A method and system for constructing a prediction model for prostate neuroendocrine cancer
By dividing time periods and extracting features of prostate neuroendocrine cancer data, calculating the comprehensive drift coefficient and combining gradient standard threshold analysis, the data drift problem is solved, improving the robustness and prediction accuracy of the model, ensuring the changing data environment Provide reliable forecasts in the
Patent Information
- Application Number
- CN202411288068.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-09-14
AI Technical Summary
The existing prostate neuroendocrine cancer prediction model has data drift problems due to the data spanning multiple years, resulting in significant differences between early and recent data, affecting the accuracy of the model in the current and future patient populations.
By dividing time periods, extracting feature and calculating comprehensive drift coefficients for prostate neuroendocrine cancer-related data, combining gradient standard threshold analysis, identifying the degree of data drift and formulating corresponding model construction and adjustment strategies to ensure that the model is in a constantly changing clinical data environment Maintain a high level of prediction accuracy and stability.
It significantly improves the robustness of the NEPC prediction model, ensuring that the model can provide reliable predictive results in a changing clinical data environment, supporting clinical decision-making and the development of personalized treatment options.
Smart Images

Figure CN118800465B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical information processing technology, and in particular to a method and system for constructing a prostate neuroendocrine cancer prediction model. Background Art
[0002] The prostate neuroendocrine carcinoma prediction model is a tool or algorithm used to assess a patient's risk of developing prostate neuroendocrine carcinoma (NEPC). NEPC is a rare and highly malignant subtype of prostate cancer that usually occurs in patients with castration-resistant prostate cancer (CRPC). The prediction model may combine multiple factors, including the patient's clinical data, gene expression characteristics, imaging results, etc., to predict the likelihood of a patient developing NEPC through calculation or analysis. This model is designed to help doctors make early diagnoses and personalized treatment decisions to improve patient outcomes.
[0003] In the existing technology, the prediction model of prostate neuroendocrine carcinoma (NEPC) is usually constructed by collecting and analyzing a large amount of patient data. These data may include gene expression profiles, protein markers, clinical characteristics (such as age, medical history, treatment response), and imaging data. Researchers use machine learning or statistical analysis methods to extract key features related to the occurrence of NEPC and combine these features to train the prediction model. After verification and optimization, the final model can be used to predict the risk of new patients developing NEPC, thereby assisting clinical decision-making.
[0004] The shortcomings of the prior art:
[0005] Because NEPC has a long development period, sample data may span multiple decades. As time goes by, diagnostic techniques and treatment options for prostate cancer continue to improve, which may lead to significant differences between early data and recent data, the so-called data drift. And if these differences are not identified and corrected in time during model training, the model may make predictions based on drifted data, thereby reducing its accuracy in current and future patient populations. Summary of the invention
[0006] The purpose of the present invention is to provide a method and system for constructing a prostate neuroendocrine cancer prediction model to address the deficiencies in the background technology.
[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for constructing a prediction model for prostate neuroendocrine cancer, comprising:
[0008] Collect and summarize data related to prostate neuroendocrine cancer for the construction of a prediction model, including clinical characteristics, gene expression, and imaging data;
[0009] The collected prostate neuroendocrine carcinoma related data are divided into several time periods, and the prostate neuroendocrine carcinoma related data in each time period are preprocessed respectively;
[0010] Feature extraction was performed on the pre-processed prostate neuroendocrine carcinoma-related data, and the rare gene mutation frequency fluctuation characteristics and abnormal immune abundance distribution change characteristics of the prostate neuroendocrine carcinoma-related data were extracted respectively;
[0011] According to the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation;
[0012] The calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data was compared and analyzed with the gradient standard threshold. According to the analysis results, the prostate neuroendocrine carcinoma related data were divided into different data drift levels, and corresponding model construction adjustment strategies were formulated.
[0013] Preferably, the collected prostate neuroendocrine carcinoma related data is divided into several time periods, including:
[0014] The data related to prostate neuroendocrine carcinoma are sorted in chronological order to form a time series data set X={x1,x2,…,xN}, where xi is the data at the i-th time point and N is the total number of data points. The mean and standard deviation of the time series are calculated and used as the basis for time series segmentation.
[0015] The time series X is divided into K intervals, each interval Sj represents a subsequence Xj={xsj,…,xej}, where sj and ej are the starting and ending positions of the jth interval, respectively. The internal changes of interval Sj are measured by the loss function C(Sj), and the loss function is: ;in, Is the interval The goal is to minimize the total loss function of the entire time series through dynamic programming , the calculation expression is: ; The task of dynamic programming is to find the change point that minimizes the total loss ; Create an array DP to store the minimum loss value of the first i data points, initialize it to DP[0]=0, and the rest are infinite, define an array P to store the best segmentation point position for each segment, for each time point i, calculate the segmentation cost between each possible previous time point j, the expression is: ; Store the corresponding best segmentation point j: P[i]=j; Find the optimal segmentation point sequence by backtracking the P array, initialize tK=N, and then backtrack: ; Continue backtracking until all The final segmentation point sequence , which is the node position where the data features change significantly. According to the identified change nodes, the data set is divided into multiple time periods, and the corresponding time period label is added to each data point.
[0016] Preferably, after respectively extracting the rare gene mutation frequency fluctuation characteristics and the abnormal change characteristics of immune abundance distribution of the prostate neuroendocrine carcinoma related data, the method further comprises generating a mutation frequency fluctuation index according to the rare gene mutation frequency fluctuation characteristics extracted from the prostate neuroendocrine carcinoma related data, and the method for obtaining the mutation frequency fluctuation index is:
[0017] The gene mutation frequency data related to prostate neuroendocrine carcinoma are sorted in chronological order to form a time series x(t), where t is time and x(t) is the mutation frequency at time t. The data are standardized and recorded as , the expression is: ;in, is the mean of the mutation frequency data, is its standard deviation, and the standardized time series is transformed using fast Fourier transform Convert from time domain to frequency domain to get spectrum , the expression is: ;in, N is the length of the time series, f is the frequency, is the kernel function of Fourier transform, and the spectrum amplitude of gene mutation frequency data is calculated. Represents the intensity of the signal at each frequency component, and the calculation expression is: ;in, yes X(f) The real part of yes X(f) The imaginary part of The amplitude distribution of the spectrum is obtained, the main frequency of the amplitude is obtained, and the total energy E in the spectrum is calculated. The expression is: ; The mutation frequency fluctuation index is calculated based on the total energy E in the calculated spectrum, and the expression is: ; In the formula, is the mutation frequency fluctuation index, Represents the total energy of the selected main frequency components.
[0018] Preferably, after respectively extracting the rare gene mutation frequency fluctuation characteristics and the abnormal change characteristics of immune abundance distribution of the prostate neuroendocrine carcinoma related data, the method further comprises generating an abnormal immune abundance distribution index according to the extracted abnormal change characteristics of immune abundance distribution of the prostate neuroendocrine carcinoma related data, and the method for obtaining the abnormal immune abundance distribution index is:
[0019] Collect immune-related data from patients with prostate neuroendocrine carcinoma, including the abundance of different immune cell types; standardize the immune abundance data and calculate the mean abundance distribution of each immune cell type in patient samples and standard deviation ; For each sample, the immune abundance value Test to determine whether it deviates from the normal distribution and determine the abnormal immune abundance value , the expression is: or ; Where k is a constant; Count the abnormal immune abundance values in each sample Number of , and the degree of deviation of outliers , the expression is: ; s is the label of the immune cell category, and the immune abundance distribution abnormality index is calculated, and the expression is: ; Where M is the total number of immune cell types, is the number of abnormal immune abundance values of the i-th type of immune cells, is the total number of samples of the i-th type of immune cells, is the sum of the deviations of the abnormal immune abundance values of the i-th type of immune cells, is the total deviation degree of the i-th type of immune cells, It is the abnormal index of immune abundance distribution.
[0020] Preferably, the weight assignment of the prostate neuroendocrine carcinoma related data in each time period is determined based on the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, and the comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is obtained after weighted average calculation, including:
[0021] The mutation frequency fluctuation index and the immune abundance distribution abnormality index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses each group of first eigenvectors to predict the weighted assignment labels of prostate neuroendocrine carcinoma related data in each time period as the prediction target, and takes minimizing the sum of prediction errors of the weighted assignment labels of prostate neuroendocrine carcinoma related data in all time periods as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined according to the model output results, wherein the machine learning model is a polynomial regression model, and the comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is obtained by weighted average calculation of the weight assignment of the prostate neuroendocrine carcinoma related data in each time period.
[0022] Preferably, the calculated comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is compared and analyzed with the gradient standard threshold, and the prostate neuroendocrine carcinoma related data is divided into different data drift levels according to the analysis results, and a corresponding model construction adjustment strategy is formulated, including:
[0023] The calculated comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the comprehensive drift coefficient is compared with the first standard threshold and the second standard threshold respectively;
[0024] If the comprehensive drift coefficient is greater than the second standard threshold, the prostate neuroendocrine cancer related data is classified as a high data drift level, and a data abnormality signal is generated; the data needs to be re-analyzed and modeled;
[0025] If the comprehensive drift coefficient is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the prostate neuroendocrine carcinoma related data is classified into a drift level in the data, and a suspected abnormal signal of the data is generated; and the parameters of the existing model are adjusted;
[0026] If the comprehensive drift coefficient is less than the first standard threshold, the prostate neuroendocrine carcinoma related data is classified as a low data drift level, and a data normal signal is generated. The existing model remains unchanged and the prediction continues.
[0027] The present invention also provides a system for constructing a prostate neuroendocrine cancer prediction model, comprising a data acquisition module, a data partitioning module, a feature extraction module, a comprehensive drift coefficient calculation module and a model adjustment module;
[0028] Data acquisition module: collects and summarizes prostate neuroendocrine cancer related data for building a prediction model, including clinical characteristics, gene expression and imaging data;
[0029] Data division module: divide the collected prostate neuroendocrine carcinoma related data into several time periods, and pre-process the prostate neuroendocrine carcinoma related data in each time period;
[0030] Feature extraction module: extracts features from the preprocessed prostate neuroendocrine carcinoma-related data, and extracts the rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics of the prostate neuroendocrine carcinoma-related data;
[0031] Comprehensive drift coefficient calculation module: According to the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation;
[0032] Model adjustment module: The calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is compared and analyzed with the gradient standard threshold, and the prostate neuroendocrine carcinoma related data is divided into different data drift levels according to the analysis results, and the corresponding model construction adjustment strategy is formulated.
[0033] In the above technical solution, the technical effects and advantages provided by the present invention are:
[0034] Through systematic steps and algorithms, the present invention effectively solves the data drift problem that the prostate neuroendocrine carcinoma prediction model may face when the data spans multiple decades. By dividing the data into time periods, extracting features, and calculating the comprehensive drift coefficient, and analyzing it in combination with the gradient standard threshold, it is possible to accurately identify the degree of data drift and formulate corresponding model adjustment strategies for different drift levels. This process ensures that the model maintains a high level of prediction accuracy and stability in a constantly changing clinical data environment.
[0035] The present invention uses a flexible model adjustment strategy to adopt corresponding model reconstruction, fine-tuning or maintenance strategies according to different levels of data drift, thereby achieving a balance between model performance and clinical practicality. Overall, this method can significantly improve the robustness of the NEPC prediction model, ensuring that the model can still provide reliable prediction results in a constantly changing clinical data environment, and ultimately provide strong support for clinical decision-making and the formulation of personalized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0037] Figure 1 The present invention is a flow chart of the method.
[0038] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION
[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] Example 1, please refer to Figure 1 As shown, the method for constructing a prostate neuroendocrine cancer prediction model described in this embodiment includes the following steps:
[0041] S1: Collect and summarize relevant data on prostate neuroendocrine cancer for the construction of a prediction model, including clinical characteristics, gene expression, and imaging data;
[0042] S2: dividing the collected prostate neuroendocrine carcinoma related data into several time periods, and preprocessing the prostate neuroendocrine carcinoma related data in each time period respectively;
[0043] S3: Feature extraction is performed on the preprocessed prostate neuroendocrine carcinoma related data, and the rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics of the prostate neuroendocrine carcinoma related data are extracted respectively;
[0044] S4: According to the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation;
[0045] S5: Compare and analyze the calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data with the gradient standard threshold, divide the prostate neuroendocrine carcinoma related data into different data drift levels according to the analysis results, and formulate corresponding model construction adjustment strategies.
[0046] In S1, the prostate neuroendocrine carcinoma related data for building a prediction model are collected and summarized, and the related data include clinical characteristics, gene expression and imaging data, specifically:
[0047] Determine the research objectives: Identify the specific types of data required for model prediction (e.g., early diagnosis of NEPC, prediction of progression, etc.).
[0048] Define data types: List the specific types of data that need to be collected, including clinical characteristics, gene expression data, imaging data, etc.
[0049] Data source identification and confirmation: Collaborate with participating hospitals and clinics to obtain clinical data and diagnostic records of patients. Access public or proprietary genomic databases (such as TCGA, GEO) to obtain gene expression profile data of NEPC patients. Obtain MRI, CT, PET and other imaging data from medical imaging databases (such as PACS systems, research imaging libraries). Extract relevant information from published research literature and clinical trial data.
[0050] Clinical characteristics data collection: Collect basic information of patients, such as age, gender, race, etc. Collect diagnostic information, including prostate cancer type, stage, grade, etc. Collect treatment information, such as surgical records, radiotherapy and chemotherapy regimens, drug treatment details, etc. Collect follow-up data, including recurrence time, metastasis, survival rate, etc.
[0051] Gene expression data collection: Extract tumor samples from patients and perform genome sequencing or expression profiling. Download or access NEPC gene expression data from public gene databases and associate them with patient clinical information. Ensure data standardization, such as normalization of expression levels and correction of batch effects.
[0052] Imaging data collection: Obtain the patient's imaging examination records, including MRI, CT, PET and other imaging data. Ensure that the DICOM format of the imaging data is complete and obtain relevant metadata, such as scan time, device type, image resolution, etc. Match the imaging data with the patient's clinical records and genetic data for multimodal analysis.
[0053] All collected data should be deduplicated to ensure that each patient's record is unique. For missing data, appropriate imputation strategies should be applied (such as mean imputation, interpolation, or estimation through similar samples). Data formats should be standardized to ensure that all data types have consistent coding methods (such as ICD coding, gene name standards, etc.). Gene expression data should be standardized, such as log2 transformation or Z-score normalization, to eliminate systematic biases between different experiments. Clinical characteristics should be categorized or normalized to facilitate integration with other types of data.
[0054] Integrate clinical features, gene expression, and imaging data into a unified dataset. Ensure that each patient has a corresponding record in all data types. Create a unique patient ID to ensure consistency and traceability between all data types.
[0055] S2: Divide the collected prostate neuroendocrine carcinoma related data into several time periods, and pre-process the prostate neuroendocrine carcinoma related data in each time period respectively.
[0056] The data related to prostate neuroendocrine carcinoma are sorted in chronological order to form a time series data set X={x1,x2,…,xN}, where xi is the data at the i-th time point and N is the total number of data points. The mean and standard deviation of the time series are calculated and used as the basis for time series segmentation. The time series X is divided into K intervals, each interval Sj represents a subsequence Xj={xsj,…,xej}, where sj and ej are the starting and ending positions of the j-th interval, respectively. In order to measure the stability of the data in each interval, a loss function C(Sj) is defined to measure the internal changes of the interval Sj (such as changes in the mean or variance). The loss function is: ;in, Is the interval The goal is to minimize the total loss function of the entire time series through dynamic programming , the calculation expression is: ; The task of dynamic programming is to find the change point that minimizes the total loss ; Create an array DP to store the minimum loss value of the first i data points, initialize it to DP[0]=0, and the rest are infinite, define an array P to store the best segmentation point position for each segment, for each time point i (from 1 to N), calculate the segmentation cost between each possible previous time point j, the expression is: ; Store the corresponding best segmentation point j: P[i]=j; Find the optimal segmentation point sequence by backtracking the P array, initialize tK=N, and then backtrack: ; Continue backtracking until all The final segmentation point sequence , which is the node position where the data features change significantly. According to the identified change nodes, the data set is divided into multiple time periods, and the corresponding time period label is added to each data point.
[0057] Check for missing data in each time period and use appropriate methods (such as mean imputation, interpolation, or nearest neighbor imputation) to handle missing values.
[0058] Outlier detection and processing: In each time period, outliers are detected and processed using statistical analysis or machine learning algorithms (such as isolation forest and Z-score) to ensure the authenticity and consistency of the data.
[0059] Data normalization: Standardize numerical data (such as gene expression values, imaging measurements) in different time periods to eliminate scale differences between time periods. Common methods include Z-score normalization or min-max normalization.
[0060] Feature scaling: For features that may change over time (such as improvements in image resolution and detection methods), they are normalized and adjusted to make data in different time periods comparable.
[0061] Correction for batch effects: Since there may be differences in technology and experimental conditions in different time periods, batch effect correction methods (such as Combat and SVA) are used to adjust gene expression data to eliminate data bias caused by differences in experimental batches.
[0062] Homogenization processing: Homogenize data from different time periods to ensure that the source and processing methods of data are consistent between different time periods.
[0063] S3: Feature extraction is performed on the preprocessed prostate neuroendocrine carcinoma related data, and the rare gene mutation frequency fluctuation characteristics and abnormal immune abundance distribution change characteristics of the prostate neuroendocrine carcinoma related data are extracted respectively.
[0064] Classify the data into clinical feature data, gene expression data, and imaging data;
[0065] Clinical characteristic data: such as patient age, gender, tumor stage, pathological grade, PSA (prostate-specific antigen) level, etc.
[0066] Gene expression data: including RNA sequencing data, DNA mutation information, gene methylation status, etc.
[0067] Imaging data: such as MRI, CT, PET imaging features, tumor volume, morphology, signal intensity, etc.
[0068] Extract basic clinical features, such as age, tumor stage, grade, PSA level, etc. Combine multiple clinical features to create new synthetic features. For example, combine PSA level and tumor stage to create a new tumor burden index.
[0069] Gene expression data feature extraction: Statistical methods (such as t-test, ANOVA) were used to screen genes with significant expression differences between NEPC patients and non-NEPC patients. The mutation frequencies of NEPC-related genes were counted to extract rare but highly predictive mutant genes.
[0070] Gene pathway analysis: Pathway enrichment analysis was performed through KEGG, GO and other databases to extract gene pathway characteristics related to NEPC.
[0071] Morphological feature extraction: Extract the morphological features of the tumor in the image, such as tumor volume, edge clarity, shape irregularity, etc.
[0072] Texture analysis: Texture analysis algorithms (such as GLCM, gray-level co-occurrence matrix) are used to extract texture features in images to assess the internal heterogeneity of tumors.
[0073] Dynamic contrast enhancement features: Extract the enhancement pattern of the tumor in dynamic contrast-enhanced images such as DCE-MRI, such as the enhancement rate and residual intensity after enhancement.
[0074] The mutation frequency fluctuation index is generated based on the rare gene mutation frequency fluctuation characteristics extracted from the prostate neuroendocrine carcinoma related data. The method for obtaining the mutation frequency fluctuation index is as follows:
[0075] The gene mutation frequency data related to prostate neuroendocrine carcinoma are sorted in chronological order to form a time series x(t), where t is time and x(t) is the mutation frequency at time t. The data are standardized and recorded as Convert from time domain to frequency domain to get spectrum , the expression is: ;in, N is the length of the time series, f is the frequency, is the kernel function of Fourier transform, and the spectrum amplitude of gene mutation frequency data is calculated. Represents the intensity of the signal at each frequency component, and the calculation expression is: ;in, yes X(f) The real part of yes X(f) The imaginary part of The amplitude distribution of the gene is obtained, and the main frequency of the amplitude is obtained. The corresponding fluctuation component reflects the periodicity or oscillation characteristics of the gene mutation frequency. The total energy E in the spectrum is calculated to evaluate the volatility of the entire time series. The expression is: ; E reflects the fluctuation intensity of the entire time series. The mutation frequency fluctuation index is calculated based on the total energy E in the calculated spectrum. The expression is: ; In the formula, is the mutation frequency fluctuation index, It represents the total energy of the selected main frequency components. The higher the MF value, the more drastic the fluctuation of the gene mutation frequency at a specific frequency, which may reflect the periodic changes of mutations or the influence of some external factors.
[0076] The larger the mutation frequency fluctuation index, the higher the possibility of drift in prostate neuroendocrine carcinoma-related data. This means that the gene mutation frequency shows significant fluctuations or instability in different time periods, which may reflect changes in data characteristics due to factors such as advances in diagnostic technology, changes in treatment methods, or differences in patient groups. This drift may affect the model's prediction accuracy on current or future data, so it is necessary to detect and adjust the model in a timely manner to cope with the impact of data drift.
[0077] The immune abundance distribution abnormality index is generated according to the abnormal change characteristics of immune abundance distribution of the extracted prostate neuroendocrine carcinoma related data. The method for obtaining the immune abundance distribution abnormality index is as follows:
[0078] Collect immune-related data from patients with prostate neuroendocrine carcinoma, including the abundance of different immune cell types (such as T cells, B cells, macrophages, etc.); standardize the immune abundance data and calculate the mean abundance distribution of each immune cell type in patient samples and standard deviation ; For each sample, the immune abundance value Test to determine whether it deviates from the normal distribution and determine the abnormal immune abundance value , the expression is: or ; Where k is a constant; the value is usually 2 or 3 to determine the criteria for outliers; the abnormal immune abundance value in each sample is counted Number of , and the degree of deviation of the outliers (i.e., the number of standard deviations from the mean) , the expression is: ; s is the label of the immune cell category, and the immune abundance distribution abnormality index is calculated, and the expression is: ; Where M is the total number of immune cell types, is the number of abnormal immune abundance values of the i-th type of immune cells, is the total number of samples of the i-th type of immune cells, is the sum of the deviations of the abnormal immune abundance values of the i-th type of immune cells, is the total deviation degree of the i-th type of immune cells, The higher the abnormal index of immune abundance distribution, the more significant the abnormal changes in immune cell abundance or marker expression in the sample, which may indicate the progression of prostate neuroendocrine carcinoma or other immune-related abnormalities.
[0079] The larger the abnormal index of immune abundance distribution, the higher the possibility of drift in prostate neuroendocrine carcinoma related data. This indicates that the immune system-related characteristics show significant abnormal changes in different time periods or samples, which may be due to differences in diagnostic technology, treatment methods or patient groups, resulting in a significant shift in immune abundance distribution. This data drift may affect the stability and accuracy of the prediction model, so it needs to be identified and adjusted in a timely manner to ensure the effectiveness of the model for the current data.
[0080] S4: Based on the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation.
[0081] The mutation frequency fluctuation index and the immune abundance distribution abnormality index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses each group of first eigenvectors to predict the weighted assignment labels of prostate neuroendocrine carcinoma related data in each time period as the prediction target, and takes minimizing the sum of prediction errors of the weighted assignment labels of prostate neuroendocrine carcinoma related data in all time periods as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined according to the model output results, wherein the machine learning model is a polynomial regression model, and the comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is obtained by weighted average calculation of the weight assignment of the prostate neuroendocrine carcinoma related data in each time period.
[0082] The method for obtaining the weight assignment of the prostate neuroendocrine carcinoma related data in each time period is as follows: from the first eigenvector training data of the trained machine learning model, the corresponding function expression is obtained: ; In the formula, is the output function of the model, is the mutation frequency fluctuation index, is the abnormal distribution index of immune abundance, Assign weights to the data related to prostate neuroendocrine cancer in each time period.
[0083] S5: Compare and analyze the calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data with the gradient standard threshold, divide the prostate neuroendocrine carcinoma related data into different data drift levels according to the analysis results, and formulate corresponding model construction adjustment strategies.
[0084] The calculated comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the comprehensive drift coefficient is compared with the first standard threshold and the second standard threshold respectively;
[0085] If the comprehensive drift coefficient is greater than the second standard threshold, the prostate neuroendocrine carcinoma-related data is classified as a high drift level and a data anomaly signal is generated; the data needs to be reanalyzed and modeled, and new feature engineering and model architectures are considered to ensure that the model can adapt to the new data characteristics. If the amount of data allows, try to add new data sets or combine recent clinical data to enhance the generalization ability of the model.
[0086] If the comprehensive drift coefficient is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the prostate neuroendocrine carcinoma-related data is classified as a medium drift level and a suspected abnormal data signal is generated; the parameters of the existing model (such as learning rate and regularization coefficient) are adjusted to better adapt to the moderate drift of the data. Re-evaluate and update features, especially those that are sensitive to drift, to ensure that they can reflect the latest data characteristics.
[0087] If the comprehensive drift coefficient is less than the first standard threshold, the prostate neuroendocrine carcinoma related data is classified as a low drift level and a data normal signal is generated. If the data only shows low drift, the existing model can be kept unchanged and predictions can continue. Although the current drift is small, the data and model performance should still be checked regularly to ensure that adjustments are made in time before the drift expands further. According to the drift trend, small-scale model parameter adjustments are made to ensure that the model continues to operate stably.
[0088] In this embodiment, by systematically collecting, processing and analyzing data related to prostate neuroendocrine carcinoma, especially the rare gene mutation frequency and abnormal changes in immune abundance distribution characteristics of data in different time periods, a comprehensive drift coefficient is generated and compared with the gradient standard threshold. Through this series of steps, the degree of drift of the data can be accurately identified, and the model building strategy can be adjusted in time to improve the adaptability and accuracy of the prediction model in different time periods. This method can not only enhance the robustness of the model, but also ensure that it maintains a high predictive performance in the face of a changing data environment, thereby more effectively supporting clinical decision-making.
[0089] Example 2, please refer to Figure 2 As shown, a system for constructing a prostate neuroendocrine cancer prediction model described in this embodiment includes a data acquisition module, a data partitioning module, a feature extraction module, a comprehensive drift coefficient calculation module and a model adjustment module;
[0090] Data acquisition module: collects and summarizes prostate neuroendocrine cancer related data for building a prediction model, including clinical characteristics, gene expression and imaging data;
[0091] Data division module: divide the collected prostate neuroendocrine carcinoma related data into several time periods, and pre-process the prostate neuroendocrine carcinoma related data in each time period;
[0092] Feature extraction module: extracts features from the preprocessed prostate neuroendocrine carcinoma-related data, and extracts the rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics of the prostate neuroendocrine carcinoma-related data;
[0093] Comprehensive drift coefficient calculation module: According to the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation;
[0094] Model adjustment module: The calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is compared and analyzed with the gradient standard threshold, and the prostate neuroendocrine carcinoma related data is divided into different data drift levels according to the analysis results, and the corresponding model construction adjustment strategy is formulated.
[0095] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.
[0096] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.
[0097] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.
[0098] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0099] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0100] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.
Claims
1. A method for constructing a prediction model for prostate neuroendocrine cancer, characterized in that: include: Collect and summarize data related to prostate neuroendocrine cancer for the construction of a prediction model, including clinical characteristics, gene expression, and imaging data; The collected prostate neuroendocrine carcinoma related data are divided into several time periods, and the prostate neuroendocrine carcinoma related data in each time period are preprocessed respectively; Feature extraction was performed on the pre-processed prostate neuroendocrine carcinoma-related data, and the rare gene mutation frequency fluctuation characteristics and abnormal immune abundance distribution change characteristics of the prostate neuroendocrine carcinoma-related data were extracted respectively; According to the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation; The calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data was compared and analyzed with the gradient standard threshold. According to the analysis results, the prostate neuroendocrine carcinoma related data were divided into different data drift levels, and corresponding model construction adjustment strategies were formulated.
2. The method for constructing a prostate neuroendocrine cancer prediction model according to claim 1, characterized in that: The collected data related to prostate neuroendocrine cancer are divided into several time periods, including: The data related to prostate neuroendocrine carcinoma are sorted in chronological order to form a time series data set X={x1,x2,…,xN}, where xi is the data at the i-th time point and N is the total number of data points. The mean and standard deviation of the time series are calculated and used as the basis for time series segmentation. The time series X is divided into K intervals, each interval Sj represents a subsequence Xj={xsj,…,xej}, where sj and ej are the starting and ending positions of the jth interval, respectively. The internal changes of interval Sj are measured by the loss function C(Sj), and the loss function is: ;in, Is the interval The goal is to minimize the total loss function of the entire time series through dynamic programming , the calculation expression is: ; The task of dynamic programming is to find the change point that minimizes the total loss ; Create an array DP to store the minimum loss value of the first i data points, initialize it to DP[0]=0, and the rest are infinite, define an array P to store the best segmentation point position for each segment, for each time point i, calculate the segmentation cost between each possible previous time point j, the expression is: ; Store the corresponding best segmentation point j: P[i]=j; Find the optimal segmentation point sequence by backtracking the P array, initialize tK=N, and then backtrack: ; Continue backtracking until all The final segmentation point sequence , which is the node position where the data features change significantly. According to the identified change nodes, the data set is divided into multiple time periods, and the corresponding time period label is added to each data point.
3. The method for constructing a prostate neuroendocrine cancer prediction model according to claim 1, characterized in that: After respectively extracting the rare gene mutation frequency fluctuation characteristics and the abnormal change characteristics of immune abundance distribution of prostate neuroendocrine carcinoma related data, the method further includes generating a mutation frequency fluctuation index according to the rare gene mutation frequency fluctuation characteristics extracted from the prostate neuroendocrine carcinoma related data. The method for obtaining the mutation frequency fluctuation index is: The gene mutation frequency data related to prostate neuroendocrine carcinoma are sorted in chronological order to form a time series x(t), where t is time and x(t) is the mutation frequency at time t. The data are standardized and recorded as , the expression is: ;in, is the mean of the mutation frequency data, is its standard deviation, and the standardized time series is transformed using fast Fourier transform Convert from time domain to frequency domain to get spectrum , the expression is: ;in, N is the length of the time series, f is the frequency, is the kernel function of Fourier transform, and the spectrum amplitude of gene mutation frequency data is calculated. Represents the intensity of the signal at each frequency component, and the calculation expression is: ;in, yes X(f) The real part of yes X(f) The imaginary part of The amplitude distribution of the spectrum is obtained, the main frequency of the amplitude is obtained, and the total energy E in the spectrum is calculated. The expression is: ; The mutation frequency fluctuation index is calculated based on the total energy E in the calculated spectrum, and the expression is: ; In the formula, is the mutation frequency fluctuation index, Represents the total energy of the selected main frequency components.
4. The method for constructing a prediction model for prostate neuroendocrine cancer according to claim 3, characterized in that: After respectively extracting the rare gene mutation frequency fluctuation characteristics and the abnormal change characteristics of immune abundance distribution of the prostate neuroendocrine carcinoma related data, the method further comprises generating an abnormal immune abundance distribution index according to the extracted abnormal change characteristics of immune abundance distribution of the prostate neuroendocrine carcinoma related data. The method for obtaining the abnormal immune abundance distribution index is: Collect immune-related data from patients with prostate neuroendocrine carcinoma, including the abundance of different immune cell types; standardize the immune abundance data and calculate the mean abundance distribution of each immune cell type in patient samples and standard deviation ; The abundance of immune cells in each sample Test to determine whether it deviates from the normal distribution and determine the abnormal immune abundance value , the expression is: or ; Where k is a constant; Count the abnormal immune abundance values in each sample Number of , and the degree of deviation of outliers , the expression is: ; s is the label of the immune cell category, and the immune abundance distribution abnormality index is calculated, and the expression is: ; Where M is the total number of immune cell types, is the number of abnormal immune abundance values of the i-th type of immune cells, is the total number of samples of the i-th type of immune cells, is the sum of the deviations of the abnormal immune abundance values of the i-th type of immune cells, is the total deviation degree of the i-th type of immune cells, It is the abnormal index of immune abundance distribution.
5. The method for constructing a prostate neuroendocrine cancer prediction model according to claim 4, characterized in that: The method determines the weight assignment of prostate neuroendocrine carcinoma related data in each time period based on the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, and obtains the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data after weighted average calculation, including: The mutation frequency fluctuation index and the immune abundance distribution abnormality index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses each group of first eigenvectors to predict the weighted assignment labels of prostate neuroendocrine carcinoma related data in each time period as the prediction target, and takes minimizing the sum of prediction errors of the weighted assignment labels of prostate neuroendocrine carcinoma related data in all time periods as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined according to the model output results, wherein the machine learning model is a polynomial regression model, and the comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is obtained by weighted average calculation of the weight assignment of the prostate neuroendocrine carcinoma related data in each time period.
6. The method for constructing a prediction model for prostate neuroendocrine cancer according to claim 5, characterized in that: The calculated comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is compared and analyzed with the gradient standard threshold, and the prostate neuroendocrine carcinoma related data is divided into different data drift levels according to the analysis results, and a corresponding model construction adjustment strategy is formulated, including: The calculated comprehensive drift coefficient of the prostate neuroendocrine carcinoma related data is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the comprehensive drift coefficient is compared with the first standard threshold and the second standard threshold respectively; If the comprehensive drift coefficient is greater than the second standard threshold, the prostate neuroendocrine cancer related data is classified as a high data drift level, and a data abnormality signal is generated; the data needs to be re-analyzed and modeled; If the comprehensive drift coefficient is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the prostate neuroendocrine carcinoma related data is classified into a drift level in the data, and a suspected abnormal signal of the data is generated; and the parameters of the existing model are adjusted; If the comprehensive drift coefficient is less than the first standard threshold, the prostate neuroendocrine carcinoma related data is classified as a low data drift level, and a data normal signal is generated. The existing model remains unchanged and the prediction continues.
7. A system for constructing a prostate neuroendocrine cancer prediction model, used to implement the method for constructing a prostate neuroendocrine cancer prediction model according to any one of claims 1 to 6, characterized in that: It includes data acquisition module, data partitioning module, feature extraction module, comprehensive drift coefficient calculation module and model adjustment module; Data acquisition module: collects and summarizes prostate neuroendocrine cancer related data for building a prediction model, including clinical characteristics, gene expression and imaging data; Data division module: divide the collected prostate neuroendocrine carcinoma related data into several time periods, and pre-process the prostate neuroendocrine carcinoma related data in each time period; Feature extraction module: extracts features from the preprocessed prostate neuroendocrine carcinoma-related data, and extracts the rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics of the prostate neuroendocrine carcinoma-related data; Comprehensive drift coefficient calculation module: According to the extracted rare gene mutation frequency fluctuation characteristics and immune abundance distribution abnormal change characteristics, the weight assignment of prostate neuroendocrine carcinoma related data in each time period is determined, and the comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is obtained by weighted average calculation; Model adjustment module: The calculated comprehensive drift coefficient of prostate neuroendocrine carcinoma related data is compared and analyzed with the gradient standard threshold, and the prostate neuroendocrine carcinoma related data is divided into different data drift levels according to the analysis results, and the corresponding model construction adjustment strategy is formulated.
