Breast cancer liver metastasis risk dynamic monitoring and early warning system based on metabonomics

Abstract

The invention relates to the field of medical health, in particular to a breast cancer liver metastasis risk dynamic monitoring and early warning system based on metabonomics, which comprises a data acquisition and preprocessing module for acquiring venous blood samples at three time points of disease progression of a patient, measuring specific metabolite concentration and performing normalization processing; the dynamic feature extraction module calculates a gain index, namely the sum of the change rates of the indexes; the deep learning modeling module uses the first and second deep networks to train sample data in different time periods, and establishes a patient disease prediction model. The comprehensive evaluation module adopts an ROC curve and an AUC value to evaluate the prediction accuracy of the model; the risk early warning module generates a liver metastasis risk early warning signal according to the evaluation result; according to the system, the breast cancer liver metastasis risk is dynamically monitored through metabonomics data acquisition and analysis at multiple time points, the metabolic change trend in disease progression is reflected more accurately, and early warning of the liver metastasis risk is achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical health, and specifically to a metabolomics-based dynamic monitoring and early warning system for the risk of liver metastasis in breast cancer patients. In particular, the system utilizes metabolomics data from patients at different time points, through deep learning algorithms and data analysis techniques, to dynamically monitor and provide early warnings for the risk of liver metastasis in breast cancer patients. Background Art

[0002] Breast cancer is one of the most common malignancies in women, and liver metastasis is a significant factor affecting patient prognosis. Currently, the detection of breast cancer liver metastasis relies primarily on imaging tests such as ultrasound, CT, and MRI. However, these methods can only detect established liver metastases and lack the ability to predict the risk of liver metastasis early.

[0003] Metabolomics, a key branch of systems biology, studies the overall changes in small molecule metabolites within an organism, reflecting its functional state and metabolic profile. Previous studies have shown that tumor cells undergo metabolic reprogramming during metastasis, and these changes can be detected using metabolomics. However, existing metabolomics-based research has largely remained at the single-shot detection stage, lacking the ability to monitor and analyze dynamic changes in metabolites.

[0004] Traditional tumor metastasis risk assessment methods generally suffer from low predictive accuracy, short warning windows, and insufficient individualization. In particular, they lack dynamic monitoring of changes in a patient's metabolic profile during disease progression, making it difficult to detect early signs of liver metastasis risk. Therefore, there is an urgent need to develop a system that can dynamically monitor and provide early warning of breast cancer liver metastasis risk, thereby improving the accuracy of breast cancer liver metastasis prediction and providing more timely and accurate support for clinical treatment decisions. Summary of the Invention

[0005] The purpose of this invention is to provide a metabolomics-based dynamic monitoring and early warning system for the risk of breast cancer liver metastasis. By collecting metabolomics data at multiple time points during the patient's disease progression, a dynamic prediction model is constructed to achieve early prediction and early warning of the risk of breast cancer liver metastasis.

[0006] The present invention proposes a metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk, including:

[0007] A data acquisition and preprocessing module is used to collect venous blood samples from patients at three time points during the progression of the patient's disease, extract the supernatant, measure and obtain metabolic concentration data of dehydrolactic acid, acetone, lactic acid, malonyl galactose, and glucose, and normalize the metabolic concentration data of dehydrolactic acid, lactic acid, and malonate;

[0008] a dynamic feature extraction module connected to the data acquisition and preprocessing module, configured to receive the normalized metabolic concentration data and calculate a gain index, wherein the gain index is the sum of the change rates of the various indicators at three time points;

[0009] a deep learning modeling module, connected to the dynamic feature extraction module, configured to receive the gain index, train a first deep network using sample data before the first time point, at the first time point, and at the second time point, train a second deep network using sample data before the second time point, at the second time point, and at the third time point, and integrate the first deep network and the second deep network to establish a patient disease prediction model;

[0010] a comprehensive evaluation module, connected to the deep learning modeling module, configured to receive the output of the patient disease prediction model, perform a comprehensive evaluation of the prediction accuracy using the ROC curve, and determine the prediction accuracy of the model using the AUC value; and

[0011] The risk warning module is connected to the comprehensive evaluation module, and is used to receive the evaluation result of the comprehensive evaluation module and generate a liver metastasis risk warning signal based on the evaluation result.

[0012] Preferably, the data acquisition and preprocessing module is specifically used for:

[0013] At three time points, 1.5 ml of venous blood was drawn from the patient, centrifuged, and the supernatant was collected and frozen in a -80 degree Celsius refrigerator. The indicators were then centrally measured and unified.

[0014] Calculating the patient's gain index based on the dynamic changes of the indicators, using the ranking results of the gain index as a prognostic indicator of the patient's disease, and establishing a dynamic model;

[0015] The calculation formula of the gain index is: Gain = |(S-1m)-(S-2m)|+|(S-1m)-(S-2n)|+...+|(S-1n)-(S-2n)|, where Gain is the gain index, S represents the average value of each indicator, m represents the third time point, and n represents the second time point.

[0016] Preferably, the deep learning modeling module is specifically used to:

[0017] Using sample data before the first time point, at the first time point, and at the second time point, the first deep network is trained to model the first deep network;

[0018] Training the second deep network with the sample data before the second time point, at the second time point, and at the third time point to complete re-modeling of the second deep network;

[0019] The first deep network and the second deep network are then integrated separately to establish a patient disease prediction model to predict three outcomes of the patient's disease progression.

[0020] Preferably, the data acquisition and preprocessing module is specifically used for:

[0021] The metabolic concentration data of the lactic acid, the acetone, the dehydrolactic acid, and the malonate were normalized to eliminate the influence of the differences and proportions between different types of samples.

[0022] Preferably, the comprehensive evaluation module is specifically used for:

[0023] The change in the ROC curve area is used as an indicator to evaluate the prediction accuracy of the system.

[0024] Preferably, the system further comprises a feature screening module, which is connected to the data acquisition and preprocessing module and is used to:

[0025] The Spearman rank correlation method was used to calculate the correlation between the metabolomics indicators of each sample.

[0026] Preferably, the feature screening module is specifically used to:

[0027] The Spearman rank correlation coefficient was used to analyze the correlation between the metabolic concentration data of the dehydrolactic acid, the acetone, the lactic acid, the malonyl galactose, and the glucose and the disease condition;

[0028] Correlation calculation is performed using linear regression, and correlation analysis is performed on the metabolic concentration data of the dehydrolactic acid, the acetone, the lactic acid, the malonyl galactose, and the glucose using linear regression correlation analysis;

[0029] Indicators with correlation coefficients of linear regression correlation analysis less than 0.05 were screened as screening conditions.

[0030] Preferably, the feature screening module is specifically used to:

[0031] The screened metabolic concentration data are processed using principal component analysis (PCA) and a Relief algorithm to calculate metabolic concentration data characteristics of the dehydrolactic acid, the acetone, the lactic acid, the malonyl galactose, and the glucose, and the screened indicators are output.

[0032] Preferably, the feature screening module is specifically used to:

[0033] The obtained metabolic concentration data features of the dehydrolactic acid, the acetone, the lactic acid, the galactose malonate, and the glucose are screened and dimensionally reduced using the Relief algorithm to eliminate attribute features that are less relevant or irrelevant to the classification effect;

[0034] Using the Relief algorithm, randomly select samples from the classified data objects, find the sample with the largest and smallest similarity to each training sample, if the category of the training sample is the same as the category of the largest sample, then add 1 point to the training sample, otherwise subtract 1 point, after completing the operation with all training samples, the scores of the training samples are accumulated;

[0035] The samples with higher scores are screened out and used for model building.

[0036] According to the metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk, the deep learning modeling module and the risk early warning module are specifically used to:

[0037] Using the metabolic concentration data of dehydrolactic acid, lactic acid, galactose malonate, and glucose as input data, and using the metabolic concentration data of dehydrolactic acid, lactic acid, and malonate after principal component analysis (PCA) as output data, a deep neural network is trained to construct a dynamic patient disease prediction model;

[0038] By inputting the metabolic concentration data of the dehydrolactic acid, the lactic acid, the malonyl galactose and the glucose before the first time point, the first time point and the second time point, the prediction results before the first time point, the first time point and the second time point are obtained, and the patient's liver metastasis risk warning results are output.

[0039] The beneficial effects of the present invention include:

[0040] 1. By collecting and analyzing metabolomics data at multiple time points, it provides dynamic monitoring capabilities for the risk of breast cancer liver metastasis, more accurately reflecting metabolic trends during disease progression than single-time testing.

[0041] 2. An innovative gain index calculation method was proposed, which effectively quantified the degree of change in metabolite concentrations at different time points, providing a more reliable quantitative indicator for risk assessment;

[0042] 3. A two-stage deep neural network modeling strategy can more comprehensively capture the characteristics of different disease development stages and improve the accuracy and reliability of the prediction model;

[0043] 4. Comprehensively utilize multiple feature screening algorithms, such as Spearman rank correlation method, principal component analysis, and Relief algorithm, to effectively screen key metabolic indicators related to the risk of breast cancer liver metastasis and reduce the impact of data noise;

[0044] 5. The prediction evaluation system based on ROC curve and AUC value provides an objective and quantitative evaluation standard for the system's prediction performance;

[0045] 6. The present invention can issue an early warning signal 3-6 months before liver metastasis occurs, providing a sufficient time window for clinical intervention and significantly improving the patient's treatment prognosis and quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram of the overall structure of the metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk of the present invention;

[0047] Figure 2 This is a workflow diagram of the data acquisition and preprocessing module of the present invention;

[0048] Figure 3 Schematic diagram of gain index calculation of the present invention;

[0049] Figure 4 This is a diagram of the modeling structure of the two-stage deep neural network of the present invention;

[0050] Figure 5 This is a workflow diagram of the feature screening module of the present invention;

[0051] Figure 6 Schematic diagram of prediction evaluation based on ROC curve of the present invention;

[0052] Figure 7 This is a workflow diagram of the risk warning module of the present invention. DETAILED DESCRIPTION

[0053] Please refer to the attached Figure 1-7 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood by those skilled in the art that these specific embodiments are only used to illustrate the present invention and should not be construed as limiting the present invention.

[0054] like Figure 1 As shown, the metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk provided by the present invention includes a data acquisition and preprocessing module 1, a dynamic feature extraction module 2, a deep learning modeling module 3, a comprehensive evaluation module 4, a risk early warning module 5 and a feature screening module 6.

[0055] The data acquisition and preprocessing module 1 is responsible for collecting venous blood samples from patients at three time points during the progression of the patient's disease, extracting the supernatant, measuring and obtaining the metabolic concentration data of dehydrolactic acid, acetone, lactic acid, malonate galactose and glucose, and normalizing the metabolic concentration data of dehydrolactic acid, lactic acid and malonate.

[0056] In a preferred embodiment of the present invention, Figure 2 As shown, the data acquisition and preprocessing module 1 specifically performs the following steps:

[0057] First, 1.5 ml of venous blood is drawn from the patient at three key time points after the patient's breast cancer diagnosis (for example, at the time of diagnosis, in the middle of treatment, and in the late stage of treatment). The blood draw volume of 1.5 ml is based on clinical experience, which can not only obtain a sufficient sample volume for metabolomics analysis, but also minimize trauma to the patient. Preferably, the interval between the three time points is 3-6 months. This time interval is sufficient to observe significant changes in metabolite concentrations, but not too long to miss the best time for intervention.

[0058] Next, centrifuge the venous blood sample (preferably at 3000 rpm for 15 minutes), collect the supernatant, and freeze it at -80°C. Freezing at -80°C effectively maintains the stability of metabolites in the sample, preventing degradation and ensuring the accuracy of the measurement results.

[0059] Then, the metabolic concentration data of dehydrolactic acid, acetone, lactic acid, galactose malonate, and glucose in the sample are centrally measured. In one embodiment of the present invention, ultra-high performance liquid chromatography-quadrupole / electrostatic field orbitrap high-resolution mass spectrometry is used for the determination. This method has high sensitivity and high specificity and can accurately determine the concentrations of the above metabolites.

[0060] Finally, the measured metabolic concentration data are normalized to eliminate the influence of differences and proportions between different types of samples. The normalization process uses the following formula:

[0061]

[0062] Among them, x norm is the normalized metabolite concentration value, X is the original metabolite concentration value, X min is the minimum concentration of the metabolite in all samples, X max is the maximum concentration value of the metabolite in all samples. Through this processing, all metabolite concentration values ​​are mapped to the [0,1] interval, which is convenient for subsequent analysis and comparison.

[0063] The dynamic feature extraction module 2 is connected to the data acquisition and preprocessing module 1 and is used to receive the normalized metabolic concentration data and calculate the gain index, which is the sum of the change rates of each indicator at three time points.

[0064] like Figure 3 As shown, the present invention innovatively proposes the concept of gain index (Gain) to quantify the degree of change in metabolite concentration at different time points. The calculation formula of gain index is:

[0065] Gain=|(S-1m)-(S-2m)|+|(S-1m)-(S-2n)|+...+|(S-1n)-(S-2n)|,

[0066] Where Gain is the gain index, S represents the average of each indicator, m represents the third time point, and n represents the second time point. This formula quantifies the magnitude of changes in metabolites between different time points. The greater the change, the higher the gain index, indicating more significant fluctuations in the patient's metabolic state and a potentially higher risk of liver metastasis.

[0067] In practical applications, for example, for the metabolite dehydrolactic acid, if the average concentration at the second time point is 3.5 nmol / mL and the average concentration at the third time point is 5.2 nmol / mL, the gain between these two time points is |3.5 - 5.2| = 1.7. Similar gain values ​​are calculated for all metabolites between all time point pairs and summed to obtain an overall gain index.

[0068] Preferably, the ranking of the gain index is used as a prognostic indicator for the patient's disease, and a dynamic model is established. For example, if all patients are ranked from high to low by gain index, the top 20% of patients may face a higher risk of liver metastasis and require focused monitoring and intervention.

[0069] The deep learning modeling module 3 is connected to the dynamic feature extraction module 2 and is used to receive the gain index and adopt a two-stage deep neural network modeling strategy to build a patient disease prediction model.

[0070] like Figure 4 As shown, the present invention adopts an innovative two-stage deep neural network modeling strategy:

[0071] Phase 1: The first deep network is trained using sample data from before, at, and after the first and second time points to build a model. This phase captures the characteristics of early and mid-stage disease changes.

[0072] Phase 2: The second deep network is trained using sample data from before, at, and after the second and third time points, completing the remodeling of the second deep network. This phase captures the changing characteristics of the disease in the middle and late stages.

[0073] Finally, the first deep network and the second deep network are integrated separately to establish a patient disease prediction model to predict three outcomes of the patient's disease progression (e.g., low risk, medium risk, and high risk).

[0074] In a preferred embodiment of the present invention, the deep neural network employs a multi-layer perceptron (MLP) structure, comprising an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer is equal to the number of metabolite indicators, the hidden layers use the ReLU activation function, and the output layer uses the Softmax function for multi-classification prediction. The network parameters are preferably set as follows:

[0075] Number of hidden layers: 3;

[0076] Number of nodes per layer: 64 nodes in the first layer, 32 nodes in the second layer, and 16 nodes in the third layer;

[0077] Learning rate: 0.001;

[0078] Batch size: 32;

[0079] Training epochs: 200

[0080] These parameter settings are based on the results of a large number of experimental verifications and can avoid overfitting problems while ensuring model accuracy.

[0081] The integration of two deep networks adopts the weighted average method, namely:

[0082] P final =α×P network1 +(1-α)×P network2 ,

[0083] Among them, P final is the final prediction result, P network1 is the prediction result of the first network, P network2 is the prediction result of the second network, and α is a weight parameter, usually set to 0.5, indicating that the prediction results of the two networks have equal importance. In practical applications, the value of α can be adjusted based on the performance of the two networks, so the network with better performance can be given a higher weight.

[0084] The comprehensive evaluation module 4 is connected to the deep learning modeling module 3, and is used to receive the output results of the patient disease prediction model, use the ROC curve to comprehensively evaluate the prediction accuracy, and use the size of the AUC value to judge the prediction accuracy of the model.

[0085] like Figure 6 As shown, the present invention uses the receiver operating characteristic (ROC) curve and the area under the curve (AUC) value to evaluate the performance of the prediction model. The ROC curve is a curve plotted with the true positive rate as the vertical axis and the false positive rate as the horizontal axis. The AUC value is the area under the ROC curve, and its value is between 0.5 and 1.

[0086] The closer the AUC value is to 1, the higher the system's prediction accuracy is; when the AUC value is 1, it means that the system can completely and correctly distinguish patients from normal people; when the AUC value is 0.5, it means that the system makes random predictions, and its ability to distinguish patients from normal people is the same as random guessing.

[0087] In one embodiment of the present invention, the comprehensive evaluation module 4 also uses the change in the area under the ROC curve as an indicator for evaluating the system's prediction accuracy. Specifically, the difference between the AUC values ​​of the models at different stages is calculated. If the difference is positive and large, it indicates a significant improvement in prediction performance. If the difference is close to zero or negative, it indicates that the prediction performance has not improved or has declined, and the model parameters or structure need to be adjusted.

[0088] Preferably, when the AUC value is greater than 0.85, the model is considered to have high predictive accuracy and can be used for clinical applications; when the AUC value is between 0.75-0.85, the model has moderate predictive accuracy and can be used under the guidance of clinicians; when the AUC value is lower than 0.75, the model predictive accuracy is insufficient and needs further optimization.

[0089] The risk warning module 5 is connected to the comprehensive evaluation module 4 and is used to receive the evaluation result of the comprehensive evaluation module and generate a liver metastasis risk warning signal based on the evaluation result.

[0090] like Figure 7 As shown, the risk warning module 5 of the present invention divides patients into three levels of low risk, medium risk and high risk according to the prediction results, and generates warning signals of different levels accordingly.

[0091] Preferably, the warning classification standards are as follows:

[0092] Low risk: predicted probability <30%, generating a green warning signal and recommending regular follow-up;

[0093] Medium risk: The predicted probability is between 30% and 70%, generating a yellow warning signal and recommending increased follow-up frequency and consideration of preventive interventions;

[0094] High risk: Prediction probability >70%, generating a red warning signal and recommending immediate intervention.

[0095] The setting of these thresholds is based on a large amount of clinical data analysis and expert consensus, which can ensure sensitivity while avoiding unnecessary intervention caused by excessive warnings.

[0096] In one embodiment of the present invention, the risk warning module 5 can also output a specific risk assessment report, including information such as the changing trend of each metabolic indicator, the overall risk score, and the predicted time window for liver metastasis, to provide a reference for clinicians to formulate individualized treatment plans.

[0097] like Figure 5 As shown, the present invention further includes a feature screening module 6, which is connected to the data acquisition and preprocessing module 1 and is used to perform correlation calculation on the metabolomics indicators of each sample using the Spearman rank correlation method.

[0098] In a preferred embodiment of the present invention, the feature screening module 6 specifically performs the following steps:

[0099] First, the Spearman rank correlation coefficient was used to analyze the correlation between the metabolic concentration data of dehydrolactic acid, acetone, lactic acid, malonyl galactose, and glucose and the disease condition. The calculation formula of the Spearman rank correlation coefficient is:

[0100]

[0101] Where ρ is the Spearman rank correlation coefficient, d i is the rank difference between the two variables in the ith sample, and n is the number of samples. The Spearman rank correlation coefficient does not require the data to follow a normal distribution and is applicable to various types of data analysis.

[0102] Secondly, linear regression was used to calculate the correlation, and linear regression correlation analysis was used to analyze the correlation of the metabolic concentration data of dehydrolactic acid, acetone, lactic acid, malonyl galactose and glucose. The basic formula of linear regression correlation analysis is:

[0103] y=β0+β1x+ε,

[0104] Among them, y is the dependent variable (disease status), x is the independent variable (metabolite concentration), β0 is the intercept, β1 is the slope, and ε is the random error term.

[0105] Finally, indicators with a linear regression correlation coefficient less than 0.05 were selected as screening criteria. A correlation coefficient less than 0.05 indicates that the metabolite is strongly correlated with the disease and may be an effective indicator for predicting liver metastasis risk.

[0106] In a further embodiment of the present invention, the feature screening module 6 also uses principal component analysis (PCA) and Relief algorithm to process the screened metabolic concentration data, calculates the metabolic concentration data characteristics of dehydrolactic acid, acetone, lactic acid, malonyl galactose and glucose, and outputs the screened indicators.

[0107] Principal component analysis (PCA) is a commonly used dimensionality reduction method. Its basic idea is to transform the original variables into a set of new uncorrelated variables (principal components), which are linear combinations of the original variables. The mathematical expression of PCA is:

[0108] Z=X×W,

[0109] Where Z is the principal component matrix, X is the original data matrix, and W is the eigenvector matrix. PCA can be used to convert high-dimensional metabolic data into a low-dimensional representation, preserving the most important information and reducing computational complexity.

[0110] The Relief algorithm is a feature weight calculation method used to evaluate the discriminative ability of features. Its basic process is as follows:

[0111] 1. Randomly select samples from the classified data objects;

[0112] 2. Find the sample with the largest similarity (nearest neighbor) and the sample with the smallest similarity (farthest neighbor) to each training sample;

[0113] 3. If the category of the training sample is the same as the category of the maximum similarity sample, then the training sample is given an extra point, otherwise it is subtracted by 1 point;

[0114] 4. After completing the operation with all training samples, the scores of the training samples are accumulated;

[0115] 5. Samples with higher scores are screened out for model building.

[0116] In practical applications, the similarity calculation of the Relief algorithm usually uses the Euclidean distance, which is defined as:

[0117]

[0118] Among them, d(x,y) is the distance between samples x and y, x i and y i are the i-th eigenvalues ​​of samples x and y respectively, and n is the number of features.

[0119] Preferably, the feature screening threshold is set as follows:

[0120] Spearman rank correlation coefficient: p value less than 0.05;

[0121] Linear regression correlation coefficient: p value less than 0.05;

[0122] PCA: retain the principal components whose explained variance rate reaches more than 85%;

[0123] Relief algorithm: features with weight greater than 0.3;

[0124] Through this multi-level feature screening strategy, key metabolic indicators related to breast cancer liver metastasis can be effectively identified, improving the accuracy and interpretability of the prediction model.

[0125] In a preferred embodiment of the present invention, the deep learning modeling module 3 and the risk warning module 5 work together to complete the entire process from metabolic data analysis to risk warning:

[0126] The deep learning modeling module 3 takes the metabolic concentration data of dehydrolactic acid, lactic acid, malonate galactose and glucose as input data, and the metabolic concentration data of dehydrolactic acid, lactic acid and malonate after principal component analysis PCA as output data, trains the deep neural network, and constructs a dynamic patient disease prediction model.

[0127] Then, by inputting the metabolic concentration data of dehydrolactic acid, lactic acid, malonate galactose and glucose before the first time point, at the first time point and at the second time point, the deep learning modeling module 3 obtains the prediction results before the first time point, at the first time point and at the second time point.

[0128] Finally, based on these prediction results, the risk warning module 5 outputs the patient's liver metastasis risk warning results to provide decision support for clinicians.

[0129] Through the collaborative work of the deep learning modeling module 3 and the risk warning module 5, the present invention can achieve accurate prediction and early warning of the risk of breast cancer liver metastasis, and provide patients with personalized risk assessment and intervention recommendations.

[0130] In summary, the metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk provided by the present invention, through the collection and analysis of metabolomics data at multiple time points, innovatively proposes a gain index calculation method, and adopts a two-stage deep neural network modeling strategy and a multi-level feature screening mechanism to achieve dynamic monitoring and early warning of breast cancer liver metastasis risk, providing important support for the precise treatment and prognosis management of breast cancer patients.

[0131] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk, characterized by: include: A data acquisition and preprocessing module is used to collect venous blood samples from patients at three time points during the progression of the patient's disease, extract the supernatant, measure and obtain metabolic concentration data of dehydrolactic acid, acetone, lactic acid, malonyl galactose, and glucose, and normalize the metabolic concentration data of dehydrolactic acid, lactic acid, and malonate; a dynamic feature extraction module connected to the data acquisition and preprocessing module, configured to receive the normalized metabolic concentration data and calculate a gain index, wherein the gain index is the sum of the change rates of the various indicators at three time points; a deep learning modeling module, connected to the dynamic feature extraction module, configured to receive the gain index, train a first deep network using sample data before the first time point, at the first time point, and at the second time point, train a second deep network using sample data before the second time point, at the second time point, and at the third time point, and integrate the first deep network and the second deep network to establish a patient disease prediction model; A comprehensive evaluation module, connected to the deep learning modeling module, is used to receive the output results of the patient disease prediction model, use the ROC curve to comprehensively evaluate the prediction accuracy, and use the AUC value to judge the prediction accuracy of the model; as well as The risk warning module is connected to the comprehensive evaluation module, and is used to receive the evaluation result of the comprehensive evaluation module and generate a liver metastasis risk warning signal based on the evaluation result.

2. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 1, characterized in that: The data acquisition and preprocessing module is specifically used for: At three time points, 1.5 ml of venous blood was drawn from the patient, centrifuged, and the supernatant was collected and frozen in a -80 degree Celsius refrigerator. The indicators were then centrally measured and unified. Calculating the patient's gain index based on the dynamic changes of the indicators, using the ranking results of the gain index as a prognostic indicator of the patient's disease, and establishing a dynamic model; The calculation formula of the gain index is: Gain = |(S-1m)-(S-2m)|+|(S-1m)-(S-2n)|+...+|(S-1n)-(S-2n)|, where Gain is the gain index, S represents the average value of each indicator, m represents the third time point, and n represents the second time point.

3. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 1, characterized in that: The deep learning modeling module is specifically used to: Using sample data before the first time point, at the first time point, and at the second time point, the first deep network is trained to model the first deep network; Training the second deep network with the sample data before the second time point, at the second time point, and at the third time point to complete re-modeling of the second deep network; The first deep network and the second deep network are then integrated separately to establish a patient disease prediction model to predict three outcomes of the patient's disease progression.

4. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 1, characterized in that: The data acquisition and preprocessing module is specifically used for: The metabolic concentration data of the lactic acid, the acetone, the dehydrolactic acid, and the malonate were normalized to eliminate the influence of the differences and proportions between different types of samples.

5. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 1, characterized in that: The comprehensive evaluation module is specifically used for: The change in the ROC curve area is used as an indicator to evaluate the prediction accuracy of the system.

6. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 1, characterized in that: The system further includes a feature screening module, which is connected to the data acquisition and preprocessing module and is used to: The Spearman rank correlation method was used to calculate the correlation between the metabolomics indicators of each sample.

7. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 6, characterized in that: The feature screening module is specifically used for: The Spearman rank correlation coefficient was used to analyze the correlation between the metabolic concentration data of the dehydrolactic acid, the acetone, the lactic acid, the malonyl galactose, and the glucose and the disease condition; Correlation calculation is performed using linear regression, and correlation analysis is performed on the metabolic concentration data of the dehydrolactic acid, the acetone, the lactic acid, the malonyl galactose, and the glucose using linear regression correlation analysis; Indicators with correlation coefficients of linear regression correlation analysis less than 0.05 were screened as screening conditions.

8. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 7, characterized in that: The feature screening module is specifically used for: The screened metabolic concentration data are processed using principal component analysis (PCA) and a Relief algorithm to calculate metabolic concentration data characteristics of the dehydrolactic acid, the acetone, the lactic acid, the malonyl galactose, and the glucose, and the screened indicators are output.

9. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 8, characterized in that: The feature screening module is specifically used for: The obtained metabolic concentration data features of the dehydrolactic acid, the acetone, the lactic acid, the galactose malonate, and the glucose are screened and dimensionally reduced using the Relief algorithm to eliminate attribute features that are less relevant or irrelevant to the classification effect; Using the Relief algorithm, randomly select samples from the classified data objects, find the sample with the largest and smallest similarity to each training sample, if the category of the training sample is the same as the category of the largest sample, then add 1 point to the training sample, otherwise subtract 1 point, after completing the operation with all training samples, the scores of the training samples are accumulated; The samples with higher scores are screened out and used for model building.

10. The metabolomics-based dynamic monitoring and early warning system for breast cancer liver metastasis risk according to claim 1, characterized in that: The deep learning modeling module and the risk warning module are specifically used to: Using the metabolic concentration data of dehydrolactic acid, lactic acid, galactose malonate, and glucose as input data, and using the metabolic concentration data of dehydrolactic acid, lactic acid, and malonate after principal component analysis (PCA) as output data, a deep neural network is trained to construct a dynamic patient disease prediction model; By inputting the metabolic concentration data of the dehydrolactic acid, the lactic acid, the malonyl galactose and the glucose before the first time point, the first time point and the second time point, the prediction results before the first time point, the first time point and the second time point are obtained, and the patient's liver metastasis risk warning results are output.

Images