Multi-modal data fusion intelligent children asthma prediction system and method

Abstract

The invention provides a multi-modal data fusion intelligent children asthma prediction system, which comprises a data acquisition and preprocessing unit, a feature extraction unit, a data fusion and intelligent prediction unit and a decision and feedback unit, and is characterized in that the data acquisition and preprocessing unit is used for acquiring multi-modal physiological signal data; the feature extraction unit is used for extracting features of the multi-modal physiological signals; the data fusion and intelligent identification unit is used for carrying out feature fusion on the extracted multi-modal physiological signal data; the decision and feedback unit is used for comprehensive judgment and dynamic feedback adjustment; in the dynamic feedback and self-adaptive adjustment part, a correction coefficient and the weight of the model are adjusted according to real-time data, so that the prediction probability is more accurate. The system effectively overcomes the problem that the existing system depends on a single data source or medical staff depends on subjective evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical treatment, in particular to a multi-modal data fusion intelligent child asthma prediction system and method. BACKGROUND

[0002] Asthma is a chronic non-specific inflammatory disease of the respiratory tract, which belongs to type I hypersensitivity disease, leading to airway contraction, spasm and hyperemia, and further causing respiratory difficulty, cough, wheezing and chest tightness and other symptoms. Asthma is one of the most common chronic respiratory diseases in childhood, affecting about 7% to 10% of children worldwide. If not detected in time, long-term ineffective control may lead to permanent changes in airway structure (airway remodeling), which in turn leads to a sustained decline in lung function in children as adults and increases the risk of other respiratory diseases. Therefore, early and accurate prediction and timely treatment of asthma are crucial for the physical and mental health development of children.

[0003] Currently, the detection of child asthma mainly relies on asthma prediction index (API), lung function detection, biomarker detection and other methods. Among them, API is a commonly used tool in clinical practice for assessing the risk of developing asthma in young children in the future. This method predicts whether children will have persistent asthma symptoms after preschool age based on specific clinical criteria and family history. However, in actual application, due to its reliance on a few clinical and family history indicators, it cannot fully reflect all factors related to the onset of asthma, i.e., there are limitations in data sources, which may lead to underestimation or overestimation of the risk of asthma in children in some cases. Other methods also have similar problems, although they can help identify high-risk children to some extent, but due to the single data or reliance on subjective evaluation, it is difficult to fully and accurately predict the occurrence of asthma.

[0004] With the development of big data technology, artificial intelligence technology is increasingly widely used in the field of medical health, not only improving treatment accuracy and optimizing resource allocation, but also effectively reducing the probability of misdiagnosis and missed diagnosis. In the prediction of child asthma, artificial intelligence technology can be used to fuse and analyze data such as chest sound, respiratory rate, heart rate, blood oxygen saturation and airflow rate, which can directly reflect the physiological state and respiratory function changes of children during asthma attacks, thereby providing more comprehensive and accurate predictions.

[0005] Based on the above description, it is necessary to address the problems of relying on single data and subjective evaluation by medical personnel in existing child asthma prediction methods, and to propose an effective method for predicting child asthma by fusing multi-modal data such as child chest sound, respiratory rate, heart rate, blood oxygen saturation and airflow rate. SUMMARY

[0006] The present application provides a multi-modal data fusion intelligent child asthma prediction system and method to effectively fuse the physiological state and respiratory function change data of children with artificial intelligence technology, thereby improving the comprehensiveness and accuracy of prediction.

[0007] The technical scheme of the present application provides a multi-modal data fusion intelligent child asthma prediction system, which comprises a data acquisition and preprocessing unit, a feature extraction unit, a data fusion and intelligent prediction unit, and a decision and feedback unit.

[0008] The data acquisition and preprocessing unit is used to acquire multi-modal physiological signal data.

[0009] The feature extraction unit is used to extract the features of multi-modal physiological signals.

[0010] The data fusion and intelligent recognition unit is used to fuse the features of the extracted multi-modal physiological signal data and construct a feature vector, and then correct the feature vector considering the influence of environment and individual differences. The intelligent recognition unit first constructs a data set of the feature vector and inputs it into a fully connected neural network for training, and finally generates a probability prediction function.

[0011] The decision and feedback unit is used for comprehensive judgment and dynamic feedback adjustment. In the comprehensive judgment part, the average probability in the time window T is calculated according to the probability prediction function to judge the asthma risk, and a threshold is set. When the probability exceeds the threshold, an asthma risk alarm is triggered. In the dynamic feedback and adaptive adjustment part, the correction coefficient and the weight of the model are adjusted according to the real-time data, so that the prediction probability is more accurate.

[0012] Further, the chest sound S(t) is measured by an electronic stethoscope, the respiratory frequency fr(t) is measured by a respiratory sensor, the heart rate HR(t) is measured by a heart rate monitoring device, the blood oxygen saturation SpO2(t) is measured by an oximeter, and the airflow velocity v(t) is measured by a flow sensor. The signal preprocessing method of multi-modal data adopts noise filtering, signal smoothing and normalization method.

[0013] Further, the feature extraction of chest sound utilizes short-time Fourier transform, the respiratory instantaneous frequency is calculated by Hilbert transform, the heart rate feature is extracted by calculating heart rate variability, the blood oxygen saturation feature is calculated by oxygenation index, and the airflow rate feature is represented by respiratory work.

[0014] The present application also provides a prediction method using the multi-modal data fusion intelligent child asthma prediction system, comprising the following steps:

[0015] Step S1: data acquisition: the data collected by the system includes chest sound S(t), respiratory frequency fr (t), heart rate HR(t), blood oxygen saturation SpO2(t) and airflow velocity v(t);

[0016] Step S2: data preprocessing: after collecting the multi-modal data, the physiological signals are preprocessed, mainly including data denoising, data smoothing and normalization processing;

[0017] The signal S(t), f r (t), HR(t), SpO2(t) and v(t) are filtered using a band-pass filter, and the transfer function of the band-pass filter is: Wherein: f0 is the center frequency, Q is the quality factor;

[0018] Step S3: feature extraction, after completing the preprocessing operation of the physiological signals, the data is extracted, and the key information that helps to predict asthma is extracted from the five physiological signal data, wherein:

[0019] Chest sound feature extraction: frequency feature is extracted by using short-time Fourier transform:

[0020] (1) Frequency component analysis: Wherein, t represents time, ω represents frequency, is the input signal, defined on the time variable τ, w(t) is the window function;

[0021] (2) Energy calculation: Wherein, ω1 and ω2 are the low and high frequency boundaries respectively, X(t, ω) represents the short-time Fourier transform result of the signal, which is the spectral distribution of the signal S(τ) at time t and frequency ω;

[0022] Respiratory rate feature extraction: the instantaneous frequency is calculated using Hilbert transform: Wherein, H[·] represents Hilbert transform, which converts the signal into its corresponding analytic signal;

[0023] Heart rate variation feature extraction: the heart rate variability is analyzed to calculate the variation rate: Wherein, HR i is the i-th heartbeat interval, is the average value of the heart rate, and N represents the number of heartbeat data, i.e. the number of heartbeats collected in a period of time;

[0024] Blood oxygen saturation feature extraction: calculate the oxygenation index: Which represents the oxygenation level at time t, wherein SpO2(t) represents blood oxygen saturation, f r,inst (t) represents instantaneous respiratory rate;

[0025] Airflow rate feature extraction: use airflow rate to calculate respiratory work measurement: Where P(τ) is the respiratory pressure, P(τ) = kv(τ) 2 , k is a constant, v(τ) represents the airflow rate, and P(τ) represents the airway pressure;

[0026] τ is an intermediate time representing the integration process; the product of the airflow rate v(τ) and the airway pressure P(τ) is calculated at each time τ, and then all τ from 0 to t are accumulated, to obtain the cumulative work W resp (t) of the entire time period;

[0027] Step S4: data fusion, fuse the multi-modal data after feature extraction to form a comprehensive multi-dimensional feature vector: Then the influence of environmental temperature T e , humidity H e , and patient's age A, weight W and other parameters also need to be considered, and the multi-dimensional feature vector is corrected: Where α i is the correction coefficient function;

[0028] Step S5: intelligent identification, the corrected feature vector x * (t) and the corresponding label form a data set, and are randomly divided into training set, validation set and test set according to the ratio of 7:1:2, the training set is used to train the model and optimize the model parameters; the validation set is used to adjust the hyperparameters, select the model, and monitor the overfitting situation; the test set is used to evaluate the final performance of the model, to ensure the generalization ability and actual application effect of the model, and then the data set information is input into the fully connected neural network; in the MLP network, the input layer is used to accept the corrected feature vector x * (t) and label information;

[0029] Step S6: make a decision, judge the risk of children's asthma through the average probability in the time window T: Then set the threshold θ, when , trigger the asthma risk alarm;

[0030] Step S7: feedback, based on the decision-making of the above step S6, add dynamic feedback for self-adaptive adjustment:

[0031] Where α i is the model correction coefficient, w is the model weight, L is the loss function, and η is the learning rate.

[0032] Furthermore, the purpose of data denoising in step S2 is to remove the influence of environmental noise, equipment noise and electromagnetic interference on physiological signals during the physiological signal acquisition process, and to retain the effective feature information in the multimodal data signals.

[0033] Furthermore, the normalization process in step S2 can standardize the data range of different signals, which facilitates the subsequent extraction of physiological signal features.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] 1. This invention uses multimodal data fusion of five physiological signals—chest sounds, respiratory rate, heart rate, blood oxygen saturation, and airflow velocity—to predict childhood asthma, effectively overcoming the problems of existing methods that rely on a single data source or subjective assessments by medical staff.

[0036] 2. Based on the fused multidimensional feature vector, this invention considers and corrects the influence of environmental temperature, humidity, and patient age, weight, and other parameters, which can effectively improve the robustness of the prediction model and improve the long-term prediction effect. Attached Figure Description

[0037] Figure 1 Structure diagram of a multimodal data fusion intelligent childhood asthma prediction system:

[0038] Figure 2 Flowchart of a multimodal data fusion-based intelligent method for predicting childhood asthma. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0040] like Figure 1 As shown, the present invention provides a multimodal data fusion intelligent childhood asthma prediction system and method. The system provided by the present invention mainly consists of four parts: a data acquisition and preprocessing unit, a feature extraction unit, a data fusion and intelligent prediction unit, and a decision and feedback unit.

[0041] The data acquisition and preprocessing unit first acquires multimodal physiological signal data; chest sound S(t) can be measured using an electronic stethoscope, and respiratory rate f... r(t) can be measured by respiratory sensors such as chest belt sensors, nasal airflow sensors, etc. Heart rate HR(t) can be measured by heart rate monitoring devices (electrocardiogram, wristband heart rate detectors, etc.). Blood oxygen saturation SpO2(t) can be measured by oximeters. Airflow velocity v(t) can be measured by flow sensors. The signal preprocessing method for the above-mentioned multi-modal data mainly adopts noise filtering, signal smoothing and normalization methods.

[0042] The feature extraction unit mainly includes extracting features of multi-modal physiological signals: the feature extraction of chest sound utilizes short-time Fourier transform (STFT), calculates respiratory instantaneous frequency by Hilbert transform, extracts heart rate features by calculating heart rate variability, and calculates oxygenation index to extract blood oxygen saturation features. The airflow rate feature is represented by respiratory work.

[0043] The data fusion and intelligent recognition unit includes a data fusion part: the extracted multi-modal physiological signal data is fused and a feature vector is constructed, and then the feature vector is corrected considering the influence of environmental and individual differences. In the intelligent recognition part, first construct a data set of the feature vector and input it into a fully connected neural network (MLP) for training, and finally generate a probability prediction function.

[0044] The decision and feedback unit includes comprehensive judgment and dynamic feedback adjustment. In the comprehensive judgment part, the average probability in the time window T is calculated according to the probability prediction function to judge the asthma risk, and a threshold is set. When the probability exceeds this threshold, an asthma risk alarm is triggered. In the dynamic feedback and adaptive adjustment part, the correction coefficient and the weight of the model are adjusted according to the real-time data, so that the prediction probability is more accurate.

[0045] The method flowchart is as shown in Figure 2 The embodiment also provides a prediction method of the intelligent child asthma prediction system using multi-modal data fusion, which includes the following steps:

[0046] Step S1 data acquisition: the data collected by the system includes chest sound S(t), respiratory frequency f r (t), heart rate HR(t), blood oxygen saturation SpO2(t) and airflow velocity v(t).

[0047] The chest sound S(t) in the above-mentioned step S1 can reflect the condition of the respiratory system, especially the wheezing, breathing difficulty and other characteristics that may occur in asthma patients, helping to identify airway stenosis or other abnormalities. The chest sound S(t) can be measured by an electronic stethoscope.

[0048] The respiratory frequency f r(t) can reflect the patient's breathing rhythm and frequency changes, help identify respiratory difficulties, shortness of breath, etc., which are usually early signs of asthma attacks. Respiratory frequency f r (t) can be measured by respiratory sensors such as chest strap sensors, nasal airflow sensors, etc.

[0049] The heart rate HR(t) described in the above step S1 reflects the frequency and intensity of heart activity, and low oxygenemia and stress response during asthma attacks usually cause heart rate to rise, so changes in heart rate can be an important indicator for assessing the severity of asthma attacks. Heart rate HR(t) can be measured by heart rate monitoring devices (electrocardiogram, wristband heart rate detector, etc.).

[0050] The blood oxygen saturation SpO2(t) described in the above step S1 is a key indicator of oxygen content in the blood, and during asthma attacks, the patient's airway narrows, leading to insufficient oxygen intake, and the blood oxygen saturation decreases. Therefore, continuous monitoring of blood oxygen saturation can help predict the risk of hypoxia in asthma patients, and SpO2(t) can be measured by an oximeter.

[0051] The airflow velocity v(t) described in the above step S1 directly reflects the smoothness of the patient's airway during breathing. Asthma patients usually experience a decrease in airflow rate during an attack, and changes in airflow rate can help quantify the degree of airway obstruction and serve as a basic evaluation data for assessing the severity of asthma attacks. Airflow velocity v(t) can be measured by a flow sensor.

[0052] Step S2: Data preprocessing: After collecting multi-modal data, the physiological signals need to be preprocessed, mainly including data denoising, data smoothing and normalization processing.

[0053] The purpose of data denoising described in the above step S2 is to remove the effects of environmental noise, device noise, and electromagnetic interference on physiological signals during signal acquisition. By filtering, these unnecessary noises can be removed, and the effective feature information in the multi-modal data signal can be retained. Usually, a band-pass filter is used to filter the signals S(t), f r (t), HR(t), SpO2(t), v(t), and the transfer function of the band-pass filter is: where f0 is the center frequency and Q is the quality factor.

[0054] The purpose of the data smoothing in step S2 is to reduce short-term signal fluctuations and highlight the long-term trend of the signal. Physiological signal data often includes many short-term random fluctuations, which may be caused by temporary noise or acquisition instability, rather than reflecting actual physiological changes. These short-term fluctuations interfere with the judgment of the overall trend of the signal, and thus affect the analysis and prediction results. In processing physiological signal data, Gaussian kernel function K(t) can be used for data smoothing. Taking chest sound S(t) as an example: Similarly, the data smoothing processing can also be performed on other signals f r (t), HR(t), SpO2(t), and v(t) in the same way.

[0055] The normalization processing in step S2 can standardize the data range of different signals, facilitating the subsequent operation of physiological signal feature extraction. In processing physiological signal data, the Min-Max normalization method can be used. Taking chest sound S(t) as an example, its normalized value S n (t) can be expressed as: where S min and S max represent the minimum and maximum values of the signal, respectively. Similarly, the normalization processing can also be performed on other signals f r (t), HR(t), SpO2(t), and v(t) in the same way.

[0056] Step S3: Feature extraction. After completing the preprocessing operation of the physiological signal, the data needs to be extracted for feature extraction, and the key information that contributes to asthma prediction needs to be extracted from the above five physiological signal data.

[0057] Chest sound feature extraction: frequency feature extraction using short-time Fourier transform (STFT):

[0058] (1) Frequency component analysis: where t represents time, ω represents frequency, is the input signal, defined on the time variable τ, and w(t) is the window function;

[0059] (2) Energy calculation: where ω1 and ω2 are the low and high frequency boundaries, respectively, and X(t, ω) represents the short-time Fourier transform result of the signal, which is the spectral distribution of the signal S(τ) at time t and frequency ω;

[0060] Respiratory rate feature extraction: calculate its instantaneous frequency using Hilbert transform: where H[·] represents the Hilbert transform, which converts the signal into its corresponding analytic signal;

[0061] Feature extraction of heart rate variation: calculate its variation rate by heart rate variability analysis: Wherein, HR i is the i th heartbeat interval, is the average value of heart rate, N represents the number of heartbeat data, that is, the number of heartbeats collected in a period of time;

[0062] Feature extraction of blood oxygen saturation: calculate the oxygenation index: Which represents the oxygenation level at time t, wherein SpO2(t) represents the blood oxygen saturation, f r,inst (t) represents the instantaneous respiratory rate;

[0063] Feature extraction of airflow rate: calculate respiratory work measurement using airflow rate: Wherein, P(τ) is the respiratory pressure, P(τ)=kv(τ) 2 , k is a constant, v(τ) represents the airflow rate, P(τ) represents the airway pressure;

[0064] τ is the intermediate time in the integration process; Calculate the product of airflow rate v(τ) and airway pressure P(τ) at each time τ, then accumulate all τ from 0 to t time points, can get the cumulative work W resp (t) in the whole period;

[0065] Step S4: data fusion, fuse the multi-modal data after feature extraction to form a comprehensive multi-dimensional feature vector: Then we also need to consider the influence of environmental temperature T e , humidity H e , and patient's age A, weight W and other parameters, correct the multi-dimensional feature vector: Wherein, α i is the correction coefficient function.

[0066] Step S5: intelligent identification, the corrected feature vector x * (t) and the corresponding label (such as asthma diagnosis result) form a data set, and are randomly divided into training set, validation set and test set according to the ratio of 7:1:2, the training set is used to train the model and optimize the model parameters; The validation set is used to adjust the hyperparameters, select the model, and monitor the overfitting situation; The test set is used to evaluate the final performance of the model to ensure the generalization ability and actual application effect of the model, and then input the data set information into the fully connected neural network (MLP). In the MLP network, the input layer is used to accept the corrected feature vector x *(t) and label information, the dimension of the input layer is determined by the length of the feature vector; the hidden layer is a key part for training the data set, which extracts and converts features from the input data through nonlinear transformation, which enables the model to learn complex patterns and relationships in the data output; the output layer generates the final prediction result. Finally, the prediction probability of children's asthma is obtained: P asthm (t) = σ(w T x * (t) + b), where w is the model weight vector, b is the bias, and sigma (·) is the activation function.

[0067] Step S6: make a decision, judge the risk of children's asthma by the average probability in the time window T: Then set the threshold value θ, when , trigger the asthma risk alarm.

[0068] Step S7: feedback, based on the decision-making in the above step S6, dynamic feedback can also be added for adaptive adjustment: Where, alpha i is the model correction coefficient, w is the model weight, L is the loss function, and eta is the learning rate.

[0069] The above only describes the preferred embodiments of the present application and is not used to limit the application of the present application. Within the scope of knowledge possessed by those skilled in the art or ordinary technical personnel, various changes can be made without departing from the purpose of the present application.

Claims

1. A multi-modal data fusion intelligent children asthma prediction system, comprising a data acquisition and preprocessing unit, a feature extraction unit, a data fusion and intelligent prediction unit, a decision and feedback unit, characterized in that: the data acquisition and preprocessing unit is used to acquire multi-modal physiological signal data; the feature extraction unit is used to extract the features of multi-modal physiological signals; the data fusion and intelligent recognition unit is used to fuse the features of the extracted multi-modal physiological signal data and construct a feature vector, and then correct the feature vector considering the influence of environment and individual differences; the intelligent recognition unit first constructs a data set of the feature vector and inputs it into a fully connected neural network for training, and finally generates a probability prediction function; the decision and feedback unit is used for comprehensive judgment and dynamic feedback adjustment; in the comprehensive judgment part, the average probability in the time window T is calculated according to the probability prediction function to judge the asthma risk, and a threshold is set; when the probability exceeds the threshold, an asthma risk alarm is triggered; in the dynamic feedback and adaptive adjustment part, the correction coefficient and the weight of the model are adjusted according to the real-time data, so that the prediction probability is more accurate.

2. The multi-modal data fusion intelligent child asthma prediction system of claim 1, wherein: The chest sound S(t) is measured by an electronic stethoscope, the respiratory frequency fr(t) is measured by a respiration sensor, the heart rate HR(t) is measured by a heart rate monitoring device, the blood oxygen saturation SpO2(t) is measured by an oximeter, and the airflow velocity v(t) is measured by a flow sensor; the signal preprocessing method of multi-modal data adopts noise filtering, signal smoothing and normalization method.

3. The multi-modal data fusion intelligent child asthma prediction system of claim 1, wherein: The feature extraction of chest sound uses short-time Fourier transform, calculates the respiratory instantaneous frequency by Hilbert transform, extracts the heart rate feature by calculating the heart rate variability, calculates the oxygen saturation feature by oxygenation index, and represents the airflow rate feature by respiratory work.

4. A prediction method of an intelligent child asthma prediction system using multi-modal data fusion, characterized in that: The method comprises the following steps: Step S1: data acquisition: the data collected by the system include chest sound S(t), respiratory frequency f r (t), heart rate HR(t), blood oxygen saturation SpO2(t) and airflow velocity v(t); Step S2: data preprocessing; after collecting multi-modal data, the physiological signals are preprocessed, mainly including data denoising, data smoothing and normalization processing; The signals S(t), f r (t), HR(t), SpO2(t), v(t) are filtered using a bandpass filter with a transfer function: where f0is the center frequency and Q is the quality factor. Step S3: feature extraction; after the preprocessing operation of the physiological signals is completed, the data is subjected to feature extraction, and the key information helpful for asthma prediction is extracted from the five physiological signal data, wherein: Feature extraction of chest sound: frequency feature is extracted by short-time Fourier transform: (1) Frequency component analysis: where t denotes time, ω denotes frequency, is the input signal, defined on the time variable τ, w(t) is a window function; (2) Energy calculation: where ω1and ω2are the low and high frequency boundaries, respectively, and X(t, ω) represents the short-time Fourier transform of the signal, which is the spectral distribution of the signal S(τ) at time t and frequency ω. Feature extraction of respiratory rate: its instantaneous frequency is computed using the Hilbert transform: where H[•] denotes the Hilbert transform, which transforms a signal into its corresponding analytic signal; Feature extraction of heart rate variation: the heart rate variability is calculated for its variation rate: wherein HR i is the i-th heartbeat interval, is the average value of the heart rate, and N represents the number of heartbeat data, i.e. the number of heartbeats collected in a period of time; Feature extraction of blood oxygen saturation: calculation of oxygenation index: which represents the oxygenation level at time t, where SpO2(t) represents the blood oxygen saturation, f r，inst (t) represents the instantaneous respiratory rate; Airflow rate feature extraction: Respiratory work count is calculated using airflow rate: where P(τ) is the respiratory pressure, P(τ) = kv(τ) 2 k is a constant, v(τ) represents the airflow rate, and P(τ) represents the airway pressure; τ is an intermediate time point during the integration; the product of the airflow rate v(τ) and the airway pressure P(τ) is calculated at each time point τ, and then summed over all time points from 0 to t to give the cumulative work W over the time period resp (t); Step S4: data fusion, fuse the multi-modal data after feature extraction to form a comprehensive multi-dimensional feature vector: Then the influence of environmental temperature T e , humidity H e , and patient's age A, weight W and other parameters also need to be considered, and the multi-dimensional feature vector is corrected: Wherein, α i is the correction coefficient function; Step S5: intelligent identification, the corrected feature vector x*(t) and the corresponding label form a data set, and are randomly divided into a training set, a validation set and a test set in a ratio of 7:1:2, the training set is used for training the model and optimizing the model parameters; the validation set is used for adjusting the hyperparameters, selecting the model and monitoring the overfitting situation; the test set is used for evaluating the final performance of the model, ensuring the generalization ability of the model and the actual application effect, and then the data set information is input into a fully connected neural network; in the MLP network, the input layer is used for accepting the corrected feature vector x * (t) and label information; Step S6: Decision is made by average probability within time window T to judge the risk of children asthma: Then set threshold θ, when Asthma risk alarm is triggered. Step S7: feedback; based on the decision judgment in the above step S6, dynamic feedback is added for adaptive adjustment: where a i is a model correction coefficient, w is a model weight, L is a loss function, and η is a learning rate. 5.The prediction method of the multi-modal data fusion intelligent children asthma prediction system according to claim 4, characterized in that: The purpose of data denoising in step S2 is to remove the influence of environmental noise, equipment noise and electromagnetic interference on physiological signals during physiological signal acquisition, and to retain the effective feature information in multi-modal data signals. 6.The prediction method of the multi-modal data fusion intelligent children asthma prediction system according to claim 4, characterized in that: Normalization processing in step S2 can standardize the data range of different signals, facilitating the subsequent operation of physiological signal feature extraction.

Images