Method for carrying out risk scoring on breast cancer
The missing data is filled with Pearson correlation coefficient method and similarity analysis algorithm, and combined with Logisitic regression and SVM model evaluation, the problem of missing data processing and model dependence in breast cancer risk assessment is solved, achieving more accurate and reliable risk assessment results .
Patent Information
- Application Number
- CN202411006983.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing breast cancer risk assessment method fails to fully analyze the correlation between attributes in the data when processing missing sample data, resulting in inaccurate model training, affecting the reliability of evaluation results, and relying too much on model evaluation, so that the results cannot be completely avoided. In line with the actual situation.
Breast cancer-related data were collected, missing data was filled with Pearson correlation coefficient method and similarity analysis algorithm, Logisitic regression and SVM models were constructed, model weights were calculated using the minimum variance method, and risk assessment was combined with multiple models to ensure the accuracy of the results and reliability.
By filling in missing data and multi-model evaluation, the accuracy and systematic flexibility of breast cancer risk assessment are improved, errors and artificial error probability are reduced, and more reliable risk scores and suggestions are provided.
Smart Images

Figure CN119028598A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of prediction and evaluation, and in particular to a method for performing risk scoring on breast cancer. Background Art
[0002] Breast cancer is a malignant tumor that occurs in breast epithelial tissue and has a certain correlation with genetic factors, daily living habits, endocrine abnormalities, and mental factors. Therefore, collecting this information can provide individuals with a risk assessment of breast cancer and formulate corresponding plans. The invention patent with application number CN202011030248.4 discloses "a breast cancer risk assessment method, system, medium, and device, the method comprising the following steps: inputting a sample of body index measurement data; constructing an S-type dynamic convergence difference neural network, setting each connection weight component between the input layer and the hidden layer of the S-type dynamic convergence difference neural network to be randomly initialized and unchanged, and each connection weight component between the hidden layer and the output layer to be randomly initialized and trained; using a variety of different types of transformation functions to train a variety of different S-type dynamic convergence difference neural network models; the trained S-type dynamic convergence difference neural network model predicts the input body index measurement data sample to obtain a prediction result; the prediction result is evaluated based on the voting principle of minority obeys majority to obtain the final breast cancer risk assessment result. The present invention improves the model training efficiency, and at the same time, the use of a comprehensive evaluation mechanism can effectively ensure the reliability of the risk assessment result. "
[0003] The above-mentioned existing technology solves the problems of low efficiency of model training and the inability of the evaluation mechanism to ensure the reliability of risk assessment results. However, when processing the collected missing sample data, the correlation between multiple attributes contained in the data is not analyzed, resulting in large errors in the filling values, making the parameters obtained during the model training process inaccurate, thereby affecting the reliability of the final risk assessment results. At the same time, when user-related information is missing, the user is required to supplement the data before analysis can be performed, which reduces the flexibility of the system and relies too much on the model to assess risks, and cannot completely avoid the situation where the results are not in line with reality. Summary of the invention
[0004] The purpose of the present invention is to provide a method for risk scoring of breast cancer to solve the problems raised in the above background technology.
[0005] To achieve the above object, the present invention provides the following technical solution: a method for risk scoring of breast cancer, comprising the following steps:
[0006] S1. Collecting sample data: Obtain all public user data information on the breast cancer family registration website, and extract relevant information corresponding to each user based on this data information, including personal information, physiological information, living habit information, health information and family information;
[0007] S2. Filling in missing sample data: Since the relevant information corresponds to sixteen indicator information, the samples with missing indicator information are stored in the first data set, and vice versa, they are stored in the second data set. The Pearson correlation coefficient method is used to calculate the correlation between the indicator information in the sample, and the correlation is converted into a weight matrix through a weight conversion algorithm. The distance values between the missing sample data and each sample data in the second data set are calculated respectively by a similarity analysis algorithm, and the sample data in the second data set are sorted from small to large according to the distance value. After setting the threshold, the distance values between the missing sample data and the sample data are added in sequence. When the ratio between the added distance value and the distance sum is less than or equal to the threshold, the sample data involved in the addition process are stored in the similar data set. The inverse of the distance between the missing sample and each sample in the similar data set is used as the proximity value between the two samples. The ratio between the proximity value of a single sample and the total proximity value is set as the weight corresponding to the sample. The filling value corresponding to the indicator information in the missing sample is calculated according to the weight;
[0008] S3. Calculate model weights: After all missing samples in the first data set are processed to obtain corresponding filling values, count the amount of user data involving breast cancer and healthy user data in all sample data, delete redundant data, store 80% of the sample data in the training data set, and store the remaining data in the test data set. Calculate the corresponding model parameters based on the sample data in the training data set by the maximum likelihood estimation method, thereby constructing a Logisitic regression model. After analyzing multiple parameters by the random search method, determine the corresponding kernel function based on the combined parameters, select the parameter combination with the highest accuracy to construct the final SVM model, use the sample data in the training data set to train the Gail model and the SVM model, as well as the Logisitic model and the SVM model, and calculate the corresponding weights of each model by the minimum variance method.
[0009] S4. Collecting user information: After receiving the breast cancer questionnaire information filled out by the user, extract the user's personal information, physiological information, lifestyle information, health information and family information based on the breast cancer questionnaire information, and upload the personal medical history, family medical history and genetic test report;
[0010] S5. Determine the risk probability: When traversing the relevant information of the user, if the data corresponding to the indicator information is missing, use the sample data to analyze the corresponding fill-in value. If the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator information has two or more fill-in values, use the Logisitic model and the SVM model to analyze the indicator information of the user respectively, multiply the weight corresponding to the model by the obtained risk probability, and then add them together to confirm the final risk probability;
[0011] S6. Generate risk score: After transmitting user-related information and risk probability to the consultant interface, the obtained recommended documents, risk scores and risk probabilities are transmitted to the user interface, waiting for the user to provide feedback based on these data. If the feedback result is unacceptable, the remaining models except the originally used model will be selected for analysis. After multiple risk probabilities are generated, they will be transmitted to multiple consultant interfaces, thereby obtaining multiple sets of recommended documents, risk scores and risk probabilities, and transmitting them to the user interface.
[0012] Preferably, S2 comprises the following steps:
[0013] S201, storing sample data: since the relevant information corresponds to sixteen indicator information, the indicator information corresponding to each user is traversed one by one. If the indicator information of the user is missing, it is stored as a missing sample in the first data set, otherwise it is stored as a sample in the second data set. The personal information includes age and body mass index, the physiological information includes age at menarche, age at first childbirth, age at menopause, number of children, breastfeeding time, duration of taking contraceptive pills, and number of abortions, the living habits include alcohol intake, number of active smoking and passive smoking, and average labor intensity, the health information includes number of breast treatments, number of hormone treatments, number of times informed of breast tissue density, and number of breast biopsies, and the family information includes the number of affected first-degree relatives and the number of affected second-degree relatives;
[0014] S202. Calculate the correlation of indicator information: Count the ratio of the number of missing indicator information of each missing sample in the first data set to the total number of indicator information, select the missing sample with the smallest ratio, and use the Pearson correlation coefficient method to calculate the correlation between each indicator information in the sample. The set parameters are used to adjust the obtained correlation, and the values with correlations lower than 0.3 are defaulted to 0. The adjusted correlation is converted into a weight matrix. The Pearson correlation coefficient method is a method for calculating the degree of correlation between vectors.
[0015] Preferably, the S2 further comprises the following steps:
[0016] S203, constructing a similar data set: using a similarity analysis algorithm to calculate the distance values between the missing sample data and each sample data in the second data set respectively, and after adding these distance values to obtain the distance sum, the samples in the second data set are sorted from small to large according to the distance value. After setting a threshold, the distance values between the missing sample data and the sample data are added in sequence. When the ratio between the added distance value and the distance sum is less than or equal to the threshold, the sample data involved in the addition process is compared with the sample data in the similar data set one by one, and the sample data that does not exist in the original similar data set will be stored in the similar data set, thereby obtaining a new similar data set. When the ratio between the added distance value and the distance sum is greater than the threshold, the sample data is stopped from being stored in the similar data set. The similarity analysis algorithm is specifically as follows:
[0017]
[0018] Among them, d i,j,n Indicates that the nth sample in the data set is missing the jth indicator column a i Samples i The distance between them, p represents the total number of indicators, s,i,j represents the parameters, c is Indicates missing samples s i Located at index a s The non-missing data, c ks Represents the sample s in the dataset k Located at index a s The non-missing data, sim js Indicates the correlation between the jth indicator and the sth indicator;
[0019] S204, determining the neighboring value: taking the inverse of the distance between the missing sample and each sample in the similar data set as the neighboring value between the two samples, adding the neighboring values of each sample to obtain the total neighboring value, setting the ratio between the neighboring value of a single sample and the total neighboring value as the weight corresponding to the sample, and using the missing value analysis algorithm to calculate according to the weight and the indicator data corresponding to each sample in the similar data set, thereby obtaining the filling value corresponding to the indicator in the missing sample, the missing value analysis algorithm includes a combination coefficient generation algorithm, and the combination coefficient generation algorithm is specifically:
[0020]
[0021] Among them, x σ represents the linear combination coefficient, It represents the inverse data matrix of the non-missing indicator information in the samples to be supplemented in the dataset SK, W represents the weight matrix, and c1 represents all the non-missing data of sample s1 in the dataset.
[0022] Preferably, S3 comprises the following steps:
[0023] S301, constructing a training data set: when all missing samples in the first data set are processed to obtain corresponding filling values, the missing samples are stored as samples in the second data set, the sample data are scanned, and the amount of user data involving breast cancer and the amount of healthy user data in the sample data are counted, and the amount of data of the two is set to be the same. After deleting redundant data, 80% of the sample data is stored in the training data set, and the remaining data is stored in the test data set;
[0024] S302, calculating model parameters: analyzing the correlation between the input variables according to the correlation, and after obtaining the correlation factor, calculating the corresponding model parameters according to the sample data in the training data set by the maximum likelihood estimation method, thereby constructing a Logisitic regression model.
[0025] Preferably, the S3 further comprises the following steps:
[0026] S303, analyzing parameter combinations: after analyzing multiple parameters using the random search method, combine these parameters, determine the corresponding kernel function according to the combined parameters, train it using the sample data in the training data set, analyze the accuracy of the model prediction results using the sample data in the test set, and select the parameter combination with the highest accuracy to construct the final SVM model;
[0027] S304, determine model weights: use the sample data in the training data set to train the Gail model and the SVM model, analyze the corresponding prediction error values according to the sample data in the test set, calculate the weights of the Gail model and the SVM model by the minimum variance method, determine the prediction error values corresponding to the Logisitic model and the SVM model, and then use the minimum variance method to calculate the corresponding weights of the two. The minimum variance method mainly adjusts different weights to minimize the variance after combination.
[0028] Preferably, S5 comprises the following steps:
[0029] S501, filling user information: when traversing the user's personal information, physiological information, living habit information, health information and family information, if there is a value corresponding to the indicator information that is a missing value, the relevant information of the user is transmitted to the first data set as a missing sample, and the corresponding filling value is obtained after analysis;
[0030] S502, directly calculating the risk probability: analyzing the indicator information corresponding to the user-related information through a processor, and if the number of affected first-degree relatives in the indicator information is not zero, transmitting the user-related information to the Claus model for calculation to obtain the corresponding risk probability, wherein the processor is specifically an Intel processor, and the Claus model is a breast cancer risk assessment model.
[0031] Preferably, the S5 further comprises the following steps:
[0032] S503, first model combination analysis: if the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator information has two or more fill-in values, the Logistic model and the SVM model are used to analyze the indicator information of the user respectively, and the weight corresponding to the model is multiplied by the obtained risk probability and then added to confirm the final risk probability;
[0033] S504, second model combination analysis: If the number of affected first-degree relatives in the indicator information is zero and the data corresponding to the indicator has two or less fill values, the Gail model and the SVM model are used to analyze the user's indicator information respectively, and the weight corresponding to the model is multiplied by the obtained risk probability and then added to confirm the final risk probability.
[0034] Preferably, S6 comprises the following steps:
[0035] S601, obtaining a suggestion document: after transmitting the user's relevant information, personal medical history, family medical history, genetic test report and risk probability to the consultant interface, waiting for the consultant to fill in the corresponding suggestion document and risk score according to these data, and transmitting the suggestion document and risk score to the database;
[0036] S602, user review: after retrieving the recommended document, risk score and risk probability corresponding to the user, transmit them to the user interface and wait for the user to provide feedback based on these data;
[0037] S603. Generate multiple groups of risk scores: If the feedback result is not accepted, the remaining models except the originally used model will be selected for analysis. After multiple risk probabilities are generated, the user-related information, personal medical history, family medical history, genetic test report and multiple risk probabilities will be transmitted to multiple consultant interfaces. After the consultant fills in the corresponding suggestion documents and risk scores based on these data, the multiple groups of suggestion documents, risk scores and risk probabilities will be transmitted to the user interface. If the feedback result is accepted, the user-related information will be transmitted to the second sample set for storage.
[0038] Compared with the prior art, the present invention has the following beneficial effects:
[0039] 1. The present invention collects sample data, constructs a first data set to store missing samples, and a second data set to store complete data samples, so that missing samples can be inferred based on the complete sample data, thereby completing the missing parts. A series of parameter values obtained during subsequent model training and analysis can be more accurate. When constructing a similar data set corresponding to the missing samples based on the similarity between the complete samples and the missing samples, not only the similarity between the sample data is analyzed, but also the correlation between the indicator information will affect the screening results, ensuring that the sample data finally obtained has a higher similarity with the missing sample data. When collecting user-related information, once missing data is found, this method can be used to quickly obtain appropriate filling values, so that large errors can be avoided in subsequent risk probability assessments.
[0040] 2. When evaluating risk probability, the present invention uses the minimum variance method to infer the weights corresponding to the model, so that the final risk probability is usually completed by the joint analysis of the two models. The results obtained in this way are more accurate, and the system automatically provides materials and a visual interface to the consultant, which is convenient for the consultant to read the materials. At the same time, it provides more reliable risk scores and recommendation documents based on the risk probability obtained by model analysis, which saves the consultant's processing time on the one hand, and reduces the probability of model and human errors on the other. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 Provide an overall method flow chart for an embodiment of the present invention;
[0042] Figure 2 A flow chart of a method for filling missing sample data is provided for an embodiment of the present invention;
[0043] Figure 3 A flow chart of a method for calculating model weights is provided for an embodiment of the present invention;
[0044] Figure 4 A flow chart of a method for determining risk probability is provided for an embodiment of the present invention;
[0045] Figure 5 A flow chart of a method for generating a risk score is provided for an embodiment of the present invention. DETAILED DESCRIPTION
[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.
[0047] See also Figure 1-5 The present invention provides a technical solution: a method for risk scoring of breast cancer, comprising the following steps:
[0048] S1. Collecting sample data: Obtain all public user data information on the breast cancer family registration website, and extract relevant information corresponding to each user based on this data information, including personal information, physiological information, living habit information, health information and family information;
[0049] S2. Filling in missing sample data: Since the relevant information corresponds to sixteen indicator information, the samples with missing indicator information are stored in the first data set, and vice versa, they are stored in the second data set. The Pearson correlation coefficient method is used to calculate the correlation between the indicator information in the sample, and the correlation is converted into a weight matrix through a weight conversion algorithm. The distance values between the missing sample data and each sample data in the second data set are calculated respectively by a similarity analysis algorithm, and the sample data in the second data set are sorted from small to large according to the distance value. After setting the threshold, the distance values between the missing sample data and the sample data are added in sequence. When the ratio between the added distance value and the distance sum is less than or equal to the threshold, the sample data involved in the addition process are stored in the similar data set. The inverse of the distance between the missing sample and each sample in the similar data set is used as the proximity value between the two samples. The ratio between the proximity value of a single sample and the total proximity value is set as the weight corresponding to the sample. The filling value corresponding to the indicator information in the missing sample is calculated according to the weight;
[0050] S3. Calculate model weights: After all missing samples in the first data set are processed to obtain corresponding filling values, count the amount of user data involving breast cancer and healthy user data in all sample data, delete redundant data, store 80% of the sample data in the training data set, and store the remaining data in the test data set. Calculate the corresponding model parameters based on the sample data in the training data set by the maximum likelihood estimation method, thereby constructing a Logisitic regression model. After analyzing multiple parameters by the random search method, determine the corresponding kernel function based on the combined parameters, select the parameter combination with the highest accuracy to construct the final SVM model, use the sample data in the training data set to train the Gail model and the SVM model, as well as the Logisitic model and the SVM model, and calculate the corresponding weights of each model by the minimum variance method.
[0051] S4. Collecting user information: After receiving the breast cancer questionnaire information filled out by the user, extract the user's personal information, physiological information, lifestyle information, health information and family information based on the breast cancer questionnaire information, and upload the personal medical history, family medical history and genetic test report;
[0052] S5. Determine the risk probability: When traversing the relevant information of the user, if the data corresponding to the indicator information is missing, use the sample data to analyze the corresponding fill-in value. If the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator information has two or more fill-in values, use the Logisitic model and the SVM model to analyze the indicator information of the user respectively, multiply the weight corresponding to the model by the obtained risk probability, and then add them together to confirm the final risk probability;
[0053] S6. Generate risk score: After transmitting user-related information and risk probability to the consultant interface, the obtained recommended documents, risk scores and risk probabilities are transmitted to the user interface, waiting for the user to provide feedback based on these data. If the feedback result is unacceptable, the remaining models except the originally used model will be selected for analysis. After multiple risk probabilities are generated, they will be transmitted to multiple consultant interfaces, thereby obtaining multiple sets of recommended documents, risk scores and risk probabilities, and transmitting them to the user interface.
[0054] S2 includes the following steps:
[0055] S201, storing sample data: since the relevant information corresponds to sixteen indicator information, the indicator information corresponding to each user is traversed one by one. If the indicator information of the user is missing, it is stored as a missing sample in the first data set, otherwise it is stored as a sample in the second data set. Personal information includes age and body mass index, physiological information includes age of menarche, age of first childbirth, age of menopause, number of children, breastfeeding time, duration of taking contraceptives, and number of abortions, living habits include alcohol intake, number of active smoking and passive smoking, and average labor intensity, health information includes number of breast treatments, number of hormone treatments, number of times informed of breast tissue density, and number of breast biopsies, and family information includes the number of affected first-degree relatives and the number of affected second-degree relatives;
[0056] S202, calculating the correlation of indicator information: counting the ratio of the number of missing indicator information of each missing sample in the first data set to the total number of indicator information, selecting the missing sample with the smallest ratio, and calculating the correlation between each indicator information in the sample using the Pearson correlation coefficient method, adjusting the obtained correlation using the set parameters, setting the values of correlation lower than 0.3 to 0 by default, and converting the adjusted correlation into a weight matrix. The Pearson correlation coefficient method is a method for calculating the degree of correlation between vectors;
[0057] S2 also includes the following steps:
[0058] S203, constructing a similar data set: using a similarity analysis algorithm to calculate the distance values between the missing sample data and each sample data in the second data set respectively, and after adding these distance values to obtain the distance sum, the samples in the second data set are sorted from small to large according to the distance value. After setting a threshold, the distance values between the missing sample data and the sample data are added in sequence. When the ratio between the added distance value and the distance sum is less than or equal to the threshold, the sample data involved in the addition process is compared with the sample data in the similar data set one by one, and the sample data that does not exist in the original similar data set will be stored in the similar data set, thereby obtaining a new similar data set. When the ratio between the added distance value and the distance sum is greater than the threshold, the sample data is stopped from being stored in the similar data set. The similarity analysis algorithm is specifically as follows:
[0059]
[0060] Among them, d i,j,n Indicates that the nth sample in the data set is missing the jth indicator column a i Samples i The distance between them, p represents the total number of indicators, s,i,j represents the parameters, c is Indicates missing samples s i Located at index a s The non-missing data, c ks Represents the sample s in the dataset k Located at index a s The non-missing data, sim js Indicates the correlation between the jth indicator and the sth indicator;
[0061] S204, determining the neighboring value: taking the inverse of the distance between the missing sample and each sample in the similar data set as the neighboring value between the two samples, adding the neighboring values of each sample to obtain the total neighboring value, setting the ratio between the neighboring value of a single sample and the total neighboring value as the weight corresponding to the sample, and using the missing value analysis algorithm to calculate according to the weight and the indicator data corresponding to each sample in the similar data set, thereby obtaining the filling value corresponding to the indicator in the missing sample, the missing value analysis algorithm includes a combination coefficient generation algorithm, and the combination coefficient generation algorithm is specifically:
[0062]
[0063] Among them, x σ represents the linear combination coefficient, represents the inverse matrix of the non-missing indicator information in the to-be-filled samples corresponding to the dataset SK, W represents the weight matrix, and c1 represents all the non-missing data of the sample s1 in the dataset;
[0064] The missing value analysis algorithm also includes a data filling algorithm, which is as follows:
[0065] m ij =x σ ·W·SK m
[0066] Among them, x σ represents the linear combination coefficient, W represents the weight matrix, SK m Indicates the data on the missing indicator information of the similar sample set SK corresponding to the to-be-complemented sample s1, m ij Represents the sample s in the data set i Located in indicator information a j missing values;
[0067] S3 includes the following steps:
[0068] S301, constructing a training data set: when all missing samples in the first data set are processed to obtain corresponding filling values, the missing samples are stored as samples in the second data set, the sample data are scanned, and the amount of user data involving breast cancer and the amount of healthy user data in the sample data are counted, and the amount of data of the two is set to be the same. After deleting redundant data, 80% of the sample data is stored in the training data set, and the remaining data is stored in the test data set;
[0069] S302, calculating model parameters: analyzing the correlation between the input variables according to the correlation, and after obtaining the correlation factor, calculating the corresponding model parameters according to the sample data in the training data set by the maximum likelihood estimation method, thereby constructing a Logisitic regression model;
[0070] S3 also includes the following steps:
[0071] S303, analyzing parameter combinations: after analyzing multiple parameters using the random search method, combine these parameters, determine the corresponding kernel function according to the combined parameters, train it using the sample data in the training data set, analyze the accuracy of the model prediction results using the sample data in the test set, and select the parameter combination with the highest accuracy to construct the final SVM model;
[0072] S304, determine model weights: train the Gail model and the SVM model using the sample data in the training data set, analyze the corresponding prediction error values based on the sample data in the test set, calculate the weights of the Gail model and the SVM model using the minimum variance method, determine the prediction error values corresponding to the Logisitic model and the SVM model, and then use the minimum variance method to calculate the corresponding weights of the two. The minimum variance method mainly adjusts different weights to minimize the variance of the combination;
[0073] S5 includes the following steps:
[0074] S501, filling user information: when traversing the user's personal information, physiological information, living habit information, health information and family information, if there is a value corresponding to the indicator information that is a missing value, the relevant information of the user is transmitted to the first data set as a missing sample, and the corresponding filling value is obtained after analysis;
[0075] S502, directly calculating the risk probability: analyzing the indicator information corresponding to the user-related information through the processor, and if the number of affected first-degree relatives in the indicator information is not zero, transmitting the user-related information to the Claus model for calculation to obtain the corresponding risk probability. The Claus model is a breast cancer risk assessment model;
[0076] S5 also includes the following steps:
[0077] S503, first model combination analysis: if the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator information has two or more fill-in values, the Logistic model and the SVM model are used to analyze the indicator information of the user respectively, and the weight corresponding to the model is multiplied by the obtained risk probability and then added to confirm the final risk probability;
[0078] S504, second model combination analysis: if the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator has two or less fill values, the Gail model and the SVM model are used to analyze the indicator information of the user respectively, and the weight corresponding to the model is multiplied by the obtained risk probability and then added to confirm the final risk probability;
[0079] S6 includes the following steps:
[0080] S601, obtaining a suggestion document: after transmitting the user's relevant information, personal medical history, family medical history, genetic test report and risk probability to the consultant interface, waiting for the consultant to fill in the corresponding suggestion document and risk score according to these data, and transmitting the suggestion document and risk score to the database;
[0081] S602, user review: after retrieving the recommended document, risk score and risk probability corresponding to the user, transmit them to the user interface and wait for the user to provide feedback based on these data;
[0082] S603. Generate multiple groups of risk scores: If the feedback result is not accepted, the remaining models except the originally used model will be selected for analysis. After multiple risk probabilities are generated, the user-related information, personal medical history, family medical history, genetic test report and multiple risk probabilities will be transmitted to multiple consultant interfaces. After the consultant fills in the corresponding suggestion documents and risk scores based on these data, the multiple groups of suggestion documents, risk scores and risk probabilities will be transmitted to the user interface. If the feedback result is accepted, the user-related information will be transmitted to the second sample set for storage.
[0083] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0084] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for risk scoring of breast cancer, characterized in that: The following steps are involved: S1. Collecting sample data: Obtain all public user data information on the breast cancer family registration website, and extract relevant information corresponding to each user based on this data information, including personal information, physiological information, living habit information, health information and family information; S2. Filling in missing sample data: Since the relevant information corresponds to sixteen indicator information, the samples with missing indicator information are stored in the first data set, and vice versa, they are stored in the second data set. The Pearson correlation coefficient method is used to calculate the correlation between the indicator information in the sample, and the correlation is converted into a weight matrix through a weight conversion algorithm. The distance values between the missing sample data and each sample data in the second data set are calculated respectively by a similarity analysis algorithm, and the sample data in the second data set are sorted from small to large according to the distance value. After setting the threshold, the distance values between the missing sample data and the sample data are added in sequence. When the ratio between the added distance value and the distance sum is less than or equal to the threshold, the sample data involved in the addition process are stored in the similar data set. The inverse of the distance between the missing sample and each sample in the similar data set is used as the proximity value between the two samples. The ratio between the proximity value of a single sample and the total proximity value is set as the weight corresponding to the sample. The filling value corresponding to the indicator information in the missing sample is calculated according to the weight; S3. Calculate model weights: After all missing samples in the first data set are processed to obtain corresponding filling values, count the amount of user data involving breast cancer and healthy user data in all sample data, delete redundant data, store 80% of the sample data in the training data set, and store the remaining data in the test data set. Calculate the corresponding model parameters based on the sample data in the training data set by the maximum likelihood estimation method, thereby constructing a Logisitic regression model. After analyzing multiple parameters by the random search method, determine the corresponding kernel function based on the combined parameters, select the parameter combination with the highest accuracy to construct the final SVM model, use the sample data in the training data set to train the Gail model and the SVM model, as well as the Logisitic model and the SVM model, and calculate the corresponding weights of each model by the minimum variance method. S4. Collecting user information: After receiving the breast cancer questionnaire information filled out by the user, extract the user's personal information, physiological information, lifestyle information, health information and family information based on the breast cancer questionnaire information, and upload the personal medical history, family medical history and genetic test report; S5. Determine the risk probability: When traversing the relevant information of the user, if the data corresponding to the indicator information is missing, use the sample data to analyze the corresponding fill-in value. If the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator information has two or more fill-in values, use the Logisitic model and the SVM model to analyze the indicator information of the user respectively, multiply the weight corresponding to the model by the obtained risk probability, and then add them together to confirm the final risk probability; S6. Generate risk score: After transmitting user-related information and risk probability to the consultant interface, the obtained recommended documents, risk scores and risk probabilities are transmitted to the user interface, waiting for the user to provide feedback based on these data. If the feedback result is unacceptable, the remaining models except the originally used model will be selected for analysis. After multiple risk probabilities are generated, they will be transmitted to multiple consultant interfaces, thereby obtaining multiple sets of recommended documents, risk scores and risk probabilities, and transmitting them to the user interface.
2. A method for risk scoring of breast cancer according to claim 1, characterized in that: The S2 comprises the following steps: S201, storing sample data: since the relevant information corresponds to sixteen indicator information, the indicator information corresponding to each user is traversed one by one. If the indicator information of the user is missing, it is stored as a missing sample in the first data set, otherwise it is stored as a sample in the second data set; S202. Calculate the correlation of indicator information: Count the ratio of the number of missing indicator information of each missing sample in the first data set to the total number of indicator information, select the missing sample with the smallest ratio, and use the Pearson correlation coefficient method to calculate the correlation between each indicator information in the sample. Adjust the obtained correlation according to the set parameters, set the correlation values lower than 0.3 to 0 by default, and then convert the adjusted correlation into a weight matrix through the weight conversion algorithm.
3. A method for risk scoring of breast cancer according to claim 2, characterized in that: The S2 further comprises the following steps: S203, constructing a similar data set: using a similarity analysis algorithm to calculate the distance values between the missing sample data and each sample data in the second data set respectively, and after adding these distance values to obtain the distance sum, the samples in the second data set are sorted in ascending order according to the distance value, and after setting a threshold, the distance values between the missing sample data and the sample data are added in sequence, and when the ratio between the added distance value and the distance sum is less than or equal to the threshold, the sample data involved in the addition process is compared with the sample data in the similar data set one by one, and the sample data that does not exist in the original similar data set will be stored in the similar data set, thereby obtaining a new similar data set, and when the ratio between the added distance value and the distance sum is greater than the threshold, the storage of the sample data in the similar data set is stopped; S204, determine the neighboring value: take the inverse of the distance between the missing sample and each sample in the similar data set as the neighboring value between the two samples, add the neighboring value of each sample to obtain the total neighboring value, set the ratio between the neighboring value of a single sample and the total neighboring value as the weight corresponding to the sample, and use the missing value analysis algorithm to calculate according to the weight and the indicator data corresponding to each sample in the similar data set, so as to obtain the filling value corresponding to the indicator in the missing sample.
4. The method for risk scoring of breast cancer according to claim 1, characterized in that: The S3 comprises the following steps: S301, constructing a training data set: when all missing samples in the first data set are processed to obtain corresponding filling values, the missing samples are stored as samples in the second data set, the sample data are scanned, and the amount of user data involving breast cancer and the amount of healthy user data in the sample data are counted, and the amount of data of the two is set to be the same. After deleting redundant data, 80% of the sample data is stored in the training data set, and the remaining data is stored in the test data set; S302, calculating model parameters: analyzing the correlation between the input variables according to the correlation, and after obtaining the correlation factor, calculating the corresponding model parameters according to the sample data in the training data set by the maximum likelihood estimation method, thereby constructing a Logisitic regression model.
5. A method for risk scoring of breast cancer according to claim 4, characterized in that: The S3 further comprises the following steps: S303, analyzing parameter combinations: after analyzing multiple parameters using the random search method, combine these parameters, determine the corresponding kernel function according to the combined parameters, train it using the sample data in the training data set, analyze the accuracy of the model prediction results using the sample data in the test set, and select the parameter combination with the highest accuracy to construct the final SVM model; S304, determine model weights: use the sample data in the training data set to train the Gail model and the SVM model, analyze the corresponding prediction error values based on the sample data in the test set, calculate the weights of the Gail model and the SVM model by the minimum variance method, determine the prediction error values corresponding to the Logisitic model and the SVM model, and then use the minimum variance method to calculate the corresponding weights of the two.
6. The method for risk scoring of breast cancer according to claim 1, characterized in that: The S5 comprises the following steps: S501, filling user information: when traversing the user's personal information, physiological information, living habit information, health information and family information, if there is a value corresponding to the indicator information that is a missing value, the relevant information of the user is transmitted to the first data set as a missing sample, and the corresponding filling value is obtained after analysis; S502, directly calculating the risk probability: analyzing the indicator information corresponding to the user-related information through the processor, and if the number of affected first-degree relatives in the indicator information is not zero, transmitting the user-related information to the Claus model for calculation to obtain the corresponding risk probability.
7. A method for risk scoring of breast cancer according to claim 6, characterized in that: The S5 further comprises the following steps: S503, first model combination analysis: if the number of first-degree relatives affected in the indicator information is zero and the data corresponding to the indicator information has two or more fill-in values, the Logistic model and the SVM model are used to analyze the indicator information of the user respectively, and the weight corresponding to the model is multiplied by the obtained risk probability and then added to confirm the final risk probability; S504, second model combination analysis: If the number of affected first-degree relatives in the indicator information is zero and the data corresponding to the indicator has two or less fill values, the Gail model and the SVM model are used to analyze the user's indicator information respectively, and the weight corresponding to the model is multiplied by the obtained risk probability and then added to confirm the final risk probability.
8. The method for risk scoring of breast cancer according to claim 1, characterized in that: The S6 comprises the following steps: S601, obtaining a suggestion document: after transmitting the user's relevant information, personal medical history, family medical history, genetic test report and risk probability to the consultant interface, waiting for the consultant to fill in the corresponding suggestion document and risk score according to these data, and transmitting the suggestion document and risk score to the database; S602, user review: after retrieving the recommended document, risk score and risk probability corresponding to the user, transmit them to the user interface and wait for the user to provide feedback based on these data; S603. Generate multiple groups of risk scores: If the feedback result is not accepted, the remaining models except the originally used model will be selected for analysis. After multiple risk probabilities are generated, the user-related information, personal medical history, family medical history, genetic test report and multiple risk probabilities will be transmitted to multiple consultant interfaces. After the consultant fills in the corresponding suggestion documents and risk scores based on these data, the multiple groups of suggestion documents, risk scores and risk probabilities will be transmitted to the user interface. If the feedback result is accepted, the user-related information will be transmitted to the second sample set for storage.
Citation Information
Cited By
Rehabilitation decision data processing system based on artificial intelligence
CN121122567A
