Method and system for continuous estimation of static compliance and airway resistance of respiratory system in mechanically ventilated patients based on signal quality assessment

Abstract

The present invention provides a method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment. The method screens mechanical ventilation waveform data before and after the end-inspiratory breath-hold maneuver to measure the static compliance and airway resistance of the respiratory system, and estimates the static compliance and airway resistance values ​​for each respiratory cycle using the respiratory mechanics motion equation and the least squares method. Respiratory cycle waveform data with a small difference from the static compliance and airway resistance values ​​obtained by the end-inspiratory breath-hold maneuver is selected to train a neural network. The trained neural network is used to screen mechanical ventilation waveforms suitable for static mechanical analysis. Based on the respiratory mechanics motion equation and the least squares algorithm, continuous static compliance and airway resistance estimates are obtained, thereby achieving continuous dynamic monitoring between the human respiratory system and the ventilator to assess potential lung damage and effective gas exchange, and ensure hemodynamic stability.

Description

Technical Field

[0001] The present invention relates to the technical field of medical respiratory apparatus, and in particular to a method for continuously estimating static respiratory system compliance and airway resistance by analyzing and screening steady-state ventilator waveforms generated during mechanical ventilation. Background Art

[0002] Mechanical ventilation uses a ventilator to create a pressure gradient between the airway and the alveoli, thereby providing respiratory support. When a patient's natural ventilation or oxygenation is impaired, mechanical ventilation can restore effective ventilation and improve oxygenation through mechanical assistance. Respiratory mechanics underpins the pathophysiology of mechanical ventilation and serves as a reference for developing new mechanical ventilation modalities and tailoring individualized lung-protective ventilation strategies. Compliance and resistance are key parameters for lung-protective ventilation. Lung compliance refers to the change in lung volume per unit pressure change, reflecting the impact of changes in thoracic pressure on lung volume. It consists of two components: static compliance, which reflects the elasticity of lung tissue, and dynamic compliance, which is influenced by both lung elasticity and airway resistance. Decreased lung compliance is commonly seen in patients with respiratory diseases such as pleural effusion, chronic obstructive pulmonary disease, emphysema, lung cancer, and pneumonia. Airway resistance, on the other hand, refers to the pressure gradient generated by a unit flow rate within the airway. Airway resistance is clinically applicable to various obstructive ventilatory disorders, as well as in mechanical ventilation and respiratory monitoring settings. For example, when an underlying disease causes pulmonary congestion, lung tissue may develop fibrosis or a decrease in pulmonary surfactant, leading to increased elastic resistance and decreased compliance, which can manifest as difficulty in breathing. In contrast, in the case of emphysema, the elastic components of the lungs are extensively destroyed, reducing both recoil and elastic resistance and increasing compliance, resulting in difficulty in breathing. In short, whether lung compliance or airway resistance increases or decreases, attention should be paid, and further testing should be performed if necessary to clarify the factors causing the change.

[0003] In clinical practice, respiratory system compliance and airway resistance measured during the end-inspiratory breath-hold operation are used as the gold standard to effectively assess the severity of the patient's lung lesions and airway obstruction. Usually, clinicians need to first eliminate the influence of the patient's spontaneous breathing on mechanical ventilation and switch the ventilator to square wave volume control ventilation mode, at which time the flow rate and tidal volume are constant, and then ask the patient to hold his breath for 3 to 5 seconds. Finally, the formula C is used to calculate the patient's respiratory rate. stat =VT / (P plat -PEEP) and R stat =(PIP-P plat ) / FLOW are used to calculate the static respiratory system compliance and airway resistance respectively. Where VT represents the inhaled tidal volume, PEEP represents the positive end-expiratory pressure, PIP represents the peak airway pressure, and P plat"FLOW" represents plateau pressure, while "P" represents peak flow rate. Furthermore, the accuracy of manual measurement is easily affected by the physician's skill level and the patient's spontaneous breathing. Monitoring lacks continuity and dynamism, and fails to fully reflect the interaction between the human respiratory system and the ventilator. Overall, continuously, automatically, and accurately estimating respiratory system compliance and airway resistance during mechanical ventilation is a key issue.

[0004] The invention patent "A method for estimating quasi-static compliance of the lungs under pressure-controlled mechanical ventilation" (CN202110049428.5) discloses a method for estimating quasi-static compliance of the lungs under pressure-controlled ventilation mode. For the respiratory waveform with a non-zero end-inspiratory flow rate, an exponential function is used to fit the waveform data of the decreasing flow rate segment in the inspiratory phase and extend it to the expiratory phase until the flow rate is close to zero. The tidal volume is then calculated to supplement the fitted flow rate waveform to obtain the quasi-static compliance value.

[0005] The invention patent "An Improved Method and Apparatus for Estimating Quasi-Static Lung Compliance Under Pressure-Controlled Mechanical Ventilation" (CN202210193857.4) addresses the shortcomings of the invention patent "A Method for Estimating Quasi-Static Lung Compliance Under Pressure-Controlled Mechanical Ventilation" (CN202110049428.5), which cannot accurately estimate quasi-static lung compliance over long periods of time and in the presence of noise interference caused by spontaneous breathing, coughing, and condensation. This patent discloses an improved method for estimating quasi-static lung compliance under pressure-controlled ventilation mode. This method designs a set of regular noise filtering criteria based on feature point thresholds. This can eliminate ventilator waveforms affected by noise over a long period of time, thereby enabling stable and continuous estimation of lung compliance. However, this method has significant limitations and cannot be widely applied to other ventilation modes. In addition, the screening criteria based on feature point thresholds have the disadvantages of low automation and weak generalization ability. The method generally requires threshold tuning on a small-scale dataset, which often reduces the effectiveness when generalized to large-scale datasets.

[0006] Therefore, designing a method that can automatically screen mechanical ventilation waveforms suitable for static mechanical analysis and accurately and automatically calculate respiratory system compliance and airway resistance in any bedside ventilation mode is of great significance for providing accurate and reliable assessment of potential lung injury and effective gas exchange during ventilation. Summary of the Invention

[0007] The present invention aims to avoid manual end-inspiratory breath-hold measurements, reduce measurement failure rates due to interference from the patient's spontaneous breathing, and improve the continuity and dynamism of respiratory system compliance and airway resistance monitoring indicators. To this end, a method and system for continuously estimating static respiratory system compliance and airway resistance in mechanically ventilated patients based on signal quality assessment are provided.

[0008] The object of the present invention is achieved like this:

[0009] A method for continuously estimating static compliance and airway resistance of a mechanically ventilated patient's respiratory system based on signal quality assessment, comprising:

[0010] The real-time mechanical ventilation waveform data to be tested is input into the trained neural network for prediction. If the prediction is a positive sample, that is, a waveform suitable for static respiratory system compliance and airway resistance analysis, the respiratory mechanics motion equation and the least squares algorithm are used to estimate the dynamic and continuous static respiratory system compliance C. qstat and airway resistance R qstat estimated value;

[0011] The neural network is trained by the following method:

[0012] Construct a data set, wherein each sample of the data set includes one breath mechanical ventilation waveform data and a corresponding label; the label is obtained by comparing the static respiratory system compliance C estimated based on the one breath mechanical ventilation waveform data using the respiratory mechanics motion equation and the least squares algorithm qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat The difference in the values ​​is marked. If the static respiratory system compliance C is estimated based on the mechanical ventilation waveform data of one breath using the respiratory mechanics motion equation and the least squares algorithm qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat If the difference between the values ​​is less than the threshold, the label is marked as a positive sample, otherwise it is marked as a negative sample;

[0013] A neural network model was constructed, using the mechanical ventilation waveform data of each sample in the dataset as input and the sample type determined by the classification probability threshold as output. The model was trained by minimizing the difference between the network output and the corresponding label to obtain a preliminarily trained neural network.

[0014] Based on the data set, the classification probability threshold is optimized with the goal of improving the positive predictive value index of the neural network model until the positive predictive value index of the neural network model meets the requirements or reaches the maximum limit, and a trained neural network is obtained.

[0015] Furthermore, the dataset is constructed and obtained by the following method:

[0016] The respiratory system compliance C obtained by the inspiratory breath-hold maneuver is obtained with the time point of the inspiratory breath-hold maneuver as the center. stat and airway resistance R stat Values, as well as mechanical ventilation waveform data before and after M hours;

[0017] The 2M-hour mechanical ventilation waveform data was divided into respiratory cycles, and the static respiratory system compliance C was calculated by applying the respiratory mechanics motion equation and the least squares algorithm to each respiratory cycle. qstat and airway resistance R qstat Estimation of value;

[0018] Comparison of respiratory system compliance C measured by the inspiratory breath-hold maneuver stat and airway resistance R stat The static respiratory system compliance c was estimated by combining the value with the respiratory mechanics motion equation with the least squares method. qstat and airway resistance R qstat The absolute value of the relative deviation is calculated, and the positive samples and negative samples are divided by the threshold T to construct the data set.

[0019] Furthermore, the respiratory system compliance C obtained by measuring the end-inspiratory breath-hold operation is obtained. stat and airway resistance R stat The value method is as follows:

[0020] Find the time point of the inspiratory breath-hold operation measurement from the continuous mechanical ventilation waveform data, and calculate the respiratory system compliance C obtained by the inspiratory breath-hold operation measurement stat and airway resistance R stat The values ​​are as follows:

[0021]

[0022]

[0023] Where VT is the inhaled tidal volume, PEEP is the positive end-expiratory pressure, PIP is the peak airway pressure, and P plat represents plateau pressure, and FLOW represents peak flow rate.

[0024] Furthermore, the mechanical ventilation waveform data includes a flow rate-time signal, a pressure-time signal, and a tidal volume-time signal.

[0025] Furthermore, the dynamic continuous static respiratory system compliance C is estimated using the respiratory mechanics motion equation and the least squares algorithm. qstat and airway resistance R qstat The estimated value is as follows:

[0026] The positive end-expiratory pressure, flow-time data, pressure-time data, and tidal volume-time data of each respiratory cycle are substituted into the respiratory mechanics motion equation, and then the static respiratory system compliance C is estimated by fitting the pressure-time data using the least squares method. qstat and airway resistance R qstat value, and the goodness of fit R according to the least squares method 2 The threshold N is used to determine whether to include the respiratory cycle in the analysis. The respiratory mechanics equation of motion is as follows:

[0027]

[0028] Where P(t) represents pressure-time data, F(t) represents flow-time data, V(t) represents tidal volume-time data, and PEEP represents positive end-expiratory pressure.

[0029] Furthermore, the neural network adopts a one-dimensional convolutional neural network based on multi-scale feature extraction and attention mechanism; the one-dimensional convolutional neural network based on multi-scale feature extraction and attention mechanism is composed of an input layer, a multi-scale feature extraction module, an adaptive multi-scale feature learning module and a classification layer, which includes a one-dimensional convolution layer, a batch normalization layer, an activation layer, a one-dimensional maximum pooling layer, a feature splicing layer, a feature fusion layer, a random dropout layer, a global average pooling layer and a channel attention mechanism.

[0030] Furthermore, in the optimization of the classification probability threshold with the goal of improving the positive predictive value index of the neural network model, the classification probability 0.5 is used as the starting point, and the step size s is used as the interval to search forward and backward to obtain the optimal classification probability threshold K to maximize the positive predictive value index of the model.

[0031] A system for continuously estimating the static compliance and airway resistance of a mechanically ventilated patient's respiratory system based on signal quality assessment, for executing a method for continuously estimating the static compliance and airway resistance of a mechanically ventilated patient's respiratory system based on signal quality assessment, comprising:

[0032] The continuous estimation module is used to input the real-time mechanical ventilation waveform data to be tested into the trained neural network for prediction. If the prediction is a positive sample, that is, a waveform suitable for static respiratory system compliance and airway resistance analysis, the respiratory mechanics motion equation and the least squares algorithm are used to estimate the dynamic and continuous static respiratory system compliance C. qstat and airway resistance R qstat estimated value;

[0033] The neural network is trained by the following method:

[0034] Construct a data set, wherein each sample of the data set includes one breath mechanical ventilation waveform data and a corresponding label; the label is obtained by comparing the static respiratory system compliance C estimated based on the one breath mechanical ventilation waveform data using the respiratory mechanics motion equation and the least squares algorithm qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat The difference in the values ​​is marked. If the static respiratory system compliance C is estimated based on the respiratory mechanics motion equation and the least squares algorithm based on the mechanical ventilation waveform data of one breath, qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat If the difference between the values ​​is less than the threshold, the label is marked as a positive sample, otherwise it is marked as a negative sample;

[0035] A neural network model was constructed, using the mechanical ventilation waveform data of each sample in the dataset as input and the sample type determined by the classification probability threshold as output. The model was trained by minimizing the difference between the network output and the corresponding label to obtain a preliminarily trained neural network.

[0036] Based on the data set, the classification probability threshold is optimized with the goal of improving the positive predictive value index of the neural network model until the positive predictive value index of the neural network model meets the requirements or reaches the maximum limit, and a trained neural network is obtained.

[0037] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment is implemented.

[0038] A storage medium containing computer-executable instructions, which, when executed by a computer processor, implements a method for continuously estimating the static compliance and airway resistance of a mechanically ventilated patient's respiratory system based on signal quality assessment.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a method for continuously estimating static respiratory compliance and airway resistance in mechanically ventilated patients based on signal quality assessment. This method achieves continuous estimation of static respiratory compliance and airway resistance using a trained neural network. The neural network filters mechanical ventilation data measured before and after an inspiratory breath-hold maneuver and estimates static respiratory compliance and airway resistance values ​​for each respiratory cycle using the respiratory mechanics equation of motion and least-squares fitting. Subsequently, respiratory cycle data with minimal absolute deviation from the static respiratory compliance and airway resistance values ​​measured during the inspiratory breath-hold maneuver are selected as positive samples for training, thereby automatically selecting mechanical ventilation waveforms suitable for static mechanical analysis without a breath-hold maneuver. Furthermore, in response to the need for continuous monitoring of respiratory compliance and airway resistance, the classification probability threshold is optimized during training for the first time with the goal of improving the positive predictive value of the neural network model, ensuring continuous monitoring of respiratory compliance and airway resistance and maximizing the positive predictive value. The present invention realizes continuous dynamic monitoring between the human respiratory system and the ventilator to evaluate potential lung damage and effective gas exchange, and ensure hemodynamic stability.

[0041] The present invention is not limited to the volume-controlled ventilation mode commonly used in clinical practice to measure static respiratory system compliance and airway resistance, but is generally applicable to estimating static respiratory system compliance and airway resistance under other types of ventilation modes. Compared with the existing technology, the automatic estimation method for static respiratory system compliance and airway resistance proposed in the present invention is based on the respiratory mechanics motion equation and the least squares algorithm. Therefore, the measurement method is not restricted by the ventilation mode and has physiological interpretability. In addition, the convolutional neural network based on the attention mechanism and multi-scale feature extraction can adaptively find the respiratory cycle suitable for static mechanical analysis to reduce the impact of the patient's spontaneous breathing or other noise interference on the estimation results, and has robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of a method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment according to an embodiment of the present invention.

[0043] Figure 2 Schematic diagram of a sample screening strategy for constructing a training data set according to an embodiment of the present invention.

[0044] Figure 3 Schematic diagram of the convolutional neural network structure based on multi-scale feature extraction and attention mechanism in one embodiment of the present invention.

[0045] Figure 4This is a structural diagram of a system for continuously estimating static respiratory system compliance and airway resistance based on signal quality assessment according to an embodiment of the present invention.

[0046] Figure 5 FIG. 1 is a linear regression analysis diagram of the estimated static respiratory system compliance and the static respiratory system compliance obtained by manual static mechanical measurement in one embodiment of the present invention.

[0047] Figure 6 FIG. 1 is a linear regression analysis diagram of the static airway resistance estimated in one embodiment of the present invention and the static airway resistance obtained by manual static mechanical measurement. DETAILED DESCRIPTION

[0048] The following embodiments of the present invention are further described in conjunction with the accompanying drawings and examples. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the invention.

[0049] In order to reduce the tediousness of manual end-inspiratory breath-hold measurement in clinical practice, reduce the measurement failure rate caused by interference from the patient's spontaneous breathing, and improve the continuity and dynamics of respiratory system compliance and airway resistance monitoring indicators, the present invention provides a method for continuously estimating the static compliance and airway resistance of the respiratory system of mechanically ventilated patients based on signal quality evaluation. The real-time mechanical ventilation waveform data to be measured is input into a trained neural network for prediction. If the prediction is a positive sample, that is, a waveform suitable for static respiratory system compliance and airway resistance analysis, the respiratory mechanics motion equation and the least squares algorithm are used to estimate the dynamic and continuous static respiratory system compliance C. qstat and airway resistance R qstat Estimated value; The neural network in this method is to screen the mechanical ventilation data before and after the static respiratory system compliance and airway resistance measured at the end of the breath hold operation, and then use the respiratory mechanics motion equation and least squares fitting to estimate the static respiratory system compliance and airway resistance values ​​of each respiratory cycle. Subsequently, the respiratory cycle data with the smallest absolute value of the relative deviation from the static respiratory system compliance and airway resistance values ​​measured at the end of the breath hold operation are selected as positive samples for training. Therefore, the trained neural network can realize automatic screening of mechanical ventilation waveforms that are suitable for static mechanical analysis without the end of the breath hold operation. At the same time, combined with the need for continuous monitoring of respiratory system compliance and airway resistance monitoring indicators, the classification probability threshold is optimized for the first time with the goal of improving the positive predictive value index of the neural network model, ensuring that the respiratory system compliance and airway resistance monitoring indicators can be continuously monitored and the positive predictive value index is maximized. In one embodiment, the neural network is first trained, and then the static respiratory system compliance and airway resistance are continuously estimated based on the trained neural network, see Figure 1, the method of the present invention specifically comprises the following steps:

[0050] S1. Taking the time point of the inspiratory breath-hold measurement as the center, obtain the respiratory system compliance C measured by the inspiratory breath-hold measurement stat and airway resistance R stat Values, as well as mechanical ventilation waveform data before and after M = 1 hour.

[0051] In a specific embodiment, the mechanical ventilation waveform data includes a flow rate-time signal, a pressure-time signal, and a tidal volume-time signal, and the respiratory system compliance C measured by the end-inspiratory breath-hold operation stat and airway resistance R stat The value is automatically output by the ventilator after the breath-hold operation. If the ventilator does not automatically output the monitored respiratory system compliance C under the compliant end-inspiratory breath-hold operation stat and airway resistance R stat The value is calculated from the waveform of the static mechanical measurement as follows:

[0052]

[0053]

[0054] Where VT is the inhaled tidal volume, PEEP is the positive end-expiratory pressure, PIP is the peak airway pressure, and P plat represents plateau pressure, and FLOW represents peak flow rate.

[0055] S2. The mechanical ventilation waveform data of 1 hour before and after the static mechanical measurement are divided into respiratory cycles, and the static respiratory system compliance C is calculated by applying the respiratory mechanical motion equation and the least squares algorithm for each respiratory cycle. qstat and airway resistance R qstat The solution of the estimated value includes the following sub-steps:

[0056] S2-1. Divide the mechanical ventilation waveform data for one hour before and after the static mechanical measurement into respiratory cycles. Substitute the positive end-expiratory pressure, flow-time data, pressure-time data, and tidal volume-time data of each respiratory cycle into the respiratory mechanical motion equation. Then, use the least squares method to fit the pressure-time data and estimate the static respiratory system compliance C. qstat and airway resistance R qstat The respiratory mechanics equation of motion is as follows:

[0057]

[0058] Where P(t) represents pressure-time data, F(t) represents flow-time data, V(t) represents tidal volume-time data, and PEEP represents positive end-expiratory pressure.

[0059] S2-2, according to the least squares goodness of fit R 2 The threshold N = 0.95 is used to decide whether to include the respiratory cycle in the analysis. If the goodness of fit between the pressure-time data of the respiratory cycle and the original pressure-time data is R 2 If the value is ≥0.95, the respiratory cycle will be included in the subsequent analysis; otherwise, the data of the respiratory cycle will be eliminated.

[0060] S3. Comprehensive comparison of respiratory system compliance C measured by the end-inspiratory breath-hold maneuver stat and airway resistance R stat The static respiratory system compliance C is estimated by combining the value with the respiratory mechanics motion equation with the least squares method. qstat and airway resistance R qstat The absolute value of the relative deviation is calculated, and the positive samples and negative samples are divided into a training set and a test set according to the threshold T to construct a data set. In this embodiment, the data set is further divided into a training set and a test set.

[0061] Figure 2 The sample screening strategy for constructing a data set in the present invention is presented, which specifically includes the following sub-steps:

[0062] S3-1. If the respiratory system compliance C obtained by the end-inspiratory breath-hold measurement operation stat and airway resistance R stat Static respiratory system compliance C estimated by combining the respiratory mechanics motion equation with the least squares method qstat and airway resistance R qstat If the absolute values ​​of the relative deviations of the values ​​are all less than or equal to the threshold T = 0.05, they are classified as positive samples in the training set, as follows:

[0063] and

[0064] S3-2. If the respiratory system compliance C obtained by the end-inspiratory breath-hold measurement operation stat Static respiratory system compliance C estimated by combining the respiratory mechanics motion equation with the least squares method qstat The absolute value of the relative deviation of the value is greater than the threshold value T = 0.05, or the airway resistance R obtained by the end-inspiratory breath-hold measurement operation is greater than stat Static airway resistance R estimated by combining the respiratory mechanics motion equation with the least squares method qstat If the absolute value of the relative deviation of the value is greater than the threshold T = 0.05, it is classified as a negative sample in the training set, as follows:

[0065] or

[0066] In this specific embodiment, there are a total of 5,872 positive samples and 89,752 negative samples. To avoid the problem of poor model training accuracy caused by imbalanced sample categories, the negative samples are downsampled and 6,000 samples are randomly selected to form a dataset for model training and testing together with the positive samples.

[0067] S4. Construct a neural network and use the training set and test set for network training and testing respectively. The specific steps include the following:

[0068] S4-1, Figure 3 The convolutional neural network structure based on multi-scale feature extraction and attention mechanism used in the embodiment of the present invention is shown. The one-dimensional convolutional neural network based on multi-scale feature extraction and attention mechanism is composed of an input layer, a multi-scale feature extraction module, an adaptive multi-scale feature learning module and a classification layer, which includes a one-dimensional convolution layer, a batch normalization layer, an activation layer, a one-dimensional maximum pooling layer, a feature splicing layer, a feature fusion layer, a random drop layer, a global average pooling layer and a channel attention mechanism. Specifically, the core of the multi-scale feature extraction module is composed of several parallel one-dimensional convolution layers, feature splicing layers, batch normalization layers and activation layers, which are shown in the figure as 3 one-dimensional convolution layers, with convolution kernel widths of 3, 5 and 7 respectively, and step sizes of 1. Convolution operations of different scales can be used to capture different feature information. Then, the feature splicing layer is used to connect the feature information extracted by the convolution operations of different scales, and output after being processed by the batch normalization layer and the activation layer. Due to the problem of gradient disappearance during training, the overall structure includes multiple multi-scale feature extraction modules (shown as one in the figure) connected in series. Furthermore, residual connections are used between the multi-scale feature extraction module and the adaptive multi-scale feature learning module, so that the network can learn residual information rather than global features, thereby preventing the entire network from focusing too much on the learning and expression of high-level features while ignoring the importance of low-level features. The adaptive multi-scale feature learning module contains several multi-scale feature extraction modules, two of which are shown in the figure. The first multi-scale feature extraction module increases the step size to 2 to reduce the size of the feature map and reduce memory usage, and a random dropout layer is added between the two multi-scale feature extraction modules to reduce the risk of overfitting during training. The channel attention layer is spliced ​​after the second multi-scale feature extraction module, which can automatically focus on the feature information most relevant to the classification result from the waveform data and suppress unimportant feature information, thereby improving the feature representation ability and the overall efficiency and accuracy of the model. The feature information output by the multi-scale feature extraction module after the input layer passes through a one-dimensional maximum pooling layer with a stride of 2 and a one-dimensional convolution layer with a convolution kernel width of 1 and a stride of 1 to form residual information, which is then fused by a feature fusion layer with the feature information output by the adaptive multi-scale feature learning module at this level. The information then enters the adaptive multi-scale feature learning module of the next level.

[0069] In a specific embodiment, after the waveform data passes through the input layer, it enters the multi-scale feature extraction module once and the adaptive multi-scale feature learning module twice, and finally enters the classification layer for classification output.

[0070] S4-2. The dataset was randomly divided into training and testing parts at a ratio of 9:1. The training part was then randomly divided into training and validation sets at a ratio of 7:3 for training and optimization of model parameters, respectively. The loss function used during training was to minimize the difference between the network output and the corresponding label. In this example, 10-fold cross-validation was used, and the model test results are shown in Table 1.

[0071] Table 1 Model test performance in the embodiment

[0072]

[0073] In the table, the positive predictive value (PPV) represents the proportion of samples predicted to be positive that are actually positive, and the negative predictive value (NPV) represents the proportion of samples predicted to be negative that are actually negative, as follows:

[0074]

[0075]

[0076] Among them, TP stands for positive, TN stands for negative, FP stands for false positive, and FN stands for false negative.

[0077] S5. The trained neural network is threshold-tuned to improve the positive predictive value index of the model, that is, starting from a classification probability of 0.5 and with a step size of s = 0.05, the optimal classification probability threshold K is searched forward and backward to maximize the positive predictive value index of the model. A high positive predictive value means that the positive samples predicted by the model are indeed positive samples, which directly affects the accuracy and effectiveness of static respiratory system compliance and airway resistance monitoring, and reduces the frequency of over-prediction of the model in practical applications. In this embodiment, the optimized classification threshold K = 0.85 can automatically obtain 4 monitoring values ​​per hour without manual static mechanical measurement. The model test results are shown in Table 2.

[0078] Table 2 Model test performance after optimizing classification threshold in the embodiment

[0079]

[0080] S6. Use the threshold-tuned neural network to predict the long-term mechanical ventilation waveform. If the predicted waveform is suitable for static respiratory system compliance and airway resistance analysis, the respiratory mechanics motion equation and the least squares algorithm are used for estimation to obtain the dynamic and continuous static respiratory system compliance C.qstat and airway resistance R qstat Table 3 shows the comparative analysis of the model's prediction results on the 1-hour review data before and after the manual static mechanics operation measurement and the manual static mechanics measurement results. Figure 5 and Figure 6 The linear regression analysis results of the static respiratory system compliance and airway resistance estimated in the embodiment of the present invention and the static respiratory system compliance and airway resistance obtained by artificial static mechanical measurement are respectively shown.

[0081] Table 3 Comparison of static respiratory system compliance and airway resistance predicted by the model with those measured by manual static mechanics

[0082]

[0083] The above results show that the method of the present invention can automatically screen steady-state waveforms to calculate static respiratory system compliance and airway resistance without the need for manual static mechanical operation measurement, and has good consistency with the manual static mechanical measurement results.

[0084] The present invention also provides a system for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment, which is used to execute the method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment, comprising:

[0085] The continuous estimation module obtains continuous mechanical ventilation waveform data and inputs each respiratory cycle into a convolutional neural network based on multi-scale feature extraction and attention mechanism constructed by the continuous estimation method of static respiratory system compliance and airway resistance based on signal quality assessment. It can determine whether the current respiratory cycle is suitable for static mechanical analysis, and then obtain continuous static respiratory system compliance and airway resistance estimation values ​​based on the respiratory mechanics motion equation and the least squares algorithm.

[0086] Furthermore, if Figure 4 As shown, the system also includes:

[0087] A data screening module, used for screening mechanical ventilation data before and after measuring the static respiratory system compliance and airway resistance time during the end-inspiratory breath-hold operation;

[0088] The parameter estimation module uses the respiratory mechanics motion equation and the least squares method to estimate the static respiratory system compliance and airway resistance values ​​of each respiratory cycle;

[0089] The model training module selects respiratory cycle data with a small absolute value of relative deviation from the static respiratory system compliance and airway resistance values ​​measured by the end-inspiratory breath hold operation, and trains a convolutional neural network based on the attention mechanism and multi-scale feature extraction to automatically screen mechanical ventilation waveforms suitable for static mechanical analysis.

[0090] Corresponding to the aforementioned embodiment of a method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment, the present invention also provides an electronic device comprising one or more processors for implementing the aforementioned embodiment of a method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment.

[0091] The electronic device of the present invention may be used in any device with data processing capability, such as a computer or other device.

[0092] The device embodiments may be implemented through software, hardware, or a combination of software and hardware. Taking software implementation as an example, as a logically defined device, a processor of any device with data processing capabilities reads the corresponding computer program instructions from a non-volatile memory into the memory and executes them. From a hardware perspective, this includes a processor, memory, a network interface, and a non-volatile memory. In addition, any device with data processing capabilities in which the device in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0093] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0094] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present invention. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0095] An embodiment of the present invention also provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment in the above embodiment is implemented.

[0096] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or memory. The computer-readable storage medium may also be any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0097] In the embodiments of the present invention, continuous static respiratory system compliance and airway resistance are predicted for pressure-controlled ventilation mode. However, the present invention follows the respiratory motion equation after screening the steady-state waveform. Therefore, this method is not limited to continuously estimating static respiratory system compliance and airway resistance from mechanical ventilation data in a specific ventilation mode and is generally applicable to the analysis of waveform data in all ventilation modes. The above embodiments are merely exemplary descriptions of the present invention. However, after reading this patent application, those skilled in the art may make various modifications to the present invention without departing from the spirit and scope of the present invention.

Claims

1. A method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment, characterized in that: include: The real-time mechanical ventilation waveform data to be tested is input into the trained neural network for prediction. If the prediction is a positive sample, that is, a waveform suitable for static respiratory system compliance and airway resistance analysis, the respiratory mechanics motion equation and the least squares algorithm are used to estimate the dynamic and continuous static respiratory system compliance C. qstat and airway resistance R qstat estimated value; The neural network is trained by the following method: Construct a data set, wherein each sample of the data set includes one breath mechanical ventilation waveform data and a corresponding label; the label is obtained by comparing the static respiratory system compliance C estimated based on the one breath mechanical ventilation waveform data using the respiratory mechanics motion equation and the least squares algorithm qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat The difference in the values ​​is marked. If the static respiratory system compliance C is estimated based on the respiratory mechanics motion equation and the least squares algorithm based on the mechanical ventilation waveform data of one breath, qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat If the difference between the values ​​is less than the threshold, the label is marked as a positive sample, otherwise it is marked as a negative sample; A neural network model was constructed, using the mechanical ventilation waveform data of each sample in the dataset as input and the sample type determined by the classification probability threshold as output. The model was trained by minimizing the difference between the network output and the corresponding label to obtain a preliminarily trained neural network. Based on the data set, the classification probability threshold is optimized with the goal of improving the positive predictive value index of the neural network model until the positive predictive value index of the neural network model meets the requirements or reaches the maximum limit, and a trained neural network is obtained.

2. The method according to claim 1, characterized in that The dataset is constructed and obtained by the following method: The respiratory system compliance C obtained by the inspiratory breath-hold maneuver is obtained with the time point of the inspiratory breath-hold maneuver as the center. stat and airway resistance R stat Values, as well as mechanical ventilation waveform data before and after M hours; The 2M-hour mechanical ventilation waveform data was divided into respiratory cycles, and the static respiratory system compliance C was calculated by applying the respiratory mechanics motion equation and the least squares algorithm to each respiratory cycle. qstat and airway resistance R qstat Estimation of value; Comparison of respiratory system compliance C measured by the inspiratory breath-hold maneuver stat and airway resistance R stat The static respiratory system compliance C is estimated by combining the value with the respiratory mechanics motion equation with the least squares method. qstat and airway resistance R qstat The absolute value of the relative deviation is calculated, and the positive samples and negative samples are divided by the threshold T to construct the data set.

3. The method according to claim 2, characterized in that The respiratory system compliance C obtained by measuring the breath-hold operation at the end of inspiration is obtained. stat and airway resistance R stat The value method is as follows: Find the time point of the inspiratory breath-hold operation measurement from the continuous mechanical ventilation waveform data, and calculate the respiratory system compliance C obtained by the inspiratory breath-hold operation measurement stat and airway resistance R stat The values ​​are as follows: Where VT is the inhaled tidal volume, PEEP is the positive end-expiratory pressure, PIR is the peak airway pressure, and P plat represents plateau pressure, and FLOW represents peak flow rate.

4. The method according to claim 1, wherein Mechanical ventilation waveform data includes flow rate-time signal, pressure-time signal and tidal volume-time signal.

5. The method according to claim 1, wherein The dynamic continuous static respiratory system compliance C is estimated using the respiratory mechanics motion equation and the least squares algorithm. qstat and airway resistance R qstat The estimated value is as follows: The positive end-expiratory pressure, flow-time data, pressure-time data, and tidal volume-time data of each respiratory cycle are substituted into the respiratory mechanics motion equation, and then the static respiratory system compliance C is estimated by fitting the pressure-time data using the least squares method. qstat and airway resistance R qstat value, and the least squares goodness of fit R 2 The threshold N is used to determine whether to include the respiratory cycle in the analysis; the respiratory mechanics equation is as follows: Where P(t) represents pressure-time data, F(t) represents flow-time data, V(t) represents tidal volume-time data, and PEEP represents positive end-expiratory pressure.

6. The method according to claim 1, characterized in that The neural network adopts a one-dimensional convolutional neural network based on multi-scale feature extraction and attention mechanism; the one-dimensional convolutional neural network based on multi-scale feature extraction and attention mechanism as a whole includes an input layer, several multi-scale feature extraction modules, several adaptive multi-scale feature learning modules and a classification layer, wherein the multi-scale feature extraction module is composed of several parallel one-dimensional convolution layers, feature splicing layers, batch normalization layers and activation layers connected in sequence; the multi-scale feature extraction module and the adaptive multi-scale feature learning module adopt residual connection; the adaptive multi-scale feature learning module includes several multi-scale feature extraction modules, random dropout layers, channel attention layers, and one-dimensional maximum pooling layers, one-dimensional convolution layers and feature fusion layers; the random dropout layers are set between the several multi-scale feature extraction modules to reduce the risk of overfitting during training; the channel attention layer is used to focus on the feature information output by the last multi-scale feature extraction module that is most relevant to the classification result; The one-dimensional maximum pooling layer and the one-dimensional convolution layer connected in sequence are used to receive the features obtained by the residual connection and form residual information; The feature fusion layer is used to fuse the output of the channel attention layer and the residual information.

7. The method according to claim 1, characterized in that In the optimization of the classification probability threshold with the goal of improving the positive predictive value index of the neural network model, a classification probability of 0.5 is used as a starting point, and a step size s is used as an interval to search forward and backward to obtain the optimal classification probability threshold K to maximize the positive predictive value index of the model.

8. A system for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment, configured to execute the method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment as claimed in any one of claims 1 to 7, characterized in that: include: The continuous estimation module is used to input the real-time mechanical ventilation waveform data to be tested into the trained neural network for prediction. If the prediction is a positive sample, that is, a waveform suitable for static respiratory system compliance and airway resistance analysis, the respiratory mechanics motion equation and the least squares algorithm are used to estimate the dynamic and continuous static respiratory system compliance C. qstat and airway resistance R qstat estimated value; The neural network is trained by the following method: Construct a data set, wherein each sample of the data set includes one breath mechanical ventilation waveform data and a corresponding label; the label is obtained by comparing the static respiratory system compliance C estimated based on the one breath mechanical ventilation waveform data using the respiratory mechanics motion equation and the least squares algorithm qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat The difference in the values ​​is marked. If the static respiratory system compliance C is estimated based on the respiratory mechanics motion equation and the least squares algorithm based on the mechanical ventilation waveform data of one breath, qstat and airway resistance R qstat The respiratory system compliance C stat and airway resistance R stat If the difference between the values ​​is less than the threshold, the label is marked as a positive sample, otherwise it is marked as a negative sample; A neural network model was constructed, using the mechanical ventilation waveform data of each sample in the dataset as input and the sample type determined by the classification probability threshold as output. The model was trained by minimizing the difference between the network output and the corresponding label to obtain a preliminarily trained neural network. Based on the data set, the classification probability threshold is optimized with the goal of improving the positive predictive value index of the neural network model until the positive predictive value index of the neural network model meets the requirements or reaches the maximum limit, and a trained neural network is obtained.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment as described in any one of claims 1 to 7 is implemented.

10. A storage medium comprising computer-executable instructions, wherein when executed by a computer processor, the computer-executable instructions implement the method for continuously estimating the static compliance and airway resistance of the respiratory system of a mechanically ventilated patient based on signal quality assessment as described in any one of claims 1 to 7.

Images