Training methods and systems for predicting the recurrence probability of thrombotic thrombocytopenic purpura; methods and systems for predicting recurrence probability.

By constructing an ensemble learning model, utilizing recursive feature elimination and SMOTETomek sampling to optimize data, and combining multiple base learners and meta-learners, the problem of accurately predicting the recurrence risk of thrombotic thrombocytopenic purpura was solved, achieving high-precision and interpretable prediction results and supporting individualized intervention.

CN123120344APending Publication Date: 2026-11-17EAST CHINA UNIV OF SCI & TECH +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611201704.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-10
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing traditional prediction methods and simple machine learning methods have limitations in terms of intuitiveness and data processing capabilities when dealing with the recurrence risk of thrombotic thrombocytopenic purpura. In particular, they are prone to technical defects such as data leakage, overfitting and boundary noise interference on small sample and extremely imbalanced rare disease datasets, making it difficult to achieve accurate prediction and early intervention.

Method used

A recursive feature elimination method is used to screen key features, and the data is optimized by combining the SMOTETomek hybrid sampling method. Multiple base learners are constructed and meta-learners are constructed through M-fold cross-validation and optimal hyperparameter configuration for iterative training. Finally, the prediction results are calibrated by Platt scaling to form an ensemble learning model.

Benefits of technology

It significantly improves the predictive accuracy and robustness of the risk of recurrence of thrombotic thrombocytopenic purpura, provides the ability to identify key influencing factors and provide individualized intervention, realizes the transformation from traditional empirical medicine to precision medicine, and provides interpretable predictive results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120344A_ABST
    Figure CN123120344A_ABST
Patent Text Reader

Abstract

This invention relates to a training method and system for a recurrence probability prediction model of thrombotic thrombocytopenic purpura. The method includes: acquiring multiple original patient samples; removing redundancy from various clinical indicator features of the original patient samples; screening the redundant samples using a recursive feature elimination method; performing oversampling using the SMOTETomek mixed sampling method to obtain training samples; standardizing each clinical indicator feature to obtain standardized samples; using M-fold cross-validation, iteratively training different types of base learners with each standardized sample to obtain the recurrence probability predicted by each type of base learner for each standardized sample; training a preset meta-learner using the recurrence probability predicted by each type of base learner for each standardized sample as input and the recurrence label of each standardized sample as output; and training each base learner using all standardized samples.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of medical big data and artificial intelligence technology, and more specifically to a training method and system for a prediction model of the recurrence probability of thrombotic thrombocytopenic purpura, and a method and system for predicting the recurrence probability of thrombotic thrombocytopenic purpura. Background Technology

[0002] Thrombotic thrombocytopenic purpura (TTP) is a critical illness caused by ADAMTS13 enzyme deficiency. Its core clinical features include microangiopathic hemolytic anemia, thrombocytopenia, and multiple organ damage. Without timely treatment, the mortality rate can reach 90%. Although first-line treatment with plasma exchange (TPE) and glucocorticoids combined with rituximab has significantly improved prognosis, the disease still faces the challenge of a high relapse rate. In particular, some patients develop refractory or short-lived treatment outcomes after relapse, leading to repeated relapses. This not only seriously threatens patient survival but also poses a challenge to clinicians. Identifying these refractory and relapsed patients in the early stages of the disease and intervening early to improve prognosis remains a clinical challenge.

[0003] With the deepening of the concept of precision medicine, building a system capable of accurately predicting relapse risk and revealing key influencing factors has become a core requirement of clinical decision support systems. Traditional prediction methods mostly rely on simple clinical scoring systems, which, while intuitive, fall short when dealing with multidimensional, heterogeneous, and nonlinear interactive modern clinical data. On the other hand, while pure machine learning methods have a natural advantage in prediction accuracy, they often face the "black box" dilemma and are prone to technical defects such as data leakage, overfitting, and boundary noise interference on small-sample, highly imbalanced rare disease datasets. Summary of the Invention

[0004] The purpose of this invention is to provide a training method and system for a prediction model of the recurrence probability of thrombotic thrombocytopenic purpura (TTP), and a method and system for predicting the recurrence probability of TTP, so as to predict the recurrence probability of TTP through a data-driven approach and identify key factors affecting recurrence, achieve early warning and individualized precise intervention, and promote the transformation of TTP diagnosis and treatment from traditional experience-based medicine to precision medicine.

[0005] To achieve the above objectives, the present invention provides, in one aspect, a training method for a predictive model of the recurrence probability of thrombotic thrombocytopenic purpura, comprising:

[0006] We obtained multiple original patient samples, each of which included multiple clinical indicators and corresponding relapse labels.

[0007] Redundant clinical indicator features were screened out from the original clinical indicator features of each patient's sample and then removed to obtain each initial screening simplified sample.

[0008] A recursive feature elimination method was used to iteratively screen the clinical indicator features of each preliminary screening simplified sample to form each core feature sample;

[0009] The SMOTETomek hybrid sampling method is used to optimize the core feature samples to obtain training samples with class balance and clean boundaries.

[0010] The clinical indicator characteristics of each training sample are standardized to obtain standardized samples.

[0011] Construct multiple different types of base learners, where each type of base learner is a machine learning model with optimal hyperparameters;

[0012] For each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of that type using each standardized sample, so as to obtain the recurrence probability predicted by the base learner of that type for each standardized sample.

[0013] The pre-defined meta-learner is trained by taking the recurrence probability of each standardized sample predicted by various types of base learners as input and the recurrence label of each standardized sample as output.

[0014] Each type of base learner is trained using all standardized samples to obtain the final trained base learner for each type.

[0015] Optionally, various types of base learners can be constructed, including:

[0016] Multiple different types of machine learning models are pre-selected, and for each type of machine learning model, multiple sets of different hyperparameter combinations are configured.

[0017] For each type of machine learning model, multiple candidate models are constructed based on its corresponding multiple sets of hyperparameters.

[0018] For all candidate models of each type, K-fold cross-validation is used to train and evaluate the models using standardized samples, where K is a positive integer greater than 1. Based on the performance evaluation results, the candidate model with the best performance is selected, and the corresponding optimal hyperparameter combination is determined.

[0019] Based on each type of machine learning model and its corresponding optimal hyperparameter combination, construct the optimal model for that type, which serves as the base learner for that type.

[0020] Optionally, for each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of that type using each standardized sample, obtaining the recurrence probability predicted by the base learner for each standardized sample, specifically including:

[0021] Each standardized sample is divided into M parts. Using the M-fold cross-validation method, one part is selected as the validation set and the remaining M-1 parts are selected as the training set. The base learner of each type is trained fold by fold and the prediction results of the validation set are output.

[0022] The prediction results of each fold validation set are summarized and spliced ​​to obtain the recurrence probability corresponding to each standardized sample predicted by the base learner of this type.

[0023] Optionally, the various types of base learners are random forest models, XGBoost models, and LightGBM models, and the meta-learner is a logistic regression model.

[0024] Another aspect of the present invention provides a method for predicting the recurrence probability of thrombotic thrombocytopenic purpura, comprising:

[0025] Obtain the core clinical indicators of the patients to be predicted;

[0026] The core clinical indicators of the patients to be predicted are standardized to obtain the standardized characteristics of the patients to be predicted.

[0027] The standardized features of the patient to be predicted are input into the various types of base learners that have been trained in the end, so as to obtain the relapse probability predicted by the various types of base learners.

[0028] The relapse probabilities predicted by various types of base learners are input into a pre-trained meta-learner to obtain the relapse probability of the patient to be predicted.

[0029] The recurrence probability of the patient to be predicted is processed by Platt scaling to obtain a calibrated recurrence probability that follows the true probability distribution;

[0030] The final trained base learner and the pre-trained meta-learner were obtained using the training method described above for the prediction model of recurrence probability of thrombotic thrombocytopenic purpura.

[0031] Another aspect of the present invention provides a training system for a predictive model of the recurrence probability of thrombotic thrombocytopenic purpura, comprising:

[0032] The first acquisition module is used to acquire multiple patient original samples. Each patient original sample includes multiple clinical indicator characteristics of the patient and the corresponding relapse label.

[0033] The elimination module is used to screen out redundant clinical indicator features from the various clinical indicator features of each patient's original sample and remove the redundant clinical indicator features to obtain each initial screening simplified sample.

[0034] The screening module is used to iteratively screen the clinical indicator features of each preliminary screening simplified sample using a recursive feature elimination method to form each core feature sample.

[0035] The sampling module is used to optimize the core feature samples using the SMOTETomek hybrid sampling method to obtain training samples with balanced classes and clean boundaries.

[0036] The first standardization module is used to standardize the clinical indicator characteristics of each training sample to obtain standardized samples.

[0037] The building blocks are used to build various types of base learners, where each type of base learner is a machine learning model with optimal hyperparameters;

[0038] The first training module is used to iteratively train each type of base learner using M-fold cross-validation and standardized samples to obtain the recurrence probability predicted by the base learner for each standardized sample.

[0039] The second training module is used to train the preset meta-learner by taking the recurrence probability of each standardized sample predicted by various types of base learners as input and the recurrence label of each standardized sample as output, so as to obtain the trained meta-learner.

[0040] The third training module is used to train each type of base learner using all standardized samples to obtain the final trained base learner for each type.

[0041] Optionally, various types of base learners can be constructed, including:

[0042] Multiple different types of machine learning models are pre-selected, and for each type of machine learning model, multiple sets of different hyperparameter combinations are configured.

[0043] For each type of machine learning model, multiple candidate models are constructed based on its corresponding multiple sets of hyperparameters.

[0044] For all candidate models of each type, K-fold cross-validation is used to train and evaluate the models using standardized samples, where K is a positive integer greater than 1. Based on the performance evaluation results, the candidate model with the best performance is selected, and the corresponding optimal hyperparameter combination is determined.

[0045] Based on each type of machine learning model and its corresponding optimal hyperparameter combination, construct the optimal model for that type, which serves as the base learner for that type.

[0046] Optionally, for each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of that type using each standardized sample, obtaining the recurrence probability predicted by the base learner for each standardized sample, specifically including:

[0047] Each standardized sample is divided into M parts. Using the M-fold cross-validation method, one part is selected as the validation set and the remaining M-1 parts are selected as the training set. The base learner of each type is trained fold by fold and the prediction results of the validation set are output.

[0048] The prediction results of each fold validation set are summarized and spliced ​​to obtain the recurrence probability corresponding to each standardized sample predicted by the base learner of this type.

[0049] Optionally, the various types of base learners are random forest models, XGBoost models, and LightGBM models, and the meta-learner is a logistic regression model.

[0050] In another aspect, the present invention provides a system for predicting the recurrence probability of thrombotic thrombocytopenic purpura, comprising:

[0051] The second acquisition module is used to acquire the core clinical indicator characteristics of the patient to be predicted.

[0052] The second standardization module is used to standardize the core clinical indicators of the patient to be predicted, so as to obtain the standardized characteristics of the patient to be predicted.

[0053] The first prediction module is used to input the standardized features of the patient to be predicted into the various types of base learners that have been trained in the end, so as to obtain the recurrence probability predicted by the various types of base learners.

[0054] The second prediction module is used to input the recurrence probability predicted by various types of base learners into the pre-trained thrombotic thrombocytopenic purpura recurrence probability prediction model to obtain the recurrence probability of the patient to be predicted.

[0055] The calibration module is used to perform Platt scaling on the recurrence probability of the patient to be predicted, so as to obtain the calibrated recurrence probability.

[0056] The final trained base learner and the pre-trained meta-learner were obtained using the training method described above for the prediction model of recurrence probability of thrombotic thrombocytopenic purpura. Attached Figure Description

[0057] Figure 1A flowchart illustrating the training method for a prediction model of the recurrence probability of thrombotic thrombocytopenic purpura according to an embodiment of the present invention;

[0058] Figure 2 A flowchart of a method for predicting the recurrence probability of thrombotic thrombocytopenic purpura according to an embodiment of the present invention;

[0059] Figure 3 This is a structural block diagram of the training system for a prediction model of the recurrence probability of thrombotic thrombocytopenic purpura according to an embodiment of the present invention.

[0060] Figure 4 This is a structural block diagram of a system for predicting the recurrence probability of thrombotic thrombocytopenic purpura according to an embodiment of the present invention. Detailed Implementation

[0061] The preferred embodiments of the present invention are given below with reference to the accompanying drawings and described in detail.

[0062] like Figure 1 As shown in the figure, this embodiment of the invention provides a training method for a predictive model of the recurrence probability of thrombotic thrombocytopenic purpura, which includes the following steps:

[0063] S110: Obtain multiple original patient samples, each of which includes multiple clinical indicators and corresponding relapse labels.

[0064] The original patient samples can be obtained from the hospital's clinical database. For example, the original clinical data can be obtained from the clinical database, which includes the patient's TTP-related clinical indicators and the patient's relapse status (i.e., relapse label, including relapse or no relapse). Then, the original clinical data is preprocessed, including: digitizing the clinical indicators and relapse status to form machine learning-readable data; and filling in missing indicators by using the mean of the effective indicator data of all patients under the corresponding clinical indicator to obtain a complete sample dataset, i.e., the original sample of each patient.

[0065] Clinical indicators refer to indicators related to TTP, including but not limited to the following:

[0066] 1) Basic clinical information:

[0067] Secondary diagnosis, creatinine <2 mg / dL (46–92), presence or absence of identifiable precipitating factors (infection / drugs / tumors / autoimmune diseases / hematopoietic stem cell transplantation, etc.).

[0068] 2) Pentadala symptoms: neurological symptoms, anemia, fever, kidney damage (46–92 umol / L), and thrombocytopenia.

[0069] 3) Laboratory indicators:

[0070] Platelet count <30, reticulocyte count >2.5%, number of TPEs, total number of CD20s during the first treatment, whether CD20 was used as a first-line drug (within 72 hours after TPE), whether it is refractory iTTP (more than 5 TPEs), whether organ function has recovered, time interval for TTP group to turn negative, time interval for the first CD20, time interval for the first platelet count above 100, time interval for the first clinical remission (i.e., plt stable above 100), and time interval for TPEs.

[0071] 4) Disease outcome: whether the patient is in the ICU, whether there is a relapse, whether there is clinical deterioration, and whether the diagnosis was misdiagnosed.

[0072] 5) Symptoms and signs (which can be further subdivided into many sub-features): neurological symptoms, other symptoms, diagnosis.

[0073] 6) Autoantibodies: ANA (antinuclear antibody), anti-SSA antibody, anti-SSB antibody, dsDNA (anti-double-stranded DNA antibody), ENA (anti-extractable nuclear antigen antibody profile), and other autoantibodies.

[0074] S120: Redundant clinical indicator features are selected from the original clinical indicator features of each patient's sample and then removed to obtain each initial screening simplified sample.

[0075] Redundant clinical indicator features refer to features that have a high similarity to any one of the retained clinical indicator data. That is, for several clinical indicator features with high similarity, only one can be retained, while the others are removed. Specifically, for each clinical indicator, the clinical indicator features of each patient sample can be formed into an indicator vector for that clinical indicator. Then, a clustering algorithm is used to classify the indicator vectors of each clinical indicator. After classification, only one clinical indicator is retained in each class as the retained clinical indicator, and the rest are redundant clinical indicators. Then, the redundant clinical indicator features of each patient sample can be removed to obtain the input feature set of that patient sample. Alternatively, the clinical indicators can be made into an original indicator set, and then each clinical indicator in the original indicator set can be traversed in a preset order. For the currently traversed clinical indicator, the similarity (e.g., Pearson correlation coefficient) between the indicator vector of that clinical indicator and the indicator vectors of each of the remaining clinical indicators is calculated. If the indicator vector of that clinical indicator is similar to the indicator vectors of the remaining clinical indicators, then the redundant clinical indicator is considered redundant. If the similarity between any indicator vector in the set is greater than or equal to a preset threshold (e.g., the Pearson correlation coefficient exceeds 0.9), then that clinical indicator is considered a redundant clinical indicator and placed in the redundant indicator set. If the similarity between the indicator vector of that clinical indicator and the indicator vectors of all other clinical indicators is less than the preset threshold, then the indicator vector of that clinical indicator is considered a retained clinical indicator and placed in the retained indicator set. After the current traversal is completed, the clinical indicator of the current traversal is deleted from the original indicator set to obtain an updated original indicator set. Then, the above traversal process is repeated on the updated original indicator set until the original indicator set is empty. Finally, a redundant indicator set and a retained indicator set can be obtained. Then, the redundant clinical indicator features of each patient's redundant indicator set are removed to obtain the initial screening simplified sample of that patient.

[0076] S130: The recursive feature elimination (RFE) method is used to iteratively screen the clinical indicator features of each preliminary screening simplified sample, remove non-critical clinical indicator features with low feature importance, retain core clinical features with high discriminative power, and form samples of each core feature.

[0077] RFE is a wrapper-style feature selection method that iteratively builds the model and gradually removes the least important features. Its core mechanism is to evaluate the marginal contribution of each feature to the model performance in each iteration and remove the feature with the smallest contribution until the preset number of features or the optimal performance is reached.

[0078] S140: The SMOTETomek hybrid sampling method is used to optimize the core feature samples to obtain training samples with balanced categories and clean boundaries.

[0079] SMOTETomek is a combined method of SMOTE (Synthetic Minority Oversampling Technique) and Tomek Links (Boundary Noise Cleaning). Addressing the class imbalance problem caused by the scarcity of relapsed samples in TTP (Transient Oversampling Technique), this method first generates synthetic samples by interpolating minority (relapsed) samples in the feature space. Then, the Tomek Links algorithm identifies and removes noisy sample pairs located near the decision boundary, thereby achieving a more balanced distribution of relapsed and non-relapsed samples and clearer class boundaries.

[0080] Step S140 may specifically include:

[0081] The core feature samples are divided into majority and minority class samples based on their recurrence labels.

[0082] The SMOTE oversampling method is used to first perform feature space interpolation and synthesis on minority class samples to generate new minority class synthetic samples. For example, random interpolation is performed on minority class samples in the k-nearest neighbor space to generate synthetic points on the line connecting two real relapse samples. The synthetic samples are distributed in the neighborhood of known relapse patients in the feature space, maintaining the local manifold structure of relapse features. Then, the Tomek Links rule is used to identify and remove overlapping noise samples across class boundaries in each core feature sample. For example, the nearest neighbor samples in the feature space with opposite labels (one relapsed and one non-relapsed) are identified. If the two samples are very close, it means that the indicators of the two cases are very similar, but the results are completely opposite. Therefore, the sample belonging to the majority class (non-relapsed) is deleted, thereby cleaning up the "boundary noise" near the decision boundary and making the boundary between the two classes clearer.

[0083] By merging newly added minority class synthetic samples with each core feature sample after noise removal, training samples with class balance and clean boundaries are constructed.

[0084] S150: Standardize the clinical indicator characteristics of each training sample to obtain standardized samples.

[0085] The purpose of standardization is to eliminate size differences between clinical indicators with different dimensions, ensuring that all clinical indicator features participate in subsequent processing at the same size. For example, Z-score standardization can be used to transform each clinical indicator feature into a distribution with a mean of 0 and a standard deviation of 1. Specifically, for each feature x, its standardized value is z, calculated using the following formula:

[0086]

[0087] Where μ is the mean of the feature in each training sample, and σ is the standard deviation of the feature in each training sample.

[0088] S160: Construct multiple different types of base learners, where each type of base learner is a machine learning model with optimal hyperparameters.

[0089] In some embodiments, step S600 specifically includes:

[0090] Multiple different types of machine learning models are pre-selected, and for each type of machine learning model, multiple sets of different hyperparameter combinations are configured; for example, multiple sets of hyperparameter combinations can be randomly generated based on Monte Carlo simulation.

[0091] For each type of machine learning model, multiple candidate models are constructed based on its corresponding multiple sets of hyperparameters.

[0092] For all candidate models of each type, K-fold cross-validation is used to train and evaluate the models using standardized samples, where K is a positive integer greater than 1. Based on the performance evaluation results, the candidate model with the best performance is selected, and the corresponding optimal hyperparameter combination is determined.

[0093] Based on each type of machine learning model and its corresponding optimal hyperparameter combination, construct the optimal model for that type, which serves as the base learner for that type.

[0094] For example, the base learner can include three types: Random Forest, XGBoost (Extreme Gradient Boosting Tree), and LightGBM (Lightweight Gradient Boosting Machine). The hyperparameters of the Random Forest model include the number of trees, the maximum tree depth, and the minimum number of samples per leaf node. The hyperparameters of the XGBoost model include the learning rate, the sample sampling ratio, and the positive and negative sample weights. The hyperparameters of the LightGBM model include the number of leaf nodes, the feature sampling ratio, and the L1 regularization coefficient. In performance evaluation, ROC-AUC is used as the performance metric for the Random Forest model, measuring its ability to distinguish between relapsed and non-relapsed samples across all thresholds. F1-Score is used as the performance metric for the XGBoost and LightGBM models, emphasizing the balance between precision and recall.

[0095] S170: For each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of this type using each standardized sample to obtain the recurrence probability predicted by the base learner of this type for each standardized sample; where M is a positive integer greater than 1.

[0096] Step S170 specifically includes:

[0097] Each standardized sample is divided into M parts. Using M-fold cross-validation, one part is selected as the validation set, and the remaining M-1 parts are used as the training set. The base learner of each type is trained fold by fold, and the prediction results of the validation set are output. The prediction results of the validation set of each fold are summarized and spliced ​​to obtain the prediction results corresponding to each standardized sample, that is, the recurrence probability of each standardized sample predicted by the base learner of that type.

[0098] S180: Using the recurrence probabilities of each standardized sample predicted by various types of base learners as input and the recurrence labels of each standardized sample as output, the preset meta-learners are trained to obtain the trained meta-learners.

[0099] The meta-learner can be a logistic regression model, which adaptively learns the optimal fusion weights of the recurrence probabilities predicted by the three base learners by maximizing the log-likelihood function. Its final prediction form is:

[0100]

[0101] Where P is the recurrence probability predicted by the meta-learner, and β RF The weights of the recurrence probability predicted by the random forest model, P RF β is the recurrence probability predicted by the random forest model. XGB P represents the weights of the recurrence probability predicted by the XGBoost model. XGB β represents the recurrence probability predicted by the XGBoost model. LGBM P represents the weights of the recurrence probability predicted by the LightGBM model. LGBM β0 represents the recurrence probability predicted by LightGBM, and β0 represents the intercept (or bias term) of logistic regression.

[0102] After training, the meta-learner can also output 95% confidence intervals for three weights, which clinicians can use to understand which base learner played a dominant role.

[0103] S190: Train each type of base learner using all standardized samples to obtain the final trained base learner for each type.

[0104] Since the base learners obtained in step S170 are only trained using a subset of standardized samples, their generalization performance is insufficient. Therefore, it is necessary to train each type of base learner separately using all standardized samples to obtain the final trained base learners of each type. In subsequent predictions, the standardized clinical indicators of the patient to be predicted are first input into each type of base learner to obtain the predicted relapse probability. Then, the predicted relapse probabilities of each type of base learner are input into the trained meta-learner, so that the meta-learner outputs the corresponding predicted relapse result, which is the final prediction result.

[0105] The training method of the recurrence probability prediction model for thrombotic thrombocytopenic purpura in this invention utilizes ensemble learning to capture high-order nonlinear relationships between features, significantly improving prediction accuracy for small samples. The SMOTETomek joint method actively cleans boundary noise while supplementing the sample size, significantly improving the accuracy and robustness of identifying key influencing factors. The Stacking+LR meta-learner architecture achieves statistical interpretability of the multi-algorithm fusion process, overcoming the "fusion black box" deficiency of existing machine learning solutions. Compared to the simple voting or equal-weighted averaging of existing solutions, the meta-learner of this invention adaptively learns the contribution weights of each base model and outputs a weight coefficient report with p-values ​​and confidence intervals, enabling clinicians to understand "how much each model contributed to this prediction" in statistical language. This reintegrates the ensemble output of machine learning into a quantifiable and verifiable statistical framework. The method organically combines factor contribution quantification, patient clustering topology verification, and algorithm fusion weight interpretation, providing clinicians with a complete analytical chain from "which factors are most critical" to "how high-risk patient groups cluster" and then to "how various algorithms collaborate to reach conclusions."

[0106] like Figure 2 As shown in the figure, this invention also provides a method for predicting the recurrence probability of thrombotic thrombocytopenic purpura, which includes the following steps:

[0107] S210: Obtain the core clinical characteristics of the patient to be predicted;

[0108] S220: Standardize the core clinical indicators of the patient to be predicted to obtain the standardized characteristics of the patient to be predicted.

[0109] S230: Input the standardized features of the patient to be predicted into the various types of base learners that have been trained in the end, so as to obtain the recurrence probability predicted by the various types of base learners; the various types of base learners that have been trained in the end are obtained by the training method in step S190;

[0110] S240: Input the recurrence probabilities predicted by various types of base learners into the pre-trained thrombotic thrombocytopenic purpura recurrence probability prediction model to obtain the recurrence probability of the patient to be predicted; the pre-trained thrombotic thrombocytopenic purpura recurrence probability prediction model is obtained by the training method in step S180.

[0111] S250: The recurrence probability of the patient to be predicted is processed by Platt scaling to obtain a calibrated recurrence probability that follows the true probability distribution.

[0112] Because the raw output probabilities of ensemble tree models often have systematic biases (too extreme or conservative) and cannot be directly used as clinical probability references, Platt scaling is used to calibrate the recurrence probability of the patients to be predicted. The calibrated recurrence probability is then quantitatively validated using reliability curves and the Brier Score to ensure clinical consistency between the "output 60% recurrence probability" and the "actual 60% recurrence rate." The calibrated recurrence probability can be directly used as an individualized recurrence risk score for clinical stratification management.

[0113] like Figure 3 As shown, this embodiment of the invention also provides a training system for a prediction model of the recurrence probability of thrombotic thrombocytopenic purpura, which includes: a first acquisition module 11, a rejection module 12, a screening module 13, a sampling module 14, a first standardization module 15, a construction module 16, a first training module 17, a second training module 18, and a third training module 19.

[0114] The first acquisition module 11 is used to acquire multiple patient original samples, each of which includes multiple clinical indicator characteristics of the patient and the corresponding relapse label.

[0115] The elimination module 12 is used to screen out redundant clinical indicator features from the various clinical indicator features of each patient's original sample, and remove the redundant clinical indicator features to obtain each initial screening simplified sample.

[0116] The screening module 13 is used to iteratively screen the clinical indicator features of each preliminary screening simplified sample using a recursive feature elimination method to form each core feature sample.

[0117] The sampling module 14 is used to optimize the core feature samples using the SMOTETomek hybrid sampling method to obtain training samples with balanced categories and clean boundaries.

[0118] The first standardization module 15 is used to standardize the clinical indicator characteristics of each training sample to obtain each standardized sample.

[0119] Module 16 is used to build various types of base learners, where each type of base learner is a machine learning model with optimal hyperparameters.

[0120] The first training module 17 is used to iteratively train the base learner of each type using M-fold cross-validation with each standardized sample to obtain the recurrence probability predicted by the base learner of each type for each standardized sample.

[0121] The second training module 18 is used to train a preset meta-learner with the recurrence probability of each standardized sample predicted by various types of base learners as input and the recurrence label of each standardized sample as output, so as to obtain a trained meta-learner as a trained model for predicting the recurrence probability of thrombotic thrombocytopenic purpura.

[0122] The third training module 19 is used to train each type of base learner using all standardized samples to obtain the final trained base learner for each type.

[0123] The first acquisition module 11, the elimination module 12, the filtering module 13, the sampling module 14, the first standardization module 15, the construction module 16, the first training module 17, the second training module 18, and the third training module 19 are functional modules corresponding to steps S110-S190, respectively. For detailed descriptions, please refer to the descriptions in steps S110-S190, which will not be repeated here.

[0124] like Figure 4 As shown, this embodiment of the invention also provides a prediction system for the recurrence probability of thrombotic thrombocytopenic purpura, which includes a second acquisition module 21, a second standardization module 22, a first prediction module 23, a second prediction module 24, and a calibration module 25.

[0125] The second acquisition module 21 is used to acquire the core clinical indicator characteristics of the patient to be predicted.

[0126] The second standardization module 22 is used to standardize the core clinical indicator characteristics of the patient to be predicted, so as to obtain the standardized characteristics of the patient to be predicted.

[0127] The first prediction module 23 is used to input the standardized features of the patient to be predicted into various types of base learners that have been trained in the end, so as to obtain the recurrence probability predicted by various types of base learners.

[0128] The second prediction module 24 is used to input the recurrence probabilities predicted by various types of base learners into a pre-trained thrombotic thrombocytopenic purpura recurrence probability prediction model to obtain the recurrence probability of the patient to be predicted.

[0129] The calibration module 25 is used to perform Platt scaling on the recurrence probability of the patient to be predicted, so as to obtain the calibrated recurrence probability.

[0130] The second acquisition module 21, the second standardization module 22, the first prediction module 23, the second prediction module 24, and the calibration module 25 are functional modules corresponding to steps S210-S250, and their detailed descriptions can be found in steps S210-S250, which will not be repeated here.

[0131] Another embodiment of the present invention provides a readable storage medium storing a computer program, which, when executed in a computer, causes the computer to perform the steps of the training method or the method for predicting the recurrence probability of thrombotic thrombocytopenic purpura recurrence as described in the above embodiments of the present invention.

[0132] Another embodiment of the present invention provides an electronic device, which includes a memory and a processor. The memory stores executable code. When the processor executes the executable code, it performs the steps of the training method or the method for predicting the recurrence probability of thrombotic thrombocytopenic purpura recurrence as described in the above embodiments of the present invention.

[0133] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0134] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0135] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] In a typical configuration, an electronic device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0140] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0141] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by electronic devices. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0142] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0143] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0145] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0146] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0147] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. That is, all simple and equivalent changes and modifications made based on the claims and description of this invention fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.

Claims

1. A training method for a predictive model of recurrence probability in thrombotic thrombocytopenic purpura, characterized in that, include: We obtained multiple original patient samples, each of which included multiple clinical indicators and corresponding relapse labels. Redundant clinical indicator features were screened out from the original clinical indicator features of each patient's sample and then removed to obtain each initial screening simplified sample. A recursive feature elimination method was used to iteratively screen the clinical indicator features of each preliminary screening simplified sample to form each core feature sample; The SMOTETomek hybrid sampling method is used to optimize the core feature samples to obtain training samples with class balance and clean boundaries. The clinical indicator characteristics of each training sample are standardized to obtain standardized samples. Construct multiple different types of base learners, where each type of base learner is a machine learning model with optimal hyperparameters; For each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of that type using each standardized sample, so as to obtain the recurrence probability predicted by the base learner of that type for each standardized sample. The pre-defined meta-learner is trained by taking the recurrence probability of each standardized sample predicted by various types of base learners as input and the recurrence label of each standardized sample as output. Each type of base learner is trained using all standardized samples to obtain the final trained base learner for each type.

2. The training method for the recurrence probability prediction model of thrombotic thrombocytopenic purpura according to claim 1, characterized in that, Construct various types of base learners, specifically including: Multiple different types of machine learning models are pre-selected, and for each type of machine learning model, multiple sets of different hyperparameter combinations are configured. For each type of machine learning model, multiple candidate models are constructed based on its corresponding multiple sets of hyperparameters. For all candidate models of each type, K-fold cross-validation is used to train and evaluate the models using standardized samples, where K is a positive integer greater than 1. Based on the performance evaluation results, the candidate model with the best performance is selected, and the corresponding optimal hyperparameter combination is determined. Based on each type of machine learning model and its corresponding optimal hyperparameter combination, construct the optimal model for that type, which serves as the base learner for that type.

3. The training method for the recurrence probability prediction model of thrombotic thrombocytopenic purpura according to claim 1, characterized in that, For each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of that type using standardized samples, obtaining the recurrence probability predicted by the base learner for each standardized sample, specifically including: Each standardized sample is divided into M parts. Using the M-fold cross-validation method, one part is selected as the validation set and the remaining M-1 parts are selected as the training set. The base learner of each type is trained fold by fold and the prediction results of the validation set are output. The prediction results of each fold validation set are summarized and spliced ​​to obtain the recurrence probability corresponding to each standardized sample predicted by the base learner of this type.

4. The training method for the recurrence probability prediction model of thrombotic thrombocytopenic purpura according to claim 1, characterized in that, The various types of base learners are random forest model, XGBoost model and LightGBM model, and the meta-learner is logistic regression model.

5. A method for predicting the recurrence probability of thrombotic thrombocytopenic purpura, characterized in that, include: Obtain the core clinical indicators of the patients to be predicted; The core clinical indicators of the patients to be predicted are standardized to obtain the standardized characteristics of the patients to be predicted. The standardized features of the patient to be predicted are input into the various types of base learners that have been trained in the end, so as to obtain the relapse probability predicted by the various types of base learners. The relapse probabilities predicted by various types of base learners are input into a pre-trained meta-learner to obtain the relapse probability of the patient to be predicted. The recurrence probability of the patient to be predicted is processed by Platt scaling to obtain a calibrated recurrence probability that follows the true probability distribution; The final trained base learner and the pre-trained meta learner are obtained using the training method of the thrombotic thrombocytopenic purpura recurrence probability prediction model as described in any one of claims 1-4.

6. A training system for a predictive model of recurrence probability of thrombotic thrombocytopenic purpura, characterized in that, include: The first acquisition module is used to acquire multiple patient original samples. Each patient original sample includes multiple clinical indicator characteristics of the patient and the corresponding relapse label. The elimination module is used to screen out redundant clinical indicator features from the various clinical indicator features of each patient's original sample and remove the redundant clinical indicator features to obtain each initial screening simplified sample. The screening module is used to iteratively screen the clinical indicator features of each preliminary screening simplified sample using a recursive feature elimination method to form each core feature sample. The sampling module is used to optimize the core feature samples using the SMOTETomek hybrid sampling method to obtain training samples with balanced classes and clean boundaries. The first standardization module is used to standardize the clinical indicator characteristics of each training sample to obtain standardized samples. The building blocks are used to build various types of base learners, where each type of base learner is a machine learning model with optimal hyperparameters; The first training module is used to iteratively train each type of base learner using M-fold cross-validation and standardized samples to obtain the recurrence probability predicted by the base learner for each standardized sample. The second training module is used to train the preset meta-learner by taking the recurrence probability of each standardized sample predicted by various types of base learners as input and the recurrence label of each standardized sample as output, so as to obtain the trained meta-learner. The third training module is used to train each type of base learner using all standardized samples to obtain the final trained base learner for each type.

7. The training system for the recurrence probability prediction model of thrombotic thrombocytopenic purpura according to claim 6, characterized in that, Construct various types of base learners, specifically including: Multiple different types of machine learning models are pre-selected, and for each type of machine learning model, multiple sets of different hyperparameter combinations are configured. For each type of machine learning model, multiple candidate models are constructed based on its corresponding multiple sets of hyperparameters. For all candidate models of each type, K-fold cross-validation is used to train and evaluate the models using standardized samples, where K is a positive integer greater than 1. Based on the performance evaluation results, the candidate model with the best performance is selected, and the corresponding optimal hyperparameter combination is determined. Based on each type of machine learning model and its corresponding optimal hyperparameter combination, construct the optimal model for that type, which serves as the base learner for that type.

8. The training system for the recurrence probability prediction model of thrombotic thrombocytopenic purpura according to claim 6, characterized in that, For each type of base learner, an M-fold cross-validation method is used to iteratively train the base learner of that type using standardized samples, obtaining the recurrence probability predicted by the base learner for each standardized sample, specifically including: Each standardized sample is divided into M parts. Using the M-fold cross-validation method, one part is selected as the validation set and the remaining M-1 parts are selected as the training set. The base learner of each type is trained fold by fold and the prediction results of the validation set are output. The prediction results of each fold validation set are summarized and spliced ​​to obtain the recurrence probability corresponding to each standardized sample predicted by the base learner of this type.

9. The training system for the recurrence probability prediction model of thrombotic thrombocytopenic purpura according to claim 6, characterized in that, The various types of base learners are random forest model, XGBoost model and LightGBM model, and the meta-learner is logistic regression model.

10. A system for predicting the recurrence probability of thrombotic thrombocytopenic purpura, characterized in that, include: The second acquisition module is used to acquire the core clinical indicator characteristics of the patient to be predicted. The second standardization module is used to standardize the core clinical indicators of the patient to be predicted, so as to obtain the standardized characteristics of the patient to be predicted. The first prediction module is used to input the standardized features of the patient to be predicted into the various types of base learners that have been trained in the end, so as to obtain the recurrence probability predicted by the various types of base learners. The second prediction module is used to input the recurrence probability predicted by various types of base learners into the pre-trained thrombotic thrombocytopenic purpura recurrence probability prediction model to obtain the recurrence probability of the patient to be predicted. The calibration module is used to perform Platt scaling on the recurrence probability of the patient to be predicted, so as to obtain the calibrated recurrence probability. The final trained base learner and the pre-trained meta learner are obtained using the training method of the thrombotic thrombocytopenic purpura recurrence probability prediction model as described in any one of claims 1-4.