Hemodynamic parameter intelligent analysis method and system

Abstract

The invention relates to the technical field of hemodynamic analysis, and discloses an intelligent analysis method and system for hemodynamic parameters, and the method comprises the steps: obtaining the hemodynamic parameter data of a patient in real time through a multi-modal data collection device, and carrying out the preprocessing, noise removal and standardization processing of the data; and analyzing the preprocessed data based on a deep learning model, predicting the variation trend of the hemodynamic parameters, and generating a prediction result. And dynamically adjusting the parameters and the structure of the deep learning model according to the real-time data and the prediction result by using an adaptive optimization algorithm, and generating a personalized treatment scheme. According to the method, unprecedented precision and flexibility are achieved in hemodynamic parameter prediction and personalized treatment. In the prior art, data fusion, dynamic prediction, real-time optimization and personalized treatment cannot be achieved at the same time, but the technical problems are solved, the prediction precision is improved, and the personalized treatment efficiency and effect of the patient are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hemodynamic analysis, and particularly to an intelligent analysis method and system for hemodynamic parameters. Background Art

[0002] With the progress of computing technology and the continuous innovation of medical devices, the monitoring and analysis of hemodynamic parameters have become one of the important research directions in modern medicine. Especially in the early diagnosis and real-time monitoring of cardiovascular diseases, the dynamic analysis of parameters such as blood flow velocity and blood pressure provides key data support for the prediction and treatment of diseases. Early hemodynamic analysis methods relied on traditional physical models and laboratory tests, usually lacking real-time performance and accuracy, and unable to dynamically adapt to changes in the patient's physiological state. In recent years, with the development of deep learning technology, more and more research has begun to use artificial intelligence (AI)-based models for real-time analysis and prediction of hemodynamic data.

[0003] Although the prior art has made some progress in the prediction of hemodynamic parameters, there are still several deficiencies. First, most of the existing blood flow analysis methods rely on a single blood flow data source (such as blood flow velocity or blood pressure), lacking comprehensive analysis of multi-modal data, which makes the prediction results often limited to a single dimension and unable to fully reflect the patient's health status. Second, traditional hemodynamic models mostly adopt static methods and cannot fully consider the dynamic changes of the patient's physiological state and external environment, resulting in prediction deviations when dealing with complex and highly variable clinical data. Moreover, the existing prediction methods often fail to fully utilize the advantages of deep learning technology in complex pattern recognition, fail to effectively handle the noise problem in blood flow data, and also fail to well adapt to the individual differences of different patients. Finally, most of the existing optimization algorithms are limited to local search and are difficult to handle complex high-dimensional parameter spaces, resulting in the optimization process falling into local optimal solutions and affecting the overall performance of the model.

[0004] Different from the prior art, the present invention combines multi-modal data (such as blood flow velocity, blood pressure, electrocardiogram, etc.) and deep learning models to achieve accurate prediction of hemodynamic parameters, and introduces an adaptive optimization algorithm for dynamic adjustment. By real-time feedback of the error between the prediction result and the actual measurement value, the system can automatically optimize the parameters of the prediction model when the patient's state changes significantly, so as to achieve higher-precision prediction and generate personalized treatment plans. This technology not only solves the deficiencies of single data source and static models in the prior art, but also overcomes the limitations of existing algorithms in optimizing complex parameter spaces, thereby improving the accuracy and stability of hemodynamic parameter prediction. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed.

[0006] Therefore, the technical problems solved by the present invention are as follows: existing hemodynamic prediction methods have problems such as single data source, static model, limited optimization, and inaccurate prediction, as well as the problem of how to achieve multi-modal data fusion and adaptive adjustment of deep learning optimization models.

[0007] To solve the above technical problems, the present invention provides the following technical solution: an intelligent analysis method for hemodynamic parameters, including: obtaining hemodynamic parameter data of a patient in real time through a multi-modal data acquisition device, and preprocessing the data to remove noise and perform standardization processing.

[0008] Analyzing the preprocessed data based on a deep learning model to predict the change trend of hemodynamic parameters and generate a prediction result.

[0009] Using an adaptive optimization algorithm to dynamically adjust the parameters and structure of the deep learning model according to real-time data and prediction results, and generate a personalized treatment plan.

[0010] As a preferred solution of the intelligent analysis method for hemodynamic parameters of the present invention, wherein: the data acquisition includes real-time collection of physiological data through wearable devices, real-time monitoring instruments, and imaging devices.

[0011] The physiological data includes blood flow velocity, blood pressure, electrocardiogram, blood oxygen saturation, vascular elasticity, respiratory rate, and body temperature.

[0012] As a preferred solution of the intelligent analysis method for hemodynamic parameters of the present invention, wherein: the data preprocessing includes using a Kalman filter to remove noise from real-time data and standardize the data to ensure data consistency and accuracy.

[0013] Using a standardization method to uniformly convert data from different sources into a consistent dimension and format.

[0014] As a preferred solution of the intelligent analysis method for hemodynamic parameters of the present invention, wherein: the analysis of the preprocessed data based on the deep learning model includes using an LSTM network to process the hemodynamic time series data of the patient, and predicting the future change trend of blood flow parameters through feature extraction and learning of the time series.

[0015] The input data is the hemodynamic parameter data of the patient and its corresponding timestamp, and time series feature extraction is performed.

[0016] In the LSTM network, the predicted blood flow parameter is the learning result based on the historical data of the previous n steps, and the predicted value is expressed as:

[0017]

[0018] Among them, represents the predicted blood flow parameter, and α i represents the weight at each time step, representing the contribution degree of each moment to the prediction result. σ represents the activation function, which is used to introduce non-linear characteristics. W i represents the weight matrix, representing the linear relationship from h t-i to the prediction. b i represents the bias term, which is used to adjust the output.

[0019] By comparing with the actual observed data, the prediction error is calculated.

[0020] To optimize the prediction performance, the prediction error is minimized, and a loss function is introduced, which is expressed as:

[0021]

[0022] Among them, e t represents the prediction error, and L represents the loss function.

[0023] Combining the predicted value of the blood flow parameter and the loss function, the final formula for the future change trend of the predicted blood flow parameter is expressed as:

[0024]

[0025] Among them, represents the future blood flow parameter of the optimized prediction, x t represents the blood flow data input at the current time step t, h t-i represents the hidden state of the LSTM, representing the characteristics of the blood flow data in the past t-i steps.

[0026] As a preferred solution of the intelligent analysis method for hemodynamic parameters described in the present invention, wherein: the LSTM network includes introducing the Swish activation function and layer normalization to improve the problem of gradient disappearance during training. After introducing the Swish activation function, the update formula of the LSTM unit is expressed as:

[0027] h t = Swish(W ih ·x t + b i ) + Swish(W hh ·h t-1 + b h )

[0028] The update process of the LSTM network after introducing layer normalization is expressed as:

[0029] h t = Swish(LayerNorm(Wih ·x t +b i ))· + Swish(LayerNorm(W hh ·h t-1 +b h ))

[0030] Among them, W ih represents the weight matrix in the LSTM network, connecting the input layer and the hidden layer. W hh represents the weight matrix in the LSTM network, connecting the hidden layer and the hidden layer. b i represents the bias, corresponding to the input state. b h represents the bias, corresponding to the hidden state. Swish represents the Swish activation function.

[0031] As a preferred solution of the hemodynamic parameter intelligent analysis method described in the present invention, wherein: the deep learning model includes training the deep learning model with historical case data and real-time patient data, and evaluating the performance of the model on different data sets through the cross-validation method.

[0032] Cross-validation divides the data set D into k subsets, conducts k times of training and validation, calculates the prediction error of the model on each subset, and finally obtains the overall performance evaluation. Let D = {D1, D2,..., D k} be the divided subsets, D k represents the k-th validation set, and the other subsets are training sets.

[0033] The error calculation formula for each cross-validation is expressed as:

[0034]

[0035] Among them, E i represents the error of the i-th validation, x j ∈D i represents each sample x j , represents the predicted value of the model for x j , y j represents the actual value, |D i | represents the number of samples in the validation set D i , and the average error of k times of cross-validation is calculated as:

[0036]

[0037] Among them, represents the performance evaluation of the overall model.

[0038] As a preferred solution of the intelligent analysis method for hemodynamic parameters of the present invention, wherein: the adaptive optimization algorithm includes automatically adjusting the key parameters of the neural network based on the error between the model prediction result and the actual measurement value. When the system detects a large prediction error, the genetic algorithm is triggered to perform a global search to optimize the parameter settings of the deep learning model.

[0039] The mean square error is used as the evaluation criterion for a large prediction error, and the calculation is expressed as:

[0040]

[0041] where y t represents the true value of the blood flow parameter prediction value, and n represents the total number of training samples.

[0042] When the mean square error is greater than the error threshold, the genetic algorithm is triggered.

[0043] The key parameters are selected as the learning rate and the number of neurons in the hidden layer. The steps of the genetic algorithm include:

[0044] Randomly generate multiple parameter combinations of different learning rates and the number of neurons in the hidden layer, train the neural network according to each group of parameters, and calculate its mean square error (MSE) as the fitness value. Select the individuals with high fitness as the parents and prepare to enter the next generation. Generate a new population through crossover and mutation, and repeat until the optimal parameter combination is found.

[0045] The optimized neural network model parameters are expressed as:

[0046]

[0047] where η opt represents the optimized learning rate, η represents the learning rate, represents the optimized number of neurons in the hidden layer, N h represents the number of neurons in the hidden layer.

[0048] As a preferred solution of the intelligent analysis method for hemodynamic parameters of the present invention, wherein: the generation of a personalized treatment plan includes, when generating a personalized treatment plan, the system provides treatment plan suggestions for doctors based on the real-time blood flow parameter prediction results, medical history, and other physiological data of the patient.

[0049] When the real-time prediction result shows that it exceeds the preset normal fluctuation range, the threshold determination mechanism is automatically triggered to generate a preliminary treatment plan and perform dynamic adjustment.

[0050] When the correlation between real-time data changes significantly, the system evaluates the patient's condition based on the prediction result and physiological correlation, and further adjusts the treatment plan. The correlation between real-time data is based on the Pearson correlation coefficient.

[0051] As a preferred solution of the intelligent analysis method for hemodynamic parameters of the present invention, wherein: the generation of the personalized treatment plan further includes that, according to the real-time prediction result and the patient's real-time health status, the system automatically generates a personalized treatment plan after each prediction. When the patient's health condition fluctuates violently within a short period of time, the system will dynamically adjust the treatment plan.

[0052] An intelligent analysis system for hemodynamic parameters, characterized in that it includes,

[0053] A preprocessing module, which obtains the hemodynamic parameter data of the patient in real time through a multi-modal data acquisition device, and preprocesses the data, removes noise and performs normalization processing.

[0054] A prediction module, which analyzes the preprocessed data based on a deep learning model, predicts the change trend of hemodynamic parameters, and generates a prediction result.

[0055] An optimization module, which uses an adaptive optimization algorithm to dynamically adjust the parameters and structure of the deep learning model according to real-time data and prediction results, and generates a personalized treatment plan.

[0056] A computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0057] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0058] The beneficial effects of the present invention: By combining multi-modal data acquisition, deep learning models, real-time optimization, and the generation of personalized treatment plans, the present invention accurately predicts hemodynamic parameters in a real-time dynamically changing medical environment, and quickly adjusts the treatment plan according to the prediction results. By combining deep learning with real-time adaptive optimization, the present invention achieves unprecedented accuracy and flexibility in hemodynamic parameter prediction and personalized treatment. The prior art cannot achieve data fusion, dynamic prediction, real-time optimization, and personalized treatment at the same time, while the present invention solves these technical problems, not only improving the prediction accuracy, but also enhancing the efficiency and effect of personalized treatment for patients. Description of the Drawings

[0059] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings. Among them:

[0060] Figure 1 It is the overall flowchart of a method and system for intelligent analysis of hemodynamic parameters provided by the first embodiment of the present invention. Specific embodiments

[0061] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0062] Example 1, referring to Figure 1 , which is an embodiment of the present invention, provides a method for intelligent analysis of hemodynamic parameters, including:

[0063] S1: Real-time acquisition of hemodynamic parameter data of patients through a multi-modal data acquisition device, and preprocessing the data to remove noise and perform standardization processing.

[0064] Real-time collection of physiological data through wearable devices, real-time monitoring instruments, and imaging devices.

[0065] The physiological data includes blood flow velocity, blood pressure, electrocardiogram, blood oxygen saturation, vascular elasticity, respiratory rate, and body temperature.

[0066] It should be noted that the blood flow velocity is measured in real time by an ultrasonic Doppler device, and the data has strong continuity, which helps to dynamically monitor the blood flow state. Blood pressure (systolic blood pressure, diastolic blood pressure) is obtained in real time through a dynamic blood pressure monitoring device or a wearable device, which can reflect vascular resistance and heart health status. Electrocardiogram (ECG) can reflect heart health status and is closely related to hemodynamics. Obtained through a wearable electrocardiogram device or a clinical device. Blood oxygen saturation (SpO2) is monitored in real time by a pulse oximeter to measure the oxygen content in the blood, reflecting the oxygenation status of the patient. Vascular elasticity is estimated by the pulse wave velocity (PWV), reflecting the hardness and elasticity of the blood vessel wall, which helps to evaluate cardiovascular health. The respiratory rate is collected in real time by a respiratory monitoring device and can be comprehensively analyzed together with other physiological data to further optimize the prediction of blood flow parameters. Body temperature changes can reflect the health status of the endocrine and immune systems and affect blood flow and blood viscosity.

[0067] Use a Kalman filter to remove noise from real-time data and standardize the data to ensure data consistency and accuracy.

[0068] Use a standardization method to uniformly convert data from different sources into a consistent dimension and format.

[0069] S2: Analyze the preprocessed data based on a deep learning model, predict the change trend of hemodynamic parameters, and generate a prediction result.

[0070] Use an LSTM network to process the hemodynamic time series data of patients, and predict the future change trend of blood flow parameters through feature extraction and learning of time series.

[0071] The input data is the hemodynamic parameter data of patients and their corresponding timestamps, and time series feature extraction is performed.

[0072] Further, the input blood flow data is represented as X = [x1, x2, …, x n , where x n represents the blood flow parameter corresponding to time step n.

[0073] It should be noted that the LSTM network is a recurrent neural network that can learn long-term dependence relationships and is very suitable for predicting time series data. In the present invention, the LSTM is used to capture the relationship between hemodynamic parameters and time, and predict the change trend of blood flow in the future for a period of time.

[0074] Further, assume that the LSTM network includes several hidden layers and memory units, and the output of each unit is a vector (hidden state). The mathematical principle of the LSTM is expressed as:

[0075] h t = f(h t-1 , x t ) = LSTM(h t-1 , x t )

[0076] where f() represents the activation function of the LSTM network, h t-1 represents the hidden state of the previous time step, and x t represents the input data of the current time step.

[0077] In the LSTM network, the predicted blood flow parameter is based on the learning result of the previous n-step historical data, and the predicted value is expressed as:

[0078]

[0079] where represents the predicted blood flow parameter, and αi represents the weight at each time step, indicating the contribution degree of each moment to the prediction result. σ represents the activation function, which is used to introduce non-linearity. W i represents the weight matrix, indicating the linear relationship from h t-i to the prediction. b i represents the bias term, which is used to adjust the output.

[0080] By comparing with the actual observed data, the prediction error is calculated, which is expressed as:

[0081]

[0082] where e t represents the prediction error.

[0083] To optimize the prediction performance, the prediction error is minimized, and the loss function is introduced, which is expressed as:

[0084]

[0085] where e t represents the prediction error, and L represents the loss function.

[0086] Combining the predicted value of the blood flow parameter and the loss function, the final formula for the future change trend of the predicted blood flow parameter is expressed as:

[0087]

[0088] where represents the future blood flow parameter of the optimized prediction, x t represents the blood flow data input at the current time step t, h t-i represents the hidden state of the LSTM, indicating the characteristics of the blood flow data in the past t - i steps.

[0089] It should be noted that the value range is the actual range of the blood flow parameter. The blood pressure is between 80 and 180 mmHg, the blood flow velocity is between 0 and 100 cm / s, and the blood oxygen saturation is between 90% and 100%. α i The value range is within 0 to 1, which is used to weight the contribution degree of historical data to the prediction.

[0090] Furthermore, by combining the error between the predicted value and the actual measured value into the loss function, the parameters of the neural network can be adjusted more accurately, enabling the model optimization process to be based on the prediction error and gradually approach the true value. By optimizing the loss function, the model can adaptively adjust the prediction strategy, reduce the error in long-term prediction, and improve the prediction accuracy of the future change trend.

[0091] This method can effectively reduce the prediction error, improve the accuracy of blood flow parameter prediction, ensure accurate prediction in complex dynamic environments (such as changes in blood pressure, blood flow velocity, etc.), and help doctors adjust treatment plans in real time and optimize clinical decisions.

[0092] The Swish activation function and layer normalization are introduced to improve the problem of gradient vanishing during training. After introducing the Swish activation function, the update formula of the LSTM unit is expressed as:

[0093] h t = Swish(W ih ·x t + b i ) + Swish(W hh ·h t-1 + b h )

[0094] After introducing layer normalization, the update process of the LSTM network is expressed as:

[0095] h t = Swish(LayerNorm(W ih ·x t + b i )) + Swish(LayerNorm(W hh ·h t-1 + b h ))

[0096] Among them, W ih represents the weight matrix in the LSTM network, connecting the input layer and the hidden layer. W hh represents the weight matrix in the LSTM network, connecting the hidden layer to the hidden layer. b i represents the bias, corresponding to the input state. b h represents the bias, corresponding to the hidden state. Swish represents the Swish activation function.

[0097] It should be noted that the Swish activation function and layer normalization are introduced to optimize the performance of the LSTM network. As an adaptive activation function, the Swish function is smoother than traditional Sigmoid or ReLU, which can reduce the problem of gradient vanishing and improve the training efficiency. Layer normalization performs normalization operations on the input data of each layer to ensure that the output of each layer remains stable and avoid excessive or too small numerical fluctuations during training.

[0098] Traditional LSTM networks are prone to the problems of vanishing gradients or exploding gradients when dealing with deep networks, leading to difficulties in optimization during the training process. The Swish activation function reduces these problems through a smoother non-linear transformation while enhancing the model's expressive power. Layer normalization, on the other hand, ensures the stability of data by normalizing the input of each layer, enabling the model to converge faster during training and effectively avoiding overfitting, thereby improving the accuracy and stability of predictions.

[0099] Furthermore, LSTM networks in the prior art encounter problems such as slow training, vanishing gradients, and overfitting in deep models, making it difficult for the network to achieve accurate predictions in complex time series data. Introducing Swish and layer normalization effectively alleviates these problems, enabling the network to learn complex hemodynamic data at a deeper level and thus improving the prediction accuracy.

[0100] Historical case data and real-time patient data are used to train the deep learning model, and the performance of the model on different data sets is evaluated through the cross-validation method.

[0101] Cross-validation divides the data set D into k subsets, conducts k times of training and validation, calculates the prediction error of the model on each subset, and finally obtains the overall performance evaluation. Let D = {D1, D2, …, D k} be the subsets after division, D k represents the k-th validation set, and the other subsets are training sets.

[0102] The error calculation formula for each cross-validation is expressed as:

[0103]

[0104] Among them, E i represents the error of the i-th validation, x j ∈ D i represents each sample x j , represents the predicted value of the model for x j , y j represents the actual value, |D i | represents the number of samples in the validation set D i , and the average error of k times of cross-validation is calculated as:

[0105]

[0106] Among them, represents the performance evaluation of the overall model.

[0107] It should be noted that cross-validation divides the dataset into multiple subsets. Each time, one subset is selected as the validation set, and the remaining part is used as the training set. Through multiple rounds of training and evaluation, the performance of the model is finally evaluated by calculating the average error of all rounds. This method can effectively avoid the deviation in model evaluation caused by different data partitions.

[0108] Cross-validation helps improve the generalization ability of the model and avoid overfitting problems. Through multiple rounds of training and validation, it can ensure that the model performs consistently on different datasets, improving the prediction accuracy and stability. Especially when the amount of data is limited, it can effectively use each part of the data for training and validation.

[0109] S3: Use an adaptive optimization algorithm to dynamically adjust the parameters and structure of the deep learning model according to real-time data and prediction results, and generate personalized treatment plans.

[0110] Based on the error between the model prediction result and the actual measurement value, automatically adjust the key parameters of the neural network. When the system detects a large prediction error, trigger the genetic algorithm to perform a global search and optimize the parameter settings of the deep learning model.

[0111] Use the mean square error as the evaluation criterion for a large prediction error, and the calculation is expressed as:

[0112]

[0113] where, y t represents the true value of the blood flow parameter prediction, and n represents the total number of training samples.

[0114] When the mean square error is greater than the error threshold, trigger the genetic algorithm.

[0115] It should be noted that the error threshold is set to 0.01 and can be adjusted in specific applications.

[0116] The key parameters are the learning rate and the number of neurons in the hidden layer. Further, the learning rate controls the step size of each parameter update of the neural network. If the learning rate is too small, the model will converge very slowly and cannot effectively learn. If the learning rate is too large, it will cause the model to skip the optimal solution and even diverge. Therefore, the learning rate is the key to controlling the stability and convergence speed of the training process. The number of neurons in the hidden layer directly determines the expressive ability of the neural network. Too few neurons may lead to underfitting and cannot fully capture the complexity of the data. Too many neurons may lead to overfitting, making the model too complex to effectively generalize. Therefore, reasonably selecting the number of neurons in the hidden layer is crucial for the model performance.

[0117] The steps of the genetic algorithm include:

[0118] Randomly generate multiple parameter combinations of different learning rates and the number of neurons in the hidden layer. Train the neural network according to each set of parameters, and calculate its mean squared error (MSE) as the fitness value. Select individuals with high fitness as parents and prepare to enter the next generation. Generate a new population through crossover and mutation, and repeat until the optimal parameter combination is found.

[0119] Further, initialize the population to randomly generate N p parameter combinations, where each combination p i contains a learning rate η i and the number of neurons in a hidden layer

[0120] Fitness evaluation: Train each individual p i and calculate the mean squared error as the fitness value:

[0121]

[0122] where MSE(p i ) represents the mean squared error of combination p i .

[0123] The larger the fitness function Fitness(p i ), the better the performance of this combination.

[0124] Selection operation: Adopt the roulette wheel selection strategy to select parents based on the fitness value:

[0125]

[0126] Crossover operation: Generate new offspring parameters:

[0127] p offspring = Crossover(p parent1 , p parent2 )

[0128] where p parent1 represents offspring 1, p parent2 represents offspring 2, and p offspring represents the offspring parameter combination generated through the crossover operation.

[0129] Mutation operation: Mutate the offspring with a certain probability to change the values of the learning rate and the number of neurons:

[0130] p mutated = Mutation(p offspring )

[0131] where p mutatedRepresents the combination of offspring parameters after mutation operation. The Mutation function represents the random change and perturbation of parameters when generating new individuals, aiming to increase the diversity of the population and prevent the algorithm from falling into local optimal solutions.

[0132] The optimized neural network model parameters are represented as:

[0133]

[0134] Among them, η opt Represents the optimized learning rate, and η represents the learning rate. Represents the optimized number of neurons in the hidden layer, and N h Represents the number of neurons in the hidden layer.

[0135] It should be noted that the key parameters (learning rate and number of neurons in the hidden layer) of the neural network are automatically adjusted based on the error between the real-time prediction result and the actual measurement value. In this way, the parameters of the model can be dynamically adjusted according to real-time data, enabling the model to always maintain an optimal state, thereby reducing the prediction error and improving the prediction ability of the model. Different from traditional fixed model optimization methods, the present invention adopts an adaptive optimization algorithm (such as a genetic algorithm) to adjust parameters by feeding back the difference between the real-time prediction result and the actual value, greatly enhancing the adaptability of the model to different patient data. In clinical applications, adjusting the parameters of the model in real time can cope with changes in the patient's health status and ensure the accuracy and stability of long-term predictions.

[0136] When generating a personalized treatment plan, the system provides treatment plan suggestions for doctors based on the real-time blood flow parameter prediction results, medical history, and other physiological data of the patient.

[0137] When the real-time prediction result shows that it exceeds the preset normal fluctuation range, a threshold determination mechanism is automatically triggered to generate a preliminary treatment plan and perform dynamic adjustment.

[0138] When there are significant changes in the correlation between real-time data, the system evaluates the patient's condition based on the prediction results and physiological correlation, and further adjusts the treatment plan. The correlation between real-time data is based on the Pearson correlation coefficient.

[0139] Furthermore, the Pearson correlation coefficient is represented as:

[0140]

[0141] Among them, x i Represents the historical blood pressure data, and y i Represents the historical blood flow velocity data. Represents the mean value of blood pressure data. Represents the mean value of blood flow velocity data.

[0142] When ρ approaches 0, the correlation between the two weakens, and intervention needs to be adjusted in a timely manner.

[0143] It should be noted that the Pearson coefficient of blood flow velocity and blood pressure is used to measure the correlation between the two. In a normal healthy state, the correlation coefficient between blood flow velocity and blood pressure is generally between 0.5 and 1.0, indicating a positive correlation between the two. The closer the value is to 1, the more consistent the changes in blood flow velocity and blood pressure are. The closer the value is to 0, the weaker the relationship between the two changes. If the Pearson coefficient is below 0.5, it may indicate an abnormal pathological condition and further clinical examinations may be required.

[0144] Based on the real-time prediction results and the patient's real-time health status, the system automatically generates a personalized treatment plan after each prediction. When the patient's health condition fluctuates violently within a short period of time, the system will dynamically adjust the treatment plan.

[0145] In the above embodiments, there is also an intelligent analysis system for hemodynamic parameters, specifically:

[0146] A preprocessing module that obtains the hemodynamic parameter data of the patient in real time through a multi-modal data acquisition device, and preprocesses the data to remove noise and perform normalization processing.

[0147] A prediction module that analyzes the preprocessed data based on a deep learning model, predicts the change trend of hemodynamic parameters, and generates a prediction result.

[0148] An optimization module that uses an adaptive optimization algorithm to dynamically adjust the parameters and structure of the deep learning model according to real-time data and prediction results, and generates a personalized treatment plan.

[0149] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0150] Embodiment 2, an embodiment of the present invention, provides a method and system for intelligent analysis of hemodynamic parameters. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0151] In order to verify the effectiveness and innovation of the present invention, the following experiment is designed. The intelligent analysis method of hemodynamic parameters of the present invention is applied to conduct comparative experiments and data collection. The main objective of this experiment is to compare the effect differences between traditional blood flow parameter prediction methods and the method of the present invention, and focus on investigating prediction accuracy, real-time adjustment ability, and the generation of personalized treatment plans.

[0152] In the experiment, 6 patients were selected as the test subjects, and their key physiological parameters such as blood flow velocity, blood pressure, and blood oxygen saturation were recorded. The data acquisition devices included a wearable blood pressure monitor, a blood flow velocity sensor, an electrocardiogram device, and a pulse oximeter, etc. Before the experiment started, all devices were calibrated to ensure the accuracy and consistency of the data. The experiment process was divided into two stages. The first stage was the prediction by traditional methods, and the second stage was the application of the method of the present invention.

[0153] Data acquisition: The blood flow velocity, blood pressure, blood oxygen saturation and other parameters of the test subjects were collected in real time through various sensors, and the time series data were recorded every 5 minutes. These data would be used as inputs for processing to facilitate the operations of subsequent deep learning models and adaptive optimization algorithms.

[0154] Application of the prediction model: In the first stage, traditional blood flow prediction methods (based on historical data and simple regression models) were used to predict blood flow parameters. Then, in the second stage, the deep learning model proposed in the present invention was adopted, specifically using the LSTM network to predict blood flow parameters. At this time, the model not only utilized historical data but also combined the physiological data collected in real time for multi-modal data fusion.

[0155] Real-time optimization: In the second stage, the system used an adaptive optimization algorithm to automatically adjust the key parameters of the deep learning model, such as the learning rate and the number of neurons, according to the error between the real-time data and the prediction results to optimize the prediction results. When the prediction error exceeded the set threshold (MSE>0.01), the system would trigger a genetic algorithm for global search to optimize the model parameters.

[0156] Generation of personalized treatment plans: Based on the prediction results, the system automatically generated personalized treatment plans. Especially when it was predicted that the patient's blood flow was abnormal, the system would automatically adjust the drug dosage or treatment method.

[0157] Recording and analysis of experimental data: During the whole experiment process, all parameter changes and treatment plan adjustment data were recorded in real time and input into the database for later comparison and analysis. The experimental data are shown in Table 1.

[0158] Table 1 Experimental data

[0159]

[0160] By analyzing the experimental data, the innovation and beneficial effects of the present invention in the prediction of hemodynamic parameters and the generation of personalized treatment plans can be intuitively seen. The data in the table show the blood flow velocity, blood pressure, blood oxygen saturation, heart rate, and prediction error (MSE) of each test subject.

[0161] By comparing with the prediction error (MSE) of the traditional method, it is found that the method of the present invention significantly reduces the prediction error. For example, the MSE of the test subject P001 is 0.015, while the error of the traditional method is usually above 0.02. By optimizing the deep learning model and adjusting the real-time data, the system can more accurately predict the change trend of future blood flow parameters, thus improving the prediction accuracy.

[0162] The adaptive optimization algorithm of the present invention can dynamically adjust the model parameters according to the prediction error. When the system detects a large prediction error, it automatically triggers the genetic algorithm for global optimization, thereby further improving the prediction accuracy. For example, in the experiment of P003, the initial prediction error is 0.020, but after optimizing with the genetic algorithm, the error drops to 0.014, significantly improving the prediction accuracy.

[0163] By combining multiple data sources such as blood flow velocity, blood pressure, heart rate, and blood oxygen saturation, the method of the present invention can more comprehensively reflect the health status of patients. This multi-modal data fusion not only improves the prediction accuracy but also provides more information support during the treatment process, making personalized treatment more precise.

[0164] From the data and analysis results in the table, it can be seen that the intelligent analysis method of hemodynamic parameters of the present invention has significant advantages, especially in the aspects of adaptive optimization algorithm and personalized treatment plan generation. By combining the deep learning model and real-time data optimization, the present invention not only improves the accuracy of blood flow parameter prediction but also enables the treatment plan to be adjusted in real time according to the actual state of the patient, overcoming the problems of static prediction and lagged treatment in traditional technologies. In addition, the present invention can effectively reduce the prediction error, improve the real-time and precision of clinical treatment, thus significantly enhancing the treatment effect. Compared with the prior art, the present invention has innovation and practicability in processing complex physiological data and can provide more intelligent decision-making support for doctors in actual medical applications.

[0165] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An intelligent analysis method for hemodynamic parameters, characterized in that: include: The patient's hemodynamic parameter data is acquired in real time through multimodal data acquisition equipment, and the data is preprocessed, noise is removed, and standardized; Analyzing the preprocessed data based on a deep learning model, predicting the change trend of hemodynamic parameters, and generating prediction results; Adaptive optimization algorithms are used to dynamically adjust the parameters and structure of deep learning models based on real-time data and prediction results, and generate personalized treatment plans.

2. The intelligent hemodynamic parameter analysis method according to claim 1, characterized in that: The data collection includes real-time collection of physiological data through wearable devices, real-time monitoring instruments and imaging devices; Physiological data include blood flow velocity, blood pressure, electrocardiogram, blood oxygen saturation, vascular elasticity, respiratory rate and body temperature.

3. The intelligent hemodynamic parameter analysis method according to claim 2, characterized in that: The data preprocessing includes using a Kalman filter to remove noise from real-time data and standardize the data to ensure data consistency and accuracy; Use standardized methods to convert data from different sources into consistent dimensions and formats.

4. The intelligent hemodynamic parameter analysis method according to claim 3, characterized in that: The analyzing the pre-processed data based on the deep learning model includes using an LSTM network to process the patient's hemodynamic time series data, and predicting the future change trend of blood flow parameters through feature extraction and learning of the time series; The input data is the patient's hemodynamic parameter data and its corresponding timestamp, and time series feature extraction is performed; In the LSTM network, the predicted blood flow parameters are the learning results based on the previous n steps of historical data, and the predicted values ​​are expressed as: in, represents the predicted blood flow parameter, α i represents the weight of each time step, representing the contribution of each moment to the prediction result; σ represents the activation function, which is used to introduce nonlinear characteristics; W i Represents the weight matrix, which represents the t-i The linear relationship to the prediction; b i Represents the bias term, which is used to adjust the output; By comparing with the actual observed data, the prediction error is calculated; In order to optimize the prediction performance and minimize the prediction error, a loss function is introduced, which is expressed as: Among them, e t represents the prediction error, and L represents the loss function; Combining the predicted value of blood flow parameters with the loss function, the final formula for predicting the future change trend of blood flow parameters is expressed as: in, represents the optimized predicted future blood flow parameter, x t represents the blood flow data input at the current time step t, h t-i Represents the hidden state of LSTM, which represents the blood flow data features of the past ti steps.

5. The intelligent hemodynamic parameter analysis method according to claim 4, characterized in that: The LSTM network includes introducing the Swish activation function and layer normalization to improve the gradient vanishing problem during the training process. After the Swish activation function is introduced, the update formula of the LSTM unit is expressed as: h t =Swish(W ih ·x t +b i )+Swish(W hh ·h t-1 +b h ) The update process of the LSTM network after introducing layer normalization is expressed as: h t =Swish(LayerNorm(W ih ·x t +b i ))+Swish(LayerNorm(W hh ·h t-1 +b h )) Among them, W ih Represents the weight matrix in the LSTM network, connecting the input layer and the hidden layer; W hh Represents the weight matrix in the LSTM network, connecting the hidden layer and the hidden layer; b i Indicates bias, corresponding to the input state; b h represents the bias, corresponding to the hidden state; Swish represents the Swish activation function.

6. The intelligent hemodynamic parameter analysis method according to claim 5, characterized in that: The deep learning model includes using historical case data and real-time patient data to train the deep learning model, and evaluating the performance of the model on different data sets through a cross-validation method; Cross-validation divides the dataset D into k subsets, performs k training and validation, calculates the prediction error of the model on each subset, and finally obtains the overall performance evaluation; let D = {D1, D2, …, D k } is the subset after division, D k represents the kth validation set, and the other subsets are training sets; The error calculation formula for each cross-validation is expressed as: Among them, E i represents the error of the i-th verification, x j ∈D i Represents each sample x in the validation set j , Represents the model for x j The predicted value, y j Indicates the actual value, |D i | represents the validation set D i The number of samples in The average error of k-fold cross validation is calculated as: in, Represents the performance evaluation of the overall model.

7. The intelligent analysis method of hemodynamic parameters according to claim 6, characterized in that: The adaptive optimization algorithm includes automatically adjusting key parameters of the neural network based on the error between the model prediction results and the actual measured values; If the system detects that the prediction error is large, it triggers the genetic algorithm to perform a global search and optimize the parameter settings of the deep learning model; The mean square error is used as the evaluation criterion for large prediction error, and the calculation is expressed as: Among them, y t represents the true value of the predicted value of the blood flow parameter, and n represents the total number of training samples; When the mean square error is greater than the error threshold, the genetic algorithm is triggered; The key parameters are learning rate and number of hidden layer neurons. The genetic algorithm steps include: Randomly generate multiple different learning rate and hidden layer neuron number parameter combinations, train the neural network according to each set of parameters, and calculate its mean square error (MSE) as the fitness value; select individuals with high fitness as parents to prepare for the next generation; generate new populations through crossover and mutation, and repeat until the optimal parameter combination is found; The optimized neural network model parameters are expressed as: Among them, η opt represents the optimized learning rate, η represents the learning rate, Represents the number of hidden layer neurons after optimization, N h Represents the number of neurons in the hidden layer.

8. The intelligent hemodynamic parameter analysis method according to claim 7, characterized in that: Generating a personalized treatment plan includes, when generating a personalized treatment plan, the system provides a doctor with treatment plan suggestions based on the patient's real-time blood flow parameter prediction results, medical history and other physiological data; When the real-time prediction results show that they exceed the preset normal fluctuation range, the threshold judgment mechanism is automatically triggered to generate a preliminary treatment plan and make dynamic adjustments; When the correlation between real-time data changes significantly, the system evaluates the patient's condition based on the predicted results and physiological correlations, and further adjusts the treatment plan; the correlation between real-time data is based on the Pearson correlation coefficient.

9. The intelligent hemodynamic parameter analysis method according to claim 8, characterized in that: Generating a personalized treatment plan also includes, based on the real-time prediction results and the patient's real-time health status, the system automatically generates a personalized treatment plan after each prediction; When a patient's health condition fluctuates dramatically in a short period of time, the system will dynamically adjust the treatment plan.

10. An intelligent hemodynamic parameter analysis system using the method according to any one of claims 1 to 9, characterized in that: The preprocessing module acquires the patient's hemodynamic parameter data in real time through a multimodal data acquisition device, and preprocesses the data to remove noise and perform standardization; A prediction module, which analyzes the preprocessed data based on a deep learning model, predicts the change trend of hemodynamic parameters, and generates prediction results; The optimization module uses an adaptive optimization algorithm to dynamically adjust the parameters and structure of the deep learning model based on real-time data and prediction results, and generate personalized treatment plans.

Images