Early screening method and system for test data of metabolic disease diabetic foot

Abstract

The invention provides a metabolic disease diabetic foot test data early screening method and system. Belongs to the technical field of medical health. The method comprises the following steps: acquiring clinical sample data, transmitting the acquired clinical sample data to a cloud space for preprocessing, performing feature extraction on the preprocessed clinical sample data through a deep learning algorithm, and extracting features related to the metabolic disease diabetic foot; constructing a metabolic disease diabetic foot screening model based on blood routine data through a deep learning algorithm, and combining obtained clinical sample data with historical data to construct a data set; the constructed metabolic disease diabetic foot screening model is evaluated through multiple indexes, and the metabolic disease diabetic foot screening model is optimized based on the evaluation result. Feature extraction is carried out through a deep learning algorithm, complex modes and subtle changes related to the metabolic disease diabetic foot can be recognized, and the screening accuracy is improved.

Description

Technical Field

[0001] The present invention proposes a method and system for early screening of metabolic disease diabetic foot test data, belonging to the field of medical health technology. Background Art

[0002] In the clinical practice of metabolic disease diabetes management, metabolic disease diabetic foot, as a serious complication of metabolic disease diabetes, has always been the focus of medical attention due to its high incidence, high disability rate and high treatment costs. The occurrence of metabolic disease diabetic foot is often closely related to multiple factors such as poor blood sugar control, vascular disease, neuropathy and infection. Its early identification and intervention are of vital importance to prevent disease deterioration, reduce the risk of amputation and improve the quality of life of patients.

[0003] However, traditional screening methods for metabolic disease diabetic foot mainly rely on the doctor's clinical experience, physical examination and imaging examination, which are not only time-consuming and laborious, but also easily affected by subjective factors, resulting in limited accuracy and efficiency of screening. Especially for early metabolic disease diabetic foot, due to its unclear symptoms, it is easy to be ignored or misdiagnosed, thus missing the best time for treatment.

[0004] With the rapid development of artificial intelligence (AI) technology, especially its increasing application in the medical field, it has provided a new solution for the early screening of metabolic disease diabetic foot. AI technology can mine potential features related to metabolic disease diabetic foot from massive clinical data through advanced algorithms such as big data analysis and deep learning, build an efficient screening model, and achieve accurate prediction and early diagnosis of metabolic disease diabetic foot. Summary of the invention

[0005] The present invention provides a method and system for early screening of metabolic disease diabetic foot test data, which is used to solve the problems mentioned in the above background technology: The present invention proposes a method for early screening of metabolic disease diabetic foot test data, the method comprising: S1. Acquire clinical sample data, and transmit the acquired clinical sample data to the cloud space for preprocessing, perform feature extraction on the preprocessed clinical sample data through a deep learning algorithm, and extract features related to the metabolic disease diabetic foot; S2. Construct a metabolic disease diabetic foot screening model based on routine blood data through deep learning algorithms, and combine the acquired clinical sample data with historical data to construct a data set; evaluate the constructed metabolic disease diabetic foot screening model through multiple indicators, and optimize the metabolic disease diabetic foot screening model based on the evaluation results; S3. Introduce an external validation dataset to conduct a blind test on the optimized metabolic disease diabetic foot screening model to verify the regional adaptability of the model and perform risk grading on the model output results based on the algorithm interpretation; S4. Deploy the constructed model to the medical information system, receive patient test data in real time, and analyze the test data in real time to obtain metabolic disease diabetic foot risk warning results.

[0006] Furthermore, the S1 includes: S11. Collect sample data from multiple data sources through a multi-data integration framework, and transmit the collected sample data to the cloud space; S12, the cloud space stores the received sample data in different storage spaces according to the data type, and divides a subspace in each storage space; S13, performing redundant backup of the sample data in the storage space, and storing the backed-up data in the subspace; preprocessing the data stored in the storage space and its corresponding subspace respectively by the processing unit; S14, comparing the preprocessed results of the storage space and its corresponding subspace to determine whether there is a difference. If there is a difference, the difference is processed again; S15, taking the result after the second processing as a reference value, comparing it with the result of the first preprocessing, and taking the sample data stored in the storage space or subspace with the smallest difference value after preprocessing as the final result; S16. Use deep learning algorithms to extract features from preprocessed data, evaluate the extracted features, and select the feature subset with the most predictive value through feature selection technology.

[0007] Furthermore, the S2 includes: S21. Merge the acquired clinical sample data with the existing historical data to form a comprehensive data set; and preprocess the merged data set; S22. Extract key features related to metabolic disease diabetic foot from preprocessed data based on deep learning algorithm; S23. Through feature importance evaluation, screen the feature subset that contributes most to the prediction of metabolic disease diabetic foot; S24, adopting a multi-layer perception mechanism, designing a model architecture according to feature types and data characteristics, using the divided training set data to train the model, and optimizing the model parameters through the back propagation algorithm; S25. Evaluate multiple evaluation indicators of the model using the validation set data, wherein the multiple indicators include accuracy, sensitivity, and specificity; S26. Based on the evaluation results, iteratively optimize the model performance by adjusting the model structure, optimizing algorithm parameters, and using regularization techniques; S27. Identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set; S28. Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model.

[0008] Furthermore, the S3 includes: S31, collecting external data sources, and integrating the collected external data sources to construct an external verification data set; S32. Perform a blind test on the tuned metabolic disease diabetic foot model using an external validation dataset, analyze the performance differences of the model on the external dataset based on the blind test results, and identify potential biases and error sources; S33. Perform cluster analysis on the external validation data set to identify differences and similarities in data distribution. Based on the differences in data distribution, fine-tune or retrain the model through transfer learning. S34. Perform sensitivity analysis to evaluate the sensitivity of the model to feature changes by adjusting the proportion or range of key features in the validation dataset; S35. Based on the model prediction results, design a risk grading standard to divide the risk of diabetic foot caused by metabolic diseases into different levels, and apply local and global interpretability to explain the model prediction results; S36, and use visualization tools to visualize the risk grading results and explanatory information.

[0009] Furthermore, the S4 includes: S41. Deploy the constructed model to the medical information system and optimize the performance of the model deployment environment. By implementing the model monitoring mechanism, the operating status and prediction performance of the model can be monitored in real time. S42, integrate real-time data stream processing algorithms to collect, clean and pre-process patient test data in real time; S43, analyzing the real-time data through the model, generating risk warning information of metabolic disease diabetic foot, setting the risk threshold, and triggering the automatic alarm mechanism based on the risk threshold; S44. Notify doctors and patients of warning information in real time through multiple means.

[0010] The present invention proposes a metabolic disease diabetic foot test data early screening system, the system comprising: Data acquisition module: obtain clinical sample data and transmit the obtained clinical sample data to the cloud space for preprocessing, perform feature extraction on the preprocessed clinical sample data through deep learning algorithms, and extract features related to the metabolic disease diabetic foot; Model tuning module: Use deep learning algorithms to build a metabolic disease diabetic foot screening model based on routine blood data, and combine the acquired clinical sample data with historical data to build a data set; use multiple indicators to evaluate the constructed metabolic disease diabetic foot screening model, and tune the metabolic disease diabetic foot screening model based on the evaluation results; Model blind test module: introduce external validation data sets to conduct blind tests on the optimized metabolic disease diabetic foot screening model, verify the regional adaptability of the model, and perform risk grading on the model output results based on algorithm interpretation; Early warning acquisition module: deploy the constructed model to the medical information system, receive patient test data in real time, and analyze the test data in real time to obtain metabolic disease diabetic foot risk early warning results.

[0011] Furthermore, the data acquisition module includes: Data transmission module: collects sample data from multiple data sources through a multi-data integration framework, and transmits the collected sample data to the cloud space; Space division module: The cloud space stores the received sample data in different storage spaces according to the data type, and divides a subspace within each storage space; Redundant backup module: perform redundant backup of sample data in the storage space and store the backed-up data in the subspace; pre-process the data stored in the storage space and its corresponding subspace respectively through the processing unit; Difference judgment module: compares the preprocessed results of the storage space and its corresponding subspace to determine whether there is a difference. If there is a difference, the difference part is processed again; Result determination module: takes the result after reprocessing as the benchmark value, compares it with the result of the first preprocessing, and takes the sample data stored in the storage space or subspace with the smallest difference value after preprocessing as the final result; Feature evaluation module: Use deep learning algorithms to extract features from preprocessed data, evaluate the extracted features, and screen out the feature subset with the most predictive value through feature selection technology.

[0012] Furthermore, the model tuning module includes: Data merging module: merge the acquired clinical sample data with the existing historical data to form a comprehensive data set; and pre-process the merged data set; Feature extraction module: Based on deep learning algorithm, it extracts key features related to metabolic disease diabetic foot from preprocessed data; Subset screening module: Through feature importance evaluation, the feature subset that contributes most to the prediction of metabolic disease diabetic foot is screened; Parameter optimization module: adopts a multi-layer perception mechanism, designs the model architecture according to the feature type and data characteristics, trains the model using the divided training set data, and optimizes the model parameters through the back propagation algorithm; Index evaluation module: Use the validation set data to evaluate multiple evaluation indicators of the model, including accuracy, sensitivity and specificity; Iterative optimization module: Based on the evaluation results, iteratively optimizes the model performance by adjusting the model structure, optimizing algorithm parameters, and using regularization techniques; Problem identification module: Identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set; Model selection module: Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model.

[0013] Furthermore, the model blind testing module includes: Data integration module: collects external data sources, integrates the collected external data sources, and constructs an external verification data set; Difference analysis module: blind test the optimized metabolic disease diabetic foot model through external validation data set, analyze the performance difference of the model on the external data set through the blind test results, and identify potential bias and error sources; Fine-tuning training module: cluster analysis is performed on the external validation data set to identify differences and similarities in data distribution. Based on the differences in data distribution, the model is fine-tuned or retrained through transfer learning. Sensitivity assessment module: performs sensitivity analysis to assess the sensitivity of the model to feature changes by adjusting the proportion or range of key features in the validation data set; Model interpretation module: Based on the model prediction results, a risk grading standard is designed to divide the risk of metabolic disease diabetic foot into different levels, and local and global interpretability are applied to explain the model prediction results; Visualization display module: The risk grading results and explanatory information are visualized through visualization tools.

[0014] Furthermore, the warning acquisition module includes: Performance prediction module: deploy the constructed model to the medical information system and optimize the performance of the model deployment environment; monitor the model's operating status and prediction performance in real time by implementing a model monitoring mechanism; Data stream processing module: Integrates real-time data stream processing algorithms to collect, clean and preprocess patient test data in real time; Threshold trigger module: Analyzes real-time data through the model, generates risk warning information for diabetic foot caused by metabolic diseases, sets risk thresholds, and triggers automatic alarm mechanisms based on risk thresholds; Early warning notification module: Notify doctors and patients of early warning information in real time through multiple methods.

[0015] Beneficial effects of the present invention: by extracting features through deep learning algorithms, complex patterns and subtle changes related to metabolic disease diabetic foot can be identified, thereby improving the accuracy of screening. The automated data processing process reduces manual intervention and improves the efficiency of screening. The use of clinical sample data and historical data to construct a data set can provide a more comprehensive and in-depth risk analysis of metabolic disease diabetic foot. Feature importance evaluation ensures that the model focuses on the features that contribute most to the prediction of metabolic disease diabetic foot, thereby improving the predictive ability of the model. The model parameters are optimized by multi-layer perceptron and back propagation algorithm to ensure the predictive performance of the model. The introduction of an external validation data set for blind testing helps to verify the generalization ability of the model and ensure the effectiveness of the model in different regions and populations. The design of risk grading standards to grade the risk of metabolic disease diabetic foot helps doctors to formulate personalized treatment plans based on the risk level. Algorithm interpretation and visualization tools make the prediction results of the model more transparent and explainable, and enhance the trust of medical professionals in the model. The model is deployed in a medical information system to realize real-time analysis and risk warning of patient test data, which is helpful for early detection and intervention. The mechanism of real-time notification of doctors and patients can speed up the response and reduce the risk of serious complications of metabolic disease diabetic foot. The multi-data integration framework and cloud storage ensure the security and accessibility of data. The comparison of data redundancy backup and pre-processing process ensures the accuracy and integrity of data. Through cluster analysis and transfer learning, the model can adapt to the characteristics of different data distributions and enhance the regional adaptability of the model. Perceptual analysis helps identify the sensitivity of the model to changes in key features, further improving the robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a step diagram of the method of the present invention; Figure 2 This is a system module diagram of the present invention; Figure 3 This is a performance analysis diagram of the model described in the present invention; Figure 4This is a schematic diagram of the model structure of the present invention. DETAILED DESCRIPTION

[0017] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0018] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. The embodiments described are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0020] One embodiment of the present invention, as Figure 1 As shown, a method for early screening of metabolic disease diabetic foot test data, the method comprising: S1. Acquire clinical sample data, and transmit the acquired clinical sample data to the cloud space for preprocessing, perform feature extraction on the preprocessed clinical sample data through a deep learning algorithm, and extract features related to the metabolic disease diabetic foot; S2. Construct a metabolic disease diabetic foot screening model based on routine blood data through a deep learning algorithm, and combine the acquired clinical sample data with historical data to construct a data set; divide the data set into a training set, a validation set, and a test set; evaluate the constructed metabolic disease diabetic foot screening model through multiple indicators, including accuracy, sensitivity, and specificity, and optimize the metabolic disease diabetic foot screening model based on the evaluation results; S3. Introduce an external validation dataset to conduct a blind test on the optimized metabolic disease diabetic foot screening model to verify the regional adaptability of the model and perform risk grading on the model output results based on the algorithm interpretation; S4. Deploy the constructed model to the medical information system, receive patient test data in real time, and analyze the test data in real time to obtain metabolic disease diabetic foot risk warning results.

[0021] The working principle of the above technical solution is as follows: first, clinical sample data is collected from medical institutions or related data sources. These data may include patients' blood routine indicators, biochemical indicators, medical history information, etc.; the collected data is transmitted to the cloud space for pre-processing operations such as cleaning, denoising, and normalization to ensure the quality and consistency of the data, laying the foundation for subsequent feature extraction and model training; deep learning algorithms (such as convolutional neural networks CNN, recurrent neural networks RNN, etc.) are used to automatically extract features related to metabolic disease diabetic foot from the pre-processed data. These features may be specific combinations or changing trends of certain blood indicators, which can reflect the risk of metabolic disease diabetic foot; based on the extracted features, a metabolic disease diabetic foot screening model is constructed using a deep learning algorithm. This model can learn the complex relationships between features to predict whether a patient may have metabolic disease diabetic foot; the acquired clinical sample data is combined with historical data to construct a comprehensive data set. Then, the data set is divided into training set, validation set and test set, so as to perform cross-validation and performance testing in the process of model training. The training set is mainly used to fit the model according to the data, the validation set is used to adjust the hyperparameters of the model and make a preliminary evaluation of the model's ability, and the test set is used to evaluate the generalization ability of the final model. The training set data is randomly divided into K parts, one of which is taken as the test set each time, and the rest of the data is used as the training sample to train the model, and then the evaluation index of the trained model is obtained on the test set. K fold means completing the above experiment K times, and finally taking the mean and standard deviation of the evaluation index. The modeling process is implemented by python; the constructed model is evaluated using multiple indicators (such as accuracy, sensitivity, and specificity). The performance of the model is continuously improved by adjusting model parameters, optimizing algorithms, etc. In particular, the best balance between sensitivity and specificity is found to ensure that the model can accurately identify high-risk patients and avoid misjudgment of low-risk patients; an external validation data set is introduced to blindly test the tuned model to verify the regional adaptability and generalization ability of the model. This step is the key to ensure that the model can work stably and reliably in practical applications; based on algorithm interpretation (such as feature importance analysis, decision path visualization, etc.), the output results of the model are risk graded. Patients are divided into different levels according to the risk of metabolic disease diabetic foot, providing doctors with an intuitive reference basis; the constructed model is deployed to the medical information system to achieve seamless integration with the existing medical process; the system receives the patient's test data in real time, and uses the deployed model to analyze the data in real time to quickly generate metabolic disease diabetic foot risk warning results; doctors formulate personalized diagnosis and treatment plans based on the warning results and the patient's specific situation. For high-risk patients, intervention and treatment can be carried out in advance to reduce the incidence and severity of metabolic disease diabetic foot.

[0022] The effect of the above technical solution is: by extracting features from clinical sample data through deep learning algorithms, it can automatically learn and identify complex features that are highly related to metabolic disease diabetic foot. Compared with traditional statistical methods, this method can more accurately capture subtle changes in data, thereby improving the accuracy of screening; after the model is deployed in the medical information system, it can receive and analyze the patient's test data in real time, and quickly generate metabolic disease diabetic foot risk warning results. This enables doctors to detect potential high-risk patients early and take corresponding intervention measures, such as improving blood sugar control and strengthening foot care, so as to effectively delay or prevent the occurrence and development of metabolic disease diabetic foot; through accurate metabolic disease diabetic foot risk warning, it can avoid excessive examination and treatment of low-risk patients, while ensuring that high-risk patients receive timely and effective attention and treatment. This precision medical model helps to optimize the allocation of medical resources and reduce unnecessary medical costs; early detection and intervention of metabolic disease diabetic foot can significantly reduce complications and disabilities caused by metabolic disease diabetic foot, such as ulcers, infections, amputations, etc. This not only alleviates the physical pain of patients, but also improves their quality of life and social participation; this method demonstrates the application potential of AI technology in the medical field and promotes the innovation and progress of medical technology. With the continuous development and improvement of technology, AI-based metabolic disease diabetic foot screening methods will become more intelligent, efficient and personalized; by introducing external validation data sets to blindly test the model, the regional adaptability of the model was verified. This means that the model is not only suitable for specific medical institutions or regions, but can also be promoted and applied in a wider range, providing more patients with accurate metabolic disease diabetic foot risk warning services.

[0023] In one embodiment of the present invention, the S1 includes: S11. Collect sample data from multiple data sources through a multi-data integration framework, and transmit the collected sample data to the cloud space; S12, the cloud space stores the received sample data in different storage spaces according to the data type, and divides a subspace in each storage space; S13, performing redundant backup of the sample data in the storage space, and storing the backed-up data in the subspace; preprocessing the data stored in the storage space and its corresponding subspace respectively by the processing unit; S14, comparing the preprocessed results of the storage space and its corresponding subspace to determine whether there is a difference. If there is a difference, the difference part is processed again; S15, taking the result after the second processing as a reference value, comparing it with the result of the first preprocessing, and taking the sample data stored in the storage space or subspace with the smallest difference value after preprocessing as the final result; S16. Use deep learning algorithms to extract features from preprocessed data, evaluate the extracted features, and select the feature subset with the most predictive value through feature selection technology.

[0024] The working principle of the above technical solution is as follows: by building a multi-data integration framework including an electronic medical record system (EMR) and a laboratory information system (LIS), a comprehensive collection of sample data from different data sources (such as different hospitals, different departments, different inspection equipment, etc.) is achieved. These data sources provide rich clinical information, including routine indicator data of patients with metabolic diseases such as diabetic foot (positive data) and non-target populations (negative data); the collected sample data is securely transmitted to the cloud space for subsequent processing and analysis. The cloud space provides powerful storage and computing capabilities, which can support the rapid processing of large-scale data; after receiving the data, the cloud space first stores the sample data in different storage spaces according to the data type (such as blood routine, biochemical indicators, imaging data, etc.). This classified storage method facilitates subsequent data management and processing; within each storage space, a subspace is further divided for redundant backup of data storage. This improves the security and reliability of the data and prevents analysis interruptions caused by data loss or damage; the sample data in the storage space is redundantly backed up, and the backed-up data is stored in the subspace. This ensures that the data can be recovered under any circumstances, and guarantees the integrity and availability of the data; the data stored in the storage space and its corresponding subspace are preprocessed separately by the processing unit. The preprocessing steps may include data cleaning (removing noise, outliers, etc.), data conversion (standardization, normalization, etc.) and data compression, etc., to improve the quality of the data and reduce the amount of calculation; the results of the preprocessing of the storage space and its corresponding subspace are compared to determine whether there are differences. Since there may be a variety of preprocessing methods and parameter settings, it is necessary to compare different processing results; if there are differences, the difference part is processed again. This may involve adjusting the parameters of the preprocessing method or adopting a new preprocessing method to eliminate or reduce the difference; the result after the reprocessing is used as a benchmark value and judged with the result of the first preprocessing. By comparing the similarities and differences between different processing results, the optimal preprocessing scheme is selected; the sample data stored in the storage space or subspace with the smallest difference value after preprocessing is taken as the final result. This result will serve as the basis for subsequent feature extraction and model construction; the preprocessed data is extracted using a deep learning algorithm (such as a convolutional neural network CNN, a recurrent neural network RNN ​​or an autoencoder). These algorithms can automatically learn high-level feature representations in the data to improve the model's predictive ability, evaluate the extracted features, and select the most predictive feature subsets through feature selection techniques (such as principal component analysis (PCA) and recursive feature elimination (RFE). This helps reduce the complexity of the model and improve the model's generalization ability.

[0025] The effect of the above technical solution is: through the multi-data integration framework, sample data can be collected from multiple data sources such as electronic medical record systems and laboratory information systems to ensure the comprehensiveness and diversity of the data. This helps to capture information of different dimensions and levels, and provide a rich data foundation for subsequent analysis and modeling; the collected data is transferred to the cloud space for storage, which not only improves the accessibility and scalability of the data, but also enhances the security of the data. The cloud space provides professional security protection measures to ensure that the data is not illegally accessed or tampered with; the sample data in the storage space is redundantly backed up, and the backed-up data is stored in the subspace, further improving the reliability and availability of the data. Even if the original data is lost or damaged, it can be restored through the backup data; the sample data is stored in different storage spaces according to the data type, and the data is preprocessed. This classification processing method helps to improve the integrity of the data, and adopts appropriate preprocessing methods for different types of data, thereby improving the accuracy of the preprocessing; by comparing the results of the storage space and its corresponding subspace preprocessing, the difference can be discovered and processed in time. This comparison mechanism helps to ensure the consistency and accuracy of the data, and provides reliable data support for subsequent feature extraction and model building. The use of deep learning algorithms to extract features from preprocessed data can automatically learn high-level feature representations in the data and improve the intelligence level of feature extraction. Compared with traditional manual feature extraction methods, deep learning algorithms can more accurately capture the inherent laws and characteristics of data. The feature selection technology can screen out the most predictive feature subsets, which helps to reduce the complexity of the model and improve the generalization ability of the model. At the same time, evaluating the extracted features can ensure the validity and reliability of the selected features. Through a series of data processing and analysis steps, the high quality of the data required for subsequent model construction is ensured. This provides accurate data support for the early screening of metabolic disease diabetic foot, which helps to improve the accuracy and reliability of screening. Through the model deployed in the medical information system, patient test data can be received and analyzed in real time, and the risk warning results of metabolic disease diabetic foot can be quickly generated, which helps to achieve early intervention and treatment.

[0026] In one embodiment of the present invention, the S2 includes: S21. Merge the acquired clinical sample data (including blood routine data and other relevant indicators) with the existing historical data to form a comprehensive data set; and pre-process the merged data set. S22. Extract key features related to metabolic disease diabetic foot from preprocessed data based on deep learning algorithm; S23. Through feature importance evaluation, screen the feature subset that contributes most to the prediction of metabolic disease diabetic foot; S24, adopting a multi-layer perception mechanism, designing a model architecture according to feature types and data characteristics, using the divided training set data to train the model, and optimizing the model parameters through the back propagation algorithm; S25. Evaluate multiple evaluation indicators of the model using the validation set data, wherein the multiple indicators include accuracy, sensitivity, and specificity; Among them, accuracy is a basic indicator to measure model performance, which represents the ratio between the number of samples correctly predicted by the model and the total number of samples. In this model, 252 cases of clinically diagnosed non-metabolic diabetic foot patients matched with low-risk model warning data, and 61 cases of clinically diagnosed metabolic diabetic foot patients matched with medium- and high-risk model warning data, a total of 295 cases of accurate data were captured, and the total number of samples was 340, that is, the accuracy rate was 86.76%.

[0027] Sensitivity usually refers to the sensitivity of the model output to changes in input data in machine learning or statistical analysis. In this model, 61 patients with clinical diagnosis of metabolic disease diabetic foot matched the model warning of medium and high risk, and 16 patients with clinical diagnosis of metabolic disease diabetic foot matched the model warning of low risk, with a model sensitivity of 79.22%.

[0028] Specificity measures the accuracy of the model when predicting the negative class. A model with high specificity is able to avoid misclassifying negative samples as positive. In this model, high specificity means that the model rarely misdiagnoses healthy people as sick, reducing false alarms and the associated psychological stress, economic costs, and potential risks to patients' health.

[0029] The calculation formula of specificity is as follows: Specificity = True Negative / (True Negative + False Positive) The true negative cases in this model were patients with diabetic foot who were clinically diagnosed with non-metabolic diseases. The model warned of 252 low-risk cases, and the true negative cases and false positive cases were 323, that is, the specificity was 78.02%.

[0030] S26. Based on the evaluation results, iteratively optimize the model performance by adjusting the model structure, optimizing algorithm parameters, and using regularization techniques; S27. Checking by fitting and underfitting: Identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set.

[0031] S28. Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model.

[0032] The working principle of the above technical solution is: merge the acquired clinical sample data (including blood routine data and other related indicators) with the existing historical data to form a comprehensive data set covering multiple dimensions (such as age group, gender, metabolic disease diabetes course, whether there is metabolic disease diabetic foot, etc.); clean the merged data set to remove data records with too many missing values, outliers or illogical data to ensure the quality and reliability of the data. At the same time, standardization is performed to convert the data to the same scale for subsequent analysis and modeling; deep learning algorithms (such as convolutional neural networks CNN, recurrent neural networks RNN or their variants) are used to automatically extract key features related to metabolic disease diabetic foot from the preprocessed data. These features may include blood routine indicators such as blood sugar level, hemoglobin, white blood cell count, platelet count, and other clinical information that may be related to the risk of metabolic disease diabetic foot; analyze the contribution of each feature to the prediction of metabolic disease diabetic foot through feature importance evaluation methods (such as feature importance based on decision trees, mutual information, etc.), and screen out the feature subset that has the greatest impact on the prediction results. This helps to simplify the model structure and improve the prediction accuracy and efficiency of the model; design the multi-layer perceptron model architecture according to the feature type and data characteristics. The multi-layer perceptron is a feedforward neural network that can handle complex nonlinear relationships. The model is trained using the divided training set data. The model parameters are optimized through the back propagation algorithm so that the model can more accurately predict the risk of diabetic foot disease caused by metabolic diseases. The model is evaluated using multiple evaluation indicators using the validation set data, including accuracy, sensitivity, and specificity. These indicators can fully reflect the predictive performance of the model, such as Figure 3 As shown; according to the evaluation results, the model performance is iteratively optimized by adjusting the model structure (such as increasing or decreasing the number of network layers, changing the activation function, etc.), optimizing algorithm parameters (such as learning rate, batch size, etc.), and using regularization techniques (such as L1 / L2 regularization, Dropout, etc.). The purpose is to improve the prediction accuracy and generalization ability of the model; identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set. Overfitting is usually manifested as a small loss on the training set but a large loss on the validation set, while underfitting is manifested as a large loss on both the training set and the validation set; for the overfitting problem, regularization, increasing the amount of training data, using Dropout and other techniques can be used to reduce the complexity of the model; for the underfitting problem, you can try to increase the model complexity (such as increasing the number of network layers), adjust the learning rate and other methods to improve the model's fitting ability; after the above tuning process, select the model with the best performance as the final metabolic disease diabetic foot screening model, such as Figure 4This model will be used for subsequent real-time data analysis and risk warning.

[0033] The effect of the above technical solution is: the clinical sample data is merged with the existing historical data to form a comprehensive data set covering multiple dimensions such as different age groups, gender, metabolic disease diabetes course, and whether or not there is metabolic disease diabetic foot. This comprehensive data set helps the model learn broader and deeper features and improve the generalization ability of the model; through data cleaning and standardization, data records with too many missing values, outliers or illogical data are removed to ensure the quality and reliability of the data. This helps to reduce noise interference during model training and improve the prediction accuracy of the model; deep learning algorithms (such as CNN, RNN or their variants) are used to automatically extract key features related to metabolic disease diabetic foot from preprocessed data. Deep learning algorithms can automatically learn high-level feature representations in data, which is more efficient and accurate than traditional manual feature extraction methods; feature importance evaluation is used to screen out the feature subset that contributes most to the prediction of metabolic disease diabetic foot. This helps to reduce the complexity of the model and improve the prediction speed and accuracy of the model; a multi-layer perceptron is used as the model architecture, and it is flexibly designed according to feature types and data characteristics. The multi-layer perceptron can handle complex nonlinear relationships and is suitable for the prediction of metabolic diseases such as diabetic foot, which is affected by multiple factors. The model is trained using the divided training set data, and the model parameters are optimized by the back propagation algorithm. At the same time, the model is iteratively optimized according to the evaluation results of the validation set data, including adjusting the model structure, optimizing the algorithm parameters, and using regularization techniques. These measures help improve the prediction performance and stability of the model. The model is evaluated using multiple evaluation indicators, including accuracy, sensitivity, and specificity, using the validation set data. These evaluation indicators can fully reflect the prediction performance of the model and provide an objective basis for the selection and optimization of the model. Overfitting or underfitting problems can be identified and handled by checking the loss curves on the training set and validation set and the performance of the model on the test set. This helps to ensure the generalization ability of the model on unknown data and improve the practicality and reliability of the model. Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model. The model performs well in terms of accuracy, sensitivity, and specificity, and can accurately predict the risk of metabolic disease diabetic foot in patients, providing strong support for early intervention and treatment.

[0034] In one embodiment of the present invention, S3 includes: S31, collecting external data sources, and integrating the collected external data sources to construct an external verification data set; S32. Perform a blind test on the tuned metabolic disease diabetic foot model using an external validation dataset, analyze the performance differences of the model on the external dataset based on the blind test results, and identify potential biases and error sources; S33. Perform cluster analysis on the external validation data set to identify differences and similarities in data distribution. Based on the differences in data distribution, fine-tune or retrain the model through transfer learning. S34. Perform sensitivity analysis to evaluate the sensitivity of the model to feature changes by adjusting the proportion or range of key features in the validation data set; the sensitivity is calculated using the following formula:

[0035] Among them, S is the comprehensive sensitivity index of the model, N is the number of key features, and P is the model performance index (such as accuracy and AUC). Features The performance of the model after adjustment, Features The adjustment amount, Features The adjustment amount, Features The weight of is the weight of feature interaction, Features The weight of It is to adjust the features simultaneously and The performance of the model after is a small constant used to avoid the denominator being zero.

[0036] S35. Based on the model prediction results, design a risk grading standard to divide the risk of diabetic foot caused by metabolic diseases into different levels, and apply local and global interpretability to explain the model prediction results; S36, and through visualization tools, the risk grading results and explanatory information are visualized.

[0037] The working principle of the above technical solution is: collect external data sources from multiple channels (such as different hospitals, different regions, and different population structures) to ensure the diversity and breadth of the validation data set; integrate the collected external data sources to form an external validation data set. This data set will be used to evaluate the performance of the model on unknown or new data; use the external validation data set to blindly test the tuned metabolic disease diabetic foot model, that is, make predictions without knowing the true label of the data; analyze the performance differences of the model on the external data set through the blind test results, including changes in evaluation indicators such as accuracy, sensitivity, and specificity. At the same time, identify potential sources of bias and error, such as data distribution differences and feature missing; perform cluster analysis on the external validation data set to identify differences and similarities in data distribution. This helps to understand the characteristics between different data subsets and provide a basis for subsequent model adjustments; fine-tune or retrain the model based on the differences in data distribution through technologies such as transfer learning. Transfer learning can use the knowledge of existing models to accelerate the learning process of new tasks and improve the performance of the model on new data; by adjusting the proportion or range of key features in the validation data set, evaluate the sensitivity of the model to feature changes. This helps to understand the stability and reliability of the model under different conditions; according to the results of sensitivity analysis, analyze which changes in features the model is more sensitive to and how these changes affect the model's prediction results; based on the model's prediction results, design risk grading standards to divide the risk of metabolic disease diabetic foot into different levels. This helps doctors and patients understand the patient's risk status more intuitively; apply local interpretability (such as LIME, SHAP) and global interpretability (such as feature importance analysis, decision tree approximation) methods to explain the model's prediction results. Local interpretability methods can explain the basis of a single prediction result, while global interpretability methods can reveal the overall working mechanism and feature importance of the model. These explanatory information helps doctors and patients understand the source of risk and the basis for prediction; through visualization tools, the risk grading results and explanatory information are displayed in the form of charts, heat maps, etc. These visualizations help to intuitively convey the model's prediction results and explanatory information, and improve the understanding and acceptance of doctors and patients.

[0038] The effect of the above technical solution is: by collecting external data sources from different hospitals, regions and population structures, and constructing an external validation data set, the performance of the model in different environments can be comprehensively evaluated. This helps to discover the potential bias and error of the model on unknown data, thereby improving the generalization ability of the model; blind testing the tuned model can objectively evaluate the performance differences of the model on the external data set, and then accurately identify the potential sources of bias and error. This provides a strong basis for subsequent model adjustments; cluster analysis is performed on the external validation data set to identify the differences and similarities in data distribution. Based on these analysis results, fine-tuning or retraining the model through transfer learning can effectively reduce the decline in model performance caused by differences in data distribution, and further improve the accuracy and stability of the model; by adjusting the proportion or range of key features in the validation data set, the sensitivity of the model to feature changes is evaluated. This helps to understand the performance of the model under different conditions and enhance the robustness of the model to feature fluctuations; designing risk grading standards based on the model prediction results, and interpreting the prediction results through local and global interpretative methods, which helps doctors and patients to more clearly understand the risk level and prediction basis of metabolic disease diabetic foot. This can assist doctors in formulating more scientific and reasonable treatment plans, improve treatment effects and patient satisfaction; through visualization tools, risk grading results and explanatory information are displayed in the form of charts, heat maps, etc., making complex model prediction results intuitive and easy to understand. This helps to improve the communication efficiency between doctors and patients and enhance patients' understanding and acceptance of disease risks and treatment plans. The above formula combines external data integration, blind test evaluation, cluster analysis, transfer learning and sensitivity analysis, which can effectively improve the reliability and generalization ability of the metabolic disease diabetic foot model, and provide strong support for model deployment and optimization in practical application scenarios. The formula considers the impact of individual features and interactions between features on model performance, and comprehensively considers the contribution and mutual influence of each feature on the model in a weighted manner. This method can help determine which features have the greatest impact on the performance of the model, thereby guiding subsequent optimization and adjustment strategies. Through cluster analysis and transfer learning of external validation data sets, the differences in data distribution can be identified, and the model can be fine-tuned or retrained in a targeted manner. Doing so can improve the generalization ability of the model on new data and reduce prediction bias and errors introduced by data bias. The formulas in the sensitivity analysis section can help determine the sensitivity of the model to changes in different features, which is crucial for further feature selection and model optimization. By analyzing the weights and interaction effects of each feature, the feature engineering process can be optimized to improve the stability and predictive ability of the model under various input conditions.

[0039] In one embodiment of the present invention, the S32 includes: Desensitize the collected external data sources and set up a verification environment in the cloud that is completely isolated from the model training environment; Divide the external validation dataset into multiple batches and conduct blind testing in a random or chronological manner to simulate the data characteristics of different time periods or different patient groups; During the blind test, key indicators such as the model's response time, prediction accuracy, recall rate, and F1 score are monitored in real time, and the performance differences of the model on different data batches are recorded; Through the anomaly detection mechanism, outliers or extreme situations in the prediction results are automatically captured and recorded; using the feature importance evaluation method, the contribution of different features in model prediction is analyzed to identify the features that affect model performance; By comparing the feature distribution of the training data set with the external validation data set, we can identify the source of data deviation, such as regional differences, differences in patient groups, etc. We can also subdivide prediction errors, such as false positives and false negatives, and analyze the causes of each type of error. Use statistical methods to quantify the degree of data deviation, combine the internal structure of the model and external data characteristics, and trace the source of the error; Develop appropriate response strategies for identified sources of deviations and errors.

[0040] The working principle of the above technical solution is as follows: in order to protect patient privacy and data security, the collected external data sources are desensitized, and sensitive information (such as patient identifiers) is deleted or replaced, while retaining the integrity of the data for model verification; a verification environment that is completely isolated from the model training environment is set up in the cloud. This environment simulates the actual clinical application scenario, including data interfaces, processing procedures, etc., to ensure the independence and fairness of the verification process; the external verification data set is divided into multiple batches, and blind tests are performed in a random or chronological manner to simulate the data characteristics of different time periods or different patient groups; during the blind test, the model's response time, prediction accuracy, recall rate, F1 score and other key indicators are monitored in real time, and the performance differences of the model on different data batches are recorded. This helps to promptly discover performance fluctuations of the model under different data distributions; automatically capture and record outliers or extreme situations in the prediction results through the anomaly detection mechanism, providing clues for subsequent analysis; use feature importance evaluation methods to analyze the contribution of different features in model prediction, and identify features that have a significant impact on model performance; subdivide prediction errors (such as false positives, false negatives, etc.), and analyze the causes of each type of error (such as model overfitting, underfitting, feature noise, etc.). At the same time, by comparing the characteristic distribution of the training data set with that of the external validation data set, the source of data deviation (such as regional differences, differences in patient groups, etc.) is identified; statistical methods (such as calculating the coefficient of deviation, Kullback-Leibler divergence, etc.) are used to quantify the degree of data deviation to clarify the severity of the deviation; combining the internal structure of the model and external data characteristics, the source of the error (such as improper data preprocessing, model structure limitations, improper algorithm selection, etc.) is traced; for the identified deviations and error sources, corresponding response strategies are formulated (such as data enhancement, model structure optimization, algorithm adjustment, etc.) to improve the generalization ability and robustness of the model; the model is iteratively optimized according to the response strategy, and blind testing and performance evaluation are re-performed to ensure that the performance of the model on the external data set is improved.

[0041] The effect of the above technical solution is: by desensitizing the collected external data sources, especially the anonymization of sensitive information, such as deleting or replacing patient identifiers, the personal privacy of patients is effectively protected. This is a crucial link in medical data processing and complies with the requirements of relevant laws and regulations; a verification environment that is completely isolated from the model training environment is set up in the cloud to simulate actual clinical application scenarios, including data interfaces, processing procedures, etc., to ensure the independence and fairness of the verification process. This isolation measure avoids possible bias and interference in the verification process and improves the reliability of the verification results; through batch, random or chronological blind testing, the data characteristics of different time periods or different patient groups are simulated, so that the model can be tested more comprehensively. At the same time, real-time monitoring of multiple key indicators of the model and recording the performance differences of the model on different data batches will help to discover the performance fluctuations of the model in different scenarios. According to the identified sources of deviation and error, corresponding response strategies are formulated, such as data enhancement, model structure optimization, algorithm adjustment, etc., which can effectively improve the generalization ability and robustness of the model, so that it can maintain stable performance in different patient groups; the abnormal detection mechanism automatically captures and records the abnormal values ​​or extreme cases in the prediction results, and the feature importance evaluation method is used to analyze the contribution of different features in model prediction, which can deeply understand the internal working mechanism and performance of the model. This in-depth analysis helps to identify the features and potential improvement points that have a significant impact on the performance of the model, and provides strong support for subsequent model optimization; the degree of data deviation is quantified by statistical methods, such as calculating the deviation coefficient, Kullback-Leibler divergence, etc., which can clarify the severity of data deviation. At the same time, tracing the source of the error in combination with the internal structure of the model and the external data characteristics helps to accurately determine the cause of the error and formulate corresponding solutions. This method of quantitative analysis and error tracing provides a scientific basis for the continuous optimization of the model; the above technical solution is used to externally verify and evaluate the performance of the metabolic disease diabetic foot screening model, which can ensure the accuracy and reliability of the model in clinical application. This will not only help doctors more accurately judge patients' conditions and risk levels, but also provide patients with more personalized and precise diagnosis and treatment recommendations, thereby improving the overall effectiveness and value of clinical applications.

[0042] In one embodiment of the present invention, the S4 includes: S41. Deploy the constructed model to the medical information system and optimize the performance of the model deployment environment. By implementing the model monitoring mechanism, the operating status and prediction performance of the model can be monitored in real time. S42, integrate real-time data stream processing algorithms to collect, clean and pre-process patient test data in real time; S43. Analyze the real-time data through the model to generate metabolic disease diabetic foot risk warning information, set a risk threshold, and trigger an automatic alarm mechanism based on the risk threshold; wherein the warning information includes a risk prediction value, and the calculation formula of the risk prediction value is as follows:

[0043] Where Y represents the risk prediction value, represents the intercept term, Represents the weight coefficient, each coefficient corresponds to an input feature.

[0044] S44. Early warning information will be notified to doctors and patients in real time through multiple means including SMS, email and APP push.

[0045] The working principle of the above technical solution is: deploy the constructed and verified metabolic disease diabetic foot screening model into the medical information system. This step involves integrating the model file (such as a trained neural network model, decision tree model, etc.) into the back-end service of the medical information system so that it can receive input data and return prediction results; optimizing the performance of the model deployment environment, including the reasonable configuration of hardware resources (such as CPU, GPU, memory) to ensure that the model can run efficiently; adapting the software environment (such as operating system, Python version, dependent library) to avoid compatibility issues; and optimizing the network environment (such as bandwidth, latency) to ensure the stability and real-time performance of data transmission; implementing a model monitoring mechanism to monitor the operating status and prediction performance of the model in real time. Through monitoring, potential problems in the operation of the model can be discovered in a timely manner, such as decreased prediction accuracy, prolonged response time, etc., and corresponding measures can be taken to solve them; integrating real-time data stream processing algorithms (such as Kafka, SparkStreaming) to achieve real-time collection of patient test data. The collected data may contain noise, outliers, etc., so it needs to be cleaned and preprocessed to ensure data quality; the cleaned data is preprocessed, including data formatting, feature extraction, normalization and other steps to meet the requirements of model input; the preprocessed real-time data is input into the metabolic disease diabetic foot screening model for analysis, and the model generates metabolic disease diabetic foot risk prediction results based on the input data; the risk threshold is set according to clinical needs and actual conditions. When the prediction result exceeds the set risk threshold, the patient is considered to have a high risk of metabolic disease diabetic foot; the automatic alarm mechanism is triggered based on the risk threshold. When the prediction result exceeds the threshold, the system automatically generates metabolic disease diabetic foot risk warning information; the warning information is notified to doctors and patients in real time through a variety of methods. This includes sending text messages, emails, and push notifications through APP. These notification methods can ensure that doctors and patients can receive warning information in a timely manner and take appropriate measures to intervene.

[0046] The effect of the above technical solution is: by integrating real-time data stream processing algorithms, it is possible to realize real-time collection, cleaning and preprocessing of patient test data, and then quickly generate metabolic disease diabetic foot risk warning information. This greatly shortens the time from data collection to risk assessment and improves the efficiency of diagnosis and treatment; deploys a fully verified model and combines performance optimization measures to ensure the stable operation and accurate prediction of the model in the medical information system. This helps doctors to more accurately assess the risk of metabolic disease diabetic foot in patients, so as to formulate more effective treatment plans; the automatic alarm mechanism triggered by risk thresholds can issue warnings in time when the patient's risk increases, so that doctors and patients can respond quickly and take corresponding measures. This helps to reduce the incidence of complications of metabolic disease diabetic foot and improve the quality of life of patients; real-time notification of doctors and patients through SMS, email and APP push and other methods ensures the timely communication and reception of warning information. This helps to enhance patients' attention and management of their own health status and improve treatment compliance; the performance of the model deployment environment is optimized in hardware, software and network, ensuring the stable operation and efficient processing of the medical information system. This helps to improve the operating efficiency of the entire medical system and reduce delays in diagnosis and treatment caused by system failures or delays; the implementation of a model monitoring mechanism can monitor the operating status and predictive performance of the model in real time, and promptly discover and solve potential problems. This helps to ensure the continued effectiveness and reliability of the model in the medical information system; through real-time data stream processing algorithms, centralized management and sharing of patient test data are achieved. This helps to achieve data interoperability and resource sharing between different medical institutions and improve the utilization efficiency of medical resources; the multi-channel early warning information notification mechanism promotes communication and collaboration between doctors, patients and other medical team members. This helps to form a closer medical team to jointly cope with the challenges of chronic diseases such as metabolic diseases and diabetic foot. The above formula outputs a probability value, which can intuitively represent the risk level of patients with metabolic diseases and diabetic foot. Generally speaking, the higher the probability value, the greater the risk, and doctors and patients can formulate preventive measures or treatment plans based on this probability value; the logistic regression model is relatively simple and easy to understand and explain, especially in the medical field, doctors and patients may be more likely to accept this risk prediction model based on linear weights; the formula combines real-time data stream processing and model monitoring mechanism, which can monitor patients' test data in real time and generate corresponding risk warning information. This real-time nature helps to intervene in time and prevent the occurrence or progression of the disease; setting risk thresholds and triggering automatic alarm mechanisms can automatically notify doctors and patients when the risk value exceeds the preset threshold, helping to detect and deal with problems early; early warning information is notified to doctors and patients in real time through a variety of methods (SMS, email, APP push), which improves the timeliness and accessibility of information and ensures that key information can be effectively communicated.

[0047] One embodiment of the present invention, as Figure 2 As shown, a metabolic disease diabetic foot test data early screening system, the system comprises: Data acquisition module: obtain clinical sample data, transfer the obtained clinical sample data to the cloud space for preprocessing, perform feature extraction on the preprocessed clinical sample data through deep learning algorithms, and extract features related to metabolic disease diabetic foot; Model tuning module: Use deep learning algorithms to build a metabolic disease diabetic foot screening model based on routine blood data, and combine the acquired clinical sample data with historical data to build a data set; use multiple indicators to evaluate the constructed metabolic disease diabetic foot screening model, and tune the metabolic disease diabetic foot screening model based on the evaluation results; Model blind test module: introduce external validation data sets to conduct blind tests on the optimized metabolic disease diabetic foot screening model, verify the regional adaptability of the model, and perform risk grading on the model output results based on algorithm interpretation; Early warning acquisition module: deploy the constructed model to the medical information system, receive patient test data in real time, and analyze the test data in real time to obtain metabolic disease diabetic foot risk early warning results.

[0048] The working principle of the above technical solution is as follows: first, clinical sample data is collected from medical institutions or related data sources. These data may include patients' blood routine indicators, biochemical indicators, medical history information, etc.; the collected data is transmitted to the cloud space for pre-processing operations such as cleaning, denoising, and normalization to ensure the quality and consistency of the data, laying the foundation for subsequent feature extraction and model training; deep learning algorithms (such as convolutional neural networks CNN, recurrent neural networks RNN, etc.) are used to automatically extract features related to metabolic disease diabetic foot from the pre-processed data. These features may be specific combinations or changing trends of certain blood indicators, which can reflect the risk of metabolic disease diabetic foot; based on the extracted features, a metabolic disease diabetic foot screening model is constructed using a deep learning algorithm. This model can learn the complex relationships between features to predict whether a patient may have metabolic disease diabetic foot; the acquired clinical sample data is combined with historical data to construct a comprehensive data set. Then, the data set is divided into training set, validation set and test set, so as to perform cross-validation and performance testing in the process of model training. The training set is mainly used to fit the model according to the data, the validation set is used to adjust the hyperparameters of the model and make a preliminary evaluation of the model's ability, and the test set is used to evaluate the generalization ability of the final model. The training set data is randomly divided into K parts, one of which is taken as the test set each time, and the rest of the data is used as the training sample to train the model, and then the evaluation index of the trained model is obtained on the test set. K fold means completing the above experiment K times, and finally taking the mean and standard deviation of the evaluation index. The modeling process is implemented by python; the constructed model is evaluated using multiple indicators (such as accuracy, sensitivity, and specificity). The performance of the model is continuously improved by adjusting model parameters, optimizing algorithms, etc. In particular, the best balance between sensitivity and specificity is found to ensure that the model can accurately identify high-risk patients and avoid misjudgment of low-risk patients; an external validation data set is introduced to blindly test the tuned model to verify the regional adaptability and generalization ability of the model. This step is the key to ensure that the model can work stably and reliably in practical applications; based on algorithm interpretation (such as feature importance analysis, decision path visualization, etc.), the output results of the model are risk graded. Patients are divided into different levels according to the risk of metabolic disease diabetic foot, providing doctors with an intuitive reference basis; the constructed model is deployed to the medical information system to achieve seamless integration with the existing medical process; the system receives the patient's test data in real time, and uses the deployed model to analyze the data in real time to quickly generate metabolic disease diabetic foot risk warning results; doctors formulate personalized diagnosis and treatment plans based on the warning results and the patient's specific situation. For high-risk patients, intervention and treatment can be carried out in advance to reduce the incidence and severity of metabolic disease diabetic foot.

[0049] The effect of the above technical solution is: by extracting features from clinical sample data through deep learning algorithms, it can automatically learn and identify complex features that are highly related to metabolic disease diabetic foot. Compared with traditional statistical methods, this method can more accurately capture subtle changes in data, thereby improving the accuracy of screening; after the model is deployed in the medical information system, it can receive and analyze the patient's test data in real time, and quickly generate metabolic disease diabetic foot risk warning results. This enables doctors to detect potential high-risk patients early and take corresponding intervention measures, such as improving blood sugar control and strengthening foot care, so as to effectively delay or prevent the occurrence and development of metabolic disease diabetic foot; through accurate metabolic disease diabetic foot risk warning, it can avoid excessive examination and treatment of low-risk patients, while ensuring that high-risk patients receive timely and effective attention and treatment. This precision medical model helps to optimize the allocation of medical resources and reduce unnecessary medical costs; early detection and intervention of metabolic disease diabetic foot can significantly reduce complications and disabilities caused by metabolic disease diabetic foot, such as ulcers, infections, amputations, etc. This not only alleviates the physical pain of patients, but also improves their quality of life and social participation; this method demonstrates the application potential of AI technology in the medical field and promotes the innovation and progress of medical technology. With the continuous development and improvement of technology, AI-based metabolic disease diabetic foot screening methods will become more intelligent, efficient and personalized; by introducing external validation data sets to blindly test the model, the regional adaptability of the model was verified. This means that the model is not only suitable for specific medical institutions or regions, but can also be promoted and applied in a wider range, providing more patients with accurate metabolic disease diabetic foot risk warning services.

[0050] In one embodiment of the present invention, the data acquisition module includes: Data transmission module: collects sample data from multiple data sources through a multi-data integration framework, and transmits the collected sample data to the cloud space; Space division module: The cloud space stores the received sample data in different storage spaces according to the data type, and divides a subspace within each storage space; Redundant backup module: perform redundant backup of sample data in the storage space and store the backed-up data in the subspace; pre-process the data stored in the storage space and its corresponding subspace respectively through the processing unit; Difference judgment module: compares the preprocessed results of the storage space and its corresponding subspace to determine whether there is a difference. If there is a difference, the difference part is processed again; Result determination module: takes the result after reprocessing as the benchmark value, compares it with the result of the first preprocessing, and takes the sample data stored in the storage space or subspace with the smallest difference value after preprocessing as the final result; Feature evaluation module: Use deep learning algorithms to extract features from preprocessed data, evaluate the extracted features, and screen out the feature subset with the most predictive value through feature selection technology.

[0051] The working principle of the above technical solution is as follows: by building a multi-data integration framework including an electronic medical record system (EMR) and a laboratory information system (LIS), a comprehensive collection of sample data from different data sources (such as different hospitals, different departments, different inspection equipment, etc.) is achieved. These data sources provide rich clinical information, including routine indicator data of patients with metabolic diseases such as diabetic foot (positive data) and non-target populations (negative data); the collected sample data is securely transmitted to the cloud space for subsequent processing and analysis. The cloud space provides powerful storage and computing capabilities, which can support the rapid processing of large-scale data; after receiving the data, the cloud space first stores the sample data in different storage spaces according to the data type (such as blood routine, biochemical indicators, imaging data, etc.). This classified storage method facilitates subsequent data management and processing; within each storage space, a subspace is further divided for redundant backup of data storage. This improves the security and reliability of the data and prevents analysis interruptions caused by data loss or damage; the sample data in the storage space is redundantly backed up, and the backed-up data is stored in the subspace. This ensures that the data can be recovered under any circumstances, and guarantees the integrity and availability of the data; the data stored in the storage space and its corresponding subspace are preprocessed separately by the processing unit. The preprocessing steps may include data cleaning (removing noise, outliers, etc.), data conversion (standardization, normalization, etc.) and data compression, etc., to improve the quality of the data and reduce the amount of calculation; the results of the preprocessing of the storage space and its corresponding subspace are compared to determine whether there are differences. Since there may be a variety of preprocessing methods and parameter settings, it is necessary to compare different processing results; if there are differences, the difference part is processed again. This may involve adjusting the parameters of the preprocessing method or adopting a new preprocessing method to eliminate or reduce the difference; the result after the reprocessing is used as a benchmark value and judged with the result of the first preprocessing. By comparing the similarities and differences between different processing results, the optimal preprocessing scheme is selected; the sample data stored in the storage space or subspace with the smallest difference value after preprocessing is taken as the final result. This result will serve as the basis for subsequent feature extraction and model construction; the preprocessed data is extracted using a deep learning algorithm (such as a convolutional neural network CNN, a recurrent neural network RNN ​​or an autoencoder). These algorithms can automatically learn high-level feature representations in the data to improve the model's predictive ability, evaluate the extracted features, and select the most predictive feature subsets through feature selection techniques (such as principal component analysis (PCA) and recursive feature elimination (RFE). This helps reduce the complexity of the model and improve the model's generalization ability.

[0052] The effect of the above technical solution is: through the multi-data integration framework, sample data can be collected from multiple data sources such as electronic medical record systems and laboratory information systems to ensure the comprehensiveness and diversity of the data. This helps to capture information of different dimensions and levels, and provide a rich data foundation for subsequent analysis and modeling; the collected data is transferred to the cloud space for storage, which not only improves the accessibility and scalability of the data, but also enhances the security of the data. The cloud space provides professional security protection measures to ensure that the data is not illegally accessed or tampered with; the sample data in the storage space is redundantly backed up, and the backed-up data is stored in the subspace, further improving the reliability and availability of the data. Even if the original data is lost or damaged, it can be restored through the backup data; the sample data is stored in different storage spaces according to the data type, and the data is preprocessed. This classification processing method helps to improve the integrity of the data, and adopts appropriate preprocessing methods for different types of data, thereby improving the accuracy of the preprocessing; by comparing the results of the storage space and its corresponding subspace preprocessing, the difference can be discovered and processed in time. This comparison mechanism helps to ensure the consistency and accuracy of the data, and provides reliable data support for subsequent feature extraction and model building. The use of deep learning algorithms to extract features from preprocessed data can automatically learn high-level feature representations in the data and improve the intelligence level of feature extraction. Compared with traditional manual feature extraction methods, deep learning algorithms can more accurately capture the inherent laws and characteristics of data. The feature selection technology can screen out the most predictive feature subsets, which helps to reduce the complexity of the model and improve the generalization ability of the model. At the same time, evaluating the extracted features can ensure the validity and reliability of the selected features. Through a series of data processing and analysis steps, the high quality of the data required for subsequent model construction is ensured. This provides accurate data support for the early screening of metabolic disease diabetic foot, which helps to improve the accuracy and reliability of screening. Through the model deployed in the medical information system, patient test data can be received and analyzed in real time, and the risk warning results of metabolic disease diabetic foot can be quickly generated, which helps to achieve early intervention and treatment.

[0053] In one embodiment of the present invention, the model tuning module includes: Data merging module: Merge the acquired clinical sample data with the existing historical data to form a comprehensive data set; and pre-process the merged data set. Feature extraction module: Based on deep learning algorithms (such as convolutional neural network CNN, recurrent neural network RNN ​​or their variants), key features related to the metabolic disease diabetic foot are extracted from the preprocessed data; Subset screening module: Through feature importance evaluation, the feature subset that contributes most to the prediction of metabolic disease diabetic foot is screened; Parameter optimization module: adopts a multi-layer perception mechanism, designs the model architecture according to the feature type and data characteristics, trains the model using the divided training set data, and optimizes the model parameters through the back propagation algorithm; Index evaluation module: Use the validation set data to evaluate multiple evaluation indicators of the model, including accuracy, sensitivity, and specificity; Iterative optimization module: Based on the evaluation results, iteratively optimizes the model performance by adjusting the model structure, optimizing algorithm parameters, and using regularization techniques; Problem identification module: Through fitting and underfitting inspection: Identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set.

[0054] Model selection module: Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model.

[0055] The working principle of the above technical solution is: merge the acquired clinical sample data (including blood routine data and other related indicators) with the existing historical data to form a comprehensive data set covering multiple dimensions (such as age group, gender, metabolic disease diabetes course, whether there is metabolic disease diabetic foot, etc.); clean the merged data set to remove data records with too many missing values, outliers or illogical data to ensure the quality and reliability of the data. At the same time, standardization is performed to convert the data to the same scale for subsequent analysis and modeling; deep learning algorithms (such as convolutional neural networks CNN, recurrent neural networks RNN or their variants) are used to automatically extract key features related to metabolic disease diabetic foot from the preprocessed data. These features may include blood routine indicators such as blood sugar level, hemoglobin, white blood cell count, platelet count, and other clinical information that may be related to the risk of metabolic disease diabetic foot; analyze the contribution of each feature to the prediction of metabolic disease diabetic foot through feature importance evaluation methods (such as feature importance based on decision trees, mutual information, etc.), and screen out the feature subset that has the greatest impact on the prediction results. This helps to simplify the model structure and improve the prediction accuracy and efficiency of the model; design the multi-layer perceptron model architecture according to the feature type and data characteristics. The multi-layer perceptron is a feedforward neural network that can handle complex nonlinear relationships. The model is trained using the divided training set data. The model parameters are optimized through the back propagation algorithm so that the model can more accurately predict the risk of diabetic foot disease caused by metabolic diseases. The model is evaluated using multiple evaluation indicators using the validation set data, including accuracy, sensitivity, and specificity. These indicators can fully reflect the predictive performance of the model, such as Figure 3 As shown; according to the evaluation results, the model performance is iteratively optimized by adjusting the model structure (such as increasing or decreasing the number of network layers, changing the activation function, etc.), optimizing algorithm parameters (such as learning rate, batch size, etc.), and using regularization techniques (such as L1 / L2 regularization, Dropout, etc.). The purpose is to improve the prediction accuracy and generalization ability of the model; identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set. Overfitting is usually manifested as a small loss on the training set but a large loss on the validation set, while underfitting is manifested as a large loss on both the training set and the validation set; for the overfitting problem, regularization, increasing the amount of training data, using Dropout and other techniques can be used to reduce the complexity of the model; for the underfitting problem, you can try to increase the model complexity (such as increasing the number of network layers), adjust the learning rate and other methods to improve the model's fitting ability; after the above tuning process, select the model with the best performance as the final metabolic disease diabetic foot screening model, such as Figure 4This model will be used for subsequent real-time data analysis and risk warning.

[0056] The effect of the above technical solution is: the clinical sample data is merged with the existing historical data to form a comprehensive data set covering multiple dimensions such as different age groups, gender, metabolic disease diabetes course, and whether or not there is metabolic disease diabetic foot. This comprehensive data set helps the model learn broader and deeper features and improve the generalization ability of the model; through data cleaning and standardization, data records with too many missing values, outliers or illogical data are removed to ensure the quality and reliability of the data. This helps to reduce noise interference during model training and improve the prediction accuracy of the model; deep learning algorithms (such as CNN, RNN or their variants) are used to automatically extract key features related to metabolic disease diabetic foot from preprocessed data. Deep learning algorithms can automatically learn high-level feature representations in data, which is more efficient and accurate than traditional manual feature extraction methods; feature importance evaluation is used to screen out the feature subset that contributes most to the prediction of metabolic disease diabetic foot. This helps to reduce the complexity of the model and improve the prediction speed and accuracy of the model; a multi-layer perceptron is used as the model architecture, and it is flexibly designed according to feature types and data characteristics. The multi-layer perceptron can handle complex nonlinear relationships and is suitable for the prediction of metabolic diseases such as diabetic foot, which is affected by multiple factors. The model is trained using the divided training set data, and the model parameters are optimized by the back propagation algorithm. At the same time, the model is iteratively optimized according to the evaluation results of the validation set data, including adjusting the model structure, optimizing the algorithm parameters, and using regularization techniques. These measures help improve the prediction performance and stability of the model. The model is evaluated using multiple evaluation indicators, including accuracy, sensitivity, and specificity, using the validation set data. These evaluation indicators can fully reflect the prediction performance of the model and provide an objective basis for the selection and optimization of the model. Overfitting or underfitting problems can be identified and handled by checking the loss curves on the training set and validation set and the performance of the model on the test set. This helps to ensure the generalization ability of the model on unknown data and improve the practicality and reliability of the model. Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model. The model performs well in terms of accuracy, sensitivity, and specificity, and can accurately predict the risk of metabolic disease diabetic foot in patients, providing strong support for early intervention and treatment.

[0057] In one embodiment of the present invention, the model blind testing module includes: Data integration module: collect external data sources, integrate the collected external data sources, and build an external verification data set; Difference analysis module: blind test the optimized metabolic disease diabetic foot model through external validation data set, analyze the performance difference of the model on the external data set through the blind test results, and identify potential bias and error sources; Fine-tuning training module: cluster analysis is performed on the external validation data set to identify differences and similarities in data distribution. Based on the differences in data distribution, the model is fine-tuned or retrained through transfer learning. Sensitivity assessment module: performs sensitivity analysis to assess the sensitivity of the model to feature changes by adjusting the proportion or range of key features in the validation data set; Model interpretation module: Based on the model prediction results, a risk grading standard is designed to divide the risk of metabolic disease diabetic foot into different levels, and local and global interpretability are applied to explain the model prediction results; Visualization display module: Through visualization tools, risk grading results and explanatory information are visualized.

[0058] The working principle of the above technical solution is: collect external data sources from multiple channels (such as different hospitals, different regions, and different population structures) to ensure the diversity and breadth of the validation data set; integrate the collected external data sources to form an external validation data set. This data set will be used to evaluate the performance of the model on unknown or new data; use the external validation data set to blindly test the tuned metabolic disease diabetic foot model, that is, make predictions without knowing the true label of the data; analyze the performance differences of the model on the external data set through the blind test results, including changes in evaluation indicators such as accuracy, sensitivity, and specificity. At the same time, identify potential sources of bias and error, such as data distribution differences and feature missing; perform cluster analysis on the external validation data set to identify differences and similarities in data distribution. This helps to understand the characteristics between different data subsets and provide a basis for subsequent model adjustments; fine-tune or retrain the model based on the differences in data distribution through technologies such as transfer learning. Transfer learning can use the knowledge of existing models to accelerate the learning process of new tasks and improve the performance of the model on new data; by adjusting the proportion or range of key features in the validation data set, evaluate the sensitivity of the model to feature changes. This helps to understand the stability and reliability of the model under different conditions; according to the results of sensitivity analysis, analyze which changes in features the model is more sensitive to and how these changes affect the model's prediction results; based on the model's prediction results, design risk grading standards to divide the risk of metabolic disease diabetic foot into different levels. This helps doctors and patients understand the patient's risk status more intuitively; apply local interpretability (such as LIME, SHAP) and global interpretability (such as feature importance analysis, decision tree approximation) methods to explain the model's prediction results. Local interpretability methods can explain the basis of a single prediction result, while global interpretability methods can reveal the overall working mechanism and feature importance of the model. These explanatory information helps doctors and patients understand the source of risk and the basis for prediction; through visualization tools, the risk grading results and explanatory information are displayed in the form of charts, heat maps, etc. These visualizations help to intuitively convey the model's prediction results and explanatory information, and improve the understanding and acceptance of doctors and patients.

[0059] The effect of the above technical solution is: by collecting external data sources from different hospitals, regions and population structures, and constructing an external validation data set, the performance of the model in different environments can be comprehensively evaluated. This helps to discover the potential bias and error of the model on unknown data, thereby improving the generalization ability of the model; blind testing the tuned model can objectively evaluate the performance differences of the model on the external data set, and then accurately identify the potential sources of bias and error. This provides a strong basis for subsequent model adjustments; cluster analysis is performed on the external validation data set to identify the differences and similarities in data distribution. Based on these analysis results, fine-tuning or retraining the model through transfer learning can effectively reduce the decline in model performance caused by differences in data distribution, and further improve the accuracy and stability of the model; by adjusting the proportion or range of key features in the validation data set, the sensitivity of the model to feature changes is evaluated. This helps to understand the performance of the model under different conditions and enhance the robustness of the model to feature fluctuations; designing risk grading standards based on the model prediction results, and interpreting the prediction results through local and global interpretative methods, which helps doctors and patients to more clearly understand the risk level and prediction basis of metabolic disease diabetic foot. This can help doctors develop more scientific and reasonable treatment plans, improve treatment effects and patient satisfaction; through visualization tools, risk grading results and explanatory information are displayed in the form of charts, heat maps, etc., making complex model prediction results intuitive and easy to understand. This helps improve the communication efficiency between doctors and patients and enhance patients' understanding and acceptance of disease risks and treatment plans.

[0060] In one embodiment of the present invention, the early warning acquisition module includes: Performance prediction module: deploy the constructed model to the medical information system and optimize the performance of the model deployment environment. By implementing the model monitoring mechanism, the operating status and prediction performance of the model can be monitored in real time. Data stream processing module: Integrates real-time data stream processing algorithms to collect, clean and preprocess patient test data in real time; Threshold trigger module: Analyzes real-time data through the model, generates risk warning information for diabetic foot caused by metabolic diseases, sets risk thresholds, and triggers automatic alarm mechanisms based on risk thresholds; Early warning notification module: Early warning information is notified to doctors and patients in real time through multiple methods.

[0061] The working principle of the above technical solution is: deploy the constructed and verified metabolic disease diabetic foot screening model into the medical information system. This step involves integrating the model file (such as a trained neural network model, decision tree model, etc.) into the back-end service of the medical information system so that it can receive input data and return prediction results; optimizing the performance of the model deployment environment, including the reasonable configuration of hardware resources (such as CPU, GPU, memory) to ensure that the model can run efficiently; adapting the software environment (such as operating system, Python version, dependent library) to avoid compatibility issues; and optimizing the network environment (such as bandwidth, latency) to ensure the stability and real-time performance of data transmission; implementing a model monitoring mechanism to monitor the operating status and prediction performance of the model in real time. Through monitoring, potential problems in the operation of the model can be discovered in a timely manner, such as decreased prediction accuracy, prolonged response time, etc., and corresponding measures can be taken to solve them; integrating real-time data stream processing algorithms (such as Kafka, SparkStreaming) to achieve real-time collection of patient test data. The collected data may contain noise, outliers, etc., so it needs to be cleaned and preprocessed to ensure data quality; the cleaned data is preprocessed, including data formatting, feature extraction, normalization and other steps to meet the requirements of model input; the preprocessed real-time data is input into the metabolic disease diabetic foot screening model for analysis, and the model generates metabolic disease diabetic foot risk prediction results based on the input data; the risk threshold is set according to clinical needs and actual conditions. When the prediction result exceeds the set risk threshold, the patient is considered to have a high risk of metabolic disease diabetic foot; the automatic alarm mechanism is triggered based on the risk threshold. When the prediction result exceeds the threshold, the system automatically generates metabolic disease diabetic foot risk warning information; the warning information is notified to doctors and patients in real time through a variety of methods. This includes sending text messages, emails, and push notifications through APP. These notification methods can ensure that doctors and patients can receive warning information in a timely manner and take appropriate measures to intervene.

[0062] The effect of the above technical solution is: by integrating real-time data stream processing algorithms, it is possible to realize real-time collection, cleaning and preprocessing of patient test data, and then quickly generate metabolic disease diabetic foot risk warning information. This greatly shortens the time from data collection to risk assessment and improves the efficiency of diagnosis and treatment; deploys a fully verified model and combines performance optimization measures to ensure the stable operation and accurate prediction of the model in the medical information system. This helps doctors to more accurately assess the risk of metabolic disease diabetic foot in patients, so as to formulate more effective treatment plans; the automatic alarm mechanism triggered by risk thresholds can issue warnings in time when the patient's risk increases, so that doctors and patients can respond quickly and take corresponding measures. This helps to reduce the incidence of complications of metabolic disease diabetic foot and improve the quality of life of patients; real-time notification of doctors and patients through SMS, email and APP push and other methods ensures the timely communication and reception of warning information. This helps to enhance patients' attention and management of their own health status and improve treatment compliance; the performance of the model deployment environment is optimized in hardware, software and network, ensuring the stable operation and efficient processing of the medical information system. This helps improve the operating efficiency of the entire medical system and reduce delays in diagnosis and treatment caused by system failures or delays; the implementation of a model monitoring mechanism can monitor the operating status and predictive performance of the model in real time, and promptly identify and resolve potential problems. This helps ensure the continued effectiveness and reliability of the model in the medical information system; through real-time data stream processing algorithms, centralized management and sharing of patient test data are achieved. This helps to achieve data interoperability and resource sharing between different medical institutions and improve the efficiency of medical resource utilization; multi-channel early warning information notification mechanisms promote communication and collaboration between doctors, patients and other medical team members. This helps to form a closer medical team to jointly cope with the challenges of chronic diseases such as metabolic diseases and diabetic foot.

[0063] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for early screening of metabolic disease diabetic foot test data, characterized in that: The method comprises: S1. Acquire clinical sample data, and transmit the acquired clinical sample data to the cloud space for preprocessing, perform feature extraction on the preprocessed clinical sample data through a deep learning algorithm, and extract features related to the metabolic disease diabetic foot; S2. Construct a metabolic disease diabetic foot screening model based on routine blood data through deep learning algorithms, and combine the acquired clinical sample data with historical data to construct a data set; evaluate the constructed metabolic disease diabetic foot screening model through multiple indicators, and optimize the metabolic disease diabetic foot screening model based on the evaluation results; S3. Introduce an external validation dataset to conduct a blind test on the optimized metabolic disease diabetic foot screening model to verify the regional adaptability of the model and perform risk grading on the model output results based on the algorithm interpretation; S4. Deploy the constructed model to the medical information system, receive patient test data in real time, and analyze the test data in real time to obtain metabolic disease diabetic foot risk warning results.

2. According to claim 1, a method for early screening of metabolic disease diabetic foot test data, characterized in that: Said S1 comprises: S11. Collect sample data from multiple data sources through a multi-data integration framework, and transmit the collected sample data to the cloud space; S12, the cloud space stores the received sample data in different storage spaces according to the data type, and divides a subspace in each storage space; S13, performing redundant backup of the sample data in the storage space, and storing the backed-up data in the subspace; preprocessing the data stored in the storage space and its corresponding subspace respectively by the processing unit; S14, comparing the preprocessed results of the storage space and its corresponding subspace to determine whether there is a difference. If there is a difference, the difference is processed again; S15, taking the result after the second processing as a reference value, comparing it with the result of the first preprocessing, and taking the sample data stored in the storage space or subspace with the smallest difference value after preprocessing as the final result; S16. Use deep learning algorithms to extract features from preprocessed data, evaluate the extracted features, and select the feature subset with the most predictive value through feature selection technology.

3. According to claim 1, a method for early screening of metabolic disease diabetic foot test data, characterized in that: The S2 comprises: S21. Merge the acquired clinical sample data with the existing historical data to form a comprehensive data set; and preprocess the merged data set; S22. Extract key features related to metabolic disease diabetic foot from preprocessed data based on deep learning algorithm; S23. Through feature importance evaluation, screen the feature subset that contributes most to the prediction of metabolic disease diabetic foot; S24, adopting a multi-layer perception mechanism, designing a model architecture according to feature types and data characteristics, using the divided training set data to train the model, and optimizing the model parameters through the back propagation algorithm; S25. Evaluate multiple evaluation indicators of the model using the validation set data, wherein the multiple indicators include accuracy, sensitivity, and specificity; S26. Based on the evaluation results, iteratively optimize the model performance by adjusting the model structure, optimizing algorithm parameters, and using regularization techniques; S27. Identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set; S28. Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model.

4. According to claim 1, a method for early screening of metabolic disease diabetic foot test data, characterized in that: The S3 includes: S31, collecting external data sources, and integrating the collected external data sources to construct an external verification data set; S32. Perform a blind test on the tuned metabolic disease diabetic foot model using an external validation dataset, analyze the performance differences of the model on the external dataset based on the blind test results, and identify potential biases and error sources; S33. Perform cluster analysis on the external validation data set to identify differences and similarities in data distribution. Based on the differences in data distribution, fine-tune or retrain the model through transfer learning. S34. Perform sensitivity analysis to evaluate the sensitivity of the model to feature changes by adjusting the proportion or range of key features in the validation dataset; S35. Based on the model prediction results, design a risk grading standard to divide the risk of diabetic foot caused by metabolic diseases into different levels, and apply local and global interpretability to explain the model prediction results; S36, and use visualization tools to visualize the risk grading results and explanatory information.

5. According to claim 1, a method for early screening of metabolic disease diabetic foot test data, characterized in that: The S4 comprises: S41. Deploy the constructed model to the medical information system and optimize the performance of the model deployment environment. By implementing the model monitoring mechanism, the operating status and prediction performance of the model can be monitored in real time. S42, integrate real-time data stream processing algorithms to collect, clean and pre-process patient test data in real time; S43, analyzing the real-time data through the model, generating metabolic disease diabetic foot risk warning information, and setting the risk threshold, and triggering the automatic alarm mechanism based on the risk threshold; S44. Notify doctors and patients of warning information in real time through multiple means.

6. A metabolic disease diabetic foot test data early screening system, characterized in that: The system comprises: Data acquisition module: obtain clinical sample data and transmit the obtained clinical sample data to the cloud space for preprocessing, perform feature extraction on the preprocessed clinical sample data through deep learning algorithms, and extract features related to the metabolic disease diabetic foot; Model tuning module: Use deep learning algorithms to build a metabolic disease diabetic foot screening model based on routine blood data, and combine the acquired clinical sample data with historical data to build a data set; use multiple indicators to evaluate the constructed metabolic disease diabetic foot screening model, and tune the metabolic disease diabetic foot screening model based on the evaluation results; Model blind test module: introduce external validation data sets to conduct blind tests on the optimized metabolic disease diabetic foot screening model, verify the regional adaptability of the model, and perform risk grading on the model output results based on algorithm interpretation; Early warning acquisition module: deploy the constructed model to the medical information system, receive patient test data in real time, and analyze the test data in real time to obtain metabolic disease diabetic foot risk early warning results.

7. According to claim 6, a metabolic disease diabetic foot test data early screening system, characterized in that: The data acquisition module comprises: Data transmission module: collects sample data from multiple data sources through a multi-data integration framework, and transmits the collected sample data to the cloud space; Space division module: The cloud space stores the received sample data in different storage spaces according to the data type, and divides a subspace within each storage space; Redundant backup module: perform redundant backup of sample data in the storage space and store the backed-up data in the subspace; pre-process the data stored in the storage space and its corresponding subspace respectively through the processing unit; Difference judgment module: compares the preprocessed results of the storage space and its corresponding subspace to determine whether there is a difference. If there is a difference, the difference part is processed again; Result determination module: takes the result after reprocessing as the benchmark value, compares it with the result of the first preprocessing, and takes the sample data stored in the storage space or subspace with the smallest difference value after preprocessing as the final result; Feature evaluation module: Use deep learning algorithms to extract features from preprocessed data, evaluate the extracted features, and screen out the feature subset with the most predictive value through feature selection technology.

8. According to claim 6, a metabolic disease diabetic foot test data early screening system is characterized in that: The model tuning module includes: Data merging module: merge the acquired clinical sample data with the existing historical data to form a comprehensive data set; and pre-process the merged data set; Feature extraction module: Based on deep learning algorithm, key features related to metabolic disease diabetic foot are extracted from preprocessed data; Subset screening module: Through feature importance evaluation, the feature subset that contributes most to the prediction of metabolic disease diabetic foot is screened; Parameter optimization module: adopts a multi-layer perception mechanism, designs the model architecture according to the feature type and data characteristics, trains the model using the divided training set data, and optimizes the model parameters through the back propagation algorithm; Index evaluation module: Use the validation set data to evaluate multiple evaluation indicators of the model, including accuracy, sensitivity and specificity; Iterative optimization module: Based on the evaluation results, iteratively optimizes the model performance by adjusting the model structure, optimizing algorithm parameters, and using regularization techniques; Problem identification module: Identify and deal with overfitting or underfitting problems by checking the loss curves on the training set and validation set, as well as the performance of the model on the test set; Model selection module: Based on the above tuning process, the model with the best performance is selected as the final metabolic disease diabetic foot screening model.

9. According to claim 6, a metabolic disease diabetic foot test data early screening system, characterized in that: The model blind testing module comprises: Data integration module: collects external data sources, integrates the collected external data sources, and constructs an external verification data set; Difference analysis module: blind test the optimized metabolic disease diabetic foot model through external validation data set, analyze the performance difference of the model on the external data set through the blind test results, and identify potential bias and error sources; Fine-tuning training module: cluster analysis is performed on the external validation data set to identify differences and similarities in data distribution. Based on the differences in data distribution, the model is fine-tuned or retrained through transfer learning. Sensitivity assessment module: performs sensitivity analysis to assess the sensitivity of the model to feature changes by adjusting the proportion or range of key features in the validation data set; Model interpretation module: Based on the model prediction results, a risk grading standard is designed to divide the risk of metabolic disease diabetic foot into different levels, and local and global interpretability are applied to explain the model prediction results; Visualization display module: The risk grading results and explanatory information are visualized through visualization tools.

10. The metabolic disease diabetic foot test data early screening system according to claim 6, characterized in that: The early warning acquisition module includes: Performance prediction module: deploy the constructed model to the medical information system and optimize the performance of the model deployment environment; monitor the model's operating status and prediction performance in real time by implementing a model monitoring mechanism; Data stream processing module: Integrates real-time data stream processing algorithms to collect, clean and preprocess patient test data in real time; Threshold trigger module: Analyzes real-time data through the model, generates risk warning information for metabolic disease diabetic foot, sets risk thresholds, and triggers automatic alarm mechanisms based on risk thresholds; Early warning notification module: Notify doctors and patients of early warning information in real time through multiple methods.

Images