Diabetic macular edema treatment effect prediction system, method and equipment

Abstract

The invention relates to a diabetic macular edema treatment effect prediction system, method and equipment, and belongs to the technical field of biomedicine. Comprising the steps of obtaining and preprocessing clinical data of a patient before treatment to obtain effective data; performing feature screening on the data to obtain feature data; constructing a prediction model, and training the prediction model by using the feature data to obtain an optimal vision recovery prediction model and an optimal edema regression prediction model; according to the two types of prediction models, a combined prediction model is constructed by adopting logic weighted fusion; inputting clinical data of a patient to be treated into the two types of prediction models to obtain a prediction result; and according to a prediction result, carrying out interpretability analysis on the optimal prediction model and the combined prediction model to obtain key characteristic variables, and further obtaining the probability of vision recovery and edema regression of the to-be-treated patient after treatment. According to the method, comprehensive and multi-dimensional sources of prediction variables are ensured, and the limitation of only depending on image data or subjective evaluation is effectively overcome, so that the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and particularly to a system, method, and device for predicting the therapeutic efficacy of diabetic macular edema. Background Art

[0002] Diabetic Macular Edema (DME) is one of the most common and serious complications of diabetic retinopathy, which severely affects the visual function and quality of life of patients. In recent years, anti-vascular endothelial growth factor (VEGF) therapy has become the first-line treatment for DME. However, in clinical practice, there are significant individual differences in the therapeutic effects of anti-VEGF therapy. Accurately predicting the treatment effect is of great significance for optimizing the treatment plan and improving the prognosis of patients.

[0003] Currently, the existing efficacy prediction technologies mainly evaluate based on single-dimensional indicators. For example, the treatment effect is judged only by the change in retinal thickness or visual acuity. However, this single-dimensional prediction method has obvious limitations. On the one hand, it cannot comprehensively consider the interaction between visual acuity recovery and retinal edema regression, which are closely related and interact with each other in DME treatment. On the other hand, the existing methods ignore the importance of multi-dimensional factors such as systemic metabolic indicators (such as blood glucose, glycated hemoglobin, etc.) and hematological indicators (such as blood cell count, inflammatory markers, etc.) in predicting the treatment effect. These factors are closely related to the overall state of diabetic patients and may have a significant impact on the pathogenesis and treatment response of DME.

[0004] In addition, the existing efficacy evaluation methods mostly rely on the subjective experience of doctors, lacking objectivity and standardization, and it is difficult to achieve precise individual efficacy prediction. This subjective evaluation method may not only lead to deviations in the prediction results but also affect the consistency of treatment decisions due to differences in doctors' experience. Therefore, there are obvious deficiencies in the existing technology for predicting the effect of anti-VEGF therapy for DME, and there is an urgent need for a prediction system that can comprehensively consider multi-dimensional factors, be objective and accurate, so as to improve the individualization and precision level of DME treatment. Summary of the Invention

[0005] To this end, the technical problems to be solved by the present invention are to overcome the limitations of single-dimensional prediction, the deficiencies of subjective evaluation, and the low prediction accuracy in the prior art.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a system for predicting the therapeutic efficacy of diabetic macular edema, including: A data acquisition module, configured to obtain multi-dimensional clinical data of a patient before treatment and preprocess the multi-dimensional clinical data to obtain effective clinical data; A data screening module for screening features of the effective clinical data to obtain feature data; A model construction and training module for constructing a prediction model and training the prediction model using the feature data to obtain an optimal prediction model; wherein, the optimal prediction model includes an optimal visual acuity recovery prediction model and an optimal edema regression prediction model; A combined model construction module for constructing a visual acuity recovery - edema regression combined prediction model by logical weighted fusion according to the optimal visual acuity recovery prediction model and the optimal edema regression prediction model; A prediction module for inputting the clinical data of a patient to be treated into the optimal visual acuity recovery prediction model and the optimal edema regression prediction model to obtain prediction results; A model interpretation module for performing interpretability analysis on the optimal prediction model and the visual acuity recovery - edema regression combined prediction model according to the prediction results to obtain key feature variables that affect the prediction results of the optimal prediction model and the visual acuity recovery - edema regression combined prediction model; A prediction result output module for obtaining the probabilities of visual acuity recovery and edema regression after treatment of the patient to be treated according to the key feature variables.

[0007] In an embodiment of the present invention, the data screening module includes: A first screening sub - module for preliminarily screening out the indicators related to visual acuity recovery and edema regression in the effective clinical data through univariate regression to obtain first effective indicator data; A second screening sub - module for removing redundant variables in the first effective indicator data through LASSO regression to obtain the feature data.

[0008] In an embodiment of the present invention, the model construction and training module includes a hyperparameter optimization sub - module for optimizing the hyperparameters of the prediction model through cross - validation or grid search.

[0009] In an embodiment of the present invention, the model construction and training module further includes: A performance evaluation sub - module for evaluating the performance of the prediction model through an external validation set; A model selection sub - module for determining the construction method of the prediction model according to the performance evaluation.

[0010] In an embodiment of the present invention, the evaluation indicators of the performance evaluation include: accuracy, sensitivity, specificity, area under the receiver operating characteristic curve, and decision curve analysis.

[0011] In one embodiment of the present invention, the method for constructing the prediction model includes logistic regression, decision tree, multi-layer perceptron, random forest, and support vector machine.

[0012] In one embodiment of the present invention, the prediction result output module includes: A conversion sub-module for converting the characteristic indexes of the patient to be treated into corresponding scores; An estimation sub-module for estimating the probabilities of visual acuity recovery and edema subsidence after treatment of the patient to be treated by means of integration according to the corresponding scores.

[0013] In a second aspect, to solve the above technical problems, the present invention provides a method for predicting the treatment efficacy of diabetic macular edema, including: S1. Obtain the multi-dimensional clinical data of the patient before treatment, and preprocess the multi-dimensional clinical data to obtain effective clinical data; S2. Perform feature screening on the effective clinical data to obtain feature data; S3. Construct a prediction model, and use the feature data to train the prediction model to obtain an optimal prediction model; wherein, the optimal prediction model includes an optimal visual acuity recovery prediction model and an optimal edema subsidence prediction model; S4. According to the optimal visual acuity recovery prediction model and the optimal edema subsidence prediction model, construct a visual acuity recovery-edema subsidence combined prediction model by logical weighted fusion; S5. Input the clinical data of the patient to be treated into the optimal visual acuity recovery prediction model and the optimal edema subsidence prediction model to obtain prediction results; S6. According to the prediction results, perform interpretability analysis on the optimal prediction model and the visual acuity recovery-edema subsidence combined prediction model to obtain the key feature variables that affect the prediction results of the optimal prediction model and the visual acuity recovery-edema subsidence combined prediction model; S7. According to the key feature variables, obtain the probabilities of visual acuity recovery and edema subsidence after treatment of the patient to be treated.

[0014] In a third aspect, to solve the above technical problems, the present invention provides a device for predicting the treatment efficacy of diabetic macular edema, including the above-mentioned system for predicting the treatment efficacy of diabetic macular edema.

[0015] In one embodiment of the present invention, it further includes an automatic interpretation interface.

[0016] The above technical solutions of the present invention have the following beneficial effects compared with the prior art: (1) A treatment efficacy prediction system, method, and device for diabetic macular edema according to the present invention construct an optimal visual acuity recovery prediction model and an optimal edema regression prediction model, optimize for different objectives, and accurately capture their respective influencing factors. At the same time, a combined prediction model is constructed by logical weighted fusion, comprehensively considering the mutual relationship between the two objectives, and providing a more comprehensive prediction result. This multi-model collaborative prediction method can more accurately predict the post-treatment situation of patients to be treated, providing strong support for clinical treatment decisions.

[0017] (2) The present invention conducts interpretability analysis on the optimal prediction model and the combined prediction model, finds out the key feature variables affecting the prediction result, enhances the interpretability of the model, and enables clinicians to understand the prediction basis of the model. In addition, based on the key feature variables, the present invention can calculate the probabilities of visual acuity recovery and edema regression after treatment for patients to be treated, providing a basis for personalized treatment, helping doctors adjust treatment plans more targeted, and thus improving the overall treatment effect.

[0018] (3) The present invention realizes the transformation from relying on a single indicator to multi-dimensional information fusion, thereby enhancing the discriminative ability of the model for individual differences. It not only predicts visual acuity recovery and retinal edema regression separately, but also predicts the combined results of the two, comprehensively covering clinically important efficacy indicators.

[0019] (4) The present invention comprehensively integrates ophthalmic structure imaging indicators, systemic metabolic indicators, and hematological indicators, constructs a multi-dimensional data model, and realizes accurate prediction of treatment efficacy. This method effectively overcomes the limitations of single, subjective, and low-accuracy existing prediction means, improves the scientificity and efficiency of clinical decision-making, and at the same time promotes the level of individualized treatment plans.

[0020] (5) The present invention has good scalability, can further introduce new variables or adjust the model structure according to the results of subsequent clinical studies, continuously optimize the prediction performance, and show broad prospects for scientific research and clinical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to make the content of the present invention easier to be clearly understood, the following further details the present invention according to specific embodiments of the present invention and in combination with the drawings, where: Figure 1 It is a structural diagram of a treatment efficacy prediction system for diabetic macular edema in a preferred embodiment of the present invention; Figure 2 It is a flowchart of the implementation of a treatment efficacy prediction system for diabetic macular edema in a preferred embodiment of the present invention; Figure 3 It is a flowchart of a treatment efficacy prediction method for diabetic macular edema in a preferred embodiment of the present invention. Detailed implementation manners

[0022] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the embodiments cited are not intended to limit the present invention.

[0023] Embodiment 1: Refer to Figure 1 As shown, the embodiment of the present invention provides a treatment efficacy prediction system for diabetic macular edema, including: A data acquisition module, configured to obtain multi-dimensional clinical data of a patient before treatment and preprocess the multi-dimensional clinical data to obtain effective clinical data; A data screening module, configured to perform feature screening on the effective clinical data to obtain feature data; A model construction and training module, configured to construct a prediction model and use the feature data to train the prediction model to obtain an optimal prediction model; wherein, the optimal prediction model includes an optimal visual acuity recovery prediction model and an optimal edema regression prediction model; A combined model construction module, configured to construct a visual acuity recovery-edema regression combined prediction model by using logical weighted fusion according to the optimal visual acuity recovery prediction model and the optimal edema regression prediction model; A prediction module, configured to input the clinical data of a patient to be treated into the optimal visual acuity recovery prediction model and the optimal edema regression prediction model to obtain a prediction result; A model interpretation module, configured to perform interpretability analysis on the optimal prediction model and the visual acuity recovery-edema regression combined prediction model according to the prediction result to obtain key feature variables that affect the optimal prediction model and the visual acuity recovery-edema regression combined prediction model to produce the prediction result; A prediction result output module, configured to obtain the probabilities of visual acuity recovery and edema regression after treatment of a patient to be treated according to the key feature variables.

[0024] An embodiment of the present invention provides a system for predicting the therapeutic efficacy of diabetic macular edema. By integrating multi-source data, it ensures that the sources of prediction variables are comprehensive and multi-dimensional, effectively overcoming the limitations of traditional prediction methods that rely only on imaging data or subjective evaluations, thereby significantly improving the prediction accuracy. Specifically, the data acquisition module preprocesses multi-dimensional clinical data, effectively improving the data quality, and then obtains accurate and effective clinical data, laying a reliable foundation for subsequent analysis. The data screening module performs feature screening on the effective clinical data, accurately extracts feature data closely related to the prediction target, reduces the interference of redundant information, and improves the model training efficiency and prediction performance. In terms of model construction and fusion, the model construction and training module constructs an optimal visual acuity recovery prediction model and an optimal edema regression prediction model respectively, optimizes for different targets, and accurately captures their respective influencing factors. At the same time, the combined model construction module constructs a combined prediction model through logical weighted fusion, comprehensively considers the mutual relationship between the two targets, and provides a more comprehensive prediction result. This multi-model collaborative prediction method enables the model to more accurately predict the post-treatment situation of patients to be treated, providing strong support for clinical treatment decisions. In terms of interpretability and clinical application, the model interpretation module performs interpretability analysis on the optimal prediction model and the combined prediction model, finds out the key feature variables affecting the prediction results, enhances the interpretability of the model, and enables clinicians to understand the prediction basis of the model. In addition, based on the key feature variables, the probabilities of visual acuity recovery and edema regression after treatment of patients to be treated are obtained, providing a basis for personalized treatment, helping doctors to more targeted adjust the treatment plan, and improving the overall treatment effect. The system provided by the embodiment of the present invention can not only predict the visual acuity recovery effect after treatment, but also evaluate the regression of retinal edema, and realize the combined prediction of visual acuity recovery and edema regression.

[0025] Specifically, in the data acquisition module, complete multi-dimensional clinical data of multiple patients before treatment are obtained. For example, 560 pieces of complete multi-dimensional clinical data of patients before treatment are obtained. These multi-dimensional clinical data include patient basic information (such as age, gender, etc.), optical coherence tomography (OCT) image data, systemic metabolic index data, and hematological index data. Among them, the OCT image data is used to evaluate the retinal structural characteristics, including central retinal thickness (CRT), abnormal inner and outer retinal layers, and the integrity of the ellipsoid zone; the systemic metabolic index data covers key indicators such as low-density lipoprotein, blood urea nitrogen, and serum creatinine; the hematological index data includes hemoglobin, etc.

[0026] Furthermore, the multi-dimensional clinical data is preprocessed. The preprocessing includes normalizing the data, filling in missing values, and detecting outliers. The specific implementation method of the preprocessing is as follows: Normalize the data by scaling the values of each indicator to a unified range (such as [0,1] or [-1,1]) to eliminate the dimensional differences between different indicators, thereby improving the consistency and comparability of the data.

[0027] Exemplarily, taking the central retinal thickness (CRT) as an example, the calculation formula for normalizing the original central retinal thickness is: ; Where, is the normalized value of the central retinal thickness, is the minimum value of CRT, is the maximum value of CRT.

[0028] Fill in the missing values, and use appropriate methods (such as mean filling, median filling or interpolation filling) to supplement the missing information to ensure the integrity of the data and avoid the impact on subsequent analysis caused by data missing.

[0029] Detect and process outliers, use statistical methods (such as the Z-score method or box plot method) to identify abnormal data points, and process the outliers according to the specific situation (such as correction or deletion) to ensure the accuracy and reliability of the data.

[0030] Exemplarily, taking blood urea nitrogen (BUN) as an example, calculate the Z-score of BUN, and the calculation formula is: ; Where, is the average value of blood urea nitrogen, is the standard deviation of blood urea nitrogen. Set a judgment threshold, compare the calculated Z-score with the judgment threshold to determine whether there are outliers in blood urea nitrogen. Once an outlier is detected, the abnormal data point can be directly deleted to avoid its misleading impact on subsequent analysis; or, replace the outlier with an adjacent value, for example, use the value of the previous or next data point to replace it, so as to ensure the accuracy of the data while retaining the integrity of the data.

[0031] Through the above preprocessing steps, the quality and effectiveness of multi-dimensional clinical data can be effectively improved, providing a solid foundation for subsequent analysis and modeling.

[0032] Specifically, in the embodiments of the present invention, the data screening module uses a method combining univariate Logistic regression and LASSO regression to screen features of effective clinical data. The data screening module includes a first screening sub-module and a second screening sub-module.

[0033] Further, in the first screening sub-module, univariate Logistic regression analysis is used to preliminarily screen all candidate indicators (i.e., valid clinical data), identify indicators significantly related to vision recovery and edema regression, where the related indicators include potential features, so as to obtain the first valid indicator data. This process can effectively exclude indicators irrelevant to the target variable and narrow the feature scope.

[0034] Further, in the second screening sub-module, the LASSO regression method is used to optimize the first valid indicator data after preliminary screening, and redundant variables are removed through its regularization characteristics, and finally the feature data that has a significant impact on the prediction result is accurately determined.

[0035] Through the first screening sub-module and the second screening sub-module, this dual screening mechanism not only improves the accuracy and efficiency of feature selection, but also enhances the interpretability and prediction performance of the model.

[0036] Specifically, in the model construction and training module, three sub-models are respectively trained using the feature data, namely: a vision recovery prediction model, an edema regression prediction model, and a vision recovery-edema regression combined prediction model. The vision recovery prediction model, the edema regression prediction model, and the vision recovery-edema regression combined prediction model are constructed based on machine learning algorithms, and the constructed algorithms include Logistic Regression, Decision Tree (abbreviated as DT), Multilayer Perceptron (abbreviated as MLP), Random Forest (abbreviated as RF), and Support Vector Machine (abbreviated as SVM).

[0037] Based on machine learning algorithms, the embodiments of the present invention comprehensively integrate ophthalmic structural imaging indicators (including central retinal thickness, abnormal inner and outer retinal layers, integrity of the ellipsoid zone, etc.), systemic metabolic indicators (such as low-density lipoprotein, blood urea nitrogen, serum creatinine, etc.), and hematological indicators to construct a multi-dimensional data model to achieve accurate prediction of treatment efficacy. This innovative design effectively overcomes the limitations of single, subjective, and low-accuracy existing prediction methods, improves the scientificity and efficiency of clinical decision-making, and at the same time promotes the individualized level of treatment plans.

[0038] Further, in order to optimize the model performance, the model construction and training module includes a hyperparameter optimization sub-module, which uses cross-validation or grid search methods to tune the hyperparameters of each sub-model.

[0039] Furthermore, K-fold cross-validation is adopted in the hyperparameter optimization sub-module to ensure the robustness of the model. K-fold cross-validation is a statistical method used to evaluate the predictive performance of machine learning models. During the process of K-fold cross-validation, the training and validation of the model will be repeated K times. Each time, a different subset is selected as the validation set, and the remaining K - 1 subsets are combined as the training set. The purpose of doing this is to ensure that each data point has the opportunity to be used for training and validation, so that the model evaluation results are more comprehensive and objective. Taking K = 10 as an example, that is, 10-fold cross-validation is adopted, and the specific steps are as follows: Step 1: Randomly divide the entire dataset into 10 subsets of equal size, and each subset is called a "fold".

[0040] Step 2: Conduct 10 cycles of training and validation. In each cycle: Select one of the folds as the validation set to evaluate the performance of the model. Combine the remaining 9 folds as the training set to train the model. Train the model on the training set and evaluate the performance of the model on the validation set, and record the performance metrics (such as accuracy, area under the receiver operating characteristic curve, etc.). Repeat the above steps until each fold has been used as a validation set once.

[0041] Step 3: Calculate the average performance metric of all 10 validation results and the standard deviation of the performance metric to obtain an overall performance estimate of the model.

[0042] By comparing the average performance metrics of different models in 10-fold cross-validation, the advantages and disadvantages of different models can be evaluated and compared, and the model with the best performance can be selected for practical applications.

[0043] To comprehensively evaluate the performance of the trained model, the model construction and training module also includes a performance evaluation sub-module and a model selection sub-module. The performance evaluation sub-module uses an independent validation set (i.e., an external validation set) to strictly evaluate the performance of the above model. During the evaluation process, a variety of key evaluation metrics are comprehensively used, including accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and decision curve analysis (DCA for short), etc. These metrics comprehensively measure the predictive ability of the model from different perspectives to ensure the scientificity and reliability of the evaluation results. The model selection sub-module then determines the construction methods of the vision recovery prediction model and the edema regression prediction model according to the evaluation metrics output by the performance evaluation sub-module.

[0044] Specifically, among the evaluation metrics, accuracy measures the proportion of the model's prediction results that are consistent with the actual results, reflecting the overall correctness of the model's predictions. Sensitivity measures the model's ability to identify positive classes (such as disease patients), that is, the proportion of actual positive classes that are correctly predicted as positive classes. Specificity measures the model's ability to identify negative classes (such as healthy individuals), that is, the proportion of actual negative classes that are correctly predicted as negative classes. The area under the receiver operating characteristic curve is obtained by plotting the receiver operating characteristic curve (ROC curve) and calculating the area under the curve, reflecting the comprehensive performance of the model at different thresholds. The closer the AUC value is to 1, the better the model performance. Decision curve analysis (DCA) evaluates the practical application value of the model in clinical decision-making. By comparing the net benefits of different models, it determines the effectiveness and practicality of the model in clinical practice.

[0045] Exemplarily, the performance evaluation sub-module outputs a comprehensive evaluation metric by fusing the comprehensive metrics of accuracy, sensitivity, specificity, AUC, and DCA. In this process, first, weights are assigned to each metric, the value of each metric is multiplied by its weight, and then summed up. After obtaining the metric values of each model, the models are ranked according to the size of the comprehensive evaluation metric, and finally, the model with the highest comprehensive evaluation metric is selected as the optimal model. The effect of this design is that it can comprehensively and objectively measure the performance of the model, avoid the one-sidedness that may be brought by a single metric, and thus improve the scientificity and reliability of model selection.

[0046] After comprehensively considering the above-mentioned metrics, the model with the best performance is selected for actual clinical prediction. This process ensures that the selected model not only has high accuracy and discrimination ability statistically, but also has good decision support value in clinical applications. It can provide reliable prediction basis for clinicians, assist in formulating precise treatment plans, and thus improve the treatment effect of patients.

[0047] Furthermore, after the model construction and training module completes the training of the visual acuity recovery prediction model and the edema regression prediction model, a logical weighted fusion strategy is adopted in the combined model construction module to construct a visual acuity recovery - edema regression combined prediction model (hereinafter simply referred to as the combined prediction model). And the combined model construction module includes: A weight assignment sub-module, which is used to assign corresponding weights to the optimal visual acuity recovery prediction model and the optimal edema regression prediction model according to their performance during independent prediction. The assignment of weights is based on metrics such as the accuracy, recall rate, and F1 score of the model, ensuring that the weights can reflect the importance and reliability of each model in the prediction task. For example, if the optimal visual acuity recovery prediction model has a higher accuracy on the validation set, a higher weight is assigned to it.

[0048] The feature set construction sub-module is used to take the prediction results of two models (i.e., the optimal visual acuity recovery prediction model and the optimal edema regression prediction model) as input features, and combine other relevant features (such as the patient's basic information, clinical indicators, etc.) to construct a new feature set. These features will be used as the input of the combined prediction model to comprehensively evaluate the possibility of visual acuity recovery and edema regression.

[0049] The model establishment sub-module uses the logistic regression algorithm to model the fused feature set. The logistic regression model can output a probability value between 0 and 1 by learning the relationship between the input features and the output results, representing the joint prediction probability of the patient's visual acuity recovery and edema regression. The training objective of the model is to maximize the similarity between the prediction result and the actual result, so as to improve the accuracy of the joint prediction.

[0050] The model optimization sub-module is used to optimize the constructed combined prediction model and adjust the model parameters to improve its performance. The generalization ability of the model is evaluated by methods such as cross-validation to ensure that it can maintain a stable prediction effect on different data sets.

[0051] Furthermore, in order to improve the interpretability of the model and facilitate clinical application, the SHAP (Shapley Additive Explanations) analysis method is adopted in the model interpretation module to deeply interpret the optimal model, and the key feature variables that affect the prediction results of the optimal prediction model and the visual acuity recovery-edema regression combined prediction model are obtained. The SHAP method is based on game theory principles and can accurately quantify the contribution degree and direction of each feature to the model prediction result. By calculating the Shapley value of each feature, the positive or negative impact of each feature on the prediction result can be clearly shown, providing an intuitive and scientific explanation basis for clinicians. For example, the global interpretation can intuitively show the overall contribution of each feature to the model prediction through a SHAP summary bar chart or dot plot, while the local interpretation uses a SHAP waterfall chart to detailedly analyze the formation process of each individual prediction result, helping doctors understand the prediction basis of specific patients.

[0052] Specifically, the prediction result output module includes a conversion sub-module and an estimation sub-module. The conversion sub-module is used to convert the patient's feature indicators into corresponding scores, and each feature is assigned different weights according to its contribution degree to the prediction result. The estimation sub-module is used to quickly calculate the probability of the patient's visual acuity recovery and edema regression after treatment through integration, so as to provide immediate support for clinical decision-making.

[0053] The system described in the embodiments of the present invention not only enhances the interpretability of the model by combining SHAP analysis and probability calculation, but also improves its practicality and convenience in actual clinical scenarios, providing strong support for the precise treatment of diabetic macular edema.

[0054] Further, as Figure 2 shown, based on the above-collected and screened data and the constructed algorithm model, corresponding application tools can be developed, such as the Shiny online computing platform, so as to realize the visual display of the prediction results.

[0055] Further, to verify the effectiveness of the system described in the embodiments of the present invention, the following implementation process is given: The data acquisition module obtains multi-dimensional clinical data: Collect the clinical data of 2042 DME patients who received anti-vascular endothelial growth factor (VEGF) treatment in a tertiary hospital from January 2023 to February 2025. Among these patients, 1331 patients did not show significant improvement in retinal thickness, while 711 patients had significant improvement in retinal thickness. Key features such as the baseline central retinal thickness (CRT), the integrity of the ellipsoid zone (EZ), and the disorder of the inner retinal structure of the patients were extracted using optical coherence tomography (OCT) technology. In addition, multiple biochemical indicators including blood lipids, renal function, and blood routine were also collected.

[0056] The data acquisition module performs data preprocessing, the data screening module performs feature screening, and the model construction and training module performs model training: After normalizing these clinical data, they are input into a logistic regression model for training. The purpose of normalization is to eliminate the dimensional differences between different indicators, making the model training more stable and efficient. Through cross-validation, the performance of the model was strictly evaluated. The results showed that the area under the receiver operating characteristic curve (AUC) of the model reached 0.90, indicating that the model has high prediction accuracy and discrimination ability. This result proves that the visual acuity recovery prediction model for DME patients has important value in actual clinical applications, and can help doctors more accurately predict the treatment response of patients, thus providing a scientific basis for the formulation of personalized treatment plans.

[0057] The combined model construction module and the prediction module are used to construct a combined prediction model and generate a clinical nomogram: Based on the above data, two independent sub-models were trained respectively: one for predicting visual acuity recovery and the other for predicting the regression of retinal edema. These two sub-models can evaluate the improvement of visual acuity and the reduction of retinal edema in patients after receiving anti-VEGF treatment respectively. After the training of the sub-models was completed, a logical weighted fusion strategy was adopted to construct a combined prediction model. To make the prediction results of the model more intuitive and easy for clinical application, the rms package (a statistical modeling package commonly used in the R language) was used to generate the nomogram interface. A nomogram is a graphical prediction tool that allows doctors to quickly obtain the predicted probability of individualized treatment response by simply inputting several key indicators (such as central retinal thickness CRT, integrity of the ellipsoid zone EZ, disorder of the inner retinal structure, presence of subretinal fluid SRF, etc.).

[0058] The model interpretation module and the prediction result output module perform interpretability analysis and display: When performing interpretability analysis, a patient with a predicted probability of 0.84 was selected as an example. The SHAP force plot was used to show the contribution degree of each feature to the prediction result. In the case of this patient, the SHAP force plot revealed that "EZ disruption", "CRT > 400 µm" and "disorder of the inner retinal structure" were the features that had the greatest impact on the predicted probability. These features played a key role in the decision-making process of the model, thus having a significant impact on the prediction of the patient's treatment response. An interpretation report was automatically generated according to the analysis results of the SHAP values, providing doctors with detailed feature contribution information. According to this report, doctors can determine that this patient may have a relatively high structural response, but the potential for visual acuity improvement may be limited. Therefore, doctors may recommend a combined treatment plan in order to achieve better treatment effects.

[0059] In the embodiment of the present invention, by integrating the optical coherence tomography (OCT) imaging features and the data of systemic metabolism and biochemical indicators, the prediction accuracy of the efficacy of anti-vascular endothelial growth factor (VEGF) treatment for patients with diabetic macular edema (DME) is improved. This design realizes the transformation from relying on a single indicator to the fusion of multi-dimensional information, thereby enhancing the discriminant ability of the model for individual differences. Not only the visual acuity recovery and the regression of retinal edema are predicted separately, but also the combined results of the two are predicted, comprehensively covering the clinically important efficacy indicators. This overcomes the limitation of the prior art that only focuses on a single anatomical or functional outcome and provides a more comprehensive evaluation of the treatment effect.

[0060] The prediction results of the embodiments of the present invention are presented in the form of intuitive visualization tools, such as nomograms and SHAP analysis charts with strong interpretability. These tools are convenient for doctors to quickly understand and operate, thus enhancing the applicability of the model in clinical practice. By introducing the SHAP interpretability algorithm, the decision-making process of the model becomes transparent, providing doctors with the direction and weight of the influence of each variable. This not only helps to establish trust between doctors and patients, but also assists doctors in formulating more accurate individualized treatment plans.

[0061] In addition, by identifying in advance those patients who may have a poor response to treatment, the system provided by the embodiments of the present invention can help clinicians consider early the adoption of combination therapy or alternative treatment strategies. This can not only reduce the waste of medical resources, but also reduce the unnecessary treatment costs of patients. This system has good scalability and can further introduce new variables or adjust the model structure according to the results of subsequent clinical studies to continuously optimize the prediction performance, showing broad prospects for scientific research and clinical applications.

[0062] Embodiment 2: Based on the same inventive concept, this embodiment provides a method for predicting the treatment efficacy of diabetic macular edema. The principle of solving the problem is similar to that of a system for predicting the treatment efficacy of diabetic macular edema provided in Embodiment 1, and the repeated parts will not be described in detail.

[0063] Refer to Figure 3 As shown, this embodiment provides a method for predicting the treatment efficacy of diabetic macular edema, including but not limited to the following steps: S1. Obtain the multi-dimensional clinical data of the patient before treatment, and preprocess the multi-dimensional clinical data to obtain effective clinical data; S2. Perform feature screening on the effective clinical data to obtain feature data; S3. Construct a prediction model, and use the feature data to train the prediction model to obtain an optimal prediction model; wherein, the optimal prediction model includes an optimal visual acuity recovery prediction model and an optimal edema regression prediction model; S4. According to the optimal visual acuity recovery prediction model and the optimal edema regression prediction model, use logical weighted fusion to construct a visual acuity recovery-edema regression combined prediction model; S5. Input the clinical data of the patient to be treated into the optimal visual acuity recovery prediction model and the optimal edema regression prediction model to obtain prediction results; S6. According to the prediction results, perform interpretability analysis on the optimal prediction model and the visual acuity recovery-edema regression combined prediction model to obtain the key feature variables that affect the optimal prediction model and the visual acuity recovery-edema regression combined prediction model to produce prediction results; S7. According to the key feature variables, obtain the probabilities of visual acuity recovery and edema regression after treatment of the patient to be treated.

[0064] This embodiment provides a more accurate, comprehensive and objective comprehensive prediction method for anti-VEGF treatment of diabetic macular edema in view of the problems of single efficacy prediction method, strong subjectivity and insufficient accuracy in the prior art. This method integrates ocular imaging indicators, systemic metabolism and hematological indicators, and uses advanced machine learning technology to achieve post-treatment visual recovery, edema regression and their joint prediction, thereby providing a more accurate basis for clinical treatment decision-making and improving treatment effects.

[0065] Embodiment 3: This embodiment provides a device for predicting the therapeutic efficacy of diabetic macular edema, which is a prediction device for clinical application, including a system for predicting the therapeutic efficacy of diabetic macular edema provided in Embodiment 1.

[0066] Specifically, the prediction device includes a prediction tool that is convenient for clinical application. The tool converts the patient's characteristic indicators into corresponding scores, and each characteristic is assigned different weights according to its contribution to the prediction result. By integrating, the probability of visual recovery and edema disappearance after treatment is quickly calculated, thereby providing immediate support for clinical decision-making. This tool is designed to simplify the complex clinical decision-making process, allowing doctors to quickly and accurately evaluate the patient's treatment prognosis, thereby improving medical efficiency and patient treatment effects.

[0067] Furthermore, the prediction device combines the efficacy prediction model to construct a visual nomogram and an automatic interpretation interface, providing doctors with a simple and intuitive prediction auxiliary tool. This innovative design not only greatly simplifies the prediction process, but also enables doctors to quickly understand and apply the prediction results through visualization and automatic interpretation functions, thereby effectively promoting the clinical transformation of technological achievements and improving the scientificity and efficiency of medical decision-making.

[0068] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0069] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce a means for implementing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a block or multiple blocks.

[0070] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means that implements the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a block or multiple blocks.

[0071] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a block or multiple blocks.

[0072] Obviously, the above embodiments are merely examples for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation manners here. And the obvious changes or modifications derived therefrom are still within the protection scope of the present invention.

Claims

1. A therapeutic efficacy prediction system for diabetic macular edema, characterized in that, including: a data acquisition module, configured to obtain multi-dimensional clinical data of a patient before treatment, and preprocess the multi-dimensional clinical data to obtain effective clinical data; a data screening module, configured to perform feature screening on the effective clinical data to obtain feature data; a model construction and training module, configured to construct a prediction model and train the prediction model using the feature data to obtain an optimal prediction model; wherein, the optimal prediction model includes an optimal visual acuity recovery prediction model and an optimal edema regression prediction model; a combined model construction module, configured to construct a visual acuity recovery-edema regression combined prediction model by logical weighted fusion according to the optimal visual acuity recovery prediction model and the optimal edema regression prediction model; a prediction module, configured to input the clinical data of a patient to be treated into the optimal visual acuity recovery prediction model and the optimal edema regression prediction model to obtain prediction results; a model interpretation module, configured to perform interpretability analysis on the optimal prediction model and the visual acuity recovery-edema regression combined prediction model according to the prediction results to obtain key feature variables that affect the optimal prediction model and the visual acuity recovery-edema regression combined prediction model to produce prediction results; a prediction result output module, configured to obtain the probabilities of visual acuity recovery and edema regression of the patient to be treated after treatment according to the key feature variables.

2. The therapeutic efficacy prediction system for diabetic macular edema according to claim 1, wherein The data screening module includes: a first screening sub-module, configured to preliminarily screen out the indicators related to visual acuity recovery and edema regression in the effective clinical data through univariate regression to obtain first effective indicator data; a second screening sub-module, configured to remove redundant variables in the first effective indicator data through LASSO regression to obtain the feature data.

3. The therapeutic efficacy prediction system for diabetic macular edema according to claim 1, wherein The model construction and training module includes a hyperparameter optimization sub-module, and the hyperparameter optimization sub-module is configured to optimize the hyperparameters of the prediction model through cross-validation or grid search.

4. A therapeutic efficacy prediction system for diabetic macular edema according to claim 1, characterized in that The model construction and training module further includes: a performance evaluation sub-module, configured to perform performance evaluation on the prediction model through an external validation set; a model selection sub-module, configured to determine the construction method of the prediction model according to the performance evaluation.

5. A therapeutic efficacy prediction system for diabetic macular edema according to claim 4, wherein, The evaluation indicators of the performance evaluation include: accuracy, sensitivity, specificity, area under the receiver operating characteristic curve, and decision curve analysis.

6. The therapeutic efficacy prediction system for diabetic macular edema according to claim 4, wherein The construction methods of the prediction model include logistic regression, decision tree, multi-layer perceptron, random forest, and support vector machine.

7. A therapeutic efficacy prediction system for diabetic macular edema according to claim 1, characterized in that, The prediction result output module includes: a conversion sub-module, configured to convert the feature indicators of the patient to be treated into corresponding scores; an estimation sub-module, configured to estimate the probabilities of visual acuity recovery and edema regression of the patient to be treated after treatment in an integral manner according to the corresponding scores.

8. A method for predicting the therapeutic efficacy of diabetic macular edema, characterized in that, including: S1. Obtain multi-dimensional clinical data of a patient before treatment, and preprocess the multi-dimensional clinical data to obtain effective clinical data; S2. Perform feature screening on the effective clinical data to obtain feature data; S3. Build a prediction model and use the feature data to train the prediction model to obtain an optimal prediction model; wherein, the optimal prediction model includes an optimal visual acuity recovery prediction model and an optimal edema regression prediction model; S4. According to the optimal visual acuity recovery prediction model and the optimal edema regression prediction model, construct a visual acuity recovery-edema regression combined prediction model by logical weighted fusion; S5. Conduct interpretability analysis on the optimal prediction model and the visual acuity recovery-edema regression combined prediction model to obtain the key feature variables that affect the results generated by the optimal prediction model and the visual acuity recovery-edema regression combined prediction model; S6. According to the key feature variables, obtain the probabilities of visual acuity recovery and edema regression after treatment for the patients to be treated.

9. A therapeutic efficacy prediction device for diabetic macular edema, characterized in that , including a diabetic macular edema treatment efficacy prediction system according to any one of claims 1 to 7.

10. A therapeutic efficacy prediction device for diabetic macular edema according to claim 9, characterized in that, It further includes an automatic interpretation interface.

Images