CAR-T therapy response prediction model optimization method and system based on interpretability analysis

Abstract

The invention discloses a CAR-T therapy response prediction model optimization method and system based on interpretability analysis. The method comprises the following steps: acquiring a trained CAR-T therapy response prediction model; performing global interpretability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result; performing local interpretability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result; and performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model. According to the method, global interpretability analysis and local interpretability analysis are performed on the trained CAR-T therapy response prediction model, and model optimization is performed on the trained CAR-T therapy response prediction model based on the analysis result, so that the interpretability and prediction accuracy of the CAR-T therapy response prediction model are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical artificial intelligence, in particular to a CAR-T therapy response prediction model optimization method and system based on explainability analysis. BACKGROUND

[0002] Currently, the commonly used CAR-T therapy response prediction models in clinical practice mainly include the scoring system based on the international prognostic index (IPI) and the traditional logistic regression model. These models usually rely on a few features (such as lactate dehydrogenase (LDH) level, number of extranodal lesions, etc.) to set thresholds for scoring, thereby predicting the treatment response of patients. However, these models have the following main limitations: deep learning models are usually considered as "black box models", lacking the ability to explain the prediction results, making it difficult to gain the trust of clinicians and patients.

[0003] With the widespread application of machine learning and artificial intelligence technology in the medical field, the explainability of the model has become a key issue. Explainability analysis aims to reveal the basis of model prediction, helping clinicians understand the decision-making process of the model, thereby increasing the trust in the model. The existing explainability analysis methods of the model can only provide global or local feature importance, and cannot fully reveal the decision-making process of the model, and the existing model cannot adjust the prediction results in real time according to the dynamic changes in the patient's treatment process, resulting in inaccurate prediction results.

[0004] Therefore, there is an urgent need for a CAR-T therapy response prediction model optimization method based on explainability analysis, which can improve the explainability and prediction accuracy of the CAR-T efficacy prediction model. SUMMARY

[0005] The main purpose of the present application is to provide a CAR-T therapy response prediction model optimization method and system based on explainability analysis, which aims to solve the technical problem of how to enhance the explainability and prediction accuracy of the CAR-T efficacy prediction model in the prior art.

[0006] To achieve the above-mentioned purpose, the present application provides a CAR-T therapy response prediction model optimization method based on explainability analysis, which comprises the following steps:

[0007] Obtaining a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data;

[0008] Performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result;

[0009] perform local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result;

[0010] perform model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model.

[0011] Optionally, the step of performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result comprises:

[0012] evaluate, by a permutation feature importance algorithm, a contribution amount of each feature in the electronic health record data to a prediction result corresponding to the trained CAR-T therapy response prediction model;

[0013] sort the contribution amount of each feature in the electronic health record data to obtain a sorting result;

[0014] determine, based on the sorting result, a key feature that has an impact on the prediction result, and take the key feature as the global analysis result.

[0015] Optionally, the step of evaluating, by a permutation feature importance algorithm, a contribution amount of each feature in the electronic health record data to a prediction result corresponding to the trained CAR-T therapy response prediction model comprises:

[0016] perform multiple random perturbations on each feature in the electronic health record data by a permutation feature importance algorithm;

[0017] determine a change in accuracy of a prediction result corresponding to the trained CAR-T therapy response prediction model after each random perturbation;

[0018] evaluate, based on the changes in accuracy, the contribution amount of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model.

[0019] Optionally, the step of performing local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result comprises:

[0020] determine all feature subsets of each patient in the electronic health record data;

[0021] determine, for each patient, a marginal contribution value of each feature in the corresponding all feature subsets by using a SHAP algorithm;

[0022] determine a SHAP value of each feature by weighted average based on the marginal contribution value;

[0023] Determining, according to the SHAP value of each feature, the individualized impact of each feature on the prediction result corresponding to the trained CAR-T therapy response prediction model;

[0024] The individualized impact corresponding to each of the features is used as the local analysis result.

[0025] Optionally, the step of optimizing the trained CAR-T therapy response prediction model based on the global analysis results and the local analysis results to obtain an optimized prediction model includes:

[0026] Screening key predictive factors corresponding to the trained CAR-T therapy response prediction model according to the global analysis results and the local analysis results;

[0027] The trained CAR-T therapy response prediction model is optimized based on the key predictive factors to obtain an optimized prediction model.

[0028] Optionally, the step of optimizing the trained CAR-T therapy response prediction model based on the key predictive factors to obtain an optimized prediction model includes:

[0029] Optimizing the training data set of the trained CAR-T therapy response prediction model based on the key predictive factors to obtain an optimized feature set;

[0030] retraining the trained CAR-T therapy response prediction model based on the optimized feature set to obtain a retraining result;

[0031] The trained CAR-T therapy response prediction model is optimized according to the retraining results to obtain an optimized prediction model.

[0032] Optionally, after the step of optimizing the trained CAR-T therapy response prediction model based on the global analysis results and the local analysis results to obtain the optimized prediction model, the method further includes:

[0033] Determining a model evaluation index of the optimized prediction model;

[0034] Based on the model evaluation index, the performance of the optimized prediction model is evaluated using a ten-fold cross validation method to obtain a model evaluation result.

[0035] In addition, to achieve the above objectives, the present invention also proposes a CAR-T therapy response prediction model optimization system based on interpretability analysis, the system comprising:

[0036] The model obtaining module is configured to obtain a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data.

[0037] The global analysis module is configured to perform global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result.

[0038] The local analysis module is configured to perform local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result.

[0039] The model optimization module is configured to perform model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model.

[0040] Optionally, the global analysis module is further configured to perform multiple random perturbations on each feature in the electronic health record data by using a permutation feature importance algorithm, to determine the accuracy rate change of the prediction result corresponding to the trained CAR-T therapy response prediction model after each random perturbation, and to evaluate the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model based on the accuracy rate changes.

[0041] In addition, to achieve the above-mentioned purposes, the present application further provides a CAR-T therapy response prediction model optimization device based on explainability analysis, which comprises a memory, a processor, and a CAR-T therapy response prediction model optimization program based on explainability analysis stored in the memory and executable on the processor, wherein the CAR-T therapy response prediction model optimization program based on explainability analysis is configured to implement the steps of the CAR-T therapy response prediction model optimization method based on explainability analysis as described above.

[0042] In addition, to achieve the above-mentioned object, the application further provides a storage medium, wherein the storage medium stores an interpretable analysis-based CAR-T therapy response prediction model optimization program, and the interpretable analysis-based CAR-T therapy response prediction model optimization program, when executed by a processor, implements the steps of the above-mentioned interpretable analysis-based CAR-T therapy response prediction model optimization method.

[0043] The application discloses obtaining a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data; performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result; performing local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result; performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model. Compared with the prior art, the application improves the explainability and prediction accuracy of the CAR-T efficacy prediction model by performing global explainability analysis and local explainability analysis on the trained CAR-T therapy response prediction model and performing model optimization on the trained CAR-T therapy response prediction model based on the analysis results to obtain an optimized prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The figure is a flowchart of an embodiment of the application, i.e., an interpretable analysis-based CAR-T therapy response prediction model optimization method.

[0045] Figure 2 The figure is a flowchart of an embodiment of the application, i.e., an interpretable analysis-based CAR-T therapy response prediction model optimization method.

[0046] Figure 3 The figure is a flowchart of an embodiment of the application, i.e., an interpretable analysis-based CAR-T therapy response prediction model optimization method.

[0047] Figure 4 The figure is a structural block diagram of an embodiment of the application, i.e., an interpretable analysis-based CAR-T therapy response prediction model optimization system.

[0048] Figure 5 The figure is a structural diagram of an interpretable analysis-based CAR-T therapy response prediction model optimization device related to a hardware running environment of an embodiment of the application.

[0049] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0050] It should be understood that the specific embodiments described herein are merely exemplary and do not limit the present application.

[0051] The embodiment of the present application provides a CAR-T therapy response prediction model optimization method based on explainability analysis. Figure 1 , Figure 1 The flowchart of the first embodiment of the CAR-T therapy response prediction model optimization method based on explainability analysis of the present application is shown in the figure.

[0052] In this embodiment, the CAR-T therapy response prediction model optimization method based on explainability analysis comprises steps S10-S40:

[0053] Step S10: obtaining a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data.

[0054] It should be noted that the execution subject of the present embodiment can be a computer server device with data processing, network communication and program running functions applied in the clinical trial data analysis scene, such as servers, tablet computers, personal computers, etc., or an electronic device capable of realizing the above functions (such as a CAR-T therapy response prediction model optimization device based on explainability analysis). The following will take the CAR-T therapy response prediction model optimization device based on explainability analysis as an example to illustrate the present embodiment and the following embodiments.

[0055] It should be explained that the CAR-T therapy response prediction model in the present embodiment can be constructed based on a long short-term memory network model (LSTM) and electronic health record data. The electronic health record data (EHR data) records the dynamic changes of patients from the first visit to the treatment, and these data can be used to construct a time series model (such as LSTM) to predict the efficacy of CAR-T therapy.

[0056] It is easy to understand that the traditional CAR-T efficacy prediction model (such as IPI score and logistic regression model based on cross-sectional data) only relies on single time point data before treatment, ignoring the longitudinal treatment history information of patients. These models cannot capture the dynamic changes of patients before treatment, resulting in incomplete and inaccurate prediction results. The trained CAR-T therapy response prediction model in the present embodiment can dynamically capture the treatment trajectory of patients, including the number of previous treatment lines, treatment response and changes of biomarkers, by processing time series data, thereby providing more comprehensive decision information.

[0057] Step S20: performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result.

[0058] It should be noted that global explainability analysis aims to understand the decision logic of the machine learning model on the whole dataset, that is, how the model makes predictions by integrating all input features. Global analysis focuses on the generalization behavior of the model to all features and samples.

[0059] It should be explained that in this embodiment, the trained CAR-T therapy response prediction model can be analyzed for global explainability using a permutation feature importance algorithm to obtain a global analysis result.

[0060] It should be understood that permutation feature importance (PFI) is a model-independent feature evaluation method that quantifies the impact of feature values on model predictions by shuffling them. The core idea is that if a feature is important to the model, shuffling its values will significantly reduce the model's performance.

[0061] It should be noted that in this embodiment, the permutation feature importance algorithm first evaluates the performance of the trained CAR-T therapy response prediction model without shuffling the features, and records the performance indicators. Then, for each feature to be evaluated, the values of the feature are randomly shuffled in all samples. The model performance is then re-evaluated using the shuffled data, and the change in performance indicators is recorded. Finally, by comparing the differences in performance indicators before and after shuffling, the importance (i.e., contribution) of the feature to the model's prediction is determined.

[0062] It should be added that before using the permutation feature importance algorithm to perform global explainability analysis on the trained CAR-T therapy response prediction model, all features used in the trained CAR-T therapy response prediction model can be identified, including clinical features (such as extranodal lesions ≥2, tumor subtype, PET-CT imaging features, etc.), laboratory data (such as ferritin, D-dimer, carbon dioxide content, etc.), and treatment dynamics (such as CAR-T product, number of previous treatment lines, and treatment response, etc.).

[0063] It should be understood that global explainability analysis of the trained CAR-T therapy response prediction model can help clinicians better understand and trust the model's predictions, thereby more effectively applying it to actual clinical decision-making.

[0064] Step S30: performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result.

[0065] It should be noted that the local interpretability analysis is another core direction in the field of machine learning interpretability, focusing on understanding the decision logic of the model for a single sample or a sub-group.

[0066] It should be explained that the embodiment can use the SHAP algorithm to perform local interpretability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result.

[0067] It should be understood that SHAP (SHapley Additive exPlanations) is a model explanation algorithm based on Shapley value in game theory, which aims to decompose the prediction value of the model into the contribution value of each feature, thereby providing local and global interpretability. The core idea is to quantify the "contribution degree" of each feature to the prediction result through cooperative game theory.

[0068] It should be noted that in the SHAP algorithm in this embodiment, first, for each patient sample, all possible feature subsets are listed; then the feature to be evaluated is added to each feature subset, and the prediction performance of the model when the feature is included is evaluated; the difference between the model performance when the feature is included and not included is calculated to quantify the marginal contribution of the feature to the model prediction; the marginal contributions of all feature subsets are combined and weighted to determine the importance of the feature in the specific sample.

[0069] In specific implementation, the influence of each feature on the prediction result can be determined according to the positive and negative of the SHAP value. A positive SHAP value indicates that the feature increases the likelihood of the patient's good response to the CAR-T therapy, and a negative SHAP value indicates that it reduces the likelihood. For example, in predicting complete remission (CR), a positive SHAP value of a certain laboratory indicator may indicate that an increase in the indicator contributes to the prediction of CR.

[0070] Step S40: performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model.

[0071] It should be noted that the global analysis result can be a feature importance ranking obtained according to the global interpretability analysis, and the top-ranked features, such as the top 20 features, are selected as a global key feature set. These features contribute most to the overall prediction of the model, and usually include clinical features such as extranodal lesions ≥2, tumor subtype, and PET-CT imaging features. The global key feature set is the global analysis result.

[0072] The local analysis result can be, for each patient, identifying features playing a dominant role in the individual patient prediction according to the local explainable analysis (such as SHAP value), such as ferritin level, ratio of suppressive to cytotoxic T cells, etc. Statistics of these features frequently appearing in the local analysis of multiple patients are included in the local key feature set, i.e. the local analysis result.

[0073] Further, the global key feature set (i.e. the global analysis result) and the local key feature set (i.e. the local analysis result) can be combined to remove duplicate features and obtain a preliminary set of comprehensive key predictors. At the same time, considering the correlation and redundancy between features, the correlation analysis of the features in the preliminary set is performed to eliminate highly correlated features, and finally the key predictors are determined.

[0074] In a specific implementation, the key predictors corresponding to the trained CAR-T therapy response prediction model can be screened according to the global analysis result and the local analysis result; the trained CAR-T therapy response prediction model is optimized based on the key predictors to obtain an optimized prediction model.

[0075] It should be noted that the step of optimizing the trained CAR-T therapy response prediction model based on the key predictors to obtain an optimized prediction model includes: optimizing the training data set of the trained CAR-T therapy response prediction model based on the key predictors to obtain an optimized feature set; retraining the trained CAR-T therapy response prediction model based on the optimized feature set to obtain a retraining result; and optimizing the trained CAR-T therapy response prediction model based on the retraining result to obtain an optimized prediction model.

[0076] It is easy to understand that the factors (i.e. key predictors) playing a key role in CAR-T therapy response prediction are screened, and the prediction model is optimized accordingly to improve its prediction performance and reliability, providing more accurate treatment decision support for clinicians.

[0077] In a specific implementation, after the step of optimizing the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model, the step further includes: determining a model evaluation index of the optimized prediction model; and based on the model evaluation index, using ten-fold cross-validation method to evaluate the performance of the optimized prediction model to obtain a model evaluation result.

[0078] The embodiment discloses obtaining a trained CAR-T therapy response prediction model, which is constructed based on electronic health record data; performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result; performing local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result; and performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model. Compared with the prior art, the embodiment improves the explainability and prediction accuracy of the CAR-T efficacy prediction model by performing global explainability analysis and local explainability analysis on the trained CAR-T therapy response prediction model and performing model optimization on the trained CAR-T therapy response prediction model based on the analysis results to obtain an optimized prediction model.

[0079] Reference Figure 2 , Figure 2 The flowchart of the second embodiment of the CAR-T therapy response prediction model optimization method based on explainability analysis.

[0080] Based on the first embodiment, in the embodiment, the step S20 includes steps S201-S203.

[0081] Step S201: The contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model is evaluated by a permutation feature importance algorithm.

[0082] Step S202: The contribution of each feature in the electronic health record data is sorted to obtain a sorting result.

[0083] Step S203: The key features that have an impact on the prediction result are determined based on the sorting result, and the key features are taken as the global analysis result.

[0084] In a specific implementation, the permutation feature importance algorithm can be used to perform multiple random disturbances on each feature in the electronic health record data; the accuracy rate change of the prediction result corresponding to the trained CAR-T therapy response prediction model after each random disturbance is determined; and the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model is evaluated based on the accuracy rate changes.

[0085] It should be noted that, in order to reduce the error caused by random permutation, the permutation of each feature and the model re-evaluation process are repeated multiple times (such as 30 times), the contribution of each feature is calculated after each repetition, and then the average value is taken as the contribution of the feature.

[0086] The embodiment discloses evaluating the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model through the permutation feature importance algorithm; sorting the contribution of each feature in the electronic health record data to obtain a sorting result; determining a key feature that has an impact on the prediction result based on the sorting result, and taking the key feature as a global analysis result. Compared with the prior art, the embodiment can clearly show the contribution degree of each feature to the model prediction through the permutation feature importance algorithm, associate the complex model prediction behavior with specific features, improve the model transparency, help subsequent model optimization, and improve the prediction accuracy and reliability of the CAR-T efficacy prediction model.

[0087] Reference Figure 3 , Figure 3 The flowchart of the third embodiment of the CAR-T therapy response prediction model optimization method based on explainability analysis.

[0088] Based on the above embodiments, in the present embodiment, the step S30 further comprises steps S301-S305:

[0089] Step S301: Determine all feature subsets of each patient in the electronic health record data.

[0090] Step S302: Using the SHAP algorithm, determine the marginal contribution value of each feature in the corresponding all feature subsets for each patient.

[0091] Step S303: Based on the marginal contribution value, determine the SHAP value of each feature by weighted average.

[0092] Step S304: Determine the individualized influence of each feature on the prediction result corresponding to the trained CAR-T therapy response prediction model based on the SHAP value of each feature.

[0093] Step S305: Take the individualized influence of each feature as a local analysis result.

[0094] It is easy to understand that the feature subset is a set composed of a part of features selected from a plurality of original features, which have strong correlation with the target variable and important value for model prediction or analysis. The purpose is to reduce the feature dimension, remove redundant, irrelevant or noise features, and improve the performance and explainability of the model.

[0095] It should be noted that in the SHAP (SHapley Additive exPlanations) algorithm, the marginal contribution value is used to quantify the contribution of each feature to the model prediction result under different feature subsets. The marginal contribution value refers to the additional impact on the model prediction result when a certain feature is added to a feature subset.

[0096] The embodiment discloses determining all feature subsets of each patient in the electronic health record data; using the SHAP algorithm, determining the marginal contribution value of each feature under the corresponding all feature subsets for each patient; determining the SHAP value of each feature by weighted average based on the marginal contribution value; determining the individualized influence of each feature on the prediction result corresponding to the trained CAR-T therapy response prediction model according to the SHAP value of each feature; and taking the individualized influence of each feature as a local analysis result. Compared with the prior art, the embodiment can analyze each sample individually by the SHAP algorithm, determine the specific contribution of each feature to the prediction result of the sample, provide detailed explanations for individual patients for doctors, researchers and the like, provide guidance for model optimization, improve the prediction performance of the model, and make the model better serve actual application scenarios.

[0097] In addition, the embodiment of the present application also proposes a storage medium, which stores a CAR-T therapy response prediction model optimization program based on explainability analysis on the storage medium. The CAR-T therapy response prediction model optimization program based on explainability analysis is executed by the processor to realize the steps of the CAR-T therapy response prediction model optimization method based on explainability analysis as described above.

[0098] Reference Figure 4 , Figure 4 The structure block diagram of the first embodiment of the CAR-T therapy response prediction model optimization system based on explainability analysis of the present application is shown in the figure.

[0099] As Figure 4 shown, the CAR-T therapy response prediction model optimization system based on explainability analysis proposed by the embodiment of the present application comprises a model acquisition module 501, a global analysis module 502, a local analysis module 503 and a model optimization module 504.

[0100] The model acquisition module 501 is configured to acquire a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data.

[0101] The global analysis module 502 is configured to perform global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result.

[0102] The local analysis module 503 is configured to perform local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result.

[0103] The model optimization module 504 is configured to perform model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model.

[0104] The model optimization module 504 is further configured to filter a key prediction factor corresponding to the trained CAR-T therapy response prediction model according to the global analysis result and the local analysis result, and perform model optimization on the trained CAR-T therapy response prediction model based on the key prediction factor to obtain an optimized prediction model.

[0105] The model optimization module 504 is further configured to optimize a training data set of the trained CAR-T therapy response prediction model based on the key prediction factor to obtain an optimized feature set, retrain the trained CAR-T therapy response prediction model based on the optimized feature set to obtain a retraining result, and perform model optimization on the trained CAR-T therapy response prediction model according to the retraining result to obtain an optimized prediction model.

[0106] The model optimization module 504 is further configured to determine a model evaluation index of the optimized prediction model, and evaluate the performance of the optimized prediction model based on the model evaluation index by using a ten-fold cross-validation method to obtain a model evaluation result.

[0107] The system embodiment disclosed in the present application obtains a trained CAR-T therapy response prediction model, which is constructed based on electronic health record data. Global explainability analysis is performed on the trained CAR-T therapy response prediction model to obtain a global analysis result. Local explainability analysis is performed on the trained CAR-T therapy response prediction model to obtain a local analysis result. Model optimization is performed on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model. Compared with the prior art, the system embodiment improves the explainability and prediction accuracy of the CAR-T efficacy prediction model.

[0108] Based on the first embodiment of the CAR-T therapy response prediction model optimization system based on explainability analysis, the second embodiment of the CAR-T therapy response prediction model optimization system based on explainability analysis is proposed.

[0109] In the embodiment, the global analysis module 502 is further configured to evaluate the contribution of each feature in the electronic health record data to the prediction result of the trained CAR-T therapy response prediction model by using a permutation feature importance algorithm, sort the contribution of each feature in the electronic health record data to obtain a sorting result, and determine a key feature that has an impact on the prediction result based on the sorting result, and take the key feature as a global analysis result.

[0110] The global analysis module 502 is further configured to perform multiple random perturbations on each feature in the electronic health record data by using a permutation feature importance algorithm, determine the accuracy change of the prediction result of the trained CAR-T therapy response prediction model after each random perturbation, and evaluate the contribution of each feature in the electronic health record data to the prediction result of the trained CAR-T therapy response prediction model based on the accuracy change.

[0111] Other embodiments or specific implementation manners of the CAR-T therapy response prediction model optimization system based on explainability analysis can refer to the above-mentioned method embodiments, which will not be described here.

[0112] The application provides a CAR-T therapy response prediction model optimization device based on explainability analysis, which comprises at least one processor and a memory connected in communication with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the CAR-T therapy response prediction model optimization method based on explainability analysis in the above-mentioned embodiment one.

[0113] Reference will be made to the following Figure 5, which shows a schematic diagram of the structure of a CAR-T therapy response prediction model optimization device based on explainability analysis suitable for implementing the embodiments of the present application. The CAR-T therapy response prediction model optimization device based on explainability analysis in the embodiments of the present application can include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The CAR-T therapy response prediction model optimization device based on interpretable analysis is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0114] like Figure 5 As shown, the CAR-T therapy response prediction model optimization device based on interpretable analysis may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory 1002 or the program loaded from the storage device 1003 to the random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the CAR-T therapy response prediction model optimization device based on interpretable analysis are also stored. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other via a bus 1005. The input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the CAR-T therapy response prediction model optimization device based on interpretable analysis to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a CAR-T therapy response prediction model optimization device based on interpretable analysis with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems can be implemented or have instead.

[0115] In particular, the processes described above with reference to the flow charts can be implemented as computer software programs in accordance with embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flow charts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of embodiments of the present disclosure are performed.

[0116] The CAR-T therapy response prediction model optimization device based on explainability analysis provided by the present application adopts the CAR-T therapy response prediction model optimization method based on explainability analysis in the above-mentioned embodiments, and can solve the technical problem of how to enhance the explainability and prediction accuracy of the CAR-T efficacy prediction model in the prior art. Compared with the prior art, the CAR-T therapy response prediction model optimization device based on explainability analysis provided by the present application has the same beneficial effects as the CAR-T therapy response prediction model optimization method based on explainability analysis provided by the above-mentioned embodiments, and other technical features in the CAR-T therapy response prediction model optimization device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0117] It should be understood that parts of the present disclosure can be realized by hardware, software, firmware or a combination thereof. In the description of the above-mentioned embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0118] The above is merely specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0119] It should be noted that in this paper, the term "including", "containing" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or system. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0120] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0121] Through the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and the necessary general hardware platform, of course, they can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, an optical disk), and includes a plurality of instructions for making a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in various embodiments of the present application.

[0122] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent flow transformation made by using the content of the specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A CAR-T therapy response prediction model optimization method based on explainability analysis, characterized in that, The method comprises: obtaining a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data; performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result; performing local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result; performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model.

2. The CAR-T therapy response prediction model optimization method based on interpretability analysis of claim 1, wherein, The step of performing global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result comprises: evaluating the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model by using a permutation feature importance algorithm; sorting the contribution of each feature in the electronic health record data to obtain a sorting result; determining key features that have an impact on the prediction result based on the sorting result, and taking the key features as the global analysis result.

3. The CAR-T therapy response prediction model optimization method based on interpretability analysis of claim 2, wherein, The step of evaluating the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model by using a permutation feature importance algorithm comprises: performing multiple random perturbations on each feature in the electronic health record data by using a permutation feature importance algorithm; determining the accuracy change of the prediction result corresponding to the trained CAR-T therapy response prediction model after each random perturbation; evaluating the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model based on the accuracy changes.

4. The CAR-T therapy response prediction model optimization method based on interpretability analysis of claim 1, wherein, The step of performing local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result comprises: determining all feature subsets of each patient in the electronic health record data; determining the marginal contribution value of each feature in the corresponding all feature subsets for each patient by using a SHAP algorithm; determining the SHAP value of each feature by weighted average based on the marginal contribution value; determining the individualized influence of each feature on the prediction result corresponding to the trained CAR-T therapy response prediction model according to the SHAP value of each feature; taking the individualized influence of each feature as the local analysis result.

5. The CAR-T therapy response prediction model optimization method based on interpretability analysis of claim 1, wherein, The step of performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model comprises: screening key prediction factors corresponding to the trained CAR-T therapy response prediction model according to the global analysis result and the local analysis result; performing model optimization on the trained CAR-T therapy response prediction model based on the key prediction factors to obtain an optimized prediction model.

6. The CAR-T therapy response prediction model optimization method based on interpretability analysis of claim 5, wherein, The step of performing model optimization on the trained CAR-T therapy response prediction model based on the key prediction factors to obtain an optimized prediction model comprises: optimizing a training data set of the trained CAR-T therapy response prediction model based on the key prediction factors to obtain an optimized feature set; retraining the trained CAR-T therapy response prediction model based on the optimized feature set to obtain a retraining result; performing model optimization on the trained CAR-T therapy response prediction model based on the retraining result to obtain an optimized prediction model.

7. The CAR-T therapy response prediction model optimization method based on interpretability analysis of claim 1, wherein, The step of performing model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model further comprises: determining a model evaluation index of the optimized prediction model; based on the model evaluation index, using a ten-fold cross-validation method to evaluate the performance of the optimized prediction model to obtain a model evaluation result.

8. A CAR-T therapy response prediction model optimization system based on explainability analysis, characterized in that, The system comprises: a model acquisition module configured to acquire a trained CAR-T therapy response prediction model, wherein the trained CAR-T therapy response prediction model is constructed based on electronic health record data; a global analysis module configured to perform global explainability analysis on the trained CAR-T therapy response prediction model to obtain a global analysis result; a local analysis module configured to perform local explainability analysis on the trained CAR-T therapy response prediction model to obtain a local analysis result; a model optimization module configured to perform model optimization on the trained CAR-T therapy response prediction model based on the global analysis result and the local analysis result to obtain an optimized prediction model.

9. The CAR-T therapy response prediction model optimization system based on interpretability analysis of claim 8, wherein, The global analysis module is further configured to evaluate the contribution of each feature in the electronic health record data to a prediction result corresponding to the trained CAR-T therapy response prediction model by using a permutation feature importance algorithm; sort the contribution of each feature in the electronic health record data to obtain a sorting result; determine key features that have an impact on the prediction result based on the sorting result, and use the key features as the global analysis result.

10. The CAR-T therapy response prediction model optimization system based on interpretability analysis of claim 9, wherein, The global analysis module is further configured to perform multiple random perturbations on each feature in the electronic health record data by using a permutation feature importance algorithm; determine the accuracy rate change of a prediction result corresponding to the trained CAR-T therapy response prediction model after each random perturbation; and evaluate the contribution of each feature in the electronic health record data to the prediction result corresponding to the trained CAR-T therapy response prediction model based on the accuracy rate changes.

Images