environment monitoring analysis module is used for carrying out traceability analysis by using a time sequence convolutional neural network; the automatic report generation module intelligently evaluates the quality of the whole process and generates a report, the data storage module manages data and model storage, sharing and calling, the system constructs a cell quality evaluation system based on multi-dimensional data, integrates multiple dimensional indexes such as morphology, surface markers and functionality, realizes comprehensive evaluation, and improves the evaluation efficiency. And the one-sidedness of single index evaluation is avoided.

Description

Technical Field

[0001] This invention relates to the field of cell preparation quality assessment technology, specifically to an intelligent quality assessment system and method for the entire cell preparation process, applicable to quality monitoring and assessment during the preparation of cell therapy products. Background Technology

[0002] With the rapid development of cell therapy technology, the assessment of cell preparation quality is crucial to ensuring the safety and efficacy of cell therapy products. Currently, quality assessment during cell preparation mainly relies on human experience, which suffers from high subjectivity, poor consistency, and low efficiency. Furthermore, traditional cell quality assessment methods are often based on single or a few indicators, failing to comprehensively reflect the quality characteristics of cells. In addition, existing technologies lack in-depth analysis of the relationship between cell culture environmental factors and cell quality, making it difficult to identify potential quality problems.

[0003] In existing technologies, most cell quality assessment systems can only detect and record certain specific indicators of cells, such as cell viability and surface marker expression, but cannot systematically assess the quality of the entire preparation process. Furthermore, these systems generally lack automation and intelligent functions; data analysis and evaluation result generation still require significant manual intervention, resulting in low efficiency and susceptibility to human error.

[0004] Therefore, there is an urgent need for a quality assessment system for the entire cell preparation process that can integrate multidimensional data and achieve automated analysis and intelligent evaluation, so as to improve the quality control level and efficiency of the cell preparation process. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent quality assessment system and method for the entire cell preparation process. By integrating flow cytometry data, high-content imaging data, environmental parameter data, and experimental parameter data, a multi-dimensional cell quality evaluation system is constructed to achieve automated and intelligent assessment of the entire cell preparation process.

[0006] This invention proposes an intelligent quality assessment system for the entire cell preparation process, comprising:

[0007] The data acquisition module is used to collect flow cytometry data, high-content imaging data, environmental parameter data, and experimental parameter data during the cell preparation process;

[0008] A data preprocessing module, which is communicatively connected to the data acquisition module, is used to receive data acquired by the data acquisition module and clean and preprocess the data.

[0009] The data mining module is communicatively connected to the data preprocessing module and is used to receive the data preprocessed by the data preprocessing module and to perform visualization exploration and analysis on the preprocessed data according to the cell product index evaluation system.

[0010] The data modeling module is communicatively connected to the data mining module and is used to receive the analysis results of the data mining module, build the foundation of the data model, construct the cell product indicator evaluation system and the product traceability model. The data modeling module includes a cell culture status evaluation model and an environmental monitoring and analysis module. The environmental monitoring and analysis module uses a temporal convolutional neural network to perform traceability analysis on environmental factors inside and outside the incubator. The cell culture status evaluation model evaluates the quality of cell preparation through an artificial intelligence model.

[0011] The report generation module is communicatively connected to the data mining module and the data modeling module. It is used to receive the analysis results of the data mining module and the data model of the data modeling module, to intelligently evaluate the quality of the entire cell preparation process, and to automatically generate an evaluation report.

[0012] The data storage module is connected in communication with the automatic report generation module and is used to manage the storage, sharing, and retrieval of data and models.

[0013] Preferably, the data acquisition module includes a flow cytometry detection and analysis module, a high-content imaging data module, an environmental parameter module, and an experimental data module.

[0014] The flow cytometry detection and analysis module is used to obtain the analysis results of the flow cytometer;

[0015] The high-content imaging data module is used to acquire cell morphology and activity images and multi-channel fluorescence images collected by the high-content imaging system.

[0016] The environmental parameter module is used to acquire environmental parameters inside and outside the incubator;

[0017] The experimental data module is used to acquire experimental parameters during sample preparation and testing.

[0018] Preferably, the flow cytometry detection and analysis module includes sample preparation parameters and cell analysis parameters;

[0019] The sample preparation parameters include cell viability and quality control data files, which are used to obtain the cell product activity rate and cell quality control standard deviation.

[0020] The cell analysis parameters include positive sample data files and negative control data files, used to obtain the fluorescence intensity of each channel of the flow cytometer.

[0021] The high-content imaging data module includes data processing of high-content cell image files.

[0022] As a preferred embodiment, the data mining module includes a sample information table, which includes multiple indicators. Each indicator is processed separately in the corresponding workflow to generate a result report.

[0023] The high-content image metrics workflow includes cell segmentation of multichannel fluorescence data;

[0024] Flow cytometry was used to analyze data to obtain cell viability, growth status, CD markers, and proliferation activity.

[0025] As a preferred embodiment, the product indicator evaluation system in the data modeling module is implemented through the following steps:

[0026] Step 1: Construction and preprocessing of the sample morphological index dataset;

[0027] Step 2: Cell morphology feature detection based on convolutional neural network;

[0028] Step 3: Calculation of the positive rate of cell surface markers;

[0029] Step 4: Calculation of surface marker anomaly rate;

[0030] Step 5: Calculation of cell function indicators;

[0031] Step 6: Sample quality assessment.

[0032] As a preferred embodiment, the environmental monitoring and analysis module uses a self-designed temporal convolutional neural network model to perform source analysis on environmental factors inside and outside the incubator.

[0033] The temporal convolutional neural network model is optimized through the following steps:

[0034] Step 1: Parameter tuning of the temporal convolutional neural network model for environmental monitoring data fusion;

[0035] Step 2: Train the temporal data using a temporal convolutional neural network model and determine the model accuracy;

[0036] Step 3: Verify the accuracy of the traceability of environmental monitoring data.

[0037] Preferably, the construction of the adaptive model for the cell multidimensional index scoring includes:

[0038] The first step is to establish an initial model for multidimensional index scoring;

[0039] The second step is to establish model evaluation metrics using cross-validation.

[0040] The third step is to select the optimal model based on the model's output parameters and model evaluation metrics.

[0041] The adaptive model for multidimensional index scoring takes cell morphology data and cell surface marker expression data as inputs. These data are processed by two fully connected layers to obtain cell morphology-based scores and cell surface marker expression-based scores. The cell multidimensional index score is obtained by summing the cell morphology-based scores and cell surface marker expression-based scores.

[0042] As a preferred embodiment, the automatic report generation module integrates the data mining results and evaluation results into the evaluation report;

[0043] The data storage module utilizes data management tools to integrate data and models generated during system processes;

[0044] The automatic report generation module includes an automatic quality grade determination unit, which is based on an SVM classifier. The SVM classifier is trained based on the negative correlation between cell multidimensional index scores and the comprehensive environmental score, and is used to calculate the cell preparation quality score and classify the quality grade.

[0045] Preferably, the quality score calculation formula used by the SVM classifier is:

[0046] ;

[0047] in, Prepare mass fractions for cells, The mean absolute error of the output of the SVM classifier is used. These are the weighting coefficients. The correlation coefficient is the output of the SVM classifier;

[0048] Before classification, the SVM classifier preprocesses the input cell multidimensional index scores and the comprehensive score of the preparation environment. The preprocessing includes normalizing the data, which involves using a feature scaling algorithm to normalize the input cell multidimensional index scores and the comprehensive score of the preparation environment to a range of 0-1.

[0049] A method for intelligent quality assessment of the entire cell preparation process includes the following steps:

[0050] Step 1: Acquire high-content cell images, cell viability, and surface marker data of the sample using a high-content imaging system;

[0051] Step 2: Process the high-content cell image files;

[0052] Step 3: Collect sample preparation parameter data using flow cytometry;

[0053] Step 4: Construct a multi-dimensional evaluation system for cell product quality. Analyze and calculate the data from Steps 1 and 3 based on the multi-dimensional evaluation system for cell product quality to obtain the cell culture status analysis results.

[0054] Step 5: Obtain environmental monitoring data inside and outside the incubator, and use the traceability model that integrates cell culture status analysis results and environmental monitoring data inside and outside the incubator to obtain the influencing factors of the environmental monitoring data inside and outside the incubator.

[0055] Step 6: Based on the negative correlation between cell multidimensional index scores and environmental comprehensive scores, calculate the cell preparation quality score using an SVM classifier, and classify the cell preparation quality according to the quality score.

[0056] Step 7: Integrate the data mining results and evaluation results into the evaluation report to automatically generate a quality evaluation report;

[0057] Step 8: Use data management tools to integrate the data and models generated in the system process for storage, sharing, and retrieval.

[0058] The present invention has the following beneficial effects:

[0059] 1. This invention constructs a cell quality evaluation system based on multidimensional data, which integrates indicators from multiple dimensions such as morphology, surface markers and function, and realizes a comprehensive assessment of cell quality, avoiding the one-sidedness that may be caused by a single indicator assessment.

[0060] 2. This invention uses a temporal convolutional neural network to perform source analysis on culture environment factors, establishes a causal relationship model between environmental parameters and cell quality, and can accurately identify key environmental factors affecting cell quality, providing data support for process optimization.

[0061] 3. This invention uses a convolutional neural network to automatically identify cell morphological features, which greatly improves the accuracy and efficiency of morphological analysis and solves the problems of subjectivity and inefficiency in traditional manual interpretation.

[0062] 4. This invention establishes an adaptive model for multidimensional cell index scoring, realizing intelligent fusion and adaptive weight allocation of cell characteristic data from different dimensions, overcoming the limitations of traditional fixed-weight scoring systems.

[0063] 5. This invention achieves automatic grading of cell preparation quality based on SVM classifier, making full use of the negative correlation between cell multidimensional index scores and environmental comprehensive scores, thereby improving the accuracy and consistency of quality grading.

[0064] 6. This invention automates the entire process from data acquisition to report generation, significantly improving the efficiency of cell preparation quality assessment, reducing human error, and providing technical support for the industrial production of cell therapy products. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the overall structure of the intelligent quality assessment system for the entire cell preparation process of the present invention;

[0066] Figure 2 This is a schematic diagram of the data acquisition module of the present invention;

[0067] Figure 3 This is a schematic diagram of the data modeling module of the present invention;

[0068] Figure 4 This is a schematic diagram of the temporal convolutional neural network model of the present invention;

[0069] Figure 5 This is a schematic diagram of the adaptive model structure for cell multidimensional index scoring in this invention;

[0070] Figure 6 This is a schematic diagram of the workflow of the SVM classifier of the present invention;

[0071] Figure 7 This is a flowchart of the intelligent quality assessment method for the entire cell preparation process of the present invention. Detailed Implementation

[0072] Please refer to the attached document. Figure 1-7 The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0073] like Figure 1 As shown, the intelligent quality assessment system for the entire cell preparation process provided by this invention includes a data acquisition module 1, a data preprocessing module 2, a data mining module 3, a data modeling module 4, an automatic report generation module 5, and a data storage module 6.

[0074] Data acquisition module 1 is used to collect flow cytometry data, high-content imaging data, environmental parameter data, and experimental parameter data during cell preparation. In a preferred embodiment of the invention, such as... Figure 2 As shown, the data acquisition module 1 includes a flow cytometry detection and analysis module 11, a high-content imaging data module 12, an environmental parameter module 13, and an experimental data module 14.

[0075] The flow cytometry detection and analysis module 11 is used to acquire the analysis results of the flow cytometer. Preferably, the flow cytometry detection data acquisition frequency is 200-500 data points every 5 minutes. After acquisition, the cells are washed with PBS to avoid residual reagents affecting the cells.

[0076] The high-content imaging data module 12 is used to acquire cell morphology and activity images and multi-channel fluorescence images collected by the high-content imaging system. Preferably, the high-content image data is acquired at a frequency of one image every 12 hours to track the dynamic process of cell morphology changes, while avoiding phototoxicity to cells caused by frequent imaging.

[0077] The environmental parameter module 13 is used to acquire environmental parameters inside and outside the incubator, including but not limited to data such as CO2 concentration, temperature, humidity, and pressure. Preferably, the environmental parameters are acquired using a timed data acquisition method, and the data is output via TCP / IP protocol to ensure the reliability and real-time performance of data transmission.

[0078] Experimental data module 14 is used to acquire experimental parameters during sample preparation and testing, including but not limited to sample preparation process parameters, culture medium formulation, cell density, etc.

[0079] The data preprocessing module 2 is communicatively connected to the data acquisition module 1 and is used to receive data acquired by the data acquisition module 1, and to clean and preprocess the data. In one embodiment of the present invention, the data preprocessing module 2 performs background subtraction processing on the streaming detection data to remove non-specific signal interference; performs standardization processing on high-content image data to unify image resolution and brightness; and performs outlier processing and missing value supplementation on environmental parameter data.

[0080] Preferably, the data preprocessing module 2 uses the quantile interval calculation method to determine outliers in the sample data. Specifically, for a given dataset X, its first quartile Q1 and third quartile Q3 are calculated, and then the quantile interval IQR = Q3 - Q1 is calculated. Values ​​in the data that are less than Q1 - 1.5 × IQR or greater than Q3 + 1.5 × IQR are identified as outliers and replaced by the mean or median of the nearest points.

[0081] The data mining module 3 is communicatively connected to the data preprocessing module 2, and is used to receive the preprocessed data from the data preprocessing module 2. The preprocessed data is then visualized and analyzed according to the cell product indicator evaluation system. In a preferred embodiment of the invention, the data mining module 3 includes a sample information table, which includes multiple indicators. Each indicator is processed individually in its corresponding workflow to generate a result report.

[0082] Preferably, the high-content image index workflow includes cell segmentation of multi-channel fluorescence data and data analysis by flow cytometry to obtain key information such as cell viability, growth status, CD markers, and proliferation activity. Cell segmentation employs a deep learning-based U-Net model, which consists of an encoder and a decoder. The encoder extracts features through convolutional and pooling layers, while the decoder restores spatial resolution through upsampling and convolution, achieving pixel-level segmentation.

[0083] The data modeling module 4 communicates with the data mining module 3, receiving the analysis results from the data mining module 3 to build the foundation for the data model, constructing a cell product indicator evaluation system and a product traceability model. For example... Figure 3 As shown, the data modeling module 4 includes a cell culture status assessment model 41 and an environmental monitoring and analysis module 42.

[0084] The environmental monitoring and analysis module 42 uses a temporal convolutional neural network (TCN) to perform source analysis on environmental factors inside and outside the incubator. TCNs have a long effective memory length, enabling them to capture long-term dependencies of environmental parameters, making them suitable for processing time-series data during the culture process. For example... Figure 4 As shown, the TCN model increases the receptive field by dilated convolution while keeping the number of parameters constant, effectively processing long sequence data.

[0085] Preferably, the temporal convolutional neural network model used in the environmental monitoring and analysis module 42 undergoes parameter tuning through the following steps:

[0086] Step 1: Parameter tuning of the temporal convolutional neural network model for environmental monitoring data fusion;

[0087] Step 2: Train the temporal data using a temporal convolutional neural network model and determine the model accuracy;

[0088] Step 3: Verify the accuracy of the traceability of environmental monitoring data.

[0089] In step 1, the main parameters to be tuned include the number of network layers (usually 2-8 layers), the dilation factor (generally set to 1, 2, 4, 8...), the kernel size (usually 3 or 5), and the learning rate (initial value set to 0.001). In this embodiment, after multiple experimental comparisons, the optimal parameter combination is: 4 network layers, dilation factor [1, 2, 4, 8], kernel size 3, learning rate 0.0005, and batch size 32.

[0090] In step 2, the incubator environmental monitoring data (including CO2 concentration, temperature, humidity, and pressure) are arranged chronologically to form time-series data. The first 80% of the data is used as the training set, and the last 20% is used as the validation set. The formula for calculating the accuracy of prediction based on the incubator's internal and external environmental monitoring data using a temporal convolutional neural network is as follows:

[0091] ,

[0092] Where n represents the length of the data sequence, Represents the actual value of the time series, i.e., the first... The actual environmental parameter values ​​for each sample This represents the prediction result, that is, the first prediction made by the model. The environmental parameter values ​​for each sample. Generally, an accuracy rate of 95% or higher can be considered a good model performance.

[0093] In step 3, the formula for calculating the goodness of fit of the incubator internal and external environmental monitoring data traceability model is as follows:

[0094] ,

[0095] in, The dimensions representing environmental monitoring data include, for example, four dimensions: CO2 concentration, temperature, humidity, and pressure. This represents the data values ​​output by different sensors after the j-th iteration. This indicates the threshold values ​​for environmental parameters. Preferably, the threshold values ​​are set as follows: CO2 concentration 0.2%, temperature 0.5℃, humidity 2%, and pressure 0.5kPa.

[0096] The formula for calculating the traceability accuracy of incubator internal and external environmental monitoring data is as follows:

[0097] ,

[0098] in, Indicates the number of times the parameter changes. This represents the threshold for parameter variation, typically set to 5% of the parameter value. This represents the parameters after the j-th iteration. Indicates the first The parameters after the next iteration. When the source tracing accuracy reaches over 90%, the model can be considered to have good source tracing ability.

[0099] The cell culture status evaluation model 41 assesses the quality of cell preparation using an artificial intelligence model. In one embodiment of the present invention, the product indicator evaluation system in the data modeling module 4 is implemented through the following steps:

[0100] Step 1: Construction and preprocessing of the sample morphological index dataset;

[0101] Step 2: Cell morphology feature detection based on convolutional neural network;

[0102] Step 3: Calculation of the positive rate of cell surface markers;

[0103] Step 4: Calculation of surface marker anomaly rate;

[0104] Step 5: Calculation of cell function indicators;

[0105] Step 6: Sample quality assessment.

[0106] In step 1, cell images from the high-content imaging system are collected, and the cell image data are labeled to obtain multiple cell images and cell morphology markers for these images. Data preprocessing methods include converting the resolution of the original images to a standard size of 256×256, while simultaneously performing brightness normalization and noise filtering.

[0107] In step 2, ResNet50 is used as the basic architecture of the convolutional neural network, and a custom fully connected layer is added for cell morphology feature detection. The model is trained with a batch size of 16, an initial learning rate of 0.001, and the Adam optimizer for 100 training epochs.

[0108] In step 3, the formula for calculating the positive rate of cell surface markers is:

[0109] ,

[0110] in, Indicates the number of cells positive for surface markers. This represents the total number of cells. The threshold for determining positivity varies depending on the cell type. For example, for T cells, the fluorescence intensity threshold for CD3 positivity is typically set to the control group mean plus two standard deviations.

[0111] In step 4, the formula for calculating the surface marker anomaly rate is:

[0112] ,

[0113] in, This indicates the number of cells that are negative for surface markers (should be positive but tested negative). This indicates the total number of cells.

[0114] In step 5, the formula for calculating cell function indicators is as follows:

[0115] ,

[0116] in, Indicates the activity rate of the sample pores. This indicates the activity rate of the experimental wells.

[0117] In step 6, the sample quality assessment adopts a comprehensive scoring method, and the quality coefficient is calculated using the following formula:

[0118] ,

[0119] in, Indicates the first The state coefficient score for each sample Indicates the first The standard coefficients for each classification are usually weighted according to the importance of different indicators. For example, the weight of activity rate can be set to 0.4, the weight of surface marker expression can be set to 0.3, and the weight of morphological characteristics can be set to 0.3.

[0120] In a preferred embodiment of the present invention, the data modeling module 4 includes an adaptive model for cell multidimensional index scoring, such as... Figure 5 As shown, the construction of this model includes:

[0121] The first step is to establish an initial model for multidimensional index scoring;

[0122] The second step is to establish model evaluation metrics using cross-validation.

[0123] The third step is to select the optimal model based on the model's output parameters and model evaluation metrics.

[0124] The adaptive model for cell multidimensional index scoring takes cell morphology data and cell surface marker expression data as inputs. These data are processed by two fully connected layers to obtain scores based on cell morphology and scores based on cell surface marker expression. The cell multidimensional index score is obtained by summing the scores based on cell morphology and scores based on cell surface marker expression.

[0125] Preferably, the fully connected layers use the ReLU activation function to avoid the vanishing gradient problem. The number of neurons in the first fully connected layer is set to twice the dimension of the input features, and the number of neurons in the second fully connected layer is set to 32. To prevent overfitting, dropout layers are added between the fully connected layers, with a dropout rate of 0.3. The model uses mean squared error (MSE) as the loss function, employs the Adam optimizer, and sets the learning rate to 0.001.

[0126] The model evaluation metrics established using cross-validation include: mean absolute error (MAE) and correlation coefficient (R²) as model evaluation metrics. The formulas for MAE and R² are as follows:

[0127] ,

[0128] ,

[0129] in, This refers to the multidimensional cell index scoring data, i.e., the actual cell quality score. This is the model output value, i.e., the cell quality score predicted by the model. The mean score of the cellular multidimensional index. The sample size is denoted by . The lower the MAE output value, the higher the R² output value, and the better the performance of the scoring model. Generally, a MAE less than 0.1 and an R² greater than 0.8 are considered to indicate good model performance.

[0130] The automatic report generation module 5 is communicatively connected to the data mining module 3 and the data modeling module 4. It receives the analysis results from the data mining module 3 and the data model from the data modeling module 4, intelligently assesses the quality of the entire cell preparation process, and automatically generates an assessment report. In a preferred embodiment of the invention, the automatic report generation module 5 integrates the data mining results and assessment results into the assessment report, including but not limited to cell viability, surface marker expression, morphological characteristics, environmental parameter analysis, and quality level determination.

[0131] Preferably, the automatic report generation module 5 includes an automatic quality grade determination unit, which is implemented based on an SVM classifier. The SVM classifier is trained based on the negative correlation between cell multidimensional index scores and the comprehensive environmental score, and is used to calculate the cell preparation quality score and classify the quality grade. Figure 6 As shown, the SVM classifier first preprocesses the input data, and then makes a classification decision based on the trained model.

[0132] The formula for calculating the quality score used in the SVM classifier is:

[0133] ,

[0134] in, Prepare mass fractions for cells, The mean absolute error of the output of the SVM classifier is used. This is a weighting coefficient, typically ranging from 0.1 to 0.5; in this embodiment, a value of 0.3 is preferred. The correlation coefficients are the outputs of the SVM classifier.

[0135] Before classification, the SVM classifier preprocesses the input cell multidimensional index scores and the comprehensive preparation environment score. This preprocessing includes data normalization, which employs a feature scaling algorithm to normalize the input cell multidimensional index scores and the comprehensive preparation environment score to a range of 0-1. The formula for the feature scaling algorithm is:

[0136] ,

[0137] in, The input includes multidimensional cell index scores and a comprehensive score of the preparation environment. The output of the feature scaling algorithm, and They are respectively The minimum and maximum values.

[0138] The choice of kernel function for an SVM classifier is a key factor determining its classification performance. In this embodiment, the RBF kernel function was chosen by comparing the performance of linear kernels, polynomial kernels, and radial basis function (RBF) kernels because it handles nonlinear relationships better. The formula for the RBF kernel function is:

[0139] ,

[0140] Here, γ is a kernel parameter that controls the flexibility of the decision boundary. The optimal value is usually determined by cross-validation. In this embodiment, γ is preferably set to 0.1. This represents the squared Euclidean distance between two sample points.

[0141] The penalty parameter C of the SVM classifier controls the degree of punishment for misclassification. A larger C value results in a tighter fit of the model to the training data, but may lead to overfitting. In this embodiment, the optimal C value of 10 is determined through grid search.

[0142] The data storage module 6 is communicatively connected to the automatic report generation module 5, and is used to manage the storage, sharing, and retrieval of data and models. In a preferred embodiment of the invention, the data storage module 6 utilizes data management tools to integrate data and models generated in the system process, including but not limited to raw data, preprocessed data, analysis results, model parameters, and evaluation reports. Data storage adopts a hierarchical storage structure, classifying and managing data according to data type and usage frequency to improve data access efficiency.

[0143] This invention also provides a method for intelligent quality assessment of the entire cell preparation process, such as... Figure 7 As shown, it includes the following steps:

[0144] Step 1: Acquire high-content cell images, cell viability, and surface marker data of the sample using a high-content imaging system;

[0145] Step 2: Process the high-content cell image files;

[0146] Step 3: Collect sample preparation parameter data using flow cytometry;

[0147] Step 4: Construct a multi-dimensional evaluation system for cell product quality. Analyze and calculate the data from Steps 1 and 3 based on the multi-dimensional evaluation system for cell product quality to obtain the cell culture status analysis results.

[0148] Step 5: Obtain environmental monitoring data inside and outside the incubator, and use the traceability model that integrates cell culture status analysis results and environmental monitoring data inside and outside the incubator to obtain the influencing factors of the environmental monitoring data inside and outside the incubator.

[0149] Step 6: Based on the negative correlation between cell multidimensional index scores and environmental comprehensive scores, calculate the cell preparation quality score using an SVM classifier, and classify the cell preparation quality according to the quality score.

[0150] Step 7: Integrate the data mining results and evaluation results into the evaluation report to automatically generate a quality evaluation report;

[0151] Step 8: Use data management tools to integrate the data and models generated in the system process for storage, sharing, and retrieval.

[0152] In step 1, the high-content imaging system acquires high-content image files of the cells, cell viability, and surface marker data of the sample. Preferably, the high-content imaging system employs a confocal microscope or a fluorescence microscope, equipped with autofocus and multi-channel fluorescence detection functions to achieve simultaneous observation of cell morphology and multiple fluorescent markers. The acquisition frequency is set to one image every 12 hours to balance the temporal resolution of data acquisition with interference to cells.

[0153] In step 2, the high-content cell image files undergo data processing. This includes image preprocessing (noise filtering, background correction, brightness normalization, etc.), cell identification and segmentation, and feature extraction. Cell identification and segmentation employ deep learning methods, using pre-trained U-Net or Mask R-CNN models to achieve accurate cell identification and boundary determination. Feature extraction includes morphological features (such as cell area, perimeter, and roundness) and fluorescence intensity features (such as average fluorescence intensity and fluorescence distribution across channels).

[0154] In step 3, sample preparation parameter data are acquired using flow cytometry. Preferably, the flow cytometry acquisition includes sample preparation parameters and flow cytometry sorting parameters. Sample preparation parameters include cell density, culture medium formulation, types and concentrations of added factors, etc.; flow cytometry sorting parameters include fluorescence channel settings, threshold values, sorting speed, etc. The acquisition frequency is 200-500 data points every 5 minutes, and the cells are washed with PBS after acquisition.

[0155] In step 4, a multi-dimensional evaluation system for cell product quality is constructed. Based on this system, the data from steps 1 and 3 are analyzed and calculated to obtain the cell culture status analysis results. The multi-dimensional evaluation system includes cell morphology indicators, cell surface marker indicators, and cell function indicators.

[0156] In step 5, environmental monitoring data inside and outside the incubator are acquired. A tracing model, which integrates cell culture status analysis results and incubator environmental monitoring data, is used to identify the influencing factors of the environmental monitoring data. Environmental monitoring data includes parameters such as CO2 concentration, temperature, humidity, and pressure, and is collected in real-time using IoT sensing technology. The tracing model is constructed based on a temporal convolutional neural network, identifying key influencing factors by analyzing the correlation between time-series changes in environmental parameters and cell quality.

[0157] In step 6, based on the negative correlation between the cell multidimensional index score and the environmental comprehensive score, an SVM classifier is used to calculate the cell preparation quality score, and the cell preparation quality is classified according to the quality score. The negative correlation means that a higher cell multidimensional index score indicates a smaller impact of environmental factors on cell quality, and a correspondingly lower environmental comprehensive score; conversely, a lower cell multidimensional index score usually indicates a greater impact of environmental factors on cell quality, and a correspondingly higher environmental comprehensive score.

[0158] Quality grades are typically classified into four levels: Excellent, Good, Acceptable, and Unacceptable. Preferably, a quality score greater than 0.9 is Excellent, 0.8-0.9 is Good, 0.7-0.8 is Acceptable, and less than 0.7 is Unacceptable. These thresholds can be adjusted according to different cell types and application requirements.

[0159] In step 7, the data mining results and evaluation results are integrated into the evaluation report, automatically generating a quality evaluation report. The evaluation report includes basic information (such as sample ID, evaluation date, etc.), cell quality parameters (such as viability, surface marker expression, etc.), environmental factor analysis, quality grade determination results, and recommendations.

[0160] In step 8, data management tools are used to integrate the data and models generated in the system process for storage, sharing, and retrieval. Preferably, a distributed storage system and version control mechanism are employed to ensure the security and traceability of data and models. Simultaneously, an access control system is established to control access permissions for sensitive data.

[0161] Through the above method, the present invention realizes intelligent quality assessment of the entire cell preparation process, providing a powerful tool for quality control of cell preparation.

[0162] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart quality assessment system for the entire cell preparation process, characterized in that, include: The data acquisition module is used to collect flow cytometry data, high-content imaging data, environmental parameter data, and experimental parameter data during the cell preparation process; A data preprocessing module, which is communicatively connected to the data acquisition module, is used to receive data acquired by the data acquisition module and clean and preprocess the data. The data mining module is communicatively connected to the data preprocessing module and is used to receive the data preprocessed by the data preprocessing module and to perform visualization exploration and analysis on the preprocessed data according to the cell product index evaluation system. The data modeling module is communicatively connected to the data mining module and is used to receive the analysis results of the data mining module, build the foundation of the data model, construct the cell product indicator evaluation system and the product traceability model. The data modeling module includes a cell culture status evaluation model and an environmental monitoring and analysis module. The environmental monitoring and analysis module uses a temporal convolutional neural network to perform traceability analysis on environmental factors inside and outside the incubator. The cell culture status evaluation model evaluates the cell preparation quality through an artificial intelligence model. The report generation module is communicatively connected to the data mining module and the data modeling module. It is used to receive the analysis results of the data mining module and the data model of the data modeling module, to intelligently evaluate the quality of the entire cell preparation process, and to automatically generate an evaluation report. The data storage module is connected in communication with the automatic report generation module and is used to manage the storage, sharing, and retrieval of data and models.

2. The intelligent quality assessment system for the entire cell preparation process according to claim 1, characterized in that: The data acquisition module includes a flow cytometry detection and analysis module, a high-content imaging data module, an environmental parameter module, and an experimental data module; The flow cytometry detection and analysis module is used to obtain the analysis results of the flow cytometer; The high-content imaging data module is used to acquire cell morphology and activity images and multi-channel fluorescence images collected by the high-content imaging system. The environmental parameter module is used to acquire environmental parameters inside and outside the incubator; The experimental data module is used to acquire experimental parameters during sample preparation and testing.

3. The intelligent quality assessment system for the entire cell preparation process according to claim 2, characterized in that: The flow cytometry detection and analysis module includes sample preparation parameters and cell analysis parameters; The sample preparation parameters include cell viability and quality control data files, which are used to obtain the cell product activity rate and cell quality control standard deviation. The cell analysis parameters include positive sample data files and negative control data files, used to obtain the fluorescence intensity of each channel of the flow cytometer. The high-content imaging data module includes data processing of high-content cell image files.

4. The intelligent quality assessment system for the entire cell preparation process according to claim 3, characterized in that: The data mining module includes a sample information table, which includes multiple indicators. Each indicator is processed separately in the corresponding workflow to generate a result report. The high-content image metrics workflow includes cell segmentation of multichannel fluorescence data; Flow cytometry was used to analyze data to obtain cell viability, growth status, CD markers, and proliferation activity.

5. The intelligent quality assessment system for the entire cell preparation process according to claim 4, characterized in that: The product indicator evaluation system in the data modeling module is implemented through the following steps: Step 1: Construction and preprocessing of the sample morphological index dataset; Step 2: Cell morphology feature detection based on convolutional neural network; Step 3: Calculation of the positive rate of cell surface markers; Step 4: Calculation of surface marker anomaly rate; Step 5: Calculation of cell function indicators; Step 6: Sample quality assessment.

6. The intelligent quality assessment system for the entire cell preparation process according to claim 5, characterized in that: The environmental monitoring and analysis module uses a self-designed temporal convolutional neural network model to perform source analysis on environmental factors inside and outside the incubator. The temporal convolutional neural network model is optimized through the following steps: Step 1: Parameter tuning of the temporal convolutional neural network model for environmental monitoring data fusion; Step 2: Train the temporal data using a temporal convolutional neural network model and determine the model accuracy; Step 3: Verify the accuracy of the traceability of environmental monitoring data.

7. The intelligent quality assessment system for the entire cell preparation process according to claim 6, characterized in that: The construction of the adaptive model for cell multidimensional index scoring includes: The first step is to establish an initial model for multidimensional index scoring; The second step is to establish model evaluation metrics using cross-validation. The third step is to select the optimal model based on the model's output parameters and model evaluation metrics. The adaptive model for multidimensional index scoring takes cell morphology data and cell surface marker expression data as inputs. These data are processed by two fully connected layers to obtain cell morphology-based scores and cell surface marker expression-based scores. The cell multidimensional index score is obtained by summing the cell morphology-based scores and cell surface marker expression-based scores.

8. The intelligent quality assessment system for the entire cell preparation process according to claim 7, characterized in that: The automatic report generation module integrates data mining results and evaluation results into the evaluation report; The data storage module utilizes data management tools to integrate data and models generated during system processes; The automatic report generation module includes an automatic quality grade determination unit, which is based on an SVM classifier. The SVM classifier is trained based on the negative correlation between cell multidimensional index scores and the comprehensive environmental score, and is used to calculate the cell preparation quality score and classify the quality grade.

9. The intelligent quality assessment system for the entire cell preparation process according to claim 8, characterized in that: The quality score calculation formula used by the SVM classifier is as follows: ； in, Prepare a mass fraction for the cells. The mean absolute error of the output of the SVM classifier is used. These are the weighting coefficients. The correlation coefficient is the output of the SVM classifier; Before classification, the SVM classifier preprocesses the input cell multidimensional index scores and the comprehensive score of the preparation environment. The preprocessing includes normalizing the data, which involves using a feature scaling algorithm to normalize the input cell multidimensional index scores and the comprehensive score of the preparation environment to a range of 0-1.

10. A method for intelligent quality assessment of the entire cell preparation process, using the system described in any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Acquire high-content cell images, cell viability, and surface marker data of the sample using a high-content imaging system; Step 2: Process the high-content cell image files; Step 3: Collect sample preparation parameter data using flow cytometry; Step 4: Construct a multi-dimensional evaluation system for cell product quality. Analyze and calculate the data from Steps 1 and 3 based on the multi-dimensional evaluation system for cell product quality to obtain the cell culture status analysis results. Step 5: Obtain environmental monitoring data inside and outside the incubator, and use the traceability model that integrates cell culture status analysis results and environmental monitoring data inside and outside the incubator to obtain the influencing factors of the environmental monitoring data inside and outside the incubator. Step 6: Based on the negative correlation between cell multidimensional index scores and environmental comprehensive scores, calculate the cell preparation quality score using an SVM classifier, and classify the cell preparation quality according to the quality score. Step 7: Integrate the data mining results and evaluation results into the evaluation report to automatically generate a quality evaluation report; Step 8: Use data management tools to integrate the data and models generated in the system process for storage, sharing, and retrieval.

Images