Method for evaluating activity of tumor sample in cold storage process

Abstract

The invention provides an activity evaluation method for a tumor sample in a cold storage process, which belongs to the technical field of medical sample cold storage and comprises the following steps: continuously monitoring and recording a plurality of parameters of the tumor sample in a freezing process, including temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration, lactic acid concentration and the like. Through data preprocessing, parameter time sequence matrix construction and matrix decomposition calculation, a steady-state correlation matrix and a change correlation matrix of each parameter are obtained. On the basis of the two correlation matrixes, a parameter change coefficient matrix is calculated, and the characteristic that the parameters change along with time is represented. And inputting the matrix into a pre-trained activity evaluation model to obtain an activity evaluation score of the tumor sample so as to determine the cold storage quality grade of the tumor sample. A nondestructive testing mode is adopted, the method is rapid and efficient, and the problems that an existing detection method needs to consume a large number of samples, the detection process is complex and tedious, and large-batch tumor samples cannot be rapidly screened and evaluated are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical sample cold storage, and in particular, relates to a method for evaluating the activity of a tumor sample during cold storage. Background Art

[0002] Cryopreservation of tumor samples is one of the important technologies in clinical medicine and biomedical research. By cryopreserving tumor tissue at low temperatures, the storage time of samples can be effectively extended, ensuring that the tissue structure and biological activity are not destroyed, and providing important experimental materials for subsequent pathological diagnosis, drug screening and treatment research. At present, common clinical methods for cryopreservation of tumor samples include liquid nitrogen freezing (-196°C) and ultra-low temperature refrigerator freezing (-80°C). Under these low-temperature environments, the water in biological samples will crystallize, thereby inhibiting the metabolic activity of cells and slowing down the aging and degradation of samples.

[0003] However, even under strict low-temperature conditions, tumor samples will inevitably suffer a certain degree of cell damage during the freezing-thawing process. This damage is mainly manifested in cell membrane rupture, protein denaturation, DNA damage, etc., which will lead to a decrease in the biological activity of the sample. Therefore, how to accurately evaluate the activity state of tumor samples during cryopreservation is very important for subsequent medical research and clinical applications.

[0004] Existing methods for evaluating the activity of tumor samples mainly include cell viability detection (MTT method), ATP content determination, immunohistochemical analysis, etc. These methods are usually performed after the sample is thawed, and cannot monitor the dynamic changes of the sample during the freezing process in real time. In addition, these methods require a large amount of sample consumption, and the detection process is complicated and cumbersome, which is not conducive to the rapid screening and evaluation of large quantities of tumor samples.

[0005] Therefore, there is an urgent need to develop a new method for evaluating the activity of tumor samples during cold storage, which can continuously monitor and analyze the changes in multiple key biochemical parameters during sample cryopreservation, so as to accurately predict the activity status of the samples and provide a reliable basis for clinical applications. Summary of the invention

[0006] In view of this, the present invention provides a method for evaluating the activity of tumor samples during cold storage, which can solve the problem that existing detection methods need to consume a large amount of samples and the detection process is complicated and cumbersome, which is not conducive to the rapid screening and evaluation of large quantities of tumor samples.

[0007] The present invention is achieved in that:

[0008] The present invention provides a method for evaluating the activity of a tumor sample during cold storage, comprising the following steps:

[0009] S10, collecting tumor samples, and continuously monitoring and recording temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration, and lactate concentration during the freezing process to form multi-parameter time series data;

[0010] S20, preprocessing the multi-parameter time series data, eliminating noise and performing standardization processing;

[0011] S30, sampling the multi-parameter time series data at preset time intervals to obtain parameter data at multiple moments, and constructing a parameter time series matrix;

[0012] S40, decomposing and calculating the parameter time series matrix to obtain a steady-state parameter matrix and a change parameter matrix;

[0013] S50, establishing a correlation matrix between the parameters in the steady-state parameter matrix, recorded as a steady-state correlation matrix, and establishing a correlation matrix between the parameters in the change parameter matrix, recorded as a change correlation matrix;

[0014] S60, calculating a parameter variation coefficient matrix based on the steady-state correlation matrix and the variation correlation matrix, wherein the parameter variation coefficient matrix represents the characteristics of each parameter varying with time;

[0015] S70, inputting the parameter variation coefficient matrix into a pre-trained activity evaluation model to obtain an activity evaluation score for the tumor sample;

[0016] S80: Determine the cold storage quality grade of the tumor sample according to the tumor sample activity evaluation score and output it.

[0017] On the basis of the above technical solution, the method for evaluating the activity of a tumor sample during cold storage of the present invention can also be improved as follows:

[0018] Wherein, the activity evaluation model adopts BP neural network structure.

[0019] Furthermore, the step of constructing the training data set of the activity evaluation model includes collecting a large amount of cold storage process data of tumor samples, annotating and preprocessing them, and dividing them into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0020] Specifically, the steps of constructing the training data set of the activity evaluation model specifically include:

[0021] 1. Data Collection:

[0022] (a) Collect a large number of tumor samples of different types and stages, including but not limited to lung cancer, liver cancer, breast cancer, colon cancer, etc.;

[0023] (b) Each tumor sample was subjected to cold storage for 48 hours;

[0024] (c) During the cold storage process, the temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration, and lactic acid concentration were recorded every 5 minutes to form multi-parameter time series data;

[0025] (d) After cold storage, cell activity was immediately tested, and the cell viability was determined using the MTT assay as a true activity marker for the sample.

[0026] 2. Data preprocessing:

[0027] (a) Perform outlier detection on the collected raw data and use the box plot method to identify and delete outliers;

[0028] (b) Use sliding window median filtering to remove high-frequency noise in the data, and the window size is set to 5;

[0029] (c) Normalize the parameter data using the Min-Max normalization method:

[0030]

[0031] In the formula, x norm is the normalized value, x is the original value, and x min and x max are the minimum and maximum values ​​of this parameter respectively.

[0032] 3. Feature extraction:

[0033] (a) The preprocessed data is divided into time windows, with the window size set to 1 hour and the step length set to 30 minutes;

[0034] (b) In each time window, the following statistical characteristics are calculated: mean, standard deviation, kurtosis, skewness, maximum, minimum, median, and interquartile range;

[0035] (c) Calculate the correlation coefficient matrix between each parameter;

[0036] (d) Use wavelet transform to extract time-frequency domain features, select Daubechies wavelet, and the decomposition level is 4;

[0037] (e) Calculate the rate of change of each parameter:

[0038]

[0039] Where Δt is the sampling interval.

[0040] 4. Label processing:

[0041] (a) The cell viability measured by the MTT method was converted into discrete activity levels, which were divided into three levels: high, medium, and low;

[0042] (b) Setting thresholds: high activity ≥ 80%, low activity ≤ 50%, and medium activity between 50% and 80%.

[0043] 5. Data Augmentation:

[0044] (a) Oversample the minority samples and use the SMOTE (Synthetic Minority Oversampling Technique) algorithm to generate new minority samples;

[0045] (b) Under-sampling the majority class samples and using the ENN (Edited Nearest Neighbor) algorithm to remove noise and overlapping samples;

[0046] (c) Use data enhancement techniques, such as adding Gaussian noise and time window shifting, to increase the diversity and robustness of the dataset.

[0047] 6. Feature Selection:

[0048] (a) Using the correlation analysis method, calculate the correlation coefficient between each feature and the active label, and eliminate the features with an absolute value of the correlation coefficient less than 0.1;

[0049] (b) Dimensionality reduction was performed using principal component analysis (PCA), retaining the principal components that explained 95% of the variance;

[0050] (c) The recursive feature elimination (RFE) algorithm is used in combination with the random forest classifier to select the optimal feature subset.

[0051] 7. Dataset division:

[0052] (a) Using the stratified sampling method, the dataset is divided into training set, validation set, and test set in a ratio of 8:1:1;

[0053] (b) Ensure that the proportion of samples with different activity levels in each subset is consistent with the original dataset;

[0054] (c) The training set is further divided using the K-fold cross-validation method (K=5) to evaluate the generalization ability of the model.

[0055] 8. Data Standardization:

[0056] The divided training set, validation set, and test set are standardized respectively, using the Z-score standardization method:

[0057]

[0058] In the formula, z is the standardized value, x is the original value, μ is the mean of the feature, and σ is the standard deviation of the feature.

[0059] Furthermore, the step of training the activity evaluation model includes initializing model parameters, performing back propagation training using an Adam optimizer, adopting a cross entropy loss function, and using an early stopping method to prevent overfitting.

[0060] The formula of the parameter timing matrix is ​​as follows:

[0061]

[0062] Where X is the parameter time series matrix; ij represents the jth parameter value at the ith moment; m is the number of sampling moments; n is the number of monitored parameters.

[0063] The formula for decomposing and calculating the parameter timing matrix is ​​as follows:

[0064] X = S + V + W;

[0065] Where S is the steady-state parameter matrix; V is the variable parameter matrix; E is the error matrix.

[0066] The formula of the steady-state parameter matrix is ​​as follows:

[0067]

[0068] In the formula, s ij represents the steady-state value of the jth parameter at the ith moment.

[0069] The formula of the variation parameter matrix is ​​as follows:

[0070]

[0071] In the formula, v ij Indicates the change value of the jth parameter at the i-th moment.

[0072] The formula for establishing the correlation matrix between the parameters in the steady-state parameter matrix is ​​as follows:

[0073]

[0074] In the formula, R S is the steady-state correlation matrix; r ij Represents the correlation coefficient between the i-th parameter and the j-th parameter, and the calculation formula is:

[0075]

[0076] in, and are the average values ​​of the i-th and j-th parameters respectively.

[0077] The formula for establishing the correlation matrix between the parameters in the variation parameter matrix is ​​as follows:

[0078]

[0079] In the formula, R V is the change correlation matrix; r ij ′ represents the change correlation coefficient between the i-th parameter and the j-th parameter, and the calculation formula is:

[0080]

[0081] in, and are the mean values ​​of the changes of the i-th and j-th parameters respectively.

[0082] The formula for calculating the parameter variation coefficient matrix is ​​as follows:

[0083]

[0084] Where C is the parameter variation coefficient matrix; c ij It represents the influence coefficient of the change of the i-th parameter on the j-th parameter. The calculation formula is:

[0085]

[0086] Among them, α, β, γ, δ are weight coefficients, is the partial derivative of parameter i with respect to parameter j, θ ij is the phase difference between parameter i and parameter j, where i is an imaginary unit.

[0087] The methods for obtaining each parameter are as follows:

[0088] 1.r ij and r ij ′ is obtained by the above correlation coefficient calculation formula;

[0089] 2. Compute by numerical differentiation method:

[0090]

[0091] Among them, Δt is the time step;

[0092] 3. θ ij Calculated by Fourier transform:

[0093]

[0094] Where F(·) represents Fourier transform, arg(·) represents the argument of the complex number;

[0095] 4.α,β,γ,δ are determined by an optimization algorithm, such as the particle swarm optimization algorithm (PSO):

[0096]

[0097] In the formula, and They represent the velocity and position of the kth particle at the tth iteration respectively; represents the historical optimal position of the kth particle; represents the global optimal position; ω is the inertia weight; c1, c2 are acceleration constants; r1, r2 are random numbers between (0, 1).

[0098] Through the above steps, a complete activity evaluation method for tumor sample cold storage process can be constructed, which comprehensively considers the steady-state and changing characteristics of multiple parameters and uses a deep learning model for the final activity evaluation.

[0099] Among them, the multi-parameter time series data is preprocessed, specifically including: first, using the box plot method to detect and delete outliers in the data; then, using the sliding window median filter method to remove high-frequency noise in the data; and finally performing normalization processing.

[0100] Compared with the prior art, the beneficial effects of the method for evaluating the activity of tumor samples during cold storage provided by the present invention are:

[0101] 1. Continuous monitoring of multi-parameter changes: During the sample freezing process, this method continuously monitors and records the time series data of multiple key biochemical parameters such as temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration and lactic acid concentration, comprehensively reflecting the biochemical change characteristics of the sample in a low-temperature environment.

[0102] 2. In-depth analysis of parameter correlation: This method conducts matrix decomposition and correlation analysis on the steady-state and change characteristics of the monitoring parameters, explores the intrinsic relationship between the parameters, and establishes a comprehensive parameter change coupling model. This is conducive to more accurate evaluation of the activity state of the sample.

[0103] 3. Intelligent activity evaluation: This method uses deep learning models such as BP neural network to perform intelligent activity evaluation based on parameter change characteristics, overcoming the limitation of existing methods that require a large amount of experimental data support. The evaluation results can directly give the cryopreservation quality grade of the sample, providing an objective basis for clinical application.

[0104] 4. Nondestructive monitoring and continuous evaluation: This method performs continuous monitoring and analysis during the sample freezing process without consuming the sample, and can perform activity evaluation repeatedly, providing the possibility for long-term tracking of sample status.

[0105] In summary, the method for evaluating the activity of tumor samples during cold storage proposed in the present invention makes full use of multi-parameter dynamic monitoring and deep learning technology, and can more efficiently and non-destructively evaluate the activity state of tumor samples during the cryopreservation process, providing important support for clinical diagnosis and biomedical research. It solves the problem that existing detection methods require a large amount of samples and the detection process is complicated and cumbersome, which is not conducive to the rapid screening and evaluation of large quantities of tumor samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 A flow chart of the method provided by the present invention;

[0107] Figure 2 This is a graph showing the changing trends of biochemical parameters of lung cancer samples of different stages within 90 minutes of freezing;

[0108] Figure 3 This is the correlation analysis diagram of sample biochemical parameters;

[0109] Figure 4 This is a comparison chart of the change rates of biochemical parameters of samples with different divergences;

[0110] Figure 5 Activity evaluation diagram of samples with different divergences. DETAILED DESCRIPTION

[0111] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0112] like Figure 1 FIG. 1 is a flow chart of a method for evaluating the activity of a tumor sample during cold storage provided by the present invention. The method comprises the following steps:

[0113] S10, collecting tumor samples, and continuously monitoring and recording temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration, and lactate concentration during the freezing process to form multi-parameter time series data;

[0114] S20, preprocessing the multi-parameter time series data, eliminating noise and performing standardization;

[0115] S30, sampling the multi-parameter time series data at preset time intervals to obtain parameter data at multiple moments, and constructing a parameter time series matrix;

[0116] S40, decomposing and calculating the parameter time series matrix to obtain a steady-state parameter matrix and a change parameter matrix;

[0117] S50, establishing a correlation matrix between the parameters in the steady-state parameter matrix, recorded as the steady-state correlation matrix, and establishing a correlation matrix between the parameters in the change parameter matrix, recorded as the change correlation matrix;

[0118] S60, calculating a parameter variation coefficient matrix based on the steady-state correlation matrix and the variation correlation matrix, wherein the parameter variation coefficient matrix represents the characteristics of each parameter varying with time;

[0119] S70, inputting the parameter variation coefficient matrix into the pre-trained activity evaluation model to obtain the activity evaluation score of the tumor sample;

[0120] S80. Determine the cold storage quality grade of the tumor sample according to the tumor sample activity evaluation score and output it.

[0121] The specific implementation methods of the above steps are described in detail below:

[0122] The specific implementation of step S10 is: first, collect tumor samples of different types and stages from the clinic S = {s1, s2, ..., s N}, where N is the total number of samples. For each tumor sample s i The cryopreservation treatment lasted for 48 hours. During the freezing process, six parameters, including temperature T, pH value p, oxygen content O, carbon dioxide concentration C, glucose concentration G, and lactic acid concentration L, were continuously monitored and recorded every 5 minutes to form multi-parameter time series data. The monitoring and recording of these parameters can be expressed in the following matrix form: Among them, m is the number of sampling moments, and the parameter value at each moment constitutes a row. The monitoring and recording of these parameters is intended to fully reflect the biochemical change characteristics of tumor samples during the freezing process.

[0123] The specific implementation of step S20 is: preprocessing the original multi-parameter time series data X obtained in step S10. First, the box plot method is used to detect and delete outliers in the data. For any parameter x, the judgment condition of its outlier is: Among them, Q1 and Q3 are the first and third quartiles of x respectively.

[0124] Subsequently, the sliding window median filter method was used to remove high-frequency noise in the data, and the window size was set to 5. The filtered data is recorded as The element in row i and column j is The calculation formula is:

[0125] In order to eliminate the differences in dimensions and value ranges between different parameters, the data of each parameter are normalized by Min-Max. The normalization formula is: Among them, x norm is the normalized value, x is the original value, and x min and x max are the minimum and maximum values ​​of the parameter respectively. The normalized data is recorded as

[0126] These preprocessing steps help to eliminate noise interference and improve the reliability of subsequent analysis.

[0127] The specific implementation of step S30 is: the preprocessed multi-parameter time series data Sampling is performed at a preset time interval Δt (e.g., 30 minutes) to obtain parameter data at multiple moments. Then, these sampled data are constructed into a parameter time series matrix X, where each row represents the parameter data at a moment, and each column represents the change of a certain parameter at different moments. The mathematical expression of the parameter time series matrix X is: in, represents the jth parameter value at the i-th moment; m is the number of sampling moments; n is the number of monitored parameters. Constructing the parameter time series matrix X can intuitively reflect the changing trend of each parameter during the freezing process.

[0128] The specific implementation of step S40 is: decomposing and calculating the parameter time series matrix X constructed in step S30 to obtain the steady-state parameter matrix S and the change parameter matrix V. The mathematical model of decomposition calculation is: X = S + V + E; where S represents the steady-state value of each parameter during the freezing process, V represents the change value of each parameter, and E is the error matrix. The use of matrix decomposition technology can effectively extract the steady-state and change characteristics of the parameters, laying the foundation for subsequent analysis.

[0129] The specific implementation of step S50 is: first, based on the steady-state parameter matrix S, the steady-state correlation matrix R is calculated. S , indicating the correlation between the parameters: Among them, r ij Represents the correlation coefficient between the i-th parameter and the j-th parameter, and the calculation formula is: in, and are the average values ​​of the i-th and j-th parameters respectively.

[0130] Similarly, the change correlation matrix R is calculated based on the change parameter matrix V V , indicating the correlation between the changes in parameters: Among them, r ij′ represents the change correlation coefficient between the i-th parameter and the j-th parameter, and the calculation formula is: in, and are the mean values ​​of the changes of the i-th and j-th parameters respectively.

[0131] The calculation of these two correlation matrices can reflect the intrinsic relationship between the parameters and provide a basis for the subsequent analysis of the impact of parameter changes.

[0132] The specific implementation of step S60 is: based on the steady-state correlation matrix R calculated in step S50 S And the change correlation matrix R V , construct the parameter change coefficient matrix C, which is used to characterize the characteristics of each parameter changing over time: Among them, c ij It represents the influence coefficient of the change of the i-th parameter on the j-th parameter, and the calculation formula is: Among them, α, β, γ, δ are weight coefficients, is the partial derivative of parameter i with respect to parameter j, θ ij is the phase difference between parameter i and parameter j, where i is an imaginary unit.

[0133] parameter Calculated by numerical differentiation: Among them, Δt is the time step.

[0134] Parameter θ ij Calculate by Fourier transform: Where F(·) represents Fourier transform and arg(·) represents the argument of the complex number.

[0135] The weight coefficients α, β, γ, and δ can be determined by an optimization algorithm (such as a particle swarm optimization algorithm): in, and They represent the velocity and position of the kth particle at the tth iteration respectively; represents the historical optimal position of the kth particle; represents the global optimal position; ω is the inertia weight; c1, c2 are acceleration constants; r1, r2 are random numbers between (0, 1).

[0136] These parameter changes affect the coefficient c ij Not only the steady-state and changing correlations between parameters are considered, but also the dynamic coupling relationship between parameters is incorporated, which can more comprehensively reflect the changing characteristics of each parameter during the freezing process.

[0137] The specific implementation method of step S70 is: inputting the parameter variation coefficient matrix C calculated in step S60 into a pre-trained BP neural network activity evaluation model to obtain the final tumor sample activity evaluation score.

[0138] The training process of the BP neural network model is as follows: 1. Dataset construction: (1) Collect a large number of tumor samples of different types and stages S = {s1, s2, ..., s N}, where N is the total number of samples. (2) For each tumor sample s i The freezing treatment lasted for 48 hours. During the freezing process, the temperature T, pH value p, oxygen content O, carbon dioxide concentration C, glucose concentration G and lactic acid concentration L were recorded every 5 minutes to form multi-parameter time series data. (3) After the freezing, the cell activity test was immediately carried out, and the cell survival rate y was determined by the MTT method. i As samples i (4) The collected data is divided into training set, validation set and test set in a ratio of 8:1:1.

[0139] 2. Data preprocessing: (1) Perform outlier detection and filtering on the collected raw data. (2) Use Min-Max normalization method for feature scaling.

[0140] 3. Feature Engineering: (1) In each time window (e.g., 1 hour), calculate the statistical characteristics of each parameter (such as mean μ, standard deviation σ, kurtosis K, skewness S, maximum value x max , minimum value x min , median x med , interquartile range (IQR). (2) Calculate the correlation coefficient matrix between each parameter. (3) Use wavelet transform to extract time-frequency domain features. (4) Calculate the change rate of each parameter:

[0141] 4. Labeling treatment: (1) The cell survival rate measured by MTT method was i Converted into discrete activity levels, divided into three levels: high, medium and low. (2) Set thresholds: high activity ≥ 80%, low activity ≤ 50%, and medium activity between 50% and 80%.

[0142] 5. Data enhancement: (1) Oversample the minority class samples and use the SMOTE (Synthetic Minority Oversampling Technique) algorithm to generate new minority class samples. (2) Undersample the majority class samples and use the ENN (Edited Nearest Neighbor) algorithm to remove noise and overlapping samples. (3) Use data enhancement techniques such as adding Gaussian noise and time window shifting to increase the diversity and robustness of the dataset.

[0143] 6. Feature selection: (1) Use the correlation analysis method to calculate the correlation coefficient between each feature and the active label, and eliminate the features with an absolute value of the correlation coefficient less than 0.1. (2) Use principal component analysis (PCA) to reduce the dimension and retain the principal component that explains 95% of the variance. (3) Use the recursive feature elimination (RFE) algorithm combined with the random forest classifier to select the optimal feature subset.

[0144] 7. Model training: (1) Initialize the BP neural network model parameters. (2) Use the Adam optimizer for back propagation training and adopt the cross entropy loss function. (3) Use the early stopping method to prevent overfitting.

[0145] Through the above steps, a BP neural network activity evaluation model with excellent performance can be trained. The parameter variation coefficient matrix C calculated in step S60 is input into the model to obtain the final tumor sample activity evaluation score.

[0146] The specific implementation of step S80 is: according to the tumor sample activity evaluation score obtained in step S70, the cryopreservation quality level of the tumor sample is determined and output. The higher the evaluation score, the better the sample activity and the higher the cryopreservation quality; conversely, the lower the evaluation score, the lower the sample activity and the poorer the cryopreservation quality. The corresponding threshold can be set to divide the sample into three quality levels: high, medium and low.

[0147] In summary, the method for evaluating the activity of tumor samples during cryopreservation proposed in the present invention extracts the steady-state and change characteristics of each parameter by analyzing multi-parameter time series data, establishes a parameter change coupling model, and finally uses a deep learning model to perform activity evaluation. This method can objectively and accurately evaluate the activity state of tumor samples during cryopreservation, providing an important reference for clinical practice.

[0148] Specifically, the principle of the present invention is to conduct an in-depth analysis of the time series data of multiple key biochemical parameters during the sample freezing process.

[0149] First, during the freezing process of the samples, biochemical parameters such as temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration, and lactic acid concentration were continuously monitored and recorded to form multi-parameter time series data. The changes in these parameters reflect the metabolic activity and physiological state of the samples in a low-temperature environment. For example, a drop in temperature will slow down cell metabolism, causing changes in pH value, oxygen content, glucose concentration, etc.; while damage to the cell membrane will lead to lactic acid exudation, which will increase the lactic acid concentration.

[0150] Secondly, the collected raw data is preprocessed, including outlier detection, noise filtering, and feature scaling, etc. These preprocessing steps help improve the reliability of subsequent analysis.

[0151] Then, the preprocessed multi-parameter time series data is constructed into a parameter time series matrix to reflect the changing trend of each parameter at different times. The parameter time series matrix can be decomposed into a steady-state parameter matrix and a changing parameter matrix by matrix decomposition. The steady-state parameter matrix describes the average value or equilibrium value of each parameter during the freezing process, while the changing parameter matrix reflects the dynamic change of the parameter over time.

[0152] Based on the steady-state parameter matrix and the changing parameter matrix, the correlation matrix between the parameters can be calculated respectively, namely the steady-state correlation matrix and the changing correlation matrix. These two correlation matrices reflect the intrinsic relationship between the parameters and provide a basis for the subsequent analysis of the impact of parameter changes.

[0153] Furthermore, a parameter variation coefficient matrix was constructed to characterize the characteristics of each parameter changing over time. This variation coefficient matrix not only considers the steady-state and variation correlation between parameters, but also incorporates the dynamic coupling relationship between parameters. These comprehensive factors can more comprehensively reflect the biochemical changes of samples during the freezing process.

[0154] Finally, the calculated parameter change coefficient matrix is ​​input into the pre-trained BP neural network activity evaluation model to obtain the activity evaluation score of the tumor sample. This deep learning model can automatically learn the complex mapping relationship between parameter change characteristics and sample activity, overcoming the limitation of traditional methods that require a large amount of experimental data support.

[0155] In general, the technical solution of the present invention makes full use of advanced technologies such as multi-parameter dynamic monitoring, matrix analysis and deep learning to construct a new method for evaluating the activity of tumor samples during cold storage. This method can evaluate the activity state of samples in real time and non-destructively during the sample freezing process, providing reliable decision support for clinical applications and biomedical research.

[0156] A specific embodiment of the present invention is provided below: A tumor research center of a hospital conducted a clinical study on lung cancer, aiming to evaluate the activity changes of tumor samples of patients with lung cancer at different stages during the cryopreservation process.

[0157] The researchers collected 10 lung cancer samples of different stages, including 3 cases of stage I, 4 cases of stage II and 3 cases of stage III. Each sample was cryopreserved for 48 hours, during which six key biochemical parameters, including temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration and lactate concentration, were continuously monitored and recorded every 5 minutes.

[0158] Table 1 lists the monitoring data of three representative samples during the freezing process. It can be seen that there are obvious differences in the changes in biochemical parameters of lung cancer samples of different stages. For example, the temperature of stage I samples dropped rapidly in the initial stage, and the pH value and oxygen content also decreased rapidly; while the temperature, pH value, and oxygen content of stage III samples changed relatively slowly throughout the process. These differences reflect the significant differences in the metabolic activity and physiological state of tumor tissues under freezing conditions.

[0159] Table 1 Biochemical parameter monitoring data of lung cancer samples of different stages during freezing

[0160]

[0161] In order to deeply analyze these monitoring data, the researchers first preprocessed the raw data. The box plot method was used to detect outliers and several obviously abnormal sampling points were filtered out. Then the sliding window median filter method was used to remove high-frequency noise, and the Min-Max normalization processing was performed on each parameter data.

[0162] Figure 2 The parameter change trends of the three samples after pretreatment within 90 minutes of freezing are shown. It can be seen that the parameters of stage I samples change most dramatically, while the changes of stage III samples are relatively gentle. This reflects that the physiological tolerance of lung cancer at different stages under freezing conditions is different. Figure 2 The results showed that the temperature of stage I lung cancer samples dropped rapidly within the first 30 minutes, from 37°C to 13.2°C, and the pH value and oxygen content also dropped rapidly. However, the changes in various parameters of stage III samples were relatively slow during the entire freezing process. This difference reflects the significant differences in the metabolic activity and physiological tolerance of lung cancer tissues of different stages under freezing conditions.

[0163] Figure 3 It is a correlation heat map showing the correlation between six main biochemical parameters. Red indicates positive correlation, blue indicates negative correlation, and the darker the color, the stronger the correlation. As can be seen from the figure, temperature is significantly negatively correlated with pH value (-0.88), while it is strongly positively correlated with glucose concentration (0.98). This correlation analysis helps to understand the physiological metabolic mechanism of samples during freezing.

[0164] Figure 4 The change rates of various biochemical parameters in samples of different stages are compared in the form of grouped bar graphs. The change rates in the figure are in units of per minute, with negative values ​​indicating a decrease in the parameter and positive values ​​indicating an increase in the parameter. It can be observed that the absolute value of the change rate of most parameters in stage I samples is the largest, which is consistent with its higher metabolic activity. Special attention should be paid to the correct superscript display of the -1 power marked on the y-axis.

[0165] In order to analyze the intrinsic relationship between parameters more deeply, the researchers then performed matrix decomposition and correlation analysis on the preprocessed monitoring data. First, the data was constructed into a parameter time series matrix X:

[0166]

[0167] Then, perform matrix decomposition on X to obtain the steady-state parameter matrix S and the variable parameter matrix V:

[0168]

[0169] Next, calculate the steady-state correlation matrix R S And the change correlation matrix R V :

[0170]

[0171]

[0172] It can be seen from the correlation matrix that there is a strong correlation between the parameters. For example, temperature is negatively correlated with pH value and oxygen content, but positively correlated with carbon dioxide concentration and lactic acid concentration. These intrinsic connections reflect the metabolic regulation mechanism of tumor tissue during the freezing process.

[0173] Based on the above analysis results, the researchers further constructed a parameter change coefficient matrix C to characterize the characteristics of each parameter changing over time:

[0174] This parameter variation coefficient matrix comprehensively considers the steady-state and variation correlation of each parameter, as well as the dynamic coupling relationship between them. It can be seen that, for example, a drop in temperature will inhibit pH and oxygen content and promote the accumulation of carbon dioxide and lactic acid. This information helps to more accurately predict the physiological response of samples under freezing conditions.

[0175] Finally, the researchers input the calculated parameter variation coefficient matrix C into the pre-trained BP neural network activity evaluation model, and obtained the activity evaluation scores of the three lung cancer samples: stage I sample: 78.2 stage II sample: 68.5 stage III sample: 53.7;

[0176] According to the preset activity level threshold, the evaluation score of stage I samples is above 80%, which is high activity; the evaluation score of stage II samples is between 50%-80%, which is medium activity; and the evaluation score of stage III samples is below 50%, which is low activity. This evaluation result is consistent with the trend of biochemical parameter changes of samples during freezing, indicating that the evaluation method proposed in the present invention can objectively and reliably predict the activity state of tumor samples during cryopreservation.

[0177] Figure 5 It is a comprehensive radar chart that evaluates the activity characteristics of samples in different stages from six dimensions. The score range for each dimension is 0-100 points. The figure shows that the scores of phase I samples in various dimensions are generally high, especially in temperature tolerance, which reaches 85 points; while the scores of phase III samples are relatively low, mostly between 50-60 points. The total activity evaluation score of each sample is also marked in the figure, which is completely consistent with the data in the example.

[0178] Through the analysis of the above three representative samples, it can be found that different stages of lung cancer show significant physiological differences under freezing conditions. The activity of stage I samples is higher, indicating that their tissue metabolism and tolerance are stronger and more suitable for long-term cryopreservation; while the activity of stage III samples is lower, so more cautious freezing operations are required. These evaluation results provide important basis for clinicians to select appropriate tumor samples for subsequent research and treatment.

[0179] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for evaluating the activity of tumor samples during cold storage, characterized in that: The following steps are involved: S10, collecting tumor samples, and continuously monitoring and recording temperature, pH value, oxygen content, carbon dioxide concentration, glucose concentration, and lactate concentration during the freezing process to form multi-parameter time series data; S20, preprocessing the multi-parameter time series data, eliminating noise and performing standardization processing; S30, sampling the multi-parameter time series data at preset time intervals to obtain parameter data at multiple moments, and constructing a parameter time series matrix; S40, decomposing and calculating the parameter time series matrix to obtain a steady-state parameter matrix and a change parameter matrix; S50, establishing a correlation matrix between the parameters in the steady-state parameter matrix, recorded as a steady-state correlation matrix, and establishing a correlation matrix between the parameters in the change parameter matrix, recorded as a change correlation matrix; S60, calculating a parameter variation coefficient matrix based on the steady-state correlation matrix and the variation correlation matrix, wherein the parameter variation coefficient matrix represents the characteristics of each parameter varying with time; S70, inputting the parameter variation coefficient matrix into a pre-trained activity evaluation model to obtain an activity evaluation score for the tumor sample; S80: Determine the cold storage quality grade of the tumor sample according to the tumor sample activity evaluation score and output it.

2. A method for evaluating the activity of a tumor sample during cold storage according to claim 1, characterized in that: The activity evaluation model adopts a BP neural network structure.

3. A method for evaluating the activity of a tumor sample during cold storage according to claim 2, characterized in that: The steps of constructing the training data set of the activity evaluation model include collecting cold storage process data of a large number of tumor samples, annotating and preprocessing them, and dividing them into a training set, a validation set, and a test set in a ratio of 8:1:

1.

4. A method for evaluating the activity of a tumor sample during cold storage according to claim 3, characterized in that: The steps of training the activity evaluation model include initializing model parameters, performing back propagation training using the Adam optimizer, adopting a cross entropy loss function, and using an early stopping method to prevent overfitting.

5. The method for evaluating the activity of tumor samples during cold storage according to claim 1, characterized in that: The formula of the parameter timing matrix is ​​as follows: Where X is the parameter time series matrix; ij represents the jth parameter value at the ith moment; m is the number of sampling moments; n is the number of monitored parameters.

6. A method for evaluating the activity of tumor samples during cold storage according to claim 5, characterized in that: The formula for decomposing and calculating the parameter timing matrix is ​​as follows: X = S + V + E; Where S is the steady-state parameter matrix; V is the variable parameter matrix; E is the error matrix; the formula of the steady-state parameter matrix is ​​as follows: In the formula, s ij represents the steady-state value of the jth parameter at the ith moment; the formula of the variable parameter matrix is ​​as follows: In the formula, v ij Indicates the change value of the jth parameter at the i-th moment.

7. A method for evaluating the activity of tumor samples during cold storage according to claim 6, characterized in that: The formula of the correlation matrix between the parameters in the steady-state parameter matrix is ​​as follows: In the formula, R S is the steady-state correlation matrix; r ij Represents the correlation coefficient between the i-th parameter and the j-th parameter, and the calculation formula is: in, and are the average values ​​of the i-th and j-th parameters respectively.

8. The method for evaluating the activity of tumor samples during cold storage according to claim 7, characterized in that: The formula of the correlation matrix between the parameters in the variation parameter matrix is ​​as follows: In the formula, R V is the change correlation matrix; r ij ′ represents the change correlation coefficient between the i-th parameter and the j-th parameter, and the calculation formula is: in, and are the mean values ​​of the changes of the i-th and j-th parameters respectively.

9. A method for evaluating the activity of a tumor sample during cold storage according to claim 8, characterized in that: The formula for calculating the parameter variation coefficient matrix is ​​as follows: Where C is the parameter variation coefficient matrix; c ij It represents the influence coefficient of the change of the i-th parameter on the j-th parameter. The calculation formula is: Among them, α, β, γ, δ are weight coefficients, is the partial derivative of parameter i with respect to parameter j, θ ij is the phase difference between parameter i and parameter j, where i is an imaginary unit.

10. The method for evaluating the activity of tumor samples during cold storage according to claim 1, characterized in that: The multi-parameter time series data is preprocessed, specifically including: first, using the box plot method to detect and delete outliers in the data; then, using the sliding window median filter method to remove high-frequency noise in the data; and finally performing normalization processing.

Images