Multi-constraint characterization learning model for missing data filling and cancer diagnosis model
Through the projection matrix and constraint terms of the learning model of multi-constraint characterization, the problems of stability and accuracy of data interpolation with high missing proportions are solved, the data classification ability in ovarian cancer diagnosis is improved, and better interpolation effect is achieved.
Patent Information
- Application Number
- CN202411565752.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-11-05
AI Technical Summary
The existing missing data interpolation methods are insufficient in processing data with high missing proportions, especially in the diagnosis of ovarian cancer. The characteristics of conventional laboratory test indicators are numerous and some of them are missing, resulting in the interpolation results. Accuracy and stability need to be improved.
Using a multi-constraint characterization learning model, the missing data is projected to the latent space through the projection matrix, combined with the consistency constraint term of feature importance, the missing position estimation constraint term and the fuzzy relationship constraint term, the projection process is optimized to improve the accuracy and stability of the interpolation result. sex.
The law of missing data is learned through the projection matrix of the multi-constrained representation learning model, avoiding the loss of important features, improving the rationality and accuracy of representation learning, improving the data classification ability, and achieving stable and accurate filling of missing data.
Smart Images

Figure CN119089125A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of data processing technology, and specifically relates to a multi-constraint representation learning model and a cancer diagnosis model for missing data filling. Background Art
[0002] With the help of artificial intelligence, we can explore the value of laboratory tests in the diagnosis and prognosis prediction of ovarian cancer, screen out early ovarian cancer patients, improve postoperative living treatment, and conduct postoperative evaluation and monitoring of patients with advanced cancer.
[0003] However, due to various reasons, such as sample quality issues, patient wishes, data transmission and preservation, etc., the laboratory index data detected by routine health examinations are missing, and the missing rate is not low. The missing rate of some items can reach 90%. It is a big problem that missing data cannot be used for classifier training using machine learning. If you simply delete the missing patient data and only keep the complete patient data, the sample size of the patient will be greatly reduced, and at the same time, it will also cause the loss of data set information, which will have a great impact on the subsequent identification of ovarian cancer. Therefore, the current mainstream approach is to interpolate the missing data set first, and then train the classifier to obtain a classification model.
[0004] The current missing data interpolation algorithms mainly include: interpolation methods based on statistics, interpolation methods based on machine learning, and interpolation methods based on deep learning. The statistical interpolation method is simply to fill in the maximum and minimum values, and further to interpolate the mean, median, and mode based on simple statistics. In addition to the above simple interpolation methods, there is also an interpolation method (SOFT) that uses singular value decomposition to keep the distribution unchanged, and the most classic and commonly used interpolation method using global information is multiple interpolation (MICE). Its main idea is to better estimate and handle the uncertainty of missing data by generating multiple completed versions of multiple data sets. In addition, there is BPCA (Bayesian Principal Component Analysis), a multiple estimation method based on principal component analysis (PCA) to handle incomplete continuous data, and in order to reflect the uncertainty of parameters from one estimate to the next, the Bayesian method is used to process the PCA model. Among the interpolation methods based on machine learning, the commonly used ones are KNN (K-Nearest Neighbors), missing data interpolation method based on random forest (MISS FOREST), XGBOOST (Extreme Gradient Boosting), etc. They use machine learning models to directly predict missing values. Due to the strong adaptability and predictive ability of deep learning, there are more and more interpolation methods based on deep learning in recent years. For example, the GAIN (Generative Adversarial ImputationNets) algorithm proposes to use generative adversarial networks to interpolate missing data, so that the generated data is as close to the real distribution as possible. In addition, the interpolation algorithm based on the autoencoder uses a partially complete data set to compress the representation of the learning data, and then decodes the compressed representation of the incomplete data to obtain the complete data. There is also a further improved method of the autoencoder, REMASK (REcurrent Masked Autoencoder), which randomly "re-masks" another set of values in addition to the missing values (i.e., natural masks), and optimizes the autoencoder by reconstructing this set of re-masked values. The interpolation algorithm based on the best mode of transportation (OT) uses the best transmission distance to quantify the differences between different distributions and converts them into a loss function to estimate the missing data value.
[0005] In the statistical interpolation method, if the mean, median and other conventional statistical variables are simply used for interpolation, the variability of the data will be greatly reduced, because all missing values are replaced with the same mean, which makes the interpolated data unable to reflect the volatility of the real data. In addition, the use of conventional statistical variables for interpolation may introduce bias, especially when the data itself has an asymmetric distribution or multimodal distribution. This method may cause distortion of the results and affect the prediction performance and accuracy of the classification model. The disadvantage of using MICE for interpolation is that its computational complexity is high. In addition, MICE is highly dependent on model assumptions. If the model is not selected properly, it may lead to inaccurate interpolation results and even introduce systematic errors. Since MICE generates interpolation values through multiple models, the stability and consistency of its results may not be as good as other simpler interpolation methods. In addition, the interpolation effect of statistical interpolation methods will be relatively poor for features with a high missing ratio.
[0006] The performance of machine learning-based interpolation methods depends on the adjustment of machine learning model parameters. On the other hand, the specific interpolation effect is often closely related to the data structure distribution. If the missing ratio is high, the interpolation effect may be poor. Finally, these machine learning models often require a certain amount of non-missing samples to build the model, and if the quality of non-missing samples is not good, the machine learning-based interpolation methods will amplify the noise or outliers in the data.
[0007] The main disadvantage of deep learning-based interpolation methods is that they usually require a large amount of data and computing resources to train complex neural networks, which is particularly time-consuming when dealing with large-scale data sets or high-dimensional data. In addition, these methods are sensitive to the choice of hyperparameters and network architectures, and improper settings may lead to unstable or inaccurate interpolation results. Deep learning models are also prone to overfitting, especially when data is sparse, and may generate unreliable interpolation results. Finally, deep learning currently has poor interpretability, and most prediction processes are "black boxes."
[0008] Therefore, the imputation results of existing missing data imputation methods are not stable and accurate enough. Summary of the invention
[0009] The purpose of the present invention is to provide a multi-constraint representation learning model and a cancer diagnosis model for missing data filling, which can stably and accurately fill in the missing data.
[0010] The first aspect of the present invention discloses a multi-constraint representation learning model for missing data filling, comprising: A data preprocessing module is used to preprocess the source data to obtain preprocessed data; A projection module, used to project the preprocessed data according to a projection matrix to obtain and output filled data, wherein the projection matrix is obtained after training the multi-constraint representation learning model; The constraint functions for training the multi-constraint representation learning model include: a projection data constraint term, a feature importance consistency constraint term, a missing position estimation constraint term, and a fuzzy relationship constraint term; The feature importance consistency constraint is used to keep the feature weight ranking consistent with the feature importance ranking determined by multiple feature selection methods during the projection process; the missing position estimation constraint is used to identify the inconsistency between feature values according to the information of the missing position during back projection; the fuzzy relationship constraint is used to optimize the redundancy between the shared fusion features obtained by projection through fuzzy relationships and the correlation between the shared fusion features and the labels.
[0011] In some embodiments, the expression of the projection data constraint term is: ,in, is the preprocessed training data, is the projection matrix, is the back projection matrix, is the latent space projected to, represents the F-norm.
[0012] In some embodiments, the feature importance consistency constraint item includes a projection weight ranking unit and a feature importance ranking unit. The projection weight ranking unit is used to calculate the weight ranking of different features in the projection matrix during the projection process. The feature importance ranking unit is used to calculate the correlation between features and labels through multiple feature selection algorithms, and determine the feature importance ranking of each feature according to the correlation between the features and labels.
[0013] In some embodiments, the calculation of the correlation between features and labels by using a plurality of feature selection algorithms, and determining the feature importance ranking of each feature according to the correlation between the features and labels, includes: Calculate the correlation between features and labels according to various feature selection algorithms, and rank them according to the correlation between features and labels; Use ranking aggregation method to convert ranking results into preference matrix; Based on the preference matrix, the cost of an optimal ranking in the preference matrix is minimized by Manhattan distance to obtain the feature importance ranking.
[0014] In some embodiments, the fuzzy relationship constraint item includes a correlation unit and a redundancy unit, the correlation unit is used to calculate the fuzzy mutual information between the shared fusion feature and the label, and the redundancy unit is used to calculate the fuzzy mutual information between the shared fusion features.
[0015] In some embodiments, the expression of the correlation unit is: , ; The expression of the redundant unit is: , ;in, To share fusion features; In addition to sharing fusion features All features except are reduced to one-dimensional vectors through PCA; is a diagonal matrix, and the elements on the diagonal are the fuzzy relationship matrix The sum of the elements in each row; is a diagonal matrix, and the elements on the diagonal are The sum of the elements in each row; n is the total number of samples; k is the dimension of shared fusion features; For label.
[0016] In some embodiments, the missing position estimation constraint term includes a mask of missing data and a conversion unit, wherein the conversion unit is used to calculate a loss for transforming the data projected into the latent space into the mask of missing data.
[0017] In some embodiments, the constraint function of the multi-constraint representation learning model is expressed as: , in, is the preprocessed data; is the projection matrix; is the back projection matrix; is the latent space projected to; is the mask for missing data; is the feature importance consistency constraint; is a linear transformation matrix used to transform the data in the latent space into a mask of missing data; and is the fuzzy relation constraint; , , , , , and is the trade-off parameter used to balance different items; is the regularization term.
[0018] A second aspect of the present invention discloses a cancer diagnosis model, comprising any one of the above-mentioned multi-constraint representation learning models and cancer classification models.
[0019] In some embodiments, the step of determining the cancer classification model comprises: Using the missing data training set to train the multi-constraint representation learning model to obtain a projection matrix; Projecting the missing data training set into a shared fusion feature data set according to the projection matrix; A plurality of classifiers are trained based on the shared fusion feature data set, and the optimal classifier is selected as the cancer classification model.
[0020] The beneficial effect of the present invention is that the interpolation of missing data is realized by using the projection matrix in the multi-constraint representation learning model. When the multi-constraint representation learning model is trained so that the projection matrix learns the law of missing data, the feature importance consistency and missing position estimation constraints are added. On the one hand, the loss of information of important features in the projection process is avoided, and the rationality of the representation learning projection is improved. On the other hand, the accuracy of the representation learning projection is improved by estimating the missing position in the projection representation learning process. And the redundancy between the shared fusion features after the projection and the correlation with the label are optimized by the fuzzy relationship constraint term, so that the classification ability of the data processed by the representation learning model is greatly improved. The missing data can be filled stably and accurately. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The accompanying drawings herein show specific examples of the technical solutions described in the present invention, and together with the specific implementation methods constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.
[0022] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.
[0023] Figure 1 It is a module diagram of a multi-constraint representation learning model disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0024] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. In combination with the technical solution of the present invention in a realistic scenario, all technical and scientific terms used herein may also have meanings corresponding to the purpose of implementing the technical solution of the present invention. The "first, second..." used herein is only used to distinguish the names and does not represent a specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0025] It should be noted that when a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there can be a central component; when an component is considered to be "connected to" another component, it can be directly connected to the other component or there can be a central component at the same time; when an component is considered to be "installed on" another component, it can be directly installed on the other component or there can be a central component at the same time. When an component is considered to be "set on" another component, it can be directly set on the other component or there can be a central component at the same time.
[0026] Unless otherwise specified or defined, the "said" and "the" used in this document refer to the technical features or technical contents mentioned or described before the corresponding position, and the technical features or technical contents may be the same as or similar to the technical features or technical contents mentioned therein. In addition, the terms "including" and "having" and any variations thereof used in this document are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.
[0027] The interpolation results of commonly used missing data interpolation methods are always not stable and accurate enough, especially when it comes to missing routine laboratory test indicators. Since routine laboratory test indicators have many characteristics, a high proportion of missing characteristics, and strong correlations between characteristics, the accuracy and stability of the interpolation results need to be improved.
[0028] In view of the above problems, the present invention proposes a multi-constraint representation learning model, which uses the projection matrix in the multi-constraint representation learning model to project the missing data into the latent space to obtain complete data and fill in the missing data. When training the multi-constraint representation learning model so that the projection matrix learns the law of missing data, feature importance consistency constraints and missing position estimation constraints are added. On the one hand, the loss of information of important features in the projection process is avoided, which improves the rationality of the representation learning projection. On the other hand, the accuracy of the representation learning projection is improved by estimating the missing position in the projection representation learning process. In addition, the redundancy between the shared fusion features obtained by the fuzzy relationship optimization projection and the correlation with the label are also greatly improved, so that the identification and classification capabilities of the data processed by the representation learning model are greatly improved.
[0029] This embodiment provides a multi-constraint representation learning model for ovarian cancer laboratory indicator data to fill in missing data in ovarian cancer laboratory indicator data. It should be noted that the multi-constraint representation learning model of the present invention is not limited to filling in laboratory indicator data, but can also be used to fill in other types of data.
[0030] Specifically, the multi-constraint representation learning model mainly includes two parts: a data preprocessing module and a projection module. Among them, the data preprocessing module is used to perform various preprocessing such as normalization on various source data such as ovarian cancer laboratory indicator data to obtain preprocessed data. By normalizing each feature of the data set, the multi-constraint representation learning model can be unaffected by the data size. Then the preprocessed data is multiplied by the projection matrix, the preprocessed data is projected, and the filled data is obtained and output. Among them, the projection matrix is a data matrix. When training the multi-constraint representation learning model, the projection matrix will be iteratively optimized. After training, the final projection matrix can be obtained.
[0031] In this embodiment, when training a multi-constraint representation learning model, a constraint function of the multi-constraint representation learning model is constructed based on projection data constraints, feature importance consistency constraints, missing position estimation constraints and fuzzy relationship constraints, wherein the feature importance consistency constraint is used to keep the weight ranking of the feature consistent with the feature importance ranking determined by multiple feature selection methods during the projection process; the missing position estimation constraint is used to identify inconsistencies between feature values based on the information of the missing position during back projection; and the fuzzy relationship constraint is used to optimize the redundancy between the shared fusion features obtained by projection through fuzzy relationships and the correlation between the shared fusion features and the labels.
[0032] Specifically, refer to Figure 1 , the construction process of the constraint function is: First, define a matrix Represents the patient data matrix, which is used as training data for training. n Represents the number of patients, m Represents the number of features and Each feature is normalized. In order to learn data To obtain the complete data, we assume that there is a latent space , and through the projection matrix The data can be Projected to the latent space and learned the shared fusion features. The dimension of the shared fusion feature projected into the latent space. At the same time, by reconstructing the matrix The process of back-projection can significantly improve the reliability of data projection in representation learning by reconstructing the latent space back to the patient data matrix. The projection data constraint term can be determined according to the bidirectional mapping process, which is defined as follows: (1) in, is the preprocessed training data, is the projection matrix, is the back projection matrix, is the latent space projected to, represents the F norm. Then the projection matrix , back projection matrix and the latent space After regularization, the formula is rewritten as follows.
[0033] (2) in, , and is a trade-off parameter used to balance different items. In this embodiment, define . , , and The meaning is similar. Regularizing these parameters can avoid overfitting problems. Formula (2) is a preliminary framework for characterizing bidirectional mapping, which encourages projecting incomplete data into the latent space to obtain complete data and learn the projection matrix for new incomplete data.
[0034] However, if the data is directly projected and represented without constraints, the shared fusion features obtained are unreasonable, because in this process, the projection weights of different features are determined according to the specific missing conditions when learning data features and the distribution law of the data, and this weight constraint is one-sided. It does not take into account that the importance of different features is often different. The impact of feature importance on projection weights should be significant, otherwise the shared fusion features may lose important feature information. Therefore, this embodiment introduces a feature importance consistency constraint term in the constraint function to convert the projection matrix The weights of different features in the projection process are ranked, and then a constraint is used to keep the weight ranking of the features in the projection process consistent with the feature importance ranking determined by multiple feature selection methods.
[0035] The feature importance consistency constraint item includes the projection weight ranking unit and the feature importance ranking unit. The projection weight ranking unit is used to calculate the weight ranking of different features in the projection matrix during the projection process. The specific steps are: first calculate the weights of different features during the projection process, and the specific calculation formula is as follows: , in, is a column vector of all 1s, is a column vector, It is The sum of the weights of the features during the projection process.
[0036] Then calculate the weight ranking of the feature. The specific formula is as follows: , in, Indicates The weight ranking of features, indicating function Represents a binary relationship ,when hour, But because the indicator function It is not a continuous function and cannot be derived and optimized in the subsequent optimization process. Therefore, this embodiment introduces the SIGMOD function , it can Mapped to 0-1.
[0037] , After the above two steps of calculation, we can finally get The weight ranking of features .
[0038] The feature importance ranking unit is used to calculate the correlation between features and labels through a variety of feature selection algorithms, and determine the feature importance ranking of each feature based on the correlation between features and labels. The specific process is: using five different feature selection algorithms: random forest, mutual information, logistic regression, gradient boosting tree, and support vector machine. These feature selection algorithms all calculate the correlation between features and labels through certain variables, and rank them based on the correlation between features and labels. The importance of features is ranked; the ranking results are converted into a preference matrix using the Kemeny-Young ranking aggregation method; the cost of an optimal ranking in the preference matrix is minimized by Manhattan distance to obtain the final feature importance ranking .
[0039] Weighted Ranking Vectors and Feature Importance Ranking The elements in the vector are one-to-one corresponding, and are ranked according to the importance of the features in a certain order. Finally, the feature importance consistency constraint is designed to make the two rankings as consistent as possible. The specific formula of the feature importance consistency constraint is as follows: (3) in, .
[0040] After reorganizing formulas (2) and (3), we can obtain the constraint function of the updated multi-constraint representation learning model: (4).
[0041] Since representation learning is projection learning on missing data, and the missing ratio of some features of ovarian cancer laboratory indicators is not low, it is very important to enable the multi-constraint representation learning model to learn the rules of missing data. After analysis and research, it is found that the information of missing data is not only in the data distribution but also in the specific missing position of the missing data. Therefore, it is particularly important to add the information of the missing position to the constraint function of the multi-constraint representation learning model. In addition, a key point of representation learning is to learn the correlation between eigenvalues. Although 0 will be filled in the blank position for calculation when projecting the missing data, simple projection and back-projection cannot make good use of the missing position information, which may lead to learning wrong information. Based on the above two points, the missing position estimation constraint item of this embodiment includes a mask and a conversion unit for missing data. By using the mask of missing data during back-projection Allowing latent space data to restore the specific missing information of the source data and the mask of missing data It is a matrix with only 0 and 1. The specific distribution of 0 and 1 is determined by the missing position in the missing data set. If it is missing, it is filled with 0, and other positions are filled with 1. When back-projecting based on the information of the missing position, the inconsistency between the eigenvalues can be identified and the accuracy of the representation learning of the missing data can be improved. The expression of the conversion unit is: , which is used to calculate the loss of transforming the data projected into the latent space into the mask of missing data, where It is a linear transformation matrix, which transforms the data in the latent space into a mask of missing data so that the shared fusion features in the latent space obtain the information of the mask.
[0042] After applying the missing position estimation constraint to formula (1), the update is: (5) After reorganizing formulas (4) and (5), the updated constraint function is as follows: (6) Since the purpose of the representation learning model is to project the missing data into the latent space to obtain complete data that can be classified. In order to enable the shared fusion features obtained by the missing data projection to have better identification and classification capabilities, this embodiment starts from the relationship between information, connects the correlation between the shared fusion features and the labels and the redundancy between the shared fusion features through fuzzy information relationships, optimizes and improves the correlation between the shared fusion features and the labels, and reduces the redundancy between the shared fusion features, thereby achieving the effect of improving the identification and classification capabilities of the shared fusion features. Therefore, a fuzzy relationship constraint term is introduced into the constraint function, and the fuzzy relationship constraint term includes a correlation unit and a redundancy unit. The correlation unit is used to calculate the fuzzy mutual information between the projected shared fusion features and the labels, and the redundancy unit is used to calculate the fuzzy mutual information between the projected shared fusion features.
[0043] The specific construction process of fuzzy relationship constraints is as follows: First, let’s define fuzzy information. In this example, we define each feature of the shared fusion feature as There will be a fuzzy relationship matrix , Represents a fuzzy equivalence relationship.
[0044] , in, is a function that measures membership, used to measure The similarity between any two samples in Represents the shared feature The first of the characteristics samples. Similarly, the labels of patient data There is also a similar fuzzy relationship matrix .
[0045] With the definition of the fuzzy relationship matrix, the fuzzy information entropy of the relevant information can be calculated: , in, is related The fuzzy equivalent estimate of .
[0046] We further derive the fuzzy joint information entropy between two fuzzy equivalent information: , the calculation formula is: , Finally, the fuzzy mutual information between the two information vectors is defined as: , The shared fusion features can be quantified by the above definition of fuzzy mutual information between information vectors. and tags The correlation between them and the redundancy between shared fusion features. The specific formula is: (7) (8) in, To share fusion features; are shared fusion features except All features except are reduced to one-dimensional vectors through PCA; Is a diagonal matrix, the elements on the diagonal are the fuzzy relationship matrix The sum of the elements in each row corresponds to the matrix derivation form: , and It is also a diagonal matrix, and the elements on the diagonal are The sum of each row in Similarly, its matrix derivation form is: , in, is a The first column of the matrix is 1 and the rest are 0. is a The identity matrix of .
[0047] Finally, formulas (7) and (8) are simplified and derived to the entire shared fusion feature to obtain the correlation unit and the redundancy unit.
[0048] The expression of the correlation unit is: (9) The expression of the redundancy unit is: (10) Combining formulas (6), (9) and (10), we can obtain the final constraint function of the multi-constraint representation learning model: , in, is the preprocessed data; is the projection matrix; is the back projection matrix; is the latent space projected to; is the mask for missing data; is the feature importance consistency constraint; is a linear transformation matrix used to transform the data in the latent space into a mask of missing data; and is the fuzzy relation constraint; , , , , , and is the trade-off parameter used to balance different items; is the regularization term.
[0049] After the constraint function of the multi-constraint representation learning model is constructed, the multi-constraint representation learning model is trained and optimized. The training process of this embodiment is as follows: 87 routine laboratory test indicators and age of 714 patients with ovarian cancer (primary ovarian cancer, fallopian tube cancer and peritoneal cancer) and non-ovarian cancer control group (benign ovarian lesions and some benign fallopian tube and uterine lesions) are collected from the hospital as features, with a total of 88 features. The collected data matrix is set as The pathological examination results are used as labels (ovarian cancer = 1; non-ovarian cancer = 0). Normalize the data, fill in the missing values with 0, and calculate the mask of the missing data. . For the parameters that need to be optimized in the constraint function Perform random initialization; under the guidance of the constraint function, perform gradient descent cycles t times or until the loss value of the constraint function is reduced to a certain level: in each cycle, calculate The derivatives of the four parameters with respect to the objective function , , , , and use the following formulas to update the parameters iteratively:
[0050]
[0051]
[0052]
[0053] in, is the learning rate of gradient descent for several parameters, The formula is Directly derived.
[0054] The ultimate goal of the iteration is to obtain the projection matrix , the projection matrix can be used for subsequent new missing data Perform projection representation learning to obtain new shared fusion features For identification and classification.
[0055] After testing the ovarian cancer laboratory index data after interpolation in this embodiment and the data processed by 10 commonly used missing data interpolation methods based on statistical theory, machine learning models and deep learning models at this stage, the area under the curve (AUC), accuracy (ACC), sensitivity (SEN), specificity (SPE) and other indicators were used to quantify the classification effect, and the obtained prediction results were compared. The AUC, ACC, SEN, and SPE of the ovarian cancer laboratory index data after interpolation in this embodiment were 0.998, 0.983, 0.968, and 0.987, respectively. It shows that the laboratory index data after interpolation by the multi-constrained representation learning model has a good prediction effect on ovarian cancer prediction, and AUC, ACC, and SEN are better than other interpolation methods, indicating that the processed data has a better identification and classification effect, especially SEN is better than other interpolation methods, which has great clinical significance in practical applications.
[0056] In summary, the multi-constraint representation learning model of this embodiment adds constraints for consistent feature importance and estimation of missing positions in view of the characteristics of many missing laboratory indicator features, strong feature correlation, and high missing ratio of some features. On the one hand, it avoids the loss of information of important features during the projection process and improves the rationality of the representation learning projection. On the other hand, it improves the accuracy of the representation learning projection by estimating the missing positions during the projection representation learning process. And by optimizing the redundancy between the shared fusion features after projection and the correlation with the labels through fuzzy relations, the data identification and classification capabilities after processing by the representation learning model are greatly improved. It can cope with the situation where there is a large amount of missing data when using laboratory indicators for early diagnosis of ovarian cancer.
[0057] Based on the above-mentioned multi-constraint representation learning model, the present invention also provides a cancer diagnosis model, which includes any one of the above-mentioned multi-constraint representation learning models and cancer classification models.
[0058] The cancer diagnosis model of this embodiment also optimizes and selects a cancer classification model based on the training results of the multi-constraint representation learning model. First, the multi-constraint representation learning model is trained with the missing data training set to obtain a projection matrix; the missing data training set is projected into a shared fusion feature data set according to the projection matrix; then, multiple classifiers are trained based on the shared fusion feature data set, and the optimal classifier is selected as the cancer classification model.
[0059] The specific process is as follows: first normalize the missing data, then fill the missing positions with 0, and then divide the missing data into 5 folds, which are the training set and test set Each fold of missing data training set is trained under the representation learning model to obtain the projection matrix corresponding to each fold , multiply the missing data training set of each fold by the projection matrix to obtain the shared fusion feature dataset corresponding to each fold . Use all shared fusion feature datasets Several conventional classifiers are trained and the optimal classifier is selected as the cancer classification model.
[0060] When testing the cancer diagnosis model, the divided test set Using the projection matrix Calculate the shared fusion feature dataset for testing , the shared fusion feature dataset for testing Input into the classifier obtained in the training phase for testing and obtain the predicted output result.
[0061] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to make the public understand the disclosed content of the present invention more thoroughly and comprehensively, and it does not limit the scope of protection of the present invention.
[0062] The above embodiments are not exhaustive enumerations of the present invention, and there may be multiple other implementations not listed. Any replacement and improvement made without violating the concept of the present invention shall fall within the protection scope of the present invention.
Claims
1. A multi-constrained representation learning model for missing data filling, characterized in that: include: A data preprocessing module is used to preprocess the source data to obtain preprocessed data; A projection module, used to project the preprocessed data according to a projection matrix to obtain and output filled data, wherein the projection matrix is obtained after training the multi-constraint representation learning model; The constraint functions for training the multi-constraint representation learning model include: a projection data constraint term, a feature importance consistency constraint term, a missing position estimation constraint term, and a fuzzy relationship constraint term; The feature importance consistency constraint is used to keep the feature weight ranking consistent with the feature importance ranking determined by multiple feature selection methods during the projection process; the missing position estimation constraint is used to identify the inconsistency between feature values according to the information of the missing position during back projection; the fuzzy relationship constraint is used to optimize the redundancy between the shared fusion features obtained by projection through fuzzy relationships and the correlation between the shared fusion features and the labels.
2. The multi-constraint representation learning model for missing data filling according to claim 1, characterized in that: The expression of the projection data constraint term is: ,in, is the preprocessed training data, is the projection matrix, is the back projection matrix, is the latent space projected to, represents the F-norm.
3. The multi-constraint representation learning model for missing data filling according to claim 1, characterized in that: The feature importance consistency constraint item includes a projection weight ranking unit and a feature importance ranking unit. The projection weight ranking unit is used to calculate the weight ranking of different features in the projection matrix during the projection process. The feature importance ranking unit is used to calculate the correlation between features and labels respectively through multiple feature selection algorithms, and determine the feature importance ranking of each feature according to the correlation between the features and labels.
4. The multi-constraint representation learning model for missing data filling according to claim 3, characterized in that: The method of calculating the correlation between features and labels respectively through a plurality of feature selection algorithms and determining the feature importance ranking of each feature according to the correlation between the features and labels includes: Calculate the correlation between features and labels according to various feature selection algorithms, and rank them according to the correlation between features and labels; Use ranking aggregation method to convert ranking results into preference matrix; Based on the preference matrix, the cost of an optimal ranking in the preference matrix is minimized by Manhattan distance to obtain the feature importance ranking.
5. The multi-constraint representation learning model for missing data filling according to claim 1, characterized in that: The fuzzy relationship constraint item includes a correlation unit and a redundancy unit, the correlation unit is used to calculate the fuzzy mutual information between the shared fusion feature and the label, and the redundancy unit is used to calculate the fuzzy mutual information between the shared fusion features.
6. The multi-constraint representation learning model for missing data filling according to claim 5, characterized in that: The expression of the correlation unit is: , ; The expression of the redundant unit is: , ;in, To share fusion features; In addition to sharing fusion features All features except are reduced to one-dimensional vectors through PCA; is a diagonal matrix, and the elements on the diagonal are the fuzzy relationship matrix The sum of the elements in each row; is a diagonal matrix, and the elements on the diagonal are The sum of the elements in each row; n is the total number of samples; k is the dimension of shared fusion features; For label.
7. The multi-constraint representation learning model for missing data filling according to claim 1, characterized in that: The missing position estimation constraint term includes a mask of missing data and a conversion unit, wherein the conversion unit is used to calculate a loss of transforming data projected into the latent space into the mask of missing data.
8. The multi-constraint representation learning model for missing data filling according to claim 1, characterized in that: The expression of the constraint function of the multi-constraint representation learning model is: , in, is the preprocessed data; is the projection matrix; is the back projection matrix; is the latent space projected to; is the mask for missing data; is the feature importance consistency constraint; is a linear transformation matrix used to transform the data in the latent space into a mask of missing data; and is the fuzzy relation constraint; , , , , , and is the trade-off parameter used to balance different items; is the regularization term.
9. A cancer diagnosis model, characterized in that: It includes a multi-constraint representation learning model and a cancer classification model as described in any one of claims 1 to 8.
10. The cancer diagnosis model according to claim 9, characterized in that The steps of determining the cancer classification model include: Using the missing data training set to train the multi-constraint representation learning model to obtain a projection matrix; Projecting the missing data training set into a shared fusion feature data set according to the projection matrix; A plurality of classifiers are trained based on the shared fusion feature data set, and the optimal classifier is selected as the cancer classification model.
Citation Information
Patent Citations
Traffic data filling method based on non-negative low-rank dynamic mode decomposition
CN110188427A
Multi-source time sequence missing data recovery method based on matrix decomposition
CN111881413A
Machine learning-based sepsis early prediction method and system
CN118507071A
