Diagnostic system
Through a two-layer diagnostic model structure, combined with LSTM and multiple machine learning models and voting classifiers, hierarchical diagnosis of cardiovascular diseases is achieved, solving the problem of the inability to accurately determine the disease category in existing technologies and improving the accuracy and efficiency of diagnosis.
Patent Information
- Application Number
- CN202510851154.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-24
- Publication Date
- 2025-10-03
AI Technical Summary
Existing technologies can only distinguish whether a person has cardiovascular disease, but cannot accurately determine the specific type of disease, resulting in insufficient accuracy in disease classification.
A two-layer diagnostic model structure is adopted. The first layer uses the long short-term memory network LSTM model to determine whether a patient has cardiovascular disease. The second layer uses at least two machine learning models and a voting classifier to determine the specific disease category. The probability is output by the machine learning model and soft voting is performed using the voting classifier to determine the final result.
It improves the accuracy of cardiovascular disease prediction, simplifies the diagnosis process, enables timely treatment of sudden illnesses, and reduces medical costs.
Smart Images
Figure CN120748686A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a diagnostic system. Background Art
[0002] Cardiovascular diseases (CVDs) are among the most dangerous diseases. Traditional diagnostic methods rely on experienced medical experts and complex equipment, limited by insufficient medical resources and professionals. The development of artificial intelligence (AI) and fifth-generation mobile communication technology (5G) provides new approaches to addressing these problems. AI has demonstrated powerful capabilities in medical image analysis and data processing, extracting key features from massive amounts of medical data to provide personalized diagnosis and treatment recommendations, greatly improving diagnostic efficiency. 5G offers high speed, low latency, and large connectivity, supporting the real-time transmission of large-scale, high-resolution medical images and real-time monitoring data, making it easier for patients in remote areas to access medical services and reducing medical costs.
[0003] Currently, AI has been widely used in the field of cardiovascular disease diagnosis. For example, deep learning models such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM) are used to process and analyze complex medical data such as electrocardiograms (ECGs) and medical images; traditional machine learning algorithms such as Support Vector Machines (SVM), random forests, and decision trees are used for data classification, regression, and clustering analysis for cardiovascular disease risk prediction and classification; Natural Language Processing (NLP) is used to analyze electronic health records (EHRs), physician notes, and patient feedback to extract valuable information from unstructured text to assist in the diagnosis and monitoring of cardiovascular diseases; and multimodal data fusion technology is used to integrate data from different sources, such as ECGs, images, EHRs, and wearable device data, to provide comprehensive cardiovascular health assessment and monitoring.
[0004] In the existing technology, there have been many studies on cardiovascular disease prediction. However, these studies can only distinguish whether a patient is sick or not, but cannot determine the specific disease category of the patient, resulting in insufficient accuracy in disease classification. Summary of the Invention
[0005] The embodiments of the present application provide a diagnostic system that solves the problem of insufficient accuracy in current cardiovascular disease prediction.
[0006] In a first aspect, to achieve the above-mentioned objectives, embodiments of the present application provide a diagnostic system, comprising a diagnostic device, a diagnostic model deployed on the diagnostic device, and the diagnostic model comprising a first-layer diagnostic model and a second-layer diagnostic model; wherein:
[0007] The first-layer diagnostic model is used to output a first diagnostic result based on the cardiovascular information of the first user, where the first diagnostic result is used to indicate whether the first user suffers from cardiovascular disease;
[0008] The second-layer diagnostic model is used to output a second diagnostic result based on the cardiovascular information when the first-layer diagnostic model diagnoses that the first user suffers from a cardiovascular disease. The second diagnostic result is used to indicate the target cardiovascular disease category to which the disease suffered by the first user belongs.
[0009] in:
[0010] The first-layer diagnostic model includes a long short-term memory network LSTM model;
[0011] And / or, the second-layer diagnostic model includes at least two machine learning models and a voting classifier, wherein the machine learning model is used to identify a first probability that the disease suffered by the first user belongs to various pre-set cardiovascular disease categories, and the voting classifier is used to vote and select the target cardiovascular disease category based on the first probability.
[0012] Among them, the machine learning model includes at least two of a decision tree model, a random forest model, a neural network model, a gradient boosting tree model, and a support vector machine SVM model.
[0013] in:
[0014] Each of the machine learning models is used to receive the cardiovascular information, and perform machine learning based on the cardiovascular information to predict and output the first probability that the disease suffered by the first user belongs to various preset cardiovascular disease categories;
[0015] The voting classifier is used to perform soft voting on the first probabilities output by various machine learning models, and determine the target cardiovascular disease category based on the soft voting results.
[0016] The voting classifier is used to perform soft voting on the first probabilities output by the various machine learning models, and determine the target cardiovascular disease category based on the soft voting results, including:
[0017] The voting classifier performs weighted averaging on the first probabilities of the same cardiovascular disease category output by various machine learning models to obtain a second probability;
[0018] The voting classifier uses the cardiovascular disease category corresponding to the largest second probability as the target cardiovascular disease category.
[0019] Wherein, the diagnostic system further includes an interactive device, wherein: the interactive device is used to obtain the cardiovascular information from a data acquisition device and send a diagnostic task to the diagnostic device, wherein the diagnostic task carries the cardiovascular information.
[0020] The interactive device is further configured to send the second diagnostic result output by the diagnostic model to a display device.
[0021] The interactive device is further configured to send the cardiovascular information and the second diagnostic result to a doctor's diagnostic workstation, and the doctor's diagnostic workstation is configured to verify the second diagnostic result.
[0022] The diagnostic system further includes a medical database, wherein the medical database is used to store the second diagnostic result verified by the doctor's diagnostic workstation and the cardiovascular information corresponding to the second diagnostic result.
[0023] The diagnostic system further includes a model training device, which is used to:
[0024] Obtain cardiovascular datasets from medical and public databases;
[0025] Training the diagnostic model using the cardiovascular dataset;
[0026] The trained diagnostic model was evaluated using a cross-validation method.
[0027] The beneficial effects of the above technical solution of this application are as follows:
[0028] The diagnostic system of the embodiment of the present application includes a diagnostic device, wherein a diagnostic model is deployed on the diagnostic device. By setting the diagnostic model to a two-layer structure including a first-layer diagnostic model and a second-layer diagnostic model, a hierarchical diagnosis of cardiovascular disease can be achieved, such as: first diagnosing whether the patient has cardiovascular disease based on the first-layer diagnostic model, and then further diagnosing the type of cardiovascular disease based on the second-layer diagnostic model based on the presence of cardiovascular disease. This staged processing simplifies the tasks of each stage, improves the overall efficiency and accuracy of the diagnostic model, solves the current problem of only being able to confirm whether the patient has cardiovascular disease but not the type of cardiovascular disease the patient suffers from, improves the accuracy of cardiovascular disease prediction, and facilitates the timely treatment of sudden illnesses. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 This is one of the structural diagrams of the diagnostic system according to an embodiment of the present application;
[0030] Figure 2 This is the second structural diagram of the diagnostic system according to an embodiment of the present application;
[0031] Figure 3 This is the third structural diagram of the diagnostic system according to an embodiment of the present application;
[0032] Figure 4 A flow chart of the workflow of the diagnostic system according to an embodiment of the present application;
[0033] Figure 5 This is the fourth structural diagram of the diagnostic system according to an embodiment of the present application. DETAILED DESCRIPTION
[0034] In order to make the technical problems, technical solutions and advantages to be solved by this application clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.
[0035] It should be understood that references throughout this specification to "one embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the present application. Therefore, the appearances of "in one embodiment" or "in an embodiment" throughout this specification do not necessarily refer to the same embodiment. Furthermore, these particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0036] In the various embodiments of the present application, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0037] Additionally, the terms "system" and "network" are often used interchangeably herein.
[0038] In the embodiments provided herein, it should be understood that "B corresponding to A" means that B is associated with A and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B based solely on A; B can also be determined based on A and / or other information.
[0039] The embodiment of the present application provides a diagnostic system, such as Figure 1 As shown, the diagnostic system includes a diagnostic device, a diagnostic model is deployed on the diagnostic device, and the diagnostic model includes a first-layer diagnostic model and a second-layer diagnostic model; wherein:
[0040] The first-layer diagnostic model is used to output a first diagnostic result based on the cardiovascular information of the first user, and the first diagnostic result is used to indicate whether the first user suffers from cardiovascular disease; wherein the cardiovascular information includes, for example, photoplethysmography (PPG), electrocardiogram (ECG), blood pressure, heart rate, etc., and the examination information may also include demographic characteristics (such as age, gender, etc.).
[0041] The second-layer diagnostic model is used to output a second diagnostic result based on the cardiovascular information when the first-layer diagnostic model diagnoses that the first user suffers from a cardiovascular disease. The second diagnostic result is used to indicate the target cardiovascular disease category to which the disease suffered by the first user belongs.
[0042] In the diagnostic device of the diagnostic system of the embodiment of the present application, by setting the diagnostic model to a two-layer structure including a first-layer diagnostic model and a second-layer diagnostic model, a hierarchical diagnosis of cardiovascular disease can be achieved, such as first diagnosing whether the patient has cardiovascular disease, and then further diagnosing the type of cardiovascular disease based on the presence of cardiovascular disease. This staged processing simplifies the tasks of each stage, improves the overall efficiency and accuracy of the diagnostic model, and solves the current problem of only confirming whether a patient has cardiovascular disease but not the type of cardiovascular disease the patient suffers from. It improves the accuracy of cardiovascular disease prediction and facilitates the timely treatment of sudden illness.
[0043] As an optional implementation, the first-layer diagnostic model includes a long short-term memory network LSTM model; and / or, the second-layer diagnostic model includes at least two machine learning models and a voting classifier, wherein the machine learning model is used to identify a first probability that the disease suffered by the first user belongs to various pre-set cardiovascular disease categories, and the voting classifier is used to vote to select the target cardiovascular disease category based on the first probability.
[0044] In this optional implementation, the diagnostic model implements a phased approach to the diagnostic process (including initial screening and refined disease classification). Specifically, an LSTM model performs initial screening to determine the presence of cardiovascular disease, followed by a detailed classification of specific disease categories using machine learning models (such as random forests and decision trees) and voting classifiers. This phased approach simplifies the tasks at each stage and improves the overall efficiency and accuracy of the model.
[0045] It's important to note that in cardiovascular disease classification, patients' physiological data is typically in the form of time series. The LSTM model is specifically designed to process and predict time series data, effectively capturing long-term dependencies. The LSTM model also possesses transfer learning capabilities, allowing it to leverage updated databases to improve classification performance, making it more adaptable than other models. The core of the LSTM model lies in its gating mechanism, which maintains long-term memory and filters out unnecessary information.
[0046] Specifically, the LSTM model structure mainly consists of an input layer, an LSTM layer, a fully connected layer, and an output layer. The LSTM layer is the core of the entire network, which captures the long-term and short-term dependencies in time series data through a gating mechanism. The output of the LSTM layer is further processed and mapped by the fully connected layer. Assume that the output of the last LSTM layer is h T , then: z=h T *W+b d ; Among them, h T is the matrix multiplication of the hidden state of the LSTM model and the weight matrix of the fully connected layer, which means mapping the hidden state to the output space, b d is the bias, which is added to the result of the linear transformation to help adjust the network output, ultimately generating the transformed result z. The output layer is responsible for converting the output of the fully connected layer into the final classification result. The final classification layer uses the softmax function to perform a binary classification (whether the patient is ill or not) and generate the predicted value y, where y = softmax(z).
[0047] The convolutional neural network approach, which has limited ability to process time series data, has limitations in capturing the temporal dependencies of data. In addition, the current approach of using support vector machines (SVMs) requires complex feature engineering when processing high-dimensional data, relies on manual feature extraction, has high computational complexity, and takes a long time to train. In particular, it requires selecting and optimizing appropriate kernel functions in nonlinear problems.
[0048] As a specific implementation method, the machine learning model includes at least two of a decision tree model, a random forest model, a neural network model, a gradient boosting tree model, and a support vector machine SVM model, but is not limited thereto. The machine learning model can also be other models.
[0049] It should be noted that the decision tree model intuitively displays the decision-making process through a tree structure, which is more interpretable. The random forest model reduces the risk of overfitting and enhances the robustness and stability of the model by integrating multiple decision trees. Compared with other models, the random forest and decision tree models can provide more reliable classification results in the case of feature diversity and data imbalance. Based on this, it is preferred that the machine learning model include at least the decision tree model and the random forest model.
[0050] As an optional implementation, when the machine learning model is used to identify a first probability that the disease suffered by the first user belongs to various preset cardiovascular disease categories, the specific implementation process is as follows:
[0051] Each of the machine learning models is used to receive the cardiovascular information and perform machine learning based on the cardiovascular information to predict and output the first probability that the disease suffered by the first user belongs to various pre-set cardiovascular disease categories; illustratively, the various pre-set cardiovascular disease categories include coronary heart disease, hypertension, arrhythmia, valvular heart disease, etc., but are not limited to this.
[0052] As an optional implementation, when the voting classifier is used to vote and select the target cardiovascular disease category based on the first probability, the specific implementation process is as follows:
[0053] The voting classifier is used to perform soft voting on the first probabilities output by the various machine learning models, and determine the target cardiovascular disease category based on the soft voting results. Soft voting is an ensemble learning strategy based on probability averaging. Based on this, when the voting classifier performs soft voting on the first probabilities output by the various machine learning models, it is specifically used to:
[0054] The voting classifier performs weighted averaging on the first probabilities of the same cardiovascular disease category output by various machine learning models to obtain a second probability;
[0055] The voting classifier uses the cardiovascular disease category corresponding to the largest second probability as the target cardiovascular disease category.
[0056] That is, in the second-tier diagnostic model, a voting classifier can be selected as an ensemble method to combine the prediction results of different machine learning models (such as decision trees and random forest models). The voting classifier improves overall prediction accuracy by integrating the prediction results of multiple models. Specifically: the soft voting method in the voting classifier is used as an ensemble learning method to make the final decision by combining the prediction results of multiple different classifiers. Each classifier will predict the sample and give a probability distribution of a category. Soft voting will weightedly average the prediction results of all classifiers and then select the category with the highest probability as the final prediction result.
[0057] That is, first, the prediction results of each machine learning model for the sample (the aforementioned cardiovascular information) are used to obtain a probability distribution for a disease category. Second, the prediction results for the same disease category from all machine learning models are weighted averaged to obtain the final probability distribution. Finally, the disease category with the highest probability is selected as the final disease prediction result. Specifically, the formula for soft voting is as follows:
[0058]
[0059] Where P(y=c) represents the probability of class c, N represents the number of classifiers, and Pi(y=c) represents the probability prediction result of classifier i for class c. In soft voting, the output class is the average prediction based on the probability of the given class.
[0060] It's also worth noting that combinations of machine learning models (such as random forests and decision trees) offer excellent performance and robustness in processing multidimensional data and classification tasks. They can handle complex decision boundaries and are highly tolerant to outliers and noise. The specific implementation described above, through the combination of ensemble learning and single models (such as decision trees and random forests), offers greater flexibility and stability in classification tasks.
[0061] Furthermore, if Figure 2 As shown, the diagnostic system further includes an interactive device, wherein the interactive device is configured to obtain the cardiovascular information from the data acquisition device and send a diagnostic task to the diagnostic device, the diagnostic task carrying the cardiovascular information. Exemplarily, the interactive device sends the diagnostic task to the diagnostic device only after ensuring that the cardiovascular information is complete.
[0062] The data acquisition device is only used to collect the user's examination information (including the aforementioned cardiovascular information, and may also include the user's demographic characteristics (such as age, gender, etc.)). In addition, the data acquisition device may be part of a diagnostic system or a separate acquisition device. After collecting the cardiovascular information, the data acquisition device will pre-process the cardiovascular information and send the pre-processed cardiovascular information to the interactive device. The pre-processing of the cardiovascular information includes: removing outliers, filling missing values, filtering, feature extraction and normalization. Specifically:
[0063] The preprocessing for removing outliers is as follows: using statistical methods (such as Z-score) to identify outliers; applying rule-based methods (threshold) to remove or correct obvious noise points.
[0064] The preprocessing for filling missing values is to use interpolation methods (such as linear interpolation or spline interpolation) to fill missing values.
[0065] The preprocessing of filtering is to remove noise from PPG and ECG signals using bandpass filtering and wavelet transform.
[0066] Since PPG and ECG are continuous waveform signals, feature extraction is required to obtain characteristic values. Therefore, the preprocessing of feature extraction is as follows: extracting heart rate, peak and trough time, power spectrum density, etc. from the periodic signal; using frequency domain analysis (such as Fast Fourier Transform (FFT)) to extract frequency features; and calculating statistical features (mean, standard deviation, maximum, minimum, etc.).
[0067] The normalization preprocessing is to scale the statistical features using the Min-Max normalization method and normalize all feature values to the range of [0, 1].
[0068] On this basis, the data acquisition device specifically sends the normalized features corresponding to the cardiovascular information to the interactive device to further improve the accuracy of subsequent disease classification.
[0069] Furthermore, as an optional implementation, Figure 2 As shown, the interactive device is further configured to transmit the second diagnostic result output by the diagnostic model to a display device. The display terminal is configured to allow relevant personnel to promptly view disease prediction results. For example, patients in remote areas can promptly learn about their disease status on a display device in their area, eliminating the need for cross-regional diagnosis and treatment, thereby reducing medical costs.
[0070] Furthermore, as another optional implementation, Figure 2As shown, the interactive device is further used to send the cardiovascular information and the second diagnosis result to a doctor's diagnosis workstation, which is used to verify the second diagnosis result. In this way, the doctor can confirm the diagnosis result of the diagnosis model.
[0071] As an optional implementation, such as Figure 2 As shown, the diagnostic system further includes: a medical database, wherein the medical database is used to store the second diagnostic result verified by the doctor's diagnostic workstation and the cardiovascular information corresponding to the second diagnostic result.
[0072] In the above optional implementation method, by storing the data of the relevant diagnostic process in which the doctor confirms the correct diagnosis into the medical database, the sample size of the medical database will gradually increase, and eventually form a larger data set with a wider application range. Model training based on these data sets can improve the accuracy of model diagnosis.
[0073] As an optional implementation, such as Figure 2 As shown, the diagnostic system further includes a model training device, which is used to:
[0074] Acquire a cardiovascular dataset from a medical database and a public database, wherein data in the cardiovascular dataset has labels of cardiovascular disease categories;
[0075] It should be noted that currently, datasets are generally constructed using a single public database, which results in poor disease classification performance when the trained model is transferred to a small dataset. Alternatively, training is performed using a self-constructed dataset, which results in a small sample size. Compared to the above two existing approaches, the above steps in this implementation utilize a combination of collected data (medical databases) and public databases to construct a dataset, improving the robustness and universality of the model. Specifically, the cardiovascular dataset can be divided into training, validation, and test sets in a ratio of 70:10:20.
[0076] The diagnostic model is trained using the cardiovascular dataset; illustratively, this step designs the model architecture of the two-layer diagnostic model based on the labeled training sample set, performs hyperparameter tuning on the validation set, and performs prediction on the test set to obtain the model's prediction label. In terms of model architecture design, the number of layers (depth) and the number of nodes (width) in each layer of the network are adjusted to improve the expressiveness of the model; different activation functions (such as ReLU, Sigmoid, Tanh) are tried to improve the nonlinear expressiveness of the model. In terms of hyperparameter tuning, the hyperparameters that need to be tuned are first determined, such as learning rate, batch size, regularization parameter, dropout rate, etc. Through the grid search method, the performance of different hyperparameter combinations is systematically evaluated within the given parameter range.
[0077] The trained diagnostic model is evaluated using a cross-validation method. This step uses a cross-validation method to evaluate the stability and generalization ability of the trained two-layer diagnostic model, and uses appropriate evaluation metrics (such as accuracy, precision, recall, F1 score, etc.) to comprehensively evaluate the model performance. The performance of the model at different thresholds is evaluated by plotting a receiver operating characteristic (ROC) curve, and the area under the curve (AUC) is calculated to evaluate the overall performance of the classifier.
[0078] Here, it should be noted that the diagnostic equipment, interactive equipment and model training equipment in the above-mentioned diagnostic system of the embodiment of the present application can be integrated into the same device, or they can be two or three independent devices that can communicate with each other, and the above-mentioned interactive equipment and data acquisition equipment, display equipment, doctor workstation, etc. can communicate through mobile networks (such as 5G), and the model training equipment and public databases and medical databases, as well as the doctor workstation and medical database can communicate through mobile networks. In this way, these devices can be deployed in different locations, thereby improving the applicability of the diagnostic system.
[0079] Below, Figure 3 Taking FIG. 1 as an example, the structure and function of a specific example of the diagnostic system in an embodiment of the present application are described.
[0080] like Figure 3 As shown, the system includes:
[0081] Data acquisition terminal (corresponding to the aforementioned data acquisition device): used to collect patient examination information, including cardiovascular information. Here, the data acquisition terminal is only used to collect information. The disease diagnosis system and its diagnostic method involved are information processing methods implemented by computers and other devices;
[0082] Medical Information Mart for Intensive Care (MIMIC3) database (corresponding to the aforementioned public database): an international open-source database;
[0083] Medical database: used to store patient examination information collected by the data acquisition terminal and final cardiovascular disease diagnosis information;
[0084] AI interactive server (corresponding to the aforementioned interactive device): used for the transfer and interaction of various types of information, for example, receiving data from the data acquisition end and transmitting it to the AI computing server, receiving the preliminary diagnosis results (corresponding to the aforementioned second diagnosis results) transmitted by the AI computing server and sending them to the display end (corresponding to the aforementioned display device) and the doctor's diagnosis workstation;
[0085] AI computing server (corresponding to the aforementioned diagnostic equipment): calculates and analyzes the patient's cardiovascular information to obtain preliminary diagnosis information of cardiovascular disease;
[0086] AI training server (corresponding to the aforementioned model training device): used to learn cardiovascular disease diagnostic information and form a diagnostic model;
[0087] Display terminal (corresponding to the aforementioned display device): used to display the preliminary diagnosis information of cardiovascular disease transmitted by the AI interactive server (corresponding to the aforementioned second diagnosis result);
[0088] Doctor's Diagnosis Service Station: used by cardiovascular doctors to verify preliminary diagnosis results.
[0089] Among them, through the above-mentioned system, the information stored in the medical database is authorized for doctors and patients to access and query.
[0090] Use Figure 3 The system shown can construct a dataset by combining medical and public databases. Data collected at the data acquisition end undergoes preprocessing and arrives at the doctor's diagnostic workstation, along with preliminary diagnostic results generated by the AI interaction server, AI computing server, and AI training server. The doctor at the diagnostic workstation then verifies the patient's cardiovascular information and diagnostic results, packaged and sent to the medical database. As the sample size at the data acquisition end continues to grow, the sample size of the medical database will also gradually increase, ultimately forming a larger dataset with a wider range of applications.
[0091] Next, combine Figure 3 and Figure 4 , the process of diagnosing a disease using the diagnostic system of an embodiment of the present application is described, wherein the diagnostic process specifically includes the following steps:
[0092] Step 410, by Figure 3The data collection end in the system obtains the patient's cardiovascular information;
[0093] Step 420, pre-processing the collected cardiovascular information of the patient; the pre-processing of this step can refer to the above-mentioned pre-processing process; this step can be specifically performed by Figure 3 Executed on the data collection end;
[0094] Step 430: Transmit the pre-processed cardiovascular information to Figure 3 The AI interactive server in the Figure 3 The AI computing server in the forwarding computing tasks;
[0095] Step 440: The AI computing server Figure 3 The diagnostic model provided by the AI training server in the algorithm performs computational analysis on cardiovascular diseases to obtain preliminary diagnostic results. The diagnostic process of this step specifically includes, for example: at the first layer, an LSTM model is established to distinguish whether it is a cardiovascular disease, forming a first-layer diagnostic model based on deep learning; at the second layer, a random Sen Ling and decision tree model is established to diagnose specific cardiovascular disease categories, forming a second-layer diagnostic model based on machine learning. In addition, the data source of the AI training server is a cardiovascular data set constructed by combining the MIMIC3 database and the medical database. In addition, the data in the constructed cardiovascular data set all have labels for whether it is a cardiovascular disease and the specific cardiovascular disease category.
[0096] Step 450: The AI computing server sends the preliminary diagnosis results to the display terminal and the doctor's diagnosis workstation via the AI interactive server;
[0097] Step 460: The doctor at the doctor diagnosis workstation verifies the preliminary diagnosis result, and after confirming that it is correct, packages the patient's cardiovascular information and diagnosis result and sends them to the medical database.
[0098] As a widely used diagnostic system, e.g. Figure 5 As shown, the AI interaction server, AI computing server and AI training server (corresponding to the aforementioned interaction device, diagnostic device and model training device) can be integrated into one server (device) or deployed in a nearby location as the background device of the diagnostic system, and the data acquisition terminal (corresponding to the aforementioned data acquisition device) and the doctor's diagnostic workstation can be deployed in the corresponding hospital, or the data acquisition terminal is a device that can be connected to the hospital system, and the display terminal can be deployed at the patient's location, and the background device communicates with the data acquisition terminal, the doctor's diagnostic workstation and the display terminal through a mobile network (such as 5G). In this way, the diagnostic system of the embodiment of the present application can be used to predict diseases for patients in different places, which improves the universality of the diagnostic system and is conducive to the widespread promotion of the diagnostic system.
[0099] Use Figures 1 to 3 and Figure 5 The cardiovascular disease diagnosis method implemented by the diagnostic system shown has the following benefits: First, AI, through machine learning and big data analysis, can extract key features from massive amounts of medical data, significantly improving the accuracy of cardiovascular disease diagnosis and providing personalized diagnosis and treatment recommendations. Second, the AI system can analyze and process large amounts of data in seconds, greatly improving diagnostic efficiency and enabling doctors to make more informed decisions in a shorter time. Third, the high speed and low latency of 5G technology enable real-time transmission and processing of large-scale, high-resolution medical imaging and real-time monitoring data, supporting telemedicine. Patients in remote areas can interact and consult with experts at central hospitals in real time via the 5G network, receiving timely medical services. Fourth, the diagnostic system combining AI and 5G can reduce unnecessary examinations and duplicate diagnoses, optimize the allocation of medical resources, and reduce overall medical costs. Furthermore, through early diagnosis and intervention, emergency room visits and hospitalizations due to cardiovascular disease can be reduced, further reducing medical expenses.
[0100] In addition, the diagnostic model in the embodiment of the present application, on the one hand, uses a combination of collected data and public databases to construct a data set for training. In this way, on the one hand, the sample distribution of the training set can be made more extensive, thereby improving the model performance and generalization ability. Secondly, as the sample size of the data acquisition end continues to increase, the sample size of the medical database will gradually increase, eventually forming a larger data set with a wider range of applications, so that the diagnostic effect of the trained model is better (such as improved diagnostic accuracy). On the other hand, first, in the first layer, an LSTM model is established to distinguish whether it is a cardiovascular disease, forming a first-layer diagnostic model based on deep learning, and then in the second layer, a random forest and decision tree model are established to diagnose specific cardiovascular disease categories, forming a two-layer diagnostic model based on machine learning, thereby improving the accuracy of the preliminary diagnostic results of cardiovascular diseases.
[0101] The above exemplary embodiments are described with reference to the accompanying drawings. Many different forms and embodiments are possible without departing from the spirit and teachings of this application. Therefore, this application should not be construed as limited to the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided so that this application will be complete and impartial and will convey the scope of this application to those skilled in the art. In the drawings, component sizes and relative sizes may be exaggerated for clarity. The terminology used herein is for purposes of describing specific exemplary embodiments only and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms "a," "an," and "the" are intended to encompass such plural forms. It will be further understood that the terms "comprising" and / or "including," when used in this specification, indicate the presence of the stated features, integers, steps, operations, components, and / or elements, but do not preclude the presence or addition of one or more other features, integers, steps, operations, components, elements, and / or groups thereof. Unless otherwise indicated, when stated, a range of values includes the upper and lower limits of that range and any subranges therebetween.
[0102] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A diagnostic system, characterized in that: The diagnostic system includes a diagnostic device, a diagnostic model is deployed on the diagnostic device, and the diagnostic model includes a first-layer diagnostic model and a second-layer diagnostic model; wherein: The first-layer diagnostic model is used to output a first diagnostic result based on the cardiovascular information of the first user, and the first diagnostic result is used to indicate whether the first user suffers from cardiovascular disease; The second-layer diagnostic model is used to output a second diagnostic result based on the cardiovascular information when the first-layer diagnostic model diagnoses that the first user suffers from a cardiovascular disease. The second diagnostic result is used to indicate the target cardiovascular disease category to which the disease suffered by the first user belongs.
2. The diagnostic system according to claim 1, wherein: The first-layer diagnostic model includes a long short-term memory network LSTM model; And / or, the second-layer diagnostic model includes at least two machine learning models and a voting classifier, wherein the machine learning model is used to identify a first probability that the disease suffered by the first user belongs to various pre-set cardiovascular disease categories, and the voting classifier is used to vote and select the target cardiovascular disease category based on the first probability.
3. The diagnostic system according to claim 2, characterized in that The machine learning model includes at least two of a decision tree model, a random forest model, a neural network model, a gradient boosting tree model, and a support vector machine (SVM) model.
4. The diagnostic system according to claim 2 or 3, characterized in that: Each of the machine learning models is used to receive the cardiovascular information, and perform machine learning based on the cardiovascular information to predict and output the first probability that the disease suffered by the first user belongs to various preset cardiovascular disease categories; The voting classifier is used to perform soft voting on the first probabilities output by various machine learning models, and determine the target cardiovascular disease category based on the soft voting results.
5. The diagnostic system according to claim 4, characterized in that The voting classifier is used to perform soft voting on the first probabilities output by the various machine learning models, and determine the target cardiovascular disease category based on the soft voting results, including: The voting classifier performs weighted averaging on the first probabilities of the same cardiovascular disease category output by various machine learning models to obtain a second probability; The voting classifier uses the cardiovascular disease category corresponding to the largest second probability as the target cardiovascular disease category.
6. The diagnostic system according to claim 1, wherein: The diagnostic system further includes an interactive device, wherein: the interactive device is used to obtain the cardiovascular information from a data acquisition device and send a diagnostic task to the diagnostic device, wherein the diagnostic task carries the cardiovascular information.
7. The diagnostic system according to claim 6, characterized in that The interactive device is further configured to send the second diagnosis result output by the diagnosis model to a display device.
8. The diagnostic system according to claim 6, characterized in that The interactive device is further configured to send the cardiovascular information and the second diagnosis result to a doctor's diagnosis workstation, and the doctor's diagnosis workstation is configured to verify the second diagnosis result.
9. The diagnostic system according to claim 8, characterized in that The diagnostic system further includes a medical database, wherein the medical database is used to store the second diagnostic result verified by the doctor's diagnostic workstation and the cardiovascular information corresponding to the second diagnostic result.
10. The diagnostic system according to claim 1, wherein: The diagnostic system further includes a model training device, which is used to: Obtain cardiovascular datasets from medical and public databases; Training the diagnostic model using the cardiovascular dataset; The trained diagnostic model was evaluated using a cross-validation method.