A method for predicting coronary heart disease by fusing heart-lung motion parameter time series information

Abstract

The application discloses a coronary heart disease prediction method fusing cardiopulmonary exercise parameter time sequence information, and comprises the following steps: collecting multidimensional health data of a patient; preprocessing the multidimensional health data; extracting features from the preprocessed data; performing time sequence analysis on the extracted features by using a deep learning model; inputting the time sequence analysis result and static features into a classification model to perform coronary heart disease prediction, outputting a prediction result and generating a related health report. By acquiring demographic data, basic medical information and cardiopulmonary exercise test data of a target object, specifically including age, gender, body mass index, socio-economic information, past history, complications, clinical examination and inspection, exercise capacity, cardiovascular function and ventilation perfusion and the like, the coronary heart disease is predicted by fusing static parameters and time sequence information, so that dynamic changes in the health data of the patient can be better captured, and the prediction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of biotechnology, in particular to a method for predicting coronary heart disease by fusing cardiopulmonary exercise parameter time sequence information. BACKGROUND

[0002] Coronary heart disease (CHD) is a heart disease caused by coronary atherosclerosis, myocardial ischemia, hypoxia or necrosis, and is one of the major diseases threatening human health. The traditional diagnosis method of coronary heart disease relies on coronary computed tomography or coronary angiography, which has certain radiation or trauma to the human body and high medical cost.

[0003] Existing prediction or diagnosis models are mostly focused on a wide range of cardiovascular diseases, and lack precise early prediction of coronary heart disease. At the same time, previous prediction methods are mostly focused on the selection and application of traditional clinical indicators and parameters, and lack comprehensiveness and precision of disease characteristics of patients with coronary heart disease (CN118098575A, CN115831374A, CN115040156A).

[0004] Cardiopulmonary exercise test (CPET) as an objective, quantitative and non-invasive examination method is gradually applied to the early diagnosis and prognosis risk stratification of coronary heart disease. The CPET examination results contain rich time sequence information, which is often used to assist diagnosis by manually interpreting the changes. However, the learning cost of CPET result interpretation is high, and the interpretation level of primary hospitals is uneven, especially for the understanding of image changes, the homogeneity between medical institutions is poor. Therefore, the applicant intends to fuse cardiopulmonary exercise static parameters and time sequence information, use the deep learning and time sequence information analysis technology of artificial intelligence, and establish a cardiopulmonary exercise prediction model for patients with coronary heart disease. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the present application provides a coronary heart disease prediction model fusing static parameters and time sequence information. By analyzing the continuous health data of patients, the present application combines traditional clinical indicators and cardiopulmonary exercise test parameters, especially time sequence information parameters, uses the fusion of cardiopulmonary exercise static parameters and time sequence information to predict coronary heart disease, and based on the characteristics of real-time data monitoring and parameter changes during exercise, more dimensional and accurate prediction of the occurrence of coronary heart disease can be made, which can better capture the dynamic changes of patients during exercise, improve the prediction accuracy, help to take timely treatment measures, and ensure the physical and mental health of patients. The accuracy and timeliness of early prediction of coronary heart disease are improved.

[0006] The present application is realized at least by one of the following technical solutions.

[0007] A method for predicting coronary artery disease by integrating temporal information of cardiopulmonary exercise parameters includes the following steps:

[0008] S1. Collect multi-dimensional health data from patients;

[0009] S2. Preprocess the multi-dimensional health data;

[0010] S3. Extract features from the preprocessed data;

[0011] S4. Perform time-series analysis on the extracted features using a deep learning model;

[0012] S5. Input the time series analysis results and static features into the classification model to predict coronary heart disease, output the prediction results and generate relevant health reports.

[0013] Furthermore, in step S1, the multidimensional health data includes the target subject's demographic data, basic medical information, static parameters and time-series information of cardiopulmonary exercise testing, and the static parameters and time-series information of cardiopulmonary exercise testing include ventilation function, gas exchange status, exercise tolerance and circulatory function.

[0014] Furthermore, the preprocessing includes data cleaning, default value processing, and normalization.

[0015] Furthermore, the extracted features include time-domain features, frequency-domain features, and other features that may be related to coronary heart disease.

[0016] Furthermore, the deep learning model is a Long Short-Term Memory (LSTM) network combined with patient temporal information for disease prediction.

[0017] Furthermore, the Long Short-Term Memory (LSTM) network includes an input gate, a forget gate, and an output gate. The calculation formula for the forget gate is as follows:

[0018] f t =σ(W f .[h t-1 ,x t ]+b f )

[0019] Among them, f t W represents the weight for forgetting cell states. f This is the weight matrix of the forget gate, b f It is a bias term, x t It is the input for the current time step, h t-1 It is the hidden state of the previous time step, and σ is the sigmoid function, which maps the input value to a probability value between 0 and 1;

[0020] The formula for calculating the input gate is as follows:

[0021] i t = σ(W i .[h t-1 ,x t ]+b i )

[0022] where i t denotes the weights for updating the cell state, W i is the weight matrix for the input gate, b i is the bias term, x t is the input at the current time step, h t-1 is the hidden state at the previous time step, and σ is the sigmoid function that maps input values to probability values between 0 and 1.

[0023] The LSTM computes a candidate cell state, c , which represents the influence of the new input at the current time step on the cell state. The formula for computing the candidate cell state is as follows:

[0024]

[0025] where W c is the weight matrix for the candidate cell state, b c is the bias term, x t is the input at the current time step, h t-1 is the hidden state at the previous time step, and tanh is the hyperbolic tangent function that maps input values to values between -1 and 1.

[0026] The cell state of the LSTM is updated and passed to the next time step. At each time step t, the update formula for the cell state is as follows:

[0027]

[0028] where c is the candidate cell state at the current time step, which represents the influence of the new input at the current time step on the cell state; and c t is the candidate cell state at the next time step.

[0029] The output gate determines the information o t that is outputted by the LSTM through a sigmoid function. At each time step t, the computation formula for the output gate is as follows:

[0030] o t = σ(W o .[h t-1 ,x t ]+b o )

[0031] where W ois the weight matrix of the output gate, b o is the bias term, x t is the input at the current time step, h t-1 is the hidden state at the previous time step, σ is the sigmoid function that maps input values to probability values between 0 and 1.

[0032] The LSTM will process the cell state C t through a tanh function to obtain the hidden state h t at the current time step:

[0033] h t = o t *tanh(C t )

[0034] where tanh is the hyperbolic tangent function that maps input values to values between -1 and 1.

[0035] Further, the long short-term memory network LSTM adopts step-by-step prediction in the time series prediction task, and the specific steps are as follows:

[0036] 1) Initial input:

[0037] The historical time step data is input into the long short-term memory network LSTM to obtain the initial hidden state and cell state, ensuring that the data is arranged in chronological order.

[0038] The initial input data is a fixed-length time window containing historical data of multiple time steps.

[0039] 2) Step-by-step prediction:

[0040] For each time step, the data at the current time step and the hidden state at the previous time step are input into the long short-term memory network LSTM to obtain the output result at the current time step and the new hidden state and cell state. The long short-term memory network LSTM outputs the time series information of the patient.

[0041] 3) Iterative process:

[0042] The output result at the current time step is used as the input for the next time step, and the step 2) process is repeated until the specified number of times is reached.

[0043] 4) Prediction strategy:

[0044] True value input: After each prediction, compare the prediction of coronary heart disease with the actual case data to evaluate the prediction error; select the true value as the input for the next time step to reduce error accumulation.

[0045] Prediction value input: using the long short-term memory network LSTM to predict whether it is coronary heart disease;

[0046] Using the cell state updated at this time step and the corresponding time sequence features at the next time step as the input at the next time step, this method is suitable for long time step prediction, but may cause error accumulation step by step;

[0047] 5) Error evaluation and adjustment:

[0048] After each prediction, the prediction result is compared with the true value, the prediction error is evaluated, and the long short-term memory network LSTM input or the long short-term memory network LSTM model is adjusted or retrained according to the error size.

[0049] Further, the classification model includes a plurality of fully connected layers, each neuron of the fully connected layer is connected with all neurons of the previous layer, and each connection has a weight for adjusting the intensity of information transmission.

[0050] Further, the fully connected layer includes two parts: parameter initialization and forward propagation;

[0051] Parameter initialization: when initializing the weight matrix, a random initialization method is adopted, and when initializing the bias vector, a zero vector is used for initialization;

[0052] Forward propagation: the fully connected layer establishes a full connection relationship between all neurons of the previous layer and all neurons of the current layer, and transmits and converts information through the weight matrix.

[0053] Further, binary cross entropy is used as a loss function for back propagation to update the parameters of the long short-term memory network LSTM.

[0054] The system for implementing the coronary heart disease prediction method fusing heart-lung exercise parameter time sequence information comprises:

[0055] Data acquisition module: acquiring multi-dimensional health data of patients, including demographic data, basic medical information, static parameters and time sequence information of cardiopulmonary exercise test of the target object, including ventilation function, gas exchange, exercise tolerance and circulation function, etc.

[0056] Data preprocessing module: preprocessing the collected multi-dimensional health data, including data cleaning, default value processing, normalization and other operations.

[0057] Feature extraction module: extracting features from the preprocessed time sequence data, including time domain features, frequency domain features and other possible coronary heart disease related features.

[0058] The time sequence analysis module: using LSTM to perform time sequence analysis on the extracted features to mine the time sequence information in the data.

[0059] The disease prediction module: inputting the time sequence analysis result and the static features into the full connection layer of the classification module to make a prediction decision of coronary heart disease.

[0060] The result output module: outputting the prediction result of coronary heart disease and generating a related health report.

[0061] Compared with the prior art, the present application has the following beneficial effects:

[0062] The present application can better capture the dynamic changes in the patient's health data and improve the prediction accuracy by fusing static parameters and time sequence information to predict coronary heart disease. At the same time, the present application can realize early warning, which is helpful for timely treatment and ensures the physical and mental health of the patient. BRIEF DESCRIPTION OF DRAWINGS

[0063] Fig. 1 is an embodiment of a coronary heart disease prediction process diagram fusing time sequence information;

[0064] Fig. 2 is a structural schematic diagram of an embodiment of a time sequence analysis module;

[0065] Fig. 3 is a structural schematic diagram of an embodiment of a classification model. DETAILED DESCRIPTION

[0066] The specific embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0067] As shown in Figs. 1-3 , a coronary heart disease prediction method fusing cardiopulmonary exercise parameter time sequence information according to the present embodiment comprises the following steps:

[0068] S1, using a data acquisition module to acquire multi-dimensional health data of a patient;

[0069] As an embodiment, the multi-dimensional health data comprises demographic data, basic medical information, static parameters and time sequence information of a cardiopulmonary exercise test of a target object, wherein the static parameters and time sequence information of the cardiopulmonary exercise test comprise ventilation function, gas exchange condition, exercise tolerance and circulation function.

[0070] 1) The demographic data and basic medical information of the target object are collected by an electronic medical record system, and the personal information of the patient is processed in a hidden manner.

[0071] 2) Cardiopulmonary Exercise Test (CPET) parameter acquisition:

[0072] ① Experimental equipment: Use cardiopulmonary exercise test equipment, including power car, respiratory gas analyzer, electrocardiograph, electronic sphygmomanometer, pulse oximeter, etc.

[0073] ② Experimental process: The subjects perform gradually increasing intensity of treadmill exercise within a certain period of time, during which real-time physiological index data are collected. The data include but are not limited to exercise capacity: respiratory exchange rate, oxygen uptake, exercise load time, maximum load; cardiovascular function: exercise electrocardiogram changes, heart rate, blood pressure, oxygen pulse; ventilation and perfusion: respiratory reserve, maximum ventilation, carbon dioxide output, etc.

[0074] S2, the data preprocessing module pre-processes the health data; the preprocessing includes data cleaning, default value processing, normalization and other operations.

[0075] 1) Time series structure of data:

[0076] ① Definition of time step: During the cardiopulmonary exercise test, data collection is performed at fixed time intervals. For CPET data, data is collected every 10 seconds, so each time step corresponds to one second of data.

[0077] ② Data preprocessing: Preprocess the collected data, including data cleaning, noise filtering, missing value filling, etc., to ensure data quality.

[0078] 2) Construction of historical time step data:

[0079] ① Time step length: According to the actual situation, select an appropriate time step length, collect data for the first 15 minutes, collect data every 10 seconds, then there are 90 time steps.

[0080] ② Data arrangement: Arrange the data of each time step in chronological order to form time series data.

[0081] S3, the feature extraction module extracts features from the preprocessed data, extracts features that can reflect the cardiovascular status of the patient, such as: oxygen pulse, oxygen uptake / power; Feature extraction includes time domain features and other possible features related to coronary heart disease.

[0082] 1) Data preprocessing:

[0083] Clean the collected cardiopulmonary exercise test (CPET) data, remove noise and outliers, and ensure data quality. Remove subjects with more than 20% missing variables. Use linear interpolation and mean filling methods to fill in missing values.

[0084] 2) Time domain feature extraction: Calculate the data of each time step and extract the time domain features. Including but not limited to:

[0085] Oxygen pulse: The ratio of oxygen uptake (VO2) and heart rate (HR) is calculated.

[0086] Oxygen uptake / power ratio (VO2 / Watt): Calculated by the ratio of oxygen uptake (VO2) and exercise power.

[0087] Heart rate (HR): The number of heartbeats per minute is directly read from electrocardiogram data.

[0088] S4, the time series analysis module uses a deep learning model to perform time series analysis on the extracted features;

[0089] As an embodiment, the deep learning model is a Long Short-Term Memory (LSTM), a special recurrent neural network (RNN) structure commonly used to process time series and extract time series information. Therefore, LSTM is a neural network model in the field of deep learning.

[0090] In deep learning, LSTM is widely used in tasks that require processing long-term dependencies, such as speech recognition, natural language processing, time series prediction, etc. Compared with traditional RNN structure, LSTM introduces a gating mechanism to better capture and utilize long-term dependencies in time series data, thereby improving the performance and generalization ability of the model.

[0091] LSTM sets two key variables:

[0092] ① Hidden state h, mainly responsible for memorizing short-term information, especially the information of the current time step;

[0093] ② Cell state C, mainly responsible for memorizing long-term information.

[0094] LSTM has three key gating units, namely input gate, forget gate and output gate. These gating units control the flow of information through learning, helping the LSTM network better handle long-term dependencies, as shown in Fig. 2 .

[0095] The calculation formulas of the three gates of LSTM are as follows:

[0096] The forget gate determines which information in the cell state should be forgotten or discarded. It contains a sigmoid activation function that outputs a value between 0 and 1 based on the input data and the previous time step's hidden state. Through a multiplication operation, the forget gate controls the degree of preservation of historical information in the cell state, determining which information will persist in the cell state. The calculation formula of the forget gate is as follows:

[0097] f t =σ(W f .[h t-1 ,x t ]+b f )

[0098] Where f t represents the weight of forgetting the cell state, W f is the weight matrix of the forget gate, b f is the bias term, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, and σ is the sigmoid function that maps the input value to a probability value between 0 and 1.

[0099] The input gate controls the degree of influence of new input information on the current cell state. It contains a sigmoid activation function that generates a value between 0 and 1, indicating which values in each cell state should be updated. Through the use of a multiplication operation, the input gate determines which parts of the cell state will be updated with new input information and passes the result to the cell state. The calculation formula of the input gate is as follows:

[0100] i t =σ(W i .[h t-1 ,x t ]+b i )

[0101] Where i t represents the weight of updating the cell state, W i is the weight matrix of the input gate, b i is the bias term, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, and σ is the sigmoid function that maps the input value to a probability value between 0 and 1.

[0102] Output gate, the output gate controls how the hidden state of the current time and the cell state of the next time affect the final output. It contains a sigmoid activation function to determine which part of the output state will be activated. The output gate also contains a tanh activation function to generate a value between -1 and 1, representing the candidate value of the cell state of the current time step. Through the multiplication operation, the output gate sends the screened and updated information in the cell state to the hidden state of the next time step, and generates the final output based on this hidden state h t . The calculation formula of the output gate is as follows:

[0103] o t =σ(W o .[h t-1 ,x t ]+b o )

[0104] h t =o t *tanh(C t )

[0105] Where W o is the weight matrix of the output gate, b o is the bias term, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, σ is the sigmoid function, which maps the input value to a probability value between 0 and 1; tanh is the hyperbolic tangent function, which maps the input value to a value between -1 and 1. LSTM usually predicts step by step in time series prediction tasks. The specific steps are as follows:

[0106] 1) Initial input:

[0107] Input the data of the historical time steps in steps S2 and S3 into the LSTM model to obtain the initial hidden state and cell state, and ensure that the data is arranged in chronological order.

[0108] The initial input data is usually a fixed-length time window containing historical data of multiple time steps.

[0109] 2) Step-by-step prediction:

[0110] For each time step, input the data of the current time step and the hidden state of the previous time step into the LSTM. Through the calculation of the LSTM, the output result of the current time step and the new hidden state and cell state are obtained. The LSTM output can be a prediction value (such as the value of the next time step) or multiple prediction values (such as the values of multiple future time steps).

[0111] 3) Iterative process:

[0112] Repeat the above process by taking the output of the current time step as input for the next time step until the predicted time range is reached or the stopping condition is met.

[0113] 4) Prediction strategy:

[0114] Training phase: Input the preprocessed cardiopulmonary exercise test (CPET) data into the LSTM model, and input the true value into the long short-term memory network LSTM as the input of the next time step, and gradually make predictions: After each prediction, compare whether the prediction is coronary heart disease with the actual case data, and evaluate the prediction error. Select the true value as the input of the next time step to reduce error accumulation.

[0115] Prediction phase: Input the predicted value into the long short-term memory network LSTM: Use the predicted value as the input of the next time step to continue the prediction of the subsequent time steps. Record the result of predicting coronary heart disease at each step until the prediction of all time steps is completed.

[0116] 5) Error evaluation and adjustment:

[0117] After each prediction, compare the prediction result with the true value to evaluate the prediction error. According to the error size, the model input or the model can be retrained. Common error evaluation indicators include mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), etc.

[0118] Long-term prediction considerations

[0119] Error accumulation:

[0120] As the prediction time step increases, the prediction error may gradually accumulate, leading to a decrease in prediction accuracy. Therefore, when performing long-term prediction, special attention should be paid to the problem of error accumulation.

[0121] Model retraining:

[0122] In order to improve the accuracy of long-term prediction, the model can be retrained periodically using the latest actual data to update the model parameters, so that it can better adapt to data changes.

[0123] Dynamic adjustment:

[0124] The input sequence and model parameters can be dynamically adjusted according to the prediction results to improve the flexibility and accuracy of the prediction. For example, use an autoregressive model combined with LSTM to enhance the prediction ability of the model.

[0125] S5, input the time series analysis results and static cardiopulmonary characteristics into the classification model for coronary heart disease prediction; the result output module outputs the prediction result and generates a related health report.

[0126] The classification model predicts coronary heart disease of the patient according to the results of time series analysis and static parameters. The classification model uses a multi-layer fully connected layer (Fully Connected Layer), also known as a dense connection layer (Dense Layer), which is a basic layer type in deep learning neural networks. Each neuron of the fully connected layer is connected to all neurons of the previous layer, and each connection has a weight to adjust the strength of information transmission.

[0127] The implementation of the fully connected layer mainly includes two parts: parameter initialization and forward propagation.

[0128] Parameter initialization: the fully connected layer is composed of a weight matrix and a bias vector. When initializing the weight matrix, random initialization is usually used to fully utilize the adaptive ability of the network. When initializing the bias vector, a zero vector is usually used for initialization.

[0129] Forward propagation: the fully connected layer establishes a full connection relationship between all neurons of the previous layer and all neurons of the current layer, and transmits and converts information through the weight matrix. The specific implementation process is as follows:

[0130] 1) Input data: input data time series analysis results and preprocessed static characteristics, usually a batch of feature vectors or feature tensors, each sample is represented as a vector. The output tensor of the previous layer is pulled into one dimension, that is, an input vector is constructed. The feature vector containing time information and the static feature vector are combined into a one-dimensional vector.

[0131] 2) Weight initialization: each neuron is connected to each neuron of the previous layer, and each connection has a weight. In the fully connected layer, these weights are usually initialized, for example, using the random initialization method to perform matrix multiplication between the input vector and the weight matrix.

[0132] 3) Add the result of matrix multiplication to the bias vector.

[0133] 4) Input the result added with the bias vector into the activation function for processing to get the output result of the current layer.

[0134] 5) Repeat: if there are more fully connected layers, repeat the above steps. The output of each layer will be used as the input of the next layer until the last layer or the output layer is reached.

[0135] 6) Loss calculation: after the output layer, the loss between the predicted value and the true label is usually calculated to measure the accuracy of the model prediction.

[0136] 7) Backpropagation and optimization: Use backpropagation algorithm to calculate the gradient of loss function for each parameter, then use optimization algorithm (such as stochastic gradient descent) to update the weights to minimize the loss function.

[0137] Sigmoid activation function: Where x represents the final output result of the model

[0138] Binary cross-entropy loss function:

[0139]

[0140] Where n represents the number of predicted samples; y i represents the sample label, y i = 1 when the patient has coronary heart disease, otherwise 0; p(y i = 1) represents the probability of the model predicting that the patient has coronary heart disease.

[0141] Missing value processing (Missing Value Imputation): Handle missing values in related data, fill missing values with appropriate methods such as mean, median, mode filling, etc. Missing value processing helps improve prediction performance.

[0142] The preferred embodiments of the present application disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details and limit the present application to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the present application. The present application selects and describes these embodiments in order to better explain the principles and practical applications of the present application, so that those skilled in the art can well understand and utilize the present application.

Claims

1. A method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters, characterized in that, Includes the following steps: S1. Collect multi-dimensional health data of patients. The multi-dimensional health data includes the target subject's demographic data, basic medical information, static parameters and time-series information of cardiopulmonary exercise test, and the static parameters and time-series information of cardiopulmonary exercise test include ventilation function, gas exchange status, exercise tolerance and circulatory function. S2. Preprocess the multi-dimensional health data; S3. Extract features from the preprocessed data; S4. Perform temporal analysis on the extracted features using a deep learning model, wherein the deep learning model is a Long Short-Term Memory (LSTM) network. S5. Input the time series analysis results and static features into the classification model to predict coronary heart disease, output the prediction results and generate relevant health reports.

2. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 1, wherein in step S2, the following is characterized in that, The preprocessing includes data cleaning, default value processing, and normalization.

3. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 1, wherein in step S3, the following is characterized in that... The extracted features include time-domain features, frequency-domain features, and other features that may be related to coronary heart disease.

4. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 1, characterized in that, The Long Short-Term Memory (LSTM) network includes an input gate, a forget gate, and an output gate. The calculation formula for the forget gate is as follows: in, The weight representing the forgetting of cell states. It is the weight matrix of the forget gate. It is a bias term. This is the input for the current time step. It is the hidden state of the previous time step. It is the sigmoid function, which maps input values ​​to probability values ​​between 0 and 1; The formula for calculating the input gate is as follows: in This represents the weight used to update the cell state. It is the weight matrix of the input gate. It is a bias term. This is the input for the current time step. It is the hidden state of the previous time step. It is the sigmoid function, which maps input values ​​to probability values ​​between 0 and 1; LSTM computes candidate cell states, representing the impact of the new input at the current time step on the cell state. The calculation formula is as follows in, It is the weight matrix of the candidate cell states. It is a bias term. This is the input for the current time step. It is the hidden state of the previous time step. It is the hyperbolic tangent function, which maps input values ​​to values ​​between -1 and 1; The cell state of the LSTM is updated and propagated to the next time step at each time step. t The formula for updating cell state is as follows: in, It represents the candidate cell states at the current time step, indicating the impact of new inputs at the current time step on the cell states; It represents the candidate cell state for the next time step; The output gate determines the information to be output using the sigmoid function. At each time step t, the formula for calculating the output gate is as follows: in, It is the weight matrix of the output gate. It is a bias term. This is the input for the current time step. It is the hidden state of the previous time step. It is the sigmoid function, which maps input values ​​to probability values ​​between 0 and 1; LSTM will measure cell state The hidden state at the current time step is obtained by processing it using the tanh function. : in, It is the hyperbolic tangent function, which maps input values ​​to values ​​between -1 and 1.

5. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 4, characterized in that, The Long Short-Term Memory (LSTM) network employs a stepwise prediction approach in time series forecasting tasks, with the following specific steps: 1) Initial input: The historical time step data is input into the Long Short-Term Memory (LSTM) network to obtain the initial hidden state and cell state, ensuring that the data is arranged in chronological order; The initial input data is typically a fixed-length time window containing historical data from multiple time steps; 2) Step-by-step prediction: For each time step, the data of the current time step and the hidden state of the previous time step are input into the Long Short-Term Memory (LSTM) network. The LSTM network then obtains the output of the current time step, the new hidden state, and the cell state. The LSTM network outputs the patient's temporal information. 3) Iterative process: Use the output of the current time step as the input of the next time step, and repeat step 2) until the specified number of times is reached; 4) Prediction strategy: True value input: After each prediction, the predicted coronary heart disease is compared with actual case data to assess the prediction error; the true value is selected as the input for the next time step to reduce error accumulation; Predicted value input: Use the Long Short-Term Memory (LSTM) network to predict whether it is coronary heart disease; Using the cell state updated at this time step and the temporal features corresponding to the next time step as input for the next time step, this approach is suitable for situations requiring long-term prediction, but may lead to the gradual accumulation of errors. 5) Error assessment and adjustment: After each prediction, the prediction result is compared with the true value to evaluate the prediction error. Based on the magnitude of the error, the input of the Long Short-Term Memory (LSTM) network is adjusted or the LSTM network model is retrained.

6. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 5, characterized in that, The classification model includes multiple fully connected layers. Each neuron in a fully connected layer is connected to all neurons in the previous layer, and each connection has a weight to adjust the intensity of information transmission.

7. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 6, characterized in that, A fully connected layer consists of two parts: parameter initialization and forward propagation; Parameter initialization: When initializing the weight matrix, random initialization is used, and when initializing the bias vector, the zero vector is used for initialization; Forward propagation: The fully connected layer establishes a fully connected relationship between all neurons in the previous layer and all neurons in the current layer, and transmits and transforms information through the weight matrix.

8. The method for predicting coronary heart disease by integrating time-series information of cardiopulmonary exercise parameters according to claim 5, characterized in that, Binary cross-entropy is used as the loss function for backpropagation to update the parameters of the Long Short-Term Memory (LSTM) network.

Images